AnotherFoxGuy/webrtc-audio-processing
1
0
Fork 0
Code Pull Requests Activity

Update common_audio

Browse Source 
Corresponds to upstream commit 524e9b043e7e86fd72353b987c9d5f6a1ebf83e1

Update notes:

 * Moved src/ to webrtc/ to easily diff against the third_party/webrtc
   in the chromium tree

 * ARM/NEON/MIPS support is not yet hooked up

 * Tests have not been copied
This commit is contained in:
Arun Raghavan
2015-10-13 12:16:16 +05:30
parent 9413986e79
commit c4fb4e38de
230 changed files with 11201 additions and 8656 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
webrtc/Makefile.am Normal file
View File

@ -0,0 +1 @@
SUBDIRS = common_audio system_wrappers modules

227
webrtc/base/checks.h Normal file
View File

@ -0,0 +1,227 @@
/*
* Copyright 2006 The WebRTC Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_BASE_CHECKS_H_
#define WEBRTC_BASE_CHECKS_H_
#include <sstream>
#include <string>
#include "webrtc/typedefs.h"
// The macros here print a message to stderr and abort under various
// conditions. All will accept additional stream messages. For example:
// RTC_DCHECK_EQ(foo, bar) << "I'm printed when foo != bar.";
//
// - RTC_CHECK(x) is an assertion that x is always true, and that if it isn't,
// it's better to terminate the process than to continue. During development,
// the reason that it's better to terminate might simply be that the error
// handling code isn't in place yet; in production, the reason might be that
// the author of the code truly believes that x will always be true, but that
// she recognizes that if she is wrong, abrupt and unpleasant process
// termination is still better than carrying on with the assumption violated.
//
// RTC_CHECK always evaluates its argument, so it's OK for x to have side
// effects.
//
// - RTC_DCHECK(x) is the same as RTC_CHECK(x)---an assertion that x is always
// true---except that x will only be evaluated in debug builds; in production
// builds, x is simply assumed to be true. This is useful if evaluating x is
// expensive and the expected cost of failing to detect the violated
// assumption is acceptable. You should not handle cases where a production
// build fails to spot a violated condition, even those that would result in
// crashes. If the code needs to cope with the error, make it cope, but don't
// call RTC_DCHECK; if the condition really can't occur, but you'd sleep
// better at night knowing that the process will suicide instead of carrying
// on in case you were wrong, use RTC_CHECK instead of RTC_DCHECK.
//
// RTC_DCHECK only evaluates its argument in debug builds, so if x has visible
// side effects, you need to write e.g.
// bool w = x; RTC_DCHECK(w);
//
// - RTC_CHECK_EQ, _NE, _GT, ..., and RTC_DCHECK_EQ, _NE, _GT, ... are
// specialized variants of RTC_CHECK and RTC_DCHECK that print prettier
// messages if the condition doesn't hold. Prefer them to raw RTC_CHECK and
// RTC_DCHECK.
//
// - FATAL() aborts unconditionally.
//
// TODO(ajm): Ideally, checks.h would be combined with logging.h, but
// consolidation with system_wrappers/logging.h should happen first.
namespace rtc {
// Helper macro which avoids evaluating the arguments to a stream if
// the condition doesn't hold.
#define RTC_LAZY_STREAM(stream, condition) \
!(condition) ? static_cast<void>(0) : rtc::FatalMessageVoidify() & (stream)
// The actual stream used isn't important. We reference condition in the code
// but don't evaluate it; this is to avoid "unused variable" warnings (we do so
// in a particularly convoluted way with an extra ?: because that appears to be
// the simplest construct that keeps Visual Studio from complaining about
// condition being unused).
#define RTC_EAT_STREAM_PARAMETERS(condition) \
(true ? true : !(condition)) \
? static_cast<void>(0) \
: rtc::FatalMessageVoidify() & rtc::FatalMessage("", 0).stream()
// RTC_CHECK dies with a fatal error if condition is not true. It is *not*
// controlled by NDEBUG, so the check will be executed regardless of
// compilation mode.
//
// We make sure RTC_CHECK et al. always evaluates their arguments, as
// doing RTC_CHECK(FunctionWithSideEffect()) is a common idiom.
#define RTC_CHECK(condition) \
RTC_LAZY_STREAM(rtc::FatalMessage(__FILE__, __LINE__).stream(), \
!(condition)) \
<< "Check failed: " #condition << std::endl << "# "
// Helper macro for binary operators.
// Don't use this macro directly in your code, use RTC_CHECK_EQ et al below.
//
// TODO(akalin): Rewrite this so that constructs like if (...)
// RTC_CHECK_EQ(...) else { ... } work properly.
#define RTC_CHECK_OP(name, op, val1, val2) \
if (std::string* _result = \
rtc::Check##name##Impl((val1), (val2), #val1 " " #op " " #val2)) \
rtc::FatalMessage(__FILE__, __LINE__, _result).stream()
// Build the error message string. This is separate from the "Impl"
// function template because it is not performance critical and so can
// be out of line, while the "Impl" code should be inline. Caller
// takes ownership of the returned string.
template<class t1, class t2>
std::string* MakeCheckOpString(const t1& v1, const t2& v2, const char* names) {
std::ostringstream ss;
ss << names << " (" << v1 << " vs. " << v2 << ")";
std::string* msg = new std::string(ss.str());
return msg;
}
// MSVC doesn't like complex extern templates and DLLs.
#if !defined(COMPILER_MSVC)
// Commonly used instantiations of MakeCheckOpString<>. Explicitly instantiated
// in logging.cc.
extern template std::string* MakeCheckOpString<int, int>(
const int&, const int&, const char* names);
extern template
std::string* MakeCheckOpString<unsigned long, unsigned long>(
const unsigned long&, const unsigned long&, const char* names);
extern template
std::string* MakeCheckOpString<unsigned long, unsigned int>(
const unsigned long&, const unsigned int&, const char* names);
extern template
std::string* MakeCheckOpString<unsigned int, unsigned long>(
const unsigned int&, const unsigned long&, const char* names);
extern template
std::string* MakeCheckOpString<std::string, std::string>(
const std::string&, const std::string&, const char* name);
#endif
// Helper functions for RTC_CHECK_OP macro.
// The (int, int) specialization works around the issue that the compiler
// will not instantiate the template version of the function on values of
// unnamed enum type - see comment below.
#define DEFINE_RTC_CHECK_OP_IMPL(name, op) \
template <class t1, class t2> \
inline std::string* Check##name##Impl(const t1& v1, const t2& v2, \
const char* names) { \
if (v1 op v2) \
return NULL; \
else \
return rtc::MakeCheckOpString(v1, v2, names); \
} \
inline std::string* Check##name##Impl(int v1, int v2, const char* names) { \
if (v1 op v2) \
return NULL; \
else \
return rtc::MakeCheckOpString(v1, v2, names); \
}
DEFINE_RTC_CHECK_OP_IMPL(EQ, ==)
DEFINE_RTC_CHECK_OP_IMPL(NE, !=)
DEFINE_RTC_CHECK_OP_IMPL(LE, <=)
DEFINE_RTC_CHECK_OP_IMPL(LT, < )
DEFINE_RTC_CHECK_OP_IMPL(GE, >=)
DEFINE_RTC_CHECK_OP_IMPL(GT, > )
#undef DEFINE_RTC_CHECK_OP_IMPL
#define RTC_CHECK_EQ(val1, val2) RTC_CHECK_OP(EQ, ==, val1, val2)
#define RTC_CHECK_NE(val1, val2) RTC_CHECK_OP(NE, !=, val1, val2)
#define RTC_CHECK_LE(val1, val2) RTC_CHECK_OP(LE, <=, val1, val2)
#define RTC_CHECK_LT(val1, val2) RTC_CHECK_OP(LT, < , val1, val2)
#define RTC_CHECK_GE(val1, val2) RTC_CHECK_OP(GE, >=, val1, val2)
#define RTC_CHECK_GT(val1, val2) RTC_CHECK_OP(GT, > , val1, val2)
// The RTC_DCHECK macro is equivalent to RTC_CHECK except that it only generates
// code in debug builds. It does reference the condition parameter in all cases,
// though, so callers won't risk getting warnings about unused variables.
#if (!defined(NDEBUG) || defined(DCHECK_ALWAYS_ON))
#define RTC_DCHECK(condition) RTC_CHECK(condition)
#define RTC_DCHECK_EQ(v1, v2) RTC_CHECK_EQ(v1, v2)
#define RTC_DCHECK_NE(v1, v2) RTC_CHECK_NE(v1, v2)
#define RTC_DCHECK_LE(v1, v2) RTC_CHECK_LE(v1, v2)
#define RTC_DCHECK_LT(v1, v2) RTC_CHECK_LT(v1, v2)
#define RTC_DCHECK_GE(v1, v2) RTC_CHECK_GE(v1, v2)
#define RTC_DCHECK_GT(v1, v2) RTC_CHECK_GT(v1, v2)
#else
#define RTC_DCHECK(condition) RTC_EAT_STREAM_PARAMETERS(condition)
#define RTC_DCHECK_EQ(v1, v2) RTC_EAT_STREAM_PARAMETERS((v1) == (v2))
#define RTC_DCHECK_NE(v1, v2) RTC_EAT_STREAM_PARAMETERS((v1) != (v2))
#define RTC_DCHECK_LE(v1, v2) RTC_EAT_STREAM_PARAMETERS((v1) <= (v2))
#define RTC_DCHECK_LT(v1, v2) RTC_EAT_STREAM_PARAMETERS((v1) < (v2))
#define RTC_DCHECK_GE(v1, v2) RTC_EAT_STREAM_PARAMETERS((v1) >= (v2))
#define RTC_DCHECK_GT(v1, v2) RTC_EAT_STREAM_PARAMETERS((v1) > (v2))
#endif
// This is identical to LogMessageVoidify but in name.
class FatalMessageVoidify {
public:
FatalMessageVoidify() { }
// This has to be an operator with a precedence lower than << but
// higher than ?:
void operator&(std::ostream&) { }
};
#define RTC_UNREACHABLE_CODE_HIT false
#define RTC_NOTREACHED() RTC_DCHECK(RTC_UNREACHABLE_CODE_HIT)
#define FATAL() rtc::FatalMessage(__FILE__, __LINE__).stream()
// TODO(ajm): Consider adding RTC_NOTIMPLEMENTED macro when
// base/logging.h and system_wrappers/logging.h are consolidated such that we
// can match the Chromium behavior.
// Like a stripped-down LogMessage from logging.h, except that it aborts.
class FatalMessage {
public:
FatalMessage(const char* file, int line);
// Used for RTC_CHECK_EQ(), etc. Takes ownership of the given string.
FatalMessage(const char* file, int line, std::string* result);
NO_RETURN ~FatalMessage();
std::ostream& stream() { return stream_; }
private:
void Init(const char* file, int line);
std::ostringstream stream_;
};
// Performs the integer division a/b and returns the result. CHECKs that the
// remainder is zero.
template <typename T>
inline T CheckedDivExact(T a, T b) {
RTC_CHECK_EQ(a % b, static_cast<T>(0));
return a / b;
}
} // namespace rtc
#endif // WEBRTC_BASE_CHECKS_H_

34
webrtc/base/constructormagic.h Normal file
View File

@ -0,0 +1,34 @@
/*
* Copyright 2004 The WebRTC Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_BASE_CONSTRUCTORMAGIC_H_
#define WEBRTC_BASE_CONSTRUCTORMAGIC_H_
// Put this in the declarations for a class to be unassignable.
#define RTC_DISALLOW_ASSIGN(TypeName) \
void operator=(const TypeName&) = delete
// A macro to disallow the copy constructor and operator= functions. This should
// be used in the declarations for a class.
#define RTC_DISALLOW_COPY_AND_ASSIGN(TypeName) \
TypeName(const TypeName&) = delete; \
RTC_DISALLOW_ASSIGN(TypeName)
// A macro to disallow all the implicit constructors, namely the default
// constructor, copy constructor and operator= functions.
//
// This should be used in the declarations for a class that wants to prevent
// anyone from instantiating it. This is especially useful for classes
// containing only static methods.
#define RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(TypeName) \
TypeName() = delete; \
RTC_DISALLOW_COPY_AND_ASSIGN(TypeName)
#endif // WEBRTC_BASE_CONSTRUCTORMAGIC_H_

636
webrtc/base/scoped_ptr.h Normal file
View File

@ -0,0 +1,636 @@
/*
* Copyright 2012 The WebRTC Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
// Borrowed from Chromium's src/base/memory/scoped_ptr.h.
// Scopers help you manage ownership of a pointer, helping you easily manage a
// pointer within a scope, and automatically destroying the pointer at the end
// of a scope. There are two main classes you will use, which correspond to the
// operators new/delete and new[]/delete[].
//
// Example usage (scoped_ptr<T>):
// {
// scoped_ptr<Foo> foo(new Foo("wee"));
// } // foo goes out of scope, releasing the pointer with it.
//
// {
// scoped_ptr<Foo> foo; // No pointer managed.
// foo.reset(new Foo("wee")); // Now a pointer is managed.
// foo.reset(new Foo("wee2")); // Foo("wee") was destroyed.
// foo.reset(new Foo("wee3")); // Foo("wee2") was destroyed.
// foo->Method(); // Foo::Method() called.
// foo.get()->Method(); // Foo::Method() called.
// SomeFunc(foo.release()); // SomeFunc takes ownership, foo no longer
// // manages a pointer.
// foo.reset(new Foo("wee4")); // foo manages a pointer again.
// foo.reset(); // Foo("wee4") destroyed, foo no longer
// // manages a pointer.
// } // foo wasn't managing a pointer, so nothing was destroyed.
//
// Example usage (scoped_ptr<T[]>):
// {
// scoped_ptr<Foo[]> foo(new Foo[100]);
// foo.get()->Method(); // Foo::Method on the 0th element.
// foo[10].Method(); // Foo::Method on the 10th element.
// }
//
// These scopers also implement part of the functionality of C++11 unique_ptr
// in that they are "movable but not copyable." You can use the scopers in
// the parameter and return types of functions to signify ownership transfer
// in to and out of a function. When calling a function that has a scoper
// as the argument type, it must be called with the result of an analogous
// scoper's Pass() function or another function that generates a temporary;
// passing by copy will NOT work. Here is an example using scoped_ptr:
//
// void TakesOwnership(scoped_ptr<Foo> arg) {
// // Do something with arg
// }
// scoped_ptr<Foo> CreateFoo() {
// // No need for calling Pass() because we are constructing a temporary
// // for the return value.
// return scoped_ptr<Foo>(new Foo("new"));
// }
// scoped_ptr<Foo> PassThru(scoped_ptr<Foo> arg) {
// return arg.Pass();
// }
//
// {
// scoped_ptr<Foo> ptr(new Foo("yay")); // ptr manages Foo("yay").
// TakesOwnership(ptr.Pass()); // ptr no longer owns Foo("yay").
// scoped_ptr<Foo> ptr2 = CreateFoo(); // ptr2 owns the return Foo.
// scoped_ptr<Foo> ptr3 = // ptr3 now owns what was in ptr2.
// PassThru(ptr2.Pass()); // ptr2 is correspondingly nullptr.
// }
//
// Notice that if you do not call Pass() when returning from PassThru(), or
// when invoking TakesOwnership(), the code will not compile because scopers
// are not copyable; they only implement move semantics which require calling
// the Pass() function to signify a destructive transfer of state. CreateFoo()
// is different though because we are constructing a temporary on the return
// line and thus can avoid needing to call Pass().
//
// Pass() properly handles upcast in initialization, i.e. you can use a
// scoped_ptr<Child> to initialize a scoped_ptr<Parent>:
//
// scoped_ptr<Foo> foo(new Foo());
// scoped_ptr<FooParent> parent(foo.Pass());
//
// PassAs<>() should be used to upcast return value in return statement:
//
// scoped_ptr<Foo> CreateFoo() {
// scoped_ptr<FooChild> result(new FooChild());
// return result.PassAs<Foo>();
// }
//
// Note that PassAs<>() is implemented only for scoped_ptr<T>, but not for
// scoped_ptr<T[]>. This is because casting array pointers may not be safe.
#ifndef WEBRTC_BASE_SCOPED_PTR_H__
#define WEBRTC_BASE_SCOPED_PTR_H__
// This is an implementation designed to match the anticipated future TR2
// implementation of the scoped_ptr class.
#include <assert.h>
#include <stddef.h>
#include <stdlib.h>
#include <algorithm> // For std::swap().
#include "webrtc/base/constructormagic.h"
#include "webrtc/base/template_util.h"
#include "webrtc/typedefs.h"
namespace rtc {
// Function object which deletes its parameter, which must be a pointer.
// If C is an array type, invokes 'delete[]' on the parameter; otherwise,
// invokes 'delete'. The default deleter for scoped_ptr<T>.
template <class T>
struct DefaultDeleter {
DefaultDeleter() {}
template <typename U> DefaultDeleter(const DefaultDeleter<U>& other) {
// IMPLEMENTATION NOTE: C++11 20.7.1.1.2p2 only provides this constructor
// if U* is implicitly convertible to T* and U is not an array type.
//
// Correct implementation should use SFINAE to disable this
// constructor. However, since there are no other 1-argument constructors,
// using a static_assert based on is_convertible<> and requiring
// complete types is simpler and will cause compile failures for equivalent
// misuses.
//
// Note, the is_convertible<U*, T*> check also ensures that U is not an
// array. T is guaranteed to be a non-array, so any U* where U is an array
// cannot convert to T*.
enum { T_must_be_complete = sizeof(T) };
enum { U_must_be_complete = sizeof(U) };
static_assert(rtc::is_convertible<U*, T*>::value,
"U* must implicitly convert to T*");
}
inline void operator()(T* ptr) const {
enum { type_must_be_complete = sizeof(T) };
delete ptr;
}
};
// Specialization of DefaultDeleter for array types.
template <class T>
struct DefaultDeleter<T[]> {
inline void operator()(T* ptr) const {
enum { type_must_be_complete = sizeof(T) };
delete[] ptr;
}
private:
// Disable this operator for any U != T because it is undefined to execute
// an array delete when the static type of the array mismatches the dynamic
// type.
//
// References:
// C++98 [expr.delete]p3
// http://cplusplus.github.com/LWG/lwg-defects.html#938
template <typename U> void operator()(U* array) const;
};
template <class T, int n>
struct DefaultDeleter<T[n]> {
// Never allow someone to declare something like scoped_ptr<int[10]>.
static_assert(sizeof(T) == -1, "do not use array with size as type");
};
// Function object which invokes 'free' on its parameter, which must be
// a pointer. Can be used to store malloc-allocated pointers in scoped_ptr:
//
// scoped_ptr<int, rtc::FreeDeleter> foo_ptr(
// static_cast<int*>(malloc(sizeof(int))));
struct FreeDeleter {
inline void operator()(void* ptr) const {
free(ptr);
}
};
namespace internal {
template <typename T>
struct ShouldAbortOnSelfReset {
template <typename U>
static rtc::internal::NoType Test(const typename U::AllowSelfReset*);
template <typename U>
static rtc::internal::YesType Test(...);
static const bool value =
sizeof(Test<T>(0)) == sizeof(rtc::internal::YesType);
};
// Minimal implementation of the core logic of scoped_ptr, suitable for
// reuse in both scoped_ptr and its specializations.
template <class T, class D>
class scoped_ptr_impl {
public:
explicit scoped_ptr_impl(T* p) : data_(p) {}
// Initializer for deleters that have data parameters.
scoped_ptr_impl(T* p, const D& d) : data_(p, d) {}
// Templated constructor that destructively takes the value from another
// scoped_ptr_impl.
template <typename U, typename V>
scoped_ptr_impl(scoped_ptr_impl<U, V>* other)
: data_(other->release(), other->get_deleter()) {
// We do not support move-only deleters. We could modify our move
// emulation to have rtc::subtle::move() and rtc::subtle::forward()
// functions that are imperfect emulations of their C++11 equivalents,
// but until there's a requirement, just assume deleters are copyable.
}
template <typename U, typename V>
void TakeState(scoped_ptr_impl<U, V>* other) {
// See comment in templated constructor above regarding lack of support
// for move-only deleters.
reset(other->release());
get_deleter() = other->get_deleter();
}
~scoped_ptr_impl() {
if (data_.ptr != nullptr) {
// Not using get_deleter() saves one function call in non-optimized
// builds.
static_cast<D&>(data_)(data_.ptr);
}
}
void reset(T* p) {
// This is a self-reset, which is no longer allowed for default deleters:
// https://crbug.com/162971
assert(!ShouldAbortOnSelfReset<D>::value || p == nullptr || p != data_.ptr);
// Note that running data_.ptr = p can lead to undefined behavior if
// get_deleter()(get()) deletes this. In order to prevent this, reset()
// should update the stored pointer before deleting its old value.
//
// However, changing reset() to use that behavior may cause current code to
// break in unexpected ways. If the destruction of the owned object
// dereferences the scoped_ptr when it is destroyed by a call to reset(),
// then it will incorrectly dispatch calls to |p| rather than the original
// value of |data_.ptr|.
//
// During the transition period, set the stored pointer to nullptr while
// deleting the object. Eventually, this safety check will be removed to
// prevent the scenario initially described from occurring and
// http://crbug.com/176091 can be closed.
T* old = data_.ptr;
data_.ptr = nullptr;
if (old != nullptr)
static_cast<D&>(data_)(old);
data_.ptr = p;
}
T* get() const { return data_.ptr; }
D& get_deleter() { return data_; }
const D& get_deleter() const { return data_; }
void swap(scoped_ptr_impl& p2) {
// Standard swap idiom: 'using std::swap' ensures that std::swap is
// present in the overload set, but we call swap unqualified so that
// any more-specific overloads can be used, if available.
using std::swap;
swap(static_cast<D&>(data_), static_cast<D&>(p2.data_));
swap(data_.ptr, p2.data_.ptr);
}
T* release() {
T* old_ptr = data_.ptr;
data_.ptr = nullptr;
return old_ptr;
}
T** accept() {
reset(nullptr);
return &(data_.ptr);
}
T** use() {
return &(data_.ptr);
}
private:
// Needed to allow type-converting constructor.
template <typename U, typename V> friend class scoped_ptr_impl;
// Use the empty base class optimization to allow us to have a D
// member, while avoiding any space overhead for it when D is an
// empty class. See e.g. http://www.cantrip.org/emptyopt.html for a good
// discussion of this technique.
struct Data : public D {
explicit Data(T* ptr_in) : ptr(ptr_in) {}
Data(T* ptr_in, const D& other) : D(other), ptr(ptr_in) {}
T* ptr;
};
Data data_;
RTC_DISALLOW_COPY_AND_ASSIGN(scoped_ptr_impl);
};
} // namespace internal
// A scoped_ptr<T> is like a T*, except that the destructor of scoped_ptr<T>
// automatically deletes the pointer it holds (if any).
// That is, scoped_ptr<T> owns the T object that it points to.
// Like a T*, a scoped_ptr<T> may hold either nullptr or a pointer to a T
// object. Also like T*, scoped_ptr<T> is thread-compatible, and once you
// dereference it, you get the thread safety guarantees of T.
//
// The size of scoped_ptr is small. On most compilers, when using the
// DefaultDeleter, sizeof(scoped_ptr<T>) == sizeof(T*). Custom deleters will
// increase the size proportional to whatever state they need to have. See
// comments inside scoped_ptr_impl<> for details.
//
// Current implementation targets having a strict subset of C++11's
// unique_ptr<> features. Known deficiencies include not supporting move-only
// deleters, function pointers as deleters, and deleters with reference
// types.
template <class T, class D = rtc::DefaultDeleter<T> >
class scoped_ptr {
// TODO(ajm): If we ever import RefCountedBase, this check needs to be
// enabled.
//static_assert(rtc::internal::IsNotRefCounted<T>::value,
// "T is refcounted type and needs scoped refptr");
public:
// The element and deleter types.
typedef T element_type;
typedef D deleter_type;
// Constructor. Defaults to initializing with nullptr.
scoped_ptr() : impl_(nullptr) {}
// Constructor. Takes ownership of p.
explicit scoped_ptr(element_type* p) : impl_(p) {}
// Constructor. Allows initialization of a stateful deleter.
scoped_ptr(element_type* p, const D& d) : impl_(p, d) {}
// Constructor. Allows construction from a nullptr.
scoped_ptr(decltype(nullptr)) : impl_(nullptr) {}
// Constructor. Allows construction from a scoped_ptr rvalue for a
// convertible type and deleter.
//
// IMPLEMENTATION NOTE: C++11 unique_ptr<> keeps this constructor distinct
// from the normal move constructor. By C++11 20.7.1.2.1.21, this constructor
// has different post-conditions if D is a reference type. Since this
// implementation does not support deleters with reference type,
// we do not need a separate move constructor allowing us to avoid one
// use of SFINAE. You only need to care about this if you modify the
// implementation of scoped_ptr.
template <typename U, typename V>
scoped_ptr(scoped_ptr<U, V>&& other)
: impl_(&other.impl_) {
static_assert(!rtc::is_array<U>::value, "U cannot be an array");
}
// operator=. Allows assignment from a scoped_ptr rvalue for a convertible
// type and deleter.
//
// IMPLEMENTATION NOTE: C++11 unique_ptr<> keeps this operator= distinct from
// the normal move assignment operator. By C++11 20.7.1.2.3.4, this templated
// form has different requirements on for move-only Deleters. Since this
// implementation does not support move-only Deleters, we do not need a
// separate move assignment operator allowing us to avoid one use of SFINAE.
// You only need to care about this if you modify the implementation of
// scoped_ptr.
template <typename U, typename V>
scoped_ptr& operator=(scoped_ptr<U, V>&& rhs) {
static_assert(!rtc::is_array<U>::value, "U cannot be an array");
impl_.TakeState(&rhs.impl_);
return *this;
}
// operator=. Allows assignment from a nullptr. Deletes the currently owned
// object, if any.
scoped_ptr& operator=(decltype(nullptr)) {
reset();
return *this;
}
// Deleted copy constructor and copy assignment, to make the type move-only.
scoped_ptr(const scoped_ptr& other) = delete;
scoped_ptr& operator=(const scoped_ptr& other) = delete;
// Get an rvalue reference. (sp.Pass() does the same thing as std::move(sp).)
scoped_ptr&& Pass() { return static_cast<scoped_ptr&&>(*this); }
// Reset. Deletes the currently owned object, if any.
// Then takes ownership of a new object, if given.
void reset(element_type* p = nullptr) { impl_.reset(p); }
// Accessors to get the owned object.
// operator* and operator-> will assert() if there is no current object.
element_type& operator*() const {
assert(impl_.get() != nullptr);
return *impl_.get();
}
element_type* operator->() const {
assert(impl_.get() != nullptr);
return impl_.get();
}
element_type* get() const { return impl_.get(); }
// Access to the deleter.
deleter_type& get_deleter() { return impl_.get_deleter(); }
const deleter_type& get_deleter() const { return impl_.get_deleter(); }
// Allow scoped_ptr<element_type> to be used in boolean expressions, but not
// implicitly convertible to a real bool (which is dangerous).
//
// Note that this trick is only safe when the == and != operators
// are declared explicitly, as otherwise "scoped_ptr1 ==
// scoped_ptr2" will compile but do the wrong thing (i.e., convert
// to Testable and then do the comparison).
private:
typedef rtc::internal::scoped_ptr_impl<element_type, deleter_type>
scoped_ptr::*Testable;
public:
operator Testable() const {
return impl_.get() ? &scoped_ptr::impl_ : nullptr;
}
// Comparison operators.
// These return whether two scoped_ptr refer to the same object, not just to
// two different but equal objects.
bool operator==(const element_type* p) const { return impl_.get() == p; }
bool operator!=(const element_type* p) const { return impl_.get() != p; }
// Swap two scoped pointers.
void swap(scoped_ptr& p2) {
impl_.swap(p2.impl_);
}
// Release a pointer.
// The return value is the current pointer held by this object. If this object
// holds a nullptr, the return value is nullptr. After this operation, this
// object will hold a nullptr, and will not own the object any more.
element_type* release() WARN_UNUSED_RESULT {
return impl_.release();
}
// Delete the currently held pointer and return a pointer
// to allow overwriting of the current pointer address.
element_type** accept() WARN_UNUSED_RESULT {
return impl_.accept();
}
// Return a pointer to the current pointer address.
element_type** use() WARN_UNUSED_RESULT {
return impl_.use();
}
private:
// Needed to reach into |impl_| in the constructor.
template <typename U, typename V> friend class scoped_ptr;
rtc::internal::scoped_ptr_impl<element_type, deleter_type> impl_;
// Forbidden for API compatibility with std::unique_ptr.
explicit scoped_ptr(int disallow_construction_from_null);
// Forbid comparison of scoped_ptr types. If U != T, it totally
// doesn't make sense, and if U == T, it still doesn't make sense
// because you should never have the same object owned by two different
// scoped_ptrs.
template <class U> bool operator==(scoped_ptr<U> const& p2) const;
template <class U> bool operator!=(scoped_ptr<U> const& p2) const;
};
template <class T, class D>
class scoped_ptr<T[], D> {
public:
// The element and deleter types.
typedef T element_type;
typedef D deleter_type;
// Constructor. Defaults to initializing with nullptr.
scoped_ptr() : impl_(nullptr) {}
// Constructor. Stores the given array. Note that the argument's type
// must exactly match T*. In particular:
// - it cannot be a pointer to a type derived from T, because it is
// inherently unsafe in the general case to access an array through a
// pointer whose dynamic type does not match its static type (eg., if
// T and the derived types had different sizes access would be
// incorrectly calculated). Deletion is also always undefined
// (C++98 [expr.delete]p3). If you're doing this, fix your code.
// - it cannot be const-qualified differently from T per unique_ptr spec
// (http://cplusplus.github.com/LWG/lwg-active.html#2118). Users wanting
// to work around this may use implicit_cast<const T*>().
// However, because of the first bullet in this comment, users MUST
// NOT use implicit_cast<Base*>() to upcast the static type of the array.
explicit scoped_ptr(element_type* array) : impl_(array) {}
// Constructor. Allows construction from a nullptr.
scoped_ptr(decltype(nullptr)) : impl_(nullptr) {}
// Constructor. Allows construction from a scoped_ptr rvalue.
scoped_ptr(scoped_ptr&& other) : impl_(&other.impl_) {}
// operator=. Allows assignment from a scoped_ptr rvalue.
scoped_ptr& operator=(scoped_ptr&& rhs) {
impl_.TakeState(&rhs.impl_);
return *this;
}
// operator=. Allows assignment from a nullptr. Deletes the currently owned
// array, if any.
scoped_ptr& operator=(decltype(nullptr)) {
reset();
return *this;
}
// Deleted copy constructor and copy assignment, to make the type move-only.
scoped_ptr(const scoped_ptr& other) = delete;
scoped_ptr& operator=(const scoped_ptr& other) = delete;
// Get an rvalue reference. (sp.Pass() does the same thing as std::move(sp).)
scoped_ptr&& Pass() { return static_cast<scoped_ptr&&>(*this); }
// Reset. Deletes the currently owned array, if any.
// Then takes ownership of a new object, if given.
void reset(element_type* array = nullptr) { impl_.reset(array); }
// Accessors to get the owned array.
element_type& operator[](size_t i) const {
assert(impl_.get() != nullptr);
return impl_.get()[i];
}
element_type* get() const { return impl_.get(); }
// Access to the deleter.
deleter_type& get_deleter() { return impl_.get_deleter(); }
const deleter_type& get_deleter() const { return impl_.get_deleter(); }
// Allow scoped_ptr<element_type> to be used in boolean expressions, but not
// implicitly convertible to a real bool (which is dangerous).
private:
typedef rtc::internal::scoped_ptr_impl<element_type, deleter_type>
scoped_ptr::*Testable;
public:
operator Testable() const {
return impl_.get() ? &scoped_ptr::impl_ : nullptr;
}
// Comparison operators.
// These return whether two scoped_ptr refer to the same object, not just to
// two different but equal objects.
bool operator==(element_type* array) const { return impl_.get() == array; }
bool operator!=(element_type* array) const { return impl_.get() != array; }
// Swap two scoped pointers.
void swap(scoped_ptr& p2) {
impl_.swap(p2.impl_);
}
// Release a pointer.
// The return value is the current pointer held by this object. If this object
// holds a nullptr, the return value is nullptr. After this operation, this
// object will hold a nullptr, and will not own the object any more.
element_type* release() WARN_UNUSED_RESULT {
return impl_.release();
}
// Delete the currently held pointer and return a pointer
// to allow overwriting of the current pointer address.
element_type** accept() WARN_UNUSED_RESULT {
return impl_.accept();
}
// Return a pointer to the current pointer address.
element_type** use() WARN_UNUSED_RESULT {
return impl_.use();
}
private:
// Force element_type to be a complete type.
enum { type_must_be_complete = sizeof(element_type) };
// Actually hold the data.
rtc::internal::scoped_ptr_impl<element_type, deleter_type> impl_;
// Disable initialization from any type other than element_type*, by
// providing a constructor that matches such an initialization, but is
// private and has no definition. This is disabled because it is not safe to
// call delete[] on an array whose static type does not match its dynamic
// type.
template <typename U> explicit scoped_ptr(U* array);
explicit scoped_ptr(int disallow_construction_from_null);
// Disable reset() from any type other than element_type*, for the same
// reasons as the constructor above.
template <typename U> void reset(U* array);
void reset(int disallow_reset_from_null);
// Forbid comparison of scoped_ptr types. If U != T, it totally
// doesn't make sense, and if U == T, it still doesn't make sense
// because you should never have the same object owned by two different
// scoped_ptrs.
template <class U> bool operator==(scoped_ptr<U> const& p2) const;
template <class U> bool operator!=(scoped_ptr<U> const& p2) const;
};
template <class T, class D>
void swap(rtc::scoped_ptr<T, D>& p1, rtc::scoped_ptr<T, D>& p2) {
p1.swap(p2);
}
} // namespace rtc
template <class T, class D>
bool operator==(T* p1, const rtc::scoped_ptr<T, D>& p2) {
return p1 == p2.get();
}
template <class T, class D>
bool operator!=(T* p1, const rtc::scoped_ptr<T, D>& p2) {
return p1 != p2.get();
}
// A function to convert T* into scoped_ptr<T>
// Doing e.g. make_scoped_ptr(new FooBarBaz<type>(arg)) is a shorter notation
// for scoped_ptr<FooBarBaz<type> >(new FooBarBaz<type>(arg))
template <typename T>
rtc::scoped_ptr<T> rtc_make_scoped_ptr(T* ptr) {
return rtc::scoped_ptr<T>(ptr);
}
#endif // #ifndef WEBRTC_BASE_SCOPED_PTR_H__

114
webrtc/base/template_util.h Normal file
View File

@ -0,0 +1,114 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
// Borrowed from Chromium's src/base/template_util.h.
#ifndef WEBRTC_BASE_TEMPLATE_UTIL_H_
#define WEBRTC_BASE_TEMPLATE_UTIL_H_
#include <stddef.h> // For size_t.
namespace rtc {
// Template definitions from tr1.
template<class T, T v>
struct integral_constant {
static const T value = v;
typedef T value_type;
typedef integral_constant<T, v> type;
};
template <class T, T v> const T integral_constant<T, v>::value;
typedef integral_constant<bool, true> true_type;
typedef integral_constant<bool, false> false_type;
template <class T> struct is_pointer : false_type {};
template <class T> struct is_pointer<T*> : true_type {};
template <class T, class U> struct is_same : public false_type {};
template <class T> struct is_same<T, T> : true_type {};
template<class> struct is_array : public false_type {};
template<class T, size_t n> struct is_array<T[n]> : public true_type {};
template<class T> struct is_array<T[]> : public true_type {};
template <class T> struct is_non_const_reference : false_type {};
template <class T> struct is_non_const_reference<T&> : true_type {};
template <class T> struct is_non_const_reference<const T&> : false_type {};
template <class T> struct is_void : false_type {};
template <> struct is_void<void> : true_type {};
namespace internal {
// Types YesType and NoType are guaranteed such that sizeof(YesType) <
// sizeof(NoType).
typedef char YesType;
struct NoType {
YesType dummy[2];
};
// This class is an implementation detail for is_convertible, and you
// don't need to know how it works to use is_convertible. For those
// who care: we declare two different functions, one whose argument is
// of type To and one with a variadic argument list. We give them
// return types of different size, so we can use sizeof to trick the
// compiler into telling us which function it would have chosen if we
// had called it with an argument of type From. See Alexandrescu's
// _Modern C++ Design_ for more details on this sort of trick.
struct ConvertHelper {
template <typename To>
static YesType Test(To);
template <typename To>
static NoType Test(...);
template <typename From>
static From& Create();
};
// Used to determine if a type is a struct/union/class. Inspired by Boost's
// is_class type_trait implementation.
struct IsClassHelper {
template <typename C>
static YesType Test(void(C::*)(void));
template <typename C>
static NoType Test(...);
};
} // namespace internal
// Inherits from true_type if From is convertible to To, false_type otherwise.
//
// Note that if the type is convertible, this will be a true_type REGARDLESS
// of whether or not the conversion would emit a warning.
template <typename From, typename To>
struct is_convertible
: integral_constant<bool,
sizeof(internal::ConvertHelper::Test<To>(
internal::ConvertHelper::Create<From>())) ==
sizeof(internal::YesType)> {
};
template <typename T>
struct is_class
: integral_constant<bool,
sizeof(internal::IsClassHelper::Test<T>(0)) ==
sizeof(internal::YesType)> {
};
} // namespace rtc
#endif // WEBRTC_BASE_TEMPLATE_UTIL_H_

243
webrtc/common_audio/BUILD.gn Normal file
View File

@ -0,0 +1,243 @@
# Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
import("//build/config/arm.gni")
import("../build/webrtc.gni")
config("common_audio_config") {
include_dirs = [
"resampler/include",
"signal_processing/include",
"vad/include",
]
}
source_set("common_audio") {
sources = [
"audio_converter.cc",
"audio_converter.h",
"audio_ring_buffer.cc",
"audio_ring_buffer.h",
"audio_util.cc",
"blocker.cc",
"blocker.h",
"channel_buffer.cc",
"channel_buffer.h",
"fft4g.c",
"fft4g.h",
"fir_filter.cc",
"fir_filter.h",
"fir_filter_neon.h",
"fir_filter_sse.h",
"include/audio_util.h",
"lapped_transform.cc",
"lapped_transform.h",
"real_fourier.cc",
"real_fourier.h",
"real_fourier_ooura.cc",
"real_fourier_ooura.h",
"resampler/include/push_resampler.h",
"resampler/include/resampler.h",
"resampler/push_resampler.cc",
"resampler/push_sinc_resampler.cc",
"resampler/push_sinc_resampler.h",
"resampler/resampler.cc",
"resampler/sinc_resampler.cc",
"resampler/sinc_resampler.h",
"ring_buffer.c",
"ring_buffer.h",
"signal_processing/auto_corr_to_refl_coef.c",
"signal_processing/auto_correlation.c",
"signal_processing/complex_fft_tables.h",
"signal_processing/copy_set_operations.c",
"signal_processing/cross_correlation.c",
"signal_processing/division_operations.c",
"signal_processing/dot_product_with_scale.c",
"signal_processing/downsample_fast.c",
"signal_processing/energy.c",
"signal_processing/filter_ar.c",
"signal_processing/filter_ma_fast_q12.c",
"signal_processing/get_hanning_window.c",
"signal_processing/get_scaling_square.c",
"signal_processing/ilbc_specific_functions.c",
"signal_processing/include/real_fft.h",
"signal_processing/include/signal_processing_library.h",
"signal_processing/include/spl_inl.h",
"signal_processing/levinson_durbin.c",
"signal_processing/lpc_to_refl_coef.c",
"signal_processing/min_max_operations.c",
"signal_processing/randomization_functions.c",
"signal_processing/real_fft.c",
"signal_processing/refl_coef_to_lpc.c",
"signal_processing/resample.c",
"signal_processing/resample_48khz.c",
"signal_processing/resample_by_2.c",
"signal_processing/resample_by_2_internal.c",
"signal_processing/resample_by_2_internal.h",
"signal_processing/resample_fractional.c",
"signal_processing/spl_init.c",
"signal_processing/spl_sqrt.c",
"signal_processing/splitting_filter.c",
"signal_processing/sqrt_of_one_minus_x_squared.c",
"signal_processing/vector_scaling_operations.c",
"sparse_fir_filter.cc",
"sparse_fir_filter.h",
"vad/include/vad.h",
"vad/include/webrtc_vad.h",
"vad/vad.cc",
"vad/vad_core.c",
"vad/vad_core.h",
"vad/vad_filterbank.c",
"vad/vad_filterbank.h",
"vad/vad_gmm.c",
"vad/vad_gmm.h",
"vad/vad_sp.c",
"vad/vad_sp.h",
"vad/webrtc_vad.c",
"wav_file.cc",
"wav_file.h",
"wav_header.cc",
"wav_header.h",
"window_generator.cc",
"window_generator.h",
]
deps = [
"../system_wrappers",
]
defines = []
if (rtc_use_openmax_dl) {
sources += [
"real_fourier_openmax.cc",
"real_fourier_openmax.h",
]
defines += [ "RTC_USE_OPENMAX_DL" ]
if (rtc_build_openmax_dl) {
deps += [ "//third_party/openmax_dl/dl" ]
}
}
if (current_cpu == "arm") {
sources += [
"signal_processing/complex_bit_reverse_arm.S",
"signal_processing/spl_sqrt_floor_arm.S",
]
if (arm_version >= 7) {
sources += [ "signal_processing/filter_ar_fast_q12_armv7.S" ]
} else {
sources += [ "signal_processing/filter_ar_fast_q12.c" ]
}
}
if (rtc_build_with_neon) {
deps += [ ":common_audio_neon" ]
}
if (current_cpu == "mipsel") {
sources += [
"signal_processing/complex_bit_reverse_mips.c",
"signal_processing/complex_fft_mips.c",
"signal_processing/cross_correlation_mips.c",
"signal_processing/downsample_fast_mips.c",
"signal_processing/filter_ar_fast_q12_mips.c",
"signal_processing/include/spl_inl_mips.h",
"signal_processing/min_max_operations_mips.c",
"signal_processing/resample_by_2_mips.c",
"signal_processing/spl_sqrt_floor_mips.c",
]
if (mips_dsp_rev > 0) {
sources += [ "signal_processing/vector_scaling_operations_mips.c" ]
}
} else {
sources += [ "signal_processing/complex_fft.c" ]
}
if (current_cpu != "arm" && current_cpu != "mipsel") {
sources += [
"signal_processing/complex_bit_reverse.c",
"signal_processing/filter_ar_fast_q12.c",
"signal_processing/spl_sqrt_floor.c",
]
}
if (is_win) {
cflags = [ "/wd4334" ] # Ignore warning on shift operator promotion.
}
configs += [ "..:common_config" ]
public_configs = [
"..:common_inherited_config",
":common_audio_config",
]
if (is_clang) {
# Suppress warnings from Chrome's Clang plugins.
# See http://code.google.com/p/webrtc/issues/detail?id=163 for details.
configs -= [ "//build/config/clang:find_bad_constructs" ]
}
if (current_cpu == "x86" || current_cpu == "x64") {
deps += [ ":common_audio_sse2" ]
}
}
if (current_cpu == "x86" || current_cpu == "x64") {
source_set("common_audio_sse2") {
sources = [
"fir_filter_sse.cc",
"resampler/sinc_resampler_sse.cc",
]
if (is_posix) {
cflags = [ "-msse2" ]
}
configs += [ "..:common_inherited_config" ]
if (is_clang) {
# Suppress warnings from Chrome's Clang plugins.
# See http://code.google.com/p/webrtc/issues/detail?id=163 for details.
configs -= [ "//build/config/clang:find_bad_constructs" ]
}
}
}
if (rtc_build_with_neon) {
source_set("common_audio_neon") {
sources = [
"fir_filter_neon.cc",
"resampler/sinc_resampler_neon.cc",
"signal_processing/cross_correlation_neon.c",
"signal_processing/downsample_fast_neon.c",
"signal_processing/min_max_operations_neon.c",
]
if (current_cpu != "arm64") {
# Enable compilation for the NEON instruction set. This is needed
# since //build/config/arm.gni only enables NEON for iOS, not Android.
# This provides the same functionality as webrtc/build/arm_neon.gypi.
configs -= [ "//build/config/compiler:compiler_arm_fpu" ]
cflags = [ "-mfpu=neon" ]
}
# Disable LTO on NEON targets due to compiler bug.
# TODO(fdegans): Enable this. See crbug.com/408997.
if (rtc_use_lto) {
cflags -= [
"-flto",
"-ffat-lto-objects",
]
}
configs += [ "..:common_config" ]
public_configs = [ "..:common_inherited_config" ]
}
}

74
webrtc/common_audio/Makefile.am Normal file
View File

@ -0,0 +1,74 @@
noinst_LTLIBRARIES = libcommon_audio.la
libcommon_audio_la_SOURCES = signal_processing/include/real_fft.h \
signal_processing/include/signal_processing_library.h \
signal_processing/include/spl_inl.h \
signal_processing/include/spl_inl_armv7.h \
signal_processing/include/spl_inl_mips.h \
signal_processing/auto_corr_to_refl_coef.c \
signal_processing/auto_correlation.c \
signal_processing/complex_bit_reverse.c \
signal_processing/complex_fft.c \
signal_processing/complex_fft_tables.h \
signal_processing/copy_set_operations.c \
signal_processing/cross_correlation.c \
signal_processing/division_operations.c \
signal_processing/dot_product_with_scale.c \
signal_processing/downsample_fast.c \
signal_processing/energy.c \
signal_processing/filter_ar.c \
signal_processing/filter_ar_fast_q12.c \
signal_processing/filter_ma_fast_q12.c \
signal_processing/get_hanning_window.c \
signal_processing/get_scaling_square.c \
signal_processing/ilbc_specific_functions.c \
signal_processing/levinson_durbin.c \
signal_processing/lpc_to_refl_coef.c \
signal_processing/min_max_operations.c \
signal_processing/randomization_functions.c \
signal_processing/real_fft.c \
signal_processing/refl_coef_to_lpc.c \
signal_processing/resample.c \
signal_processing/resample_48khz.c \
signal_processing/resample_by_2.c \
signal_processing/resample_by_2_internal.c \
signal_processing/resample_by_2_internal.h \
signal_processing/resample_fractional.c \
signal_processing/spl_init.c \
signal_processing/spl_sqrt.c \
signal_processing/spl_sqrt_floor.c \
signal_processing/splitting_filter.c \
signal_processing/sqrt_of_one_minus_x_squared.c \
signal_processing/vector_scaling_operations.c \
vad/include/vad.h \
vad/include/webrtc_vad.h \
vad/vad.cc \
vad/vad_core.c \
vad/vad_core.h \
vad/vad_filterbank.c \
vad/vad_filterbank.h \
vad/vad_gmm.c \
vad/vad_gmm.h \
vad/vad_sp.c \
vad/vad_sp.h \
vad/webrtc_vad.c
libcommon_audio_la_CFLAGS = $(AM_CFLAGS) $(COMMON_CFLAGS)
libcommon_audio_la_CXXFLAGS = $(AM_CXXFLAGS) $(COMMON_CXXFLAGS)
# FIXME:
# if ARM - signal_processing/complex_bit_reverse_arm.S
# signal_processing/spl_sqrt_floor_arm.S
# ARM7 - signal_processing/filter_ar_fast_q12_armv7.S
# NEON - signal_processing/cross_correlation_neon.c
# signal_processing/downsample_fast_neon.c
# signal_processing/min_max_operations_neon.c
# if MIPS - signal_processing/complex_bit_reverse_mips.c
# signal_processing/complex_fft_mips.c
# signal_processing/cross_correlation_mips.c
# signal_processing/downsample_fast_mips.c
# signal_processing/filter_ar_fast_q12_mips.c
# signal_processing/min_max_operations_mips.c
# signal_processing/resample_by_2_mips.c
# signal_processing/spl_sqrt_floor_mips.c
# signal_processing/vector_scaling_operations_mips.c

103
webrtc/common_audio/signal_processing/auto_corr_to_refl_coef.c Normal file
View File

@ -0,0 +1,103 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_AutoCorrToReflCoef().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_AutoCorrToReflCoef(const int32_t *R, int use_order, int16_t *K)
{
int i, n;
int16_t tmp;
const int32_t *rptr;
int32_t L_num, L_den;
int16_t *acfptr, *pptr, *wptr, *p1ptr, *w1ptr, ACF[WEBRTC_SPL_MAX_LPC_ORDER],
P[WEBRTC_SPL_MAX_LPC_ORDER], W[WEBRTC_SPL_MAX_LPC_ORDER];
// Initialize loop and pointers.
acfptr = ACF;
rptr = R;
pptr = P;
p1ptr = &P[1];
w1ptr = &W[1];
wptr = w1ptr;
// First loop; n=0. Determine shifting.
tmp = WebRtcSpl_NormW32(*R);
*acfptr = (int16_t)((*rptr++ << tmp) >> 16);
*pptr++ = *acfptr++;
// Initialize ACF, P and W.
for (i = 1; i <= use_order; i++)
{
*acfptr = (int16_t)((*rptr++ << tmp) >> 16);
*wptr++ = *acfptr;
*pptr++ = *acfptr++;
}
// Compute reflection coefficients.
for (n = 1; n <= use_order; n++, K++)
{
tmp = WEBRTC_SPL_ABS_W16(*p1ptr);
if (*P < tmp)
{
for (i = n; i <= use_order; i++)
*K++ = 0;
return;
}
// Division: WebRtcSpl_div(tmp, *P)
*K = 0;
if (tmp != 0)
{
L_num = tmp;
L_den = *P;
i = 15;
while (i--)
{
(*K) <<= 1;
L_num <<= 1;
if (L_num >= L_den)
{
L_num -= L_den;
(*K)++;
}
}
if (*p1ptr > 0)
*K = -*K;
}
// Last iteration; don't do Schur recursion.
if (n == use_order)
return;
// Schur recursion.
pptr = P;
wptr = w1ptr;
tmp = (int16_t)(((int32_t)*p1ptr * (int32_t)*K + 16384) >> 15);
*pptr = WebRtcSpl_AddSatW16(*pptr, tmp);
pptr++;
for (i = 1; i <= use_order - n; i++)
{
tmp = (int16_t)(((int32_t)*wptr * (int32_t)*K + 16384) >> 15);
*pptr = WebRtcSpl_AddSatW16(*(pptr + 1), tmp);
pptr++;
tmp = (int16_t)(((int32_t)*pptr * (int32_t)*K + 16384) >> 15);
*wptr = WebRtcSpl_AddSatW16(*wptr, tmp);
wptr++;
}
}
}

65
webrtc/common_audio/signal_processing/auto_correlation.c Normal file
View File

@ -0,0 +1,65 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include <assert.h>
size_t WebRtcSpl_AutoCorrelation(const int16_t* in_vector,
size_t in_vector_length,
size_t order,
int32_t* result,
int* scale) {
int32_t sum = 0;
size_t i = 0, j = 0;
int16_t smax = 0;
int scaling = 0;
assert(order <= in_vector_length);
// Find the maximum absolute value of the samples.
smax = WebRtcSpl_MaxAbsValueW16(in_vector, in_vector_length);
// In order to avoid overflow when computing the sum we should scale the
// samples so that (in_vector_length * smax * smax) will not overflow.
if (smax == 0) {
scaling = 0;
} else {
// Number of bits in the sum loop.
int nbits = WebRtcSpl_GetSizeInBits((uint32_t)in_vector_length);
// Number of bits to normalize smax.
int t = WebRtcSpl_NormW32(WEBRTC_SPL_MUL(smax, smax));
if (t > nbits) {
scaling = 0;
} else {
scaling = nbits - t;
}
}
// Perform the actual correlation calculation.
for (i = 0; i < order + 1; i++) {
sum = 0;
/* Unroll the loop to improve performance. */
for (j = 0; i + j + 3 < in_vector_length; j += 4) {
sum += (in_vector[j + 0] * in_vector[i + j + 0]) >> scaling;
sum += (in_vector[j + 1] * in_vector[i + j + 1]) >> scaling;
sum += (in_vector[j + 2] * in_vector[i + j + 2]) >> scaling;
sum += (in_vector[j + 3] * in_vector[i + j + 3]) >> scaling;
}
for (; j < in_vector_length - i; j++) {
sum += (in_vector[j] * in_vector[i + j]) >> scaling;
}
*result++ = sum;
}
*scale = scaling;
return order + 1;
}

108
webrtc/common_audio/signal_processing/complex_bit_reverse.c Normal file
View File

@ -0,0 +1,108 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
/* Tables for data buffer indexes that are bit reversed and thus need to be
* swapped. Note that, index_7[{0, 2, 4, ...}] are for the left side of the swap
* operations, while index_7[{1, 3, 5, ...}] are for the right side of the
* operation. Same for index_8.
*/
/* Indexes for the case of stages == 7. */
static const int16_t index_7[112] = {
1, 64, 2, 32, 3, 96, 4, 16, 5, 80, 6, 48, 7, 112, 9, 72, 10, 40, 11, 104,
12, 24, 13, 88, 14, 56, 15, 120, 17, 68, 18, 36, 19, 100, 21, 84, 22, 52,
23, 116, 25, 76, 26, 44, 27, 108, 29, 92, 30, 60, 31, 124, 33, 66, 35, 98,
37, 82, 38, 50, 39, 114, 41, 74, 43, 106, 45, 90, 46, 58, 47, 122, 49, 70,
51, 102, 53, 86, 55, 118, 57, 78, 59, 110, 61, 94, 63, 126, 67, 97, 69,
81, 71, 113, 75, 105, 77, 89, 79, 121, 83, 101, 87, 117, 91, 109, 95, 125,
103, 115, 111, 123
};
/* Indexes for the case of stages == 8. */
static const int16_t index_8[240] = {
1, 128, 2, 64, 3, 192, 4, 32, 5, 160, 6, 96, 7, 224, 8, 16, 9, 144, 10, 80,
11, 208, 12, 48, 13, 176, 14, 112, 15, 240, 17, 136, 18, 72, 19, 200, 20,
40, 21, 168, 22, 104, 23, 232, 25, 152, 26, 88, 27, 216, 28, 56, 29, 184,
30, 120, 31, 248, 33, 132, 34, 68, 35, 196, 37, 164, 38, 100, 39, 228, 41,
148, 42, 84, 43, 212, 44, 52, 45, 180, 46, 116, 47, 244, 49, 140, 50, 76,
51, 204, 53, 172, 54, 108, 55, 236, 57, 156, 58, 92, 59, 220, 61, 188, 62,
124, 63, 252, 65, 130, 67, 194, 69, 162, 70, 98, 71, 226, 73, 146, 74, 82,
75, 210, 77, 178, 78, 114, 79, 242, 81, 138, 83, 202, 85, 170, 86, 106, 87,
234, 89, 154, 91, 218, 93, 186, 94, 122, 95, 250, 97, 134, 99, 198, 101,
166, 103, 230, 105, 150, 107, 214, 109, 182, 110, 118, 111, 246, 113, 142,
115, 206, 117, 174, 119, 238, 121, 158, 123, 222, 125, 190, 127, 254, 131,
193, 133, 161, 135, 225, 137, 145, 139, 209, 141, 177, 143, 241, 147, 201,
149, 169, 151, 233, 155, 217, 157, 185, 159, 249, 163, 197, 167, 229, 171,
213, 173, 181, 175, 245, 179, 205, 183, 237, 187, 221, 191, 253, 199, 227,
203, 211, 207, 243, 215, 235, 223, 251, 239, 247
};
void WebRtcSpl_ComplexBitReverse(int16_t* __restrict complex_data, int stages) {
/* For any specific value of stages, we know exactly the indexes that are
* bit reversed. Currently (Feb. 2012) in WebRTC the only possible values of
* stages are 7 and 8, so we use tables to save unnecessary iterations and
* calculations for these two cases.
*/
if (stages == 7 || stages == 8) {
int m = 0;
int length = 112;
const int16_t* index = index_7;
if (stages == 8) {
length = 240;
index = index_8;
}
/* Decimation in time. Swap the elements with bit-reversed indexes. */
for (m = 0; m < length; m += 2) {
/* We declare a int32_t* type pointer, to load both the 16-bit real
* and imaginary elements from complex_data in one instruction, reducing
* complexity.
*/
int32_t* complex_data_ptr = (int32_t*)complex_data;
int32_t temp = 0;
temp = complex_data_ptr[index[m]]; /* Real and imaginary */
complex_data_ptr[index[m]] = complex_data_ptr[index[m + 1]];
complex_data_ptr[index[m + 1]] = temp;
}
}
else {
int m = 0, mr = 0, l = 0;
int n = 1 << stages;
int nn = n - 1;
/* Decimation in time - re-order data */
for (m = 1; m <= nn; ++m) {
int32_t* complex_data_ptr = (int32_t*)complex_data;
int32_t temp = 0;
/* Find out indexes that are bit-reversed. */
l = n;
do {
l >>= 1;
} while (l > nn - mr);
mr = (mr & (l - 1)) + l;
if (mr <= m) {
continue;
}
/* Swap the elements with bit-reversed indexes.
* This is similar to the loop in the stages == 7 or 8 cases.
*/
temp = complex_data_ptr[m]; /* Real and imaginary */
complex_data_ptr[m] = complex_data_ptr[mr];
complex_data_ptr[mr] = temp;
}
}
}

119
webrtc/common_audio/signal_processing/complex_bit_reverse_arm.S Normal file
View File

@ -0,0 +1,119 @@
@
@ Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
@
@ Use of this source code is governed by a BSD-style license
@ that can be found in the LICENSE file in the root of the source
@ tree. An additional intellectual property rights grant can be found
@ in the file PATENTS. All contributing project authors may
@ be found in the AUTHORS file in the root of the source tree.
@
@ This file contains the function WebRtcSpl_ComplexBitReverse(), optimized
@ for ARMv5 platforms.
@ Reference C code is in file complex_bit_reverse.c. Bit-exact.
#include "webrtc/system_wrappers/interface/asm_defines.h"
GLOBAL_FUNCTION WebRtcSpl_ComplexBitReverse
.align 2
DEFINE_FUNCTION WebRtcSpl_ComplexBitReverse
push {r4-r7}
cmp r1, #7
adr r3, index_7 @ Table pointer.
mov r4, #112 @ Number of interations.
beq PRE_LOOP_STAGES_7_OR_8
cmp r1, #8
adr r3, index_8 @ Table pointer.
mov r4, #240 @ Number of interations.
beq PRE_LOOP_STAGES_7_OR_8
mov r3, #1 @ Initialize m.
mov r1, r3, asl r1 @ n = 1 << stages;
subs r6, r1, #1 @ nn = n - 1;
ble END
mov r5, r0 @ &complex_data
mov r4, #0 @ ml
LOOP_GENERIC:
rsb r12, r4, r6 @ l > nn - mr
mov r2, r1 @ n
LOOP_SHIFT:
asr r2, #1 @ l >>= 1;
cmp r2, r12
bgt LOOP_SHIFT
sub r12, r2, #1
and r4, r12, r4
add r4, r2 @ mr = (mr & (l - 1)) + l;
cmp r4, r3 @ mr <= m ?
ble UPDATE_REGISTERS
mov r12, r4, asl #2
ldr r7, [r5, #4] @ complex_data[2 * m, 2 * m + 1].
@ Offset 4 due to m incrementing from 1.
ldr r2, [r0, r12] @ complex_data[2 * mr, 2 * mr + 1].
str r7, [r0, r12]
str r2, [r5, #4]
UPDATE_REGISTERS:
add r3, r3, #1
add r5, #4
cmp r3, r1
bne LOOP_GENERIC
b END
PRE_LOOP_STAGES_7_OR_8:
add r4, r3, r4, asl #1
LOOP_STAGES_7_OR_8:
ldrsh r2, [r3], #2 @ index[m]
ldrsh r5, [r3], #2 @ index[m + 1]
ldr r1, [r0, r2] @ complex_data[index[m], index[m] + 1]
ldr r12, [r0, r5] @ complex_data[index[m + 1], index[m + 1] + 1]
cmp r3, r4
str r1, [r0, r5]
str r12, [r0, r2]
bne LOOP_STAGES_7_OR_8
END:
pop {r4-r7}
bx lr
@ The index tables. Note the values are doubles of the actual indexes for 16-bit
@ elements, different from the generic C code. It actually provides byte offsets
@ for the indexes.
.align 2
index_7: @ Indexes for stages == 7.
.short 4, 256, 8, 128, 12, 384, 16, 64, 20, 320, 24, 192, 28, 448, 36, 288
.short 40, 160, 44, 416, 48, 96, 52, 352, 56, 224, 60, 480, 68, 272, 72, 144
.short 76, 400, 84, 336, 88, 208, 92, 464, 100, 304, 104, 176, 108, 432, 116
.short 368, 120, 240, 124, 496, 132, 264, 140, 392, 148, 328, 152, 200, 156
.short 456, 164, 296, 172, 424, 180, 360, 184, 232, 188, 488, 196, 280, 204
.short 408, 212, 344, 220, 472, 228, 312, 236, 440, 244, 376, 252, 504, 268
.short 388, 276, 324, 284, 452, 300, 420, 308, 356, 316, 484, 332, 404, 348
.short 468, 364, 436, 380, 500, 412, 460, 444, 492
index_8: @ Indexes for stages == 8.
.short 4, 512, 8, 256, 12, 768, 16, 128, 20, 640, 24, 384, 28, 896, 32, 64
.short 36, 576, 40, 320, 44, 832, 48, 192, 52, 704, 56, 448, 60, 960, 68, 544
.short 72, 288, 76, 800, 80, 160, 84, 672, 88, 416, 92, 928, 100, 608, 104
.short 352, 108, 864, 112, 224, 116, 736, 120, 480, 124, 992, 132, 528, 136
.short 272, 140, 784, 148, 656, 152, 400, 156, 912, 164, 592, 168, 336, 172
.short 848, 176, 208, 180, 720, 184, 464, 188, 976, 196, 560, 200, 304, 204
.short 816, 212, 688, 216, 432, 220, 944, 228, 624, 232, 368, 236, 880, 244
.short 752, 248, 496, 252, 1008, 260, 520, 268, 776, 276, 648, 280, 392, 284
.short 904, 292, 584, 296, 328, 300, 840, 308, 712, 312, 456, 316, 968, 324
.short 552, 332, 808, 340, 680, 344, 424, 348, 936, 356, 616, 364, 872, 372
.short 744, 376, 488, 380, 1000, 388, 536, 396, 792, 404, 664, 412, 920, 420
.short 600, 428, 856, 436, 728, 440, 472, 444, 984, 452, 568, 460, 824, 468
.short 696, 476, 952, 484, 632, 492, 888, 500, 760, 508, 1016, 524, 772, 532
.short 644, 540, 900, 548, 580, 556, 836, 564, 708, 572, 964, 588, 804, 596
.short 676, 604, 932, 620, 868, 628, 740, 636, 996, 652, 788, 668, 916, 684
.short 852, 692, 724, 700, 980, 716, 820, 732, 948, 748, 884, 764, 1012, 796
.short 908, 812, 844, 828, 972, 860, 940, 892, 1004, 956, 988

176
webrtc/common_audio/signal_processing/complex_bit_reverse_mips.c Normal file
View File

@ -0,0 +1,176 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
static int16_t coefTable_7[] = {
4, 256, 8, 128, 12, 384, 16, 64,
20, 320, 24, 192, 28, 448, 36, 288,
40, 160, 44, 416, 48, 96, 52, 352,
56, 224, 60, 480, 68, 272, 72, 144,
76, 400, 84, 336, 88, 208, 92, 464,
100, 304, 104, 176, 108, 432, 116, 368,
120, 240, 124, 496, 132, 264, 140, 392,
148, 328, 152, 200, 156, 456, 164, 296,
172, 424, 180, 360, 184, 232, 188, 488,
196, 280, 204, 408, 212, 344, 220, 472,
228, 312, 236, 440, 244, 376, 252, 504,
268, 388, 276, 324, 284, 452, 300, 420,
308, 356, 316, 484, 332, 404, 348, 468,
364, 436, 380, 500, 412, 460, 444, 492
};
static int16_t coefTable_8[] = {
4, 512, 8, 256, 12, 768, 16, 128,
20, 640, 24, 384, 28, 896, 32, 64,
36, 576, 40, 320, 44, 832, 48, 192,
52, 704, 56, 448, 60, 960, 68, 544,
72, 288, 76, 800, 80, 160, 84, 672,
88, 416, 92, 928, 100, 608, 104, 352,
108, 864, 112, 224, 116, 736, 120, 480,
124, 992, 132, 528, 136, 272, 140, 784,
148, 656, 152, 400, 156, 912, 164, 592,
168, 336, 172, 848, 176, 208, 180, 720,
184, 464, 188, 976, 196, 560, 200, 304,
204, 816, 212, 688, 216, 432, 220, 944,
228, 624, 232, 368, 236, 880, 244, 752,
248, 496, 252, 1008, 260, 520, 268, 776,
276, 648, 280, 392, 284, 904, 292, 584,
296, 328, 300, 840, 308, 712, 312, 456,
316, 968, 324, 552, 332, 808, 340, 680,
344, 424, 348, 936, 356, 616, 364, 872,
372, 744, 376, 488, 380, 1000, 388, 536,
396, 792, 404, 664, 412, 920, 420, 600,
428, 856, 436, 728, 440, 472, 444, 984,
452, 568, 460, 824, 468, 696, 476, 952,
484, 632, 492, 888, 500, 760, 508, 1016,
524, 772, 532, 644, 540, 900, 548, 580,
556, 836, 564, 708, 572, 964, 588, 804,
596, 676, 604, 932, 620, 868, 628, 740,
636, 996, 652, 788, 668, 916, 684, 852,
692, 724, 700, 980, 716, 820, 732, 948,
748, 884, 764, 1012, 796, 908, 812, 844,
828, 972, 860, 940, 892, 1004, 956, 988
};
void WebRtcSpl_ComplexBitReverse(int16_t frfi[], int stages) {
int l;
int16_t tr, ti;
int32_t tmp1, tmp2, tmp3, tmp4;
int32_t* ptr_i;
int32_t* ptr_j;
if (stages == 8) {
int16_t* pcoeftable_8 = coefTable_8;
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"addiu %[l], $zero, 120 \n\t"
"1: \n\t"
"addiu %[l], %[l], -4 \n\t"
"lh %[tr], 0(%[pcoeftable_8]) \n\t"
"lh %[ti], 2(%[pcoeftable_8]) \n\t"
"lh %[tmp3], 4(%[pcoeftable_8]) \n\t"
"lh %[tmp4], 6(%[pcoeftable_8]) \n\t"
"addu %[ptr_i], %[frfi], %[tr] \n\t"
"addu %[ptr_j], %[frfi], %[ti] \n\t"
"addu %[tr], %[frfi], %[tmp3] \n\t"
"addu %[ti], %[frfi], %[tmp4] \n\t"
"ulw %[tmp1], 0(%[ptr_i]) \n\t"
"ulw %[tmp2], 0(%[ptr_j]) \n\t"
"ulw %[tmp3], 0(%[tr]) \n\t"
"ulw %[tmp4], 0(%[ti]) \n\t"
"usw %[tmp1], 0(%[ptr_j]) \n\t"
"usw %[tmp2], 0(%[ptr_i]) \n\t"
"usw %[tmp4], 0(%[tr]) \n\t"
"usw %[tmp3], 0(%[ti]) \n\t"
"lh %[tmp1], 8(%[pcoeftable_8]) \n\t"
"lh %[tmp2], 10(%[pcoeftable_8]) \n\t"
"lh %[tr], 12(%[pcoeftable_8]) \n\t"
"lh %[ti], 14(%[pcoeftable_8]) \n\t"
"addu %[ptr_i], %[frfi], %[tmp1] \n\t"
"addu %[ptr_j], %[frfi], %[tmp2] \n\t"
"addu %[tr], %[frfi], %[tr] \n\t"
"addu %[ti], %[frfi], %[ti] \n\t"
"ulw %[tmp1], 0(%[ptr_i]) \n\t"
"ulw %[tmp2], 0(%[ptr_j]) \n\t"
"ulw %[tmp3], 0(%[tr]) \n\t"
"ulw %[tmp4], 0(%[ti]) \n\t"
"usw %[tmp1], 0(%[ptr_j]) \n\t"
"usw %[tmp2], 0(%[ptr_i]) \n\t"
"usw %[tmp4], 0(%[tr]) \n\t"
"usw %[tmp3], 0(%[ti]) \n\t"
"bgtz %[l], 1b \n\t"
" addiu %[pcoeftable_8], %[pcoeftable_8], 16 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [ptr_i] "=&r" (ptr_i),
[ptr_j] "=&r" (ptr_j), [tr] "=&r" (tr), [l] "=&r" (l),
[tmp3] "=&r" (tmp3), [pcoeftable_8] "+r" (pcoeftable_8),
[ti] "=&r" (ti), [tmp4] "=&r" (tmp4)
: [frfi] "r" (frfi)
: "memory"
);
} else if (stages == 7) {
int16_t* pcoeftable_7 = coefTable_7;
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"addiu %[l], $zero, 56 \n\t"
"1: \n\t"
"addiu %[l], %[l], -4 \n\t"
"lh %[tr], 0(%[pcoeftable_7]) \n\t"
"lh %[ti], 2(%[pcoeftable_7]) \n\t"
"lh %[tmp3], 4(%[pcoeftable_7]) \n\t"
"lh %[tmp4], 6(%[pcoeftable_7]) \n\t"
"addu %[ptr_i], %[frfi], %[tr] \n\t"
"addu %[ptr_j], %[frfi], %[ti] \n\t"
"addu %[tr], %[frfi], %[tmp3] \n\t"
"addu %[ti], %[frfi], %[tmp4] \n\t"
"ulw %[tmp1], 0(%[ptr_i]) \n\t"
"ulw %[tmp2], 0(%[ptr_j]) \n\t"
"ulw %[tmp3], 0(%[tr]) \n\t"
"ulw %[tmp4], 0(%[ti]) \n\t"
"usw %[tmp1], 0(%[ptr_j]) \n\t"
"usw %[tmp2], 0(%[ptr_i]) \n\t"
"usw %[tmp4], 0(%[tr]) \n\t"
"usw %[tmp3], 0(%[ti]) \n\t"
"lh %[tmp1], 8(%[pcoeftable_7]) \n\t"
"lh %[tmp2], 10(%[pcoeftable_7]) \n\t"
"lh %[tr], 12(%[pcoeftable_7]) \n\t"
"lh %[ti], 14(%[pcoeftable_7]) \n\t"
"addu %[ptr_i], %[frfi], %[tmp1] \n\t"
"addu %[ptr_j], %[frfi], %[tmp2] \n\t"
"addu %[tr], %[frfi], %[tr] \n\t"
"addu %[ti], %[frfi], %[ti] \n\t"
"ulw %[tmp1], 0(%[ptr_i]) \n\t"
"ulw %[tmp2], 0(%[ptr_j]) \n\t"
"ulw %[tmp3], 0(%[tr]) \n\t"
"ulw %[tmp4], 0(%[ti]) \n\t"
"usw %[tmp1], 0(%[ptr_j]) \n\t"
"usw %[tmp2], 0(%[ptr_i]) \n\t"
"usw %[tmp4], 0(%[tr]) \n\t"
"usw %[tmp3], 0(%[ti]) \n\t"
"bgtz %[l], 1b \n\t"
" addiu %[pcoeftable_7], %[pcoeftable_7], 16 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [ptr_i] "=&r" (ptr_i),
[ptr_j] "=&r" (ptr_j), [ti] "=&r" (ti), [tr] "=&r" (tr),
[l] "=&r" (l), [pcoeftable_7] "+r" (pcoeftable_7),
[tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4)
: [frfi] "r" (frfi)
: "memory"
);
}
}

298
webrtc/common_audio/signal_processing/complex_fft.c Normal file
View File

@ -0,0 +1,298 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_ComplexFFT().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/complex_fft_tables.h"
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#define CFFTSFT 14
#define CFFTRND 1
#define CFFTRND2 16384
#define CIFFTSFT 14
#define CIFFTRND 1
int WebRtcSpl_ComplexFFT(int16_t frfi[], int stages, int mode)
{
int i, j, l, k, istep, n, m;
int16_t wr, wi;
int32_t tr32, ti32, qr32, qi32;
/* The 1024-value is a constant given from the size of kSinTable1024[],
* and should not be changed depending on the input parameter 'stages'
*/
n = 1 << stages;
if (n > 1024)
return -1;
l = 1;
k = 10 - 1; /* Constant for given kSinTable1024[]. Do not change
depending on the input parameter 'stages' */
if (mode == 0)
{
// mode==0: Low-complexity and Low-accuracy mode
while (l < n)
{
istep = l << 1;
for (m = 0; m < l; ++m)
{
j = m << k;
/* The 256-value is a constant given as 1/4 of the size of
* kSinTable1024[], and should not be changed depending on the input
* parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
*/
wr = kSinTable1024[j + 256];
wi = -kSinTable1024[j];
for (i = m; i < n; i += istep)
{
j = i + l;
tr32 = (wr * frfi[2 * j] - wi * frfi[2 * j + 1]) >> 15;
ti32 = (wr * frfi[2 * j + 1] + wi * frfi[2 * j]) >> 15;
qr32 = (int32_t)frfi[2 * i];
qi32 = (int32_t)frfi[2 * i + 1];
frfi[2 * j] = (int16_t)((qr32 - tr32) >> 1);
frfi[2 * j + 1] = (int16_t)((qi32 - ti32) >> 1);
frfi[2 * i] = (int16_t)((qr32 + tr32) >> 1);
frfi[2 * i + 1] = (int16_t)((qi32 + ti32) >> 1);
}
}
--k;
l = istep;
}
} else
{
// mode==1: High-complexity and High-accuracy mode
while (l < n)
{
istep = l << 1;
for (m = 0; m < l; ++m)
{
j = m << k;
/* The 256-value is a constant given as 1/4 of the size of
* kSinTable1024[], and should not be changed depending on the input
* parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
*/
wr = kSinTable1024[j + 256];
wi = -kSinTable1024[j];
#ifdef WEBRTC_ARCH_ARM_V7
int32_t wri = 0;
__asm __volatile("pkhbt %0, %1, %2, lsl #16" : "=r"(wri) :
"r"((int32_t)wr), "r"((int32_t)wi));
#endif
for (i = m; i < n; i += istep)
{
j = i + l;
#ifdef WEBRTC_ARCH_ARM_V7
register int32_t frfi_r;
__asm __volatile(
"pkhbt %[frfi_r], %[frfi_even], %[frfi_odd],"
" lsl #16\n\t"
"smlsd %[tr32], %[wri], %[frfi_r], %[cfftrnd]\n\t"
"smladx %[ti32], %[wri], %[frfi_r], %[cfftrnd]\n\t"
:[frfi_r]"=&r"(frfi_r),
[tr32]"=&r"(tr32),
[ti32]"=r"(ti32)
:[frfi_even]"r"((int32_t)frfi[2*j]),
[frfi_odd]"r"((int32_t)frfi[2*j +1]),
[wri]"r"(wri),
[cfftrnd]"r"(CFFTRND));
#else
tr32 = wr * frfi[2 * j] - wi * frfi[2 * j + 1] + CFFTRND;
ti32 = wr * frfi[2 * j + 1] + wi * frfi[2 * j] + CFFTRND;
#endif
tr32 >>= 15 - CFFTSFT;
ti32 >>= 15 - CFFTSFT;
qr32 = ((int32_t)frfi[2 * i]) << CFFTSFT;
qi32 = ((int32_t)frfi[2 * i + 1]) << CFFTSFT;
frfi[2 * j] = (int16_t)(
(qr32 - tr32 + CFFTRND2) >> (1 + CFFTSFT));
frfi[2 * j + 1] = (int16_t)(
(qi32 - ti32 + CFFTRND2) >> (1 + CFFTSFT));
frfi[2 * i] = (int16_t)(
(qr32 + tr32 + CFFTRND2) >> (1 + CFFTSFT));
frfi[2 * i + 1] = (int16_t)(
(qi32 + ti32 + CFFTRND2) >> (1 + CFFTSFT));
}
}
--k;
l = istep;
}
}
return 0;
}
int WebRtcSpl_ComplexIFFT(int16_t frfi[], int stages, int mode)
{
size_t i, j, l, istep, n, m;
int k, scale, shift;
int16_t wr, wi;
int32_t tr32, ti32, qr32, qi32;
int32_t tmp32, round2;
/* The 1024-value is a constant given from the size of kSinTable1024[],
* and should not be changed depending on the input parameter 'stages'
*/
n = 1 << stages;
if (n > 1024)
return -1;
scale = 0;
l = 1;
k = 10 - 1; /* Constant for given kSinTable1024[]. Do not change
depending on the input parameter 'stages' */
while (l < n)
{
// variable scaling, depending upon data
shift = 0;
round2 = 8192;
tmp32 = WebRtcSpl_MaxAbsValueW16(frfi, 2 * n);
if (tmp32 > 13573)
{
shift++;
scale++;
round2 <<= 1;
}
if (tmp32 > 27146)
{
shift++;
scale++;
round2 <<= 1;
}
istep = l << 1;
if (mode == 0)
{
// mode==0: Low-complexity and Low-accuracy mode
for (m = 0; m < l; ++m)
{
j = m << k;
/* The 256-value is a constant given as 1/4 of the size of
* kSinTable1024[], and should not be changed depending on the input
* parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
*/
wr = kSinTable1024[j + 256];
wi = kSinTable1024[j];
for (i = m; i < n; i += istep)
{
j = i + l;
tr32 = (wr * frfi[2 * j] - wi * frfi[2 * j + 1]) >> 15;
ti32 = (wr * frfi[2 * j + 1] + wi * frfi[2 * j]) >> 15;
qr32 = (int32_t)frfi[2 * i];
qi32 = (int32_t)frfi[2 * i + 1];
frfi[2 * j] = (int16_t)((qr32 - tr32) >> shift);
frfi[2 * j + 1] = (int16_t)((qi32 - ti32) >> shift);
frfi[2 * i] = (int16_t)((qr32 + tr32) >> shift);
frfi[2 * i + 1] = (int16_t)((qi32 + ti32) >> shift);
}
}
} else
{
// mode==1: High-complexity and High-accuracy mode
for (m = 0; m < l; ++m)
{
j = m << k;
/* The 256-value is a constant given as 1/4 of the size of
* kSinTable1024[], and should not be changed depending on the input
* parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
*/
wr = kSinTable1024[j + 256];
wi = kSinTable1024[j];
#ifdef WEBRTC_ARCH_ARM_V7
int32_t wri = 0;
__asm __volatile("pkhbt %0, %1, %2, lsl #16" : "=r"(wri) :
"r"((int32_t)wr), "r"((int32_t)wi));
#endif
for (i = m; i < n; i += istep)
{
j = i + l;
#ifdef WEBRTC_ARCH_ARM_V7
register int32_t frfi_r;
__asm __volatile(
"pkhbt %[frfi_r], %[frfi_even], %[frfi_odd], lsl #16\n\t"
"smlsd %[tr32], %[wri], %[frfi_r], %[cifftrnd]\n\t"
"smladx %[ti32], %[wri], %[frfi_r], %[cifftrnd]\n\t"
:[frfi_r]"=&r"(frfi_r),
[tr32]"=&r"(tr32),
[ti32]"=r"(ti32)
:[frfi_even]"r"((int32_t)frfi[2*j]),
[frfi_odd]"r"((int32_t)frfi[2*j +1]),
[wri]"r"(wri),
[cifftrnd]"r"(CIFFTRND)
);
#else
tr32 = wr * frfi[2 * j] - wi * frfi[2 * j + 1] + CIFFTRND;
ti32 = wr * frfi[2 * j + 1] + wi * frfi[2 * j] + CIFFTRND;
#endif
tr32 >>= 15 - CIFFTSFT;
ti32 >>= 15 - CIFFTSFT;
qr32 = ((int32_t)frfi[2 * i]) << CIFFTSFT;
qi32 = ((int32_t)frfi[2 * i + 1]) << CIFFTSFT;
frfi[2 * j] = (int16_t)(
(qr32 - tr32 + round2) >> (shift + CIFFTSFT));
frfi[2 * j + 1] = (int16_t)(
(qi32 - ti32 + round2) >> (shift + CIFFTSFT));
frfi[2 * i] = (int16_t)(
(qr32 + tr32 + round2) >> (shift + CIFFTSFT));
frfi[2 * i + 1] = (int16_t)(
(qi32 + ti32 + round2) >> (shift + CIFFTSFT));
}
}
}
--k;
l = istep;
}
return scale;
}

328
webrtc/common_audio/signal_processing/complex_fft_mips.c Normal file
View File

@ -0,0 +1,328 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/complex_fft_tables.h"
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#define CFFTSFT 14
#define CFFTRND 1
#define CFFTRND2 16384
#define CIFFTSFT 14
#define CIFFTRND 1
int WebRtcSpl_ComplexFFT(int16_t frfi[], int stages, int mode) {
int i = 0;
int l = 0;
int k = 0;
int istep = 0;
int n = 0;
int m = 0;
int32_t wr = 0, wi = 0;
int32_t tmp1 = 0;
int32_t tmp2 = 0;
int32_t tmp3 = 0;
int32_t tmp4 = 0;
int32_t tmp5 = 0;
int32_t tmp6 = 0;
int32_t tmp = 0;
int16_t* ptr_j = NULL;
int16_t* ptr_i = NULL;
n = 1 << stages;
if (n > 1024) {
return -1;
}
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"addiu %[k], $zero, 10 \n\t"
"addiu %[l], $zero, 1 \n\t"
"3: \n\t"
"sll %[istep], %[l], 1 \n\t"
"move %[m], $zero \n\t"
"sll %[tmp], %[l], 2 \n\t"
"move %[i], $zero \n\t"
"2: \n\t"
#if defined(MIPS_DSP_R1_LE)
"sllv %[tmp3], %[m], %[k] \n\t"
"addiu %[tmp2], %[tmp3], 512 \n\t"
"addiu %[m], %[m], 1 \n\t"
"lhx %[wi], %[tmp3](%[kSinTable1024]) \n\t"
"lhx %[wr], %[tmp2](%[kSinTable1024]) \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"sllv %[tmp3], %[m], %[k] \n\t"
"addu %[ptr_j], %[tmp3], %[kSinTable1024] \n\t"
"addiu %[ptr_i], %[ptr_j], 512 \n\t"
"addiu %[m], %[m], 1 \n\t"
"lh %[wi], 0(%[ptr_j]) \n\t"
"lh %[wr], 0(%[ptr_i]) \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"1: \n\t"
"sll %[tmp1], %[i], 2 \n\t"
"addu %[ptr_i], %[frfi], %[tmp1] \n\t"
"addu %[ptr_j], %[ptr_i], %[tmp] \n\t"
"lh %[tmp6], 0(%[ptr_i]) \n\t"
"lh %[tmp5], 2(%[ptr_i]) \n\t"
"lh %[tmp3], 0(%[ptr_j]) \n\t"
"lh %[tmp4], 2(%[ptr_j]) \n\t"
"addu %[i], %[i], %[istep] \n\t"
#if defined(MIPS_DSP_R2_LE)
"mult %[wr], %[tmp3] \n\t"
"madd %[wi], %[tmp4] \n\t"
"mult $ac1, %[wr], %[tmp4] \n\t"
"msub $ac1, %[wi], %[tmp3] \n\t"
"mflo %[tmp1] \n\t"
"mflo %[tmp2], $ac1 \n\t"
"sll %[tmp6], %[tmp6], 14 \n\t"
"sll %[tmp5], %[tmp5], 14 \n\t"
"shra_r.w %[tmp1], %[tmp1], 1 \n\t"
"shra_r.w %[tmp2], %[tmp2], 1 \n\t"
"subu %[tmp4], %[tmp6], %[tmp1] \n\t"
"addu %[tmp1], %[tmp6], %[tmp1] \n\t"
"addu %[tmp6], %[tmp5], %[tmp2] \n\t"
"subu %[tmp5], %[tmp5], %[tmp2] \n\t"
"shra_r.w %[tmp1], %[tmp1], 15 \n\t"
"shra_r.w %[tmp6], %[tmp6], 15 \n\t"
"shra_r.w %[tmp4], %[tmp4], 15 \n\t"
"shra_r.w %[tmp5], %[tmp5], 15 \n\t"
#else // #if defined(MIPS_DSP_R2_LE)
"mul %[tmp2], %[wr], %[tmp4] \n\t"
"mul %[tmp1], %[wr], %[tmp3] \n\t"
"mul %[tmp4], %[wi], %[tmp4] \n\t"
"mul %[tmp3], %[wi], %[tmp3] \n\t"
"sll %[tmp6], %[tmp6], 14 \n\t"
"sll %[tmp5], %[tmp5], 14 \n\t"
"addiu %[tmp6], %[tmp6], 16384 \n\t"
"addiu %[tmp5], %[tmp5], 16384 \n\t"
"addu %[tmp1], %[tmp1], %[tmp4] \n\t"
"subu %[tmp2], %[tmp2], %[tmp3] \n\t"
"addiu %[tmp1], %[tmp1], 1 \n\t"
"addiu %[tmp2], %[tmp2], 1 \n\t"
"sra %[tmp1], %[tmp1], 1 \n\t"
"sra %[tmp2], %[tmp2], 1 \n\t"
"subu %[tmp4], %[tmp6], %[tmp1] \n\t"
"addu %[tmp1], %[tmp6], %[tmp1] \n\t"
"addu %[tmp6], %[tmp5], %[tmp2] \n\t"
"subu %[tmp5], %[tmp5], %[tmp2] \n\t"
"sra %[tmp4], %[tmp4], 15 \n\t"
"sra %[tmp1], %[tmp1], 15 \n\t"
"sra %[tmp6], %[tmp6], 15 \n\t"
"sra %[tmp5], %[tmp5], 15 \n\t"
#endif // #if defined(MIPS_DSP_R2_LE)
"sh %[tmp1], 0(%[ptr_i]) \n\t"
"sh %[tmp6], 2(%[ptr_i]) \n\t"
"sh %[tmp4], 0(%[ptr_j]) \n\t"
"blt %[i], %[n], 1b \n\t"
" sh %[tmp5], 2(%[ptr_j]) \n\t"
"blt %[m], %[l], 2b \n\t"
" addu %[i], $zero, %[m] \n\t"
"move %[l], %[istep] \n\t"
"blt %[l], %[n], 3b \n\t"
" addiu %[k], %[k], -1 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3),
[tmp4] "=&r" (tmp4), [tmp5] "=&r" (tmp5), [tmp6] "=&r" (tmp6),
[ptr_i] "=&r" (ptr_i), [i] "=&r" (i), [wi] "=&r" (wi), [wr] "=&r" (wr),
[m] "=&r" (m), [istep] "=&r" (istep), [l] "=&r" (l), [k] "=&r" (k),
[ptr_j] "=&r" (ptr_j), [tmp] "=&r" (tmp)
: [n] "r" (n), [frfi] "r" (frfi), [kSinTable1024] "r" (kSinTable1024)
: "hi", "lo", "memory"
#if defined(MIPS_DSP_R2_LE)
, "$ac1hi", "$ac1lo"
#endif // #if defined(MIPS_DSP_R2_LE)
);
return 0;
}
int WebRtcSpl_ComplexIFFT(int16_t frfi[], int stages, int mode) {
int i = 0, l = 0, k = 0;
int istep = 0, n = 0, m = 0;
int scale = 0, shift = 0;
int32_t wr = 0, wi = 0;
int32_t tmp1 = 0, tmp2 = 0, tmp3 = 0, tmp4 = 0;
int32_t tmp5 = 0, tmp6 = 0, tmp = 0, tempMax = 0, round2 = 0;
int16_t* ptr_j = NULL;
int16_t* ptr_i = NULL;
n = 1 << stages;
if (n > 1024) {
return -1;
}
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"addiu %[k], $zero, 10 \n\t"
"addiu %[l], $zero, 1 \n\t"
"move %[scale], $zero \n\t"
"3: \n\t"
"addiu %[shift], $zero, 14 \n\t"
"addiu %[round2], $zero, 8192 \n\t"
"move %[ptr_i], %[frfi] \n\t"
"move %[tempMax], $zero \n\t"
"addu %[i], %[n], %[n] \n\t"
"5: \n\t"
"lh %[tmp1], 0(%[ptr_i]) \n\t"
"lh %[tmp2], 2(%[ptr_i]) \n\t"
"lh %[tmp3], 4(%[ptr_i]) \n\t"
"lh %[tmp4], 6(%[ptr_i]) \n\t"
#if defined(MIPS_DSP_R1_LE)
"absq_s.w %[tmp1], %[tmp1] \n\t"
"absq_s.w %[tmp2], %[tmp2] \n\t"
"absq_s.w %[tmp3], %[tmp3] \n\t"
"absq_s.w %[tmp4], %[tmp4] \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"slt %[tmp5], %[tmp1], $zero \n\t"
"subu %[tmp6], $zero, %[tmp1] \n\t"
"movn %[tmp1], %[tmp6], %[tmp5] \n\t"
"slt %[tmp5], %[tmp2], $zero \n\t"
"subu %[tmp6], $zero, %[tmp2] \n\t"
"movn %[tmp2], %[tmp6], %[tmp5] \n\t"
"slt %[tmp5], %[tmp3], $zero \n\t"
"subu %[tmp6], $zero, %[tmp3] \n\t"
"movn %[tmp3], %[tmp6], %[tmp5] \n\t"
"slt %[tmp5], %[tmp4], $zero \n\t"
"subu %[tmp6], $zero, %[tmp4] \n\t"
"movn %[tmp4], %[tmp6], %[tmp5] \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"slt %[tmp5], %[tempMax], %[tmp1] \n\t"
"movn %[tempMax], %[tmp1], %[tmp5] \n\t"
"addiu %[i], %[i], -4 \n\t"
"slt %[tmp5], %[tempMax], %[tmp2] \n\t"
"movn %[tempMax], %[tmp2], %[tmp5] \n\t"
"slt %[tmp5], %[tempMax], %[tmp3] \n\t"
"movn %[tempMax], %[tmp3], %[tmp5] \n\t"
"slt %[tmp5], %[tempMax], %[tmp4] \n\t"
"movn %[tempMax], %[tmp4], %[tmp5] \n\t"
"bgtz %[i], 5b \n\t"
" addiu %[ptr_i], %[ptr_i], 8 \n\t"
"addiu %[tmp1], $zero, 13573 \n\t"
"addiu %[tmp2], $zero, 27146 \n\t"
#if !defined(MIPS32_R2_LE)
"sll %[tempMax], %[tempMax], 16 \n\t"
"sra %[tempMax], %[tempMax], 16 \n\t"
#else // #if !defined(MIPS32_R2_LE)
"seh %[tempMax] \n\t"
#endif // #if !defined(MIPS32_R2_LE)
"slt %[tmp1], %[tmp1], %[tempMax] \n\t"
"slt %[tmp2], %[tmp2], %[tempMax] \n\t"
"addu %[tmp1], %[tmp1], %[tmp2] \n\t"
"addu %[shift], %[shift], %[tmp1] \n\t"
"addu %[scale], %[scale], %[tmp1] \n\t"
"sllv %[round2], %[round2], %[tmp1] \n\t"
"sll %[istep], %[l], 1 \n\t"
"move %[m], $zero \n\t"
"sll %[tmp], %[l], 2 \n\t"
"2: \n\t"
#if defined(MIPS_DSP_R1_LE)
"sllv %[tmp3], %[m], %[k] \n\t"
"addiu %[tmp2], %[tmp3], 512 \n\t"
"addiu %[m], %[m], 1 \n\t"
"lhx %[wi], %[tmp3](%[kSinTable1024]) \n\t"
"lhx %[wr], %[tmp2](%[kSinTable1024]) \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"sllv %[tmp3], %[m], %[k] \n\t"
"addu %[ptr_j], %[tmp3], %[kSinTable1024] \n\t"
"addiu %[ptr_i], %[ptr_j], 512 \n\t"
"addiu %[m], %[m], 1 \n\t"
"lh %[wi], 0(%[ptr_j]) \n\t"
"lh %[wr], 0(%[ptr_i]) \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"1: \n\t"
"sll %[tmp1], %[i], 2 \n\t"
"addu %[ptr_i], %[frfi], %[tmp1] \n\t"
"addu %[ptr_j], %[ptr_i], %[tmp] \n\t"
"lh %[tmp3], 0(%[ptr_j]) \n\t"
"lh %[tmp4], 2(%[ptr_j]) \n\t"
"lh %[tmp6], 0(%[ptr_i]) \n\t"
"lh %[tmp5], 2(%[ptr_i]) \n\t"
"addu %[i], %[i], %[istep] \n\t"
#if defined(MIPS_DSP_R2_LE)
"mult %[wr], %[tmp3] \n\t"
"msub %[wi], %[tmp4] \n\t"
"mult $ac1, %[wr], %[tmp4] \n\t"
"madd $ac1, %[wi], %[tmp3] \n\t"
"mflo %[tmp1] \n\t"
"mflo %[tmp2], $ac1 \n\t"
"sll %[tmp6], %[tmp6], 14 \n\t"
"sll %[tmp5], %[tmp5], 14 \n\t"
"shra_r.w %[tmp1], %[tmp1], 1 \n\t"
"shra_r.w %[tmp2], %[tmp2], 1 \n\t"
"addu %[tmp6], %[tmp6], %[round2] \n\t"
"addu %[tmp5], %[tmp5], %[round2] \n\t"
"subu %[tmp4], %[tmp6], %[tmp1] \n\t"
"addu %[tmp1], %[tmp6], %[tmp1] \n\t"
"addu %[tmp6], %[tmp5], %[tmp2] \n\t"
"subu %[tmp5], %[tmp5], %[tmp2] \n\t"
"srav %[tmp4], %[tmp4], %[shift] \n\t"
"srav %[tmp1], %[tmp1], %[shift] \n\t"
"srav %[tmp6], %[tmp6], %[shift] \n\t"
"srav %[tmp5], %[tmp5], %[shift] \n\t"
#else // #if defined(MIPS_DSP_R2_LE)
"mul %[tmp1], %[wr], %[tmp3] \n\t"
"mul %[tmp2], %[wr], %[tmp4] \n\t"
"mul %[tmp4], %[wi], %[tmp4] \n\t"
"mul %[tmp3], %[wi], %[tmp3] \n\t"
"sll %[tmp6], %[tmp6], 14 \n\t"
"sll %[tmp5], %[tmp5], 14 \n\t"
"sub %[tmp1], %[tmp1], %[tmp4] \n\t"
"addu %[tmp2], %[tmp2], %[tmp3] \n\t"
"addiu %[tmp1], %[tmp1], 1 \n\t"
"addiu %[tmp2], %[tmp2], 1 \n\t"
"sra %[tmp2], %[tmp2], 1 \n\t"
"sra %[tmp1], %[tmp1], 1 \n\t"
"addu %[tmp6], %[tmp6], %[round2] \n\t"
"addu %[tmp5], %[tmp5], %[round2] \n\t"
"subu %[tmp4], %[tmp6], %[tmp1] \n\t"
"addu %[tmp1], %[tmp6], %[tmp1] \n\t"
"addu %[tmp6], %[tmp5], %[tmp2] \n\t"
"subu %[tmp5], %[tmp5], %[tmp2] \n\t"
"sra %[tmp4], %[tmp4], %[shift] \n\t"
"sra %[tmp1], %[tmp1], %[shift] \n\t"
"sra %[tmp6], %[tmp6], %[shift] \n\t"
"sra %[tmp5], %[tmp5], %[shift] \n\t"
#endif // #if defined(MIPS_DSP_R2_LE)
"sh %[tmp1], 0(%[ptr_i]) \n\t"
"sh %[tmp6], 2(%[ptr_i]) \n\t"
"sh %[tmp4], 0(%[ptr_j]) \n\t"
"blt %[i], %[n], 1b \n\t"
" sh %[tmp5], 2(%[ptr_j]) \n\t"
"blt %[m], %[l], 2b \n\t"
" addu %[i], $zero, %[m] \n\t"
"move %[l], %[istep] \n\t"
"blt %[l], %[n], 3b \n\t"
" addiu %[k], %[k], -1 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3),
[tmp4] "=&r" (tmp4), [tmp5] "=&r" (tmp5), [tmp6] "=&r" (tmp6),
[ptr_i] "=&r" (ptr_i), [i] "=&r" (i), [m] "=&r" (m), [tmp] "=&r" (tmp),
[istep] "=&r" (istep), [wi] "=&r" (wi), [wr] "=&r" (wr), [l] "=&r" (l),
[k] "=&r" (k), [round2] "=&r" (round2), [ptr_j] "=&r" (ptr_j),
[shift] "=&r" (shift), [scale] "=&r" (scale), [tempMax] "=&r" (tempMax)
: [n] "r" (n), [frfi] "r" (frfi), [kSinTable1024] "r" (kSinTable1024)
: "hi", "lo", "memory"
#if defined(MIPS_DSP_R2_LE)
, "$ac1hi", "$ac1lo"
#endif // #if defined(MIPS_DSP_R2_LE)
);
return scale;
}

148
webrtc/common_audio/signal_processing/complex_fft_tables.h Normal file
View File

@ -0,0 +1,148 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_COMMON_AUDIO_SIGNAL_PROCESSING_COMPLEX_FFT_TABLES_H_
#define WEBRTC_COMMON_AUDIO_SIGNAL_PROCESSING_COMPLEX_FFT_TABLES_H_
#include "webrtc/typedefs.h"
static const int16_t kSinTable1024[] = {
0, 201, 402, 603, 804, 1005, 1206, 1406,
1607, 1808, 2009, 2209, 2410, 2610, 2811, 3011,
3211, 3411, 3611, 3811, 4011, 4210, 4409, 4608,
4807, 5006, 5205, 5403, 5601, 5799, 5997, 6195,
6392, 6589, 6786, 6982, 7179, 7375, 7571, 7766,
7961, 8156, 8351, 8545, 8739, 8932, 9126, 9319,
9511, 9703, 9895, 10087, 10278, 10469, 10659, 10849,
11038, 11227, 11416, 11604, 11792, 11980, 12166, 12353,
12539, 12724, 12909, 13094, 13278, 13462, 13645, 13827,
14009, 14191, 14372, 14552, 14732, 14911, 15090, 15268,
15446, 15623, 15799, 15975, 16150, 16325, 16499, 16672,
16845, 17017, 17189, 17360, 17530, 17699, 17868, 18036,
18204, 18371, 18537, 18702, 18867, 19031, 19194, 19357,
19519, 19680, 19840, 20000, 20159, 20317, 20474, 20631,
20787, 20942, 21096, 21249, 21402, 21554, 21705, 21855,
22004, 22153, 22301, 22448, 22594, 22739, 22883, 23027,
23169, 23311, 23452, 23592, 23731, 23869, 24006, 24143,
24278, 24413, 24546, 24679, 24811, 24942, 25072, 25201,
25329, 25456, 25582, 25707, 25831, 25954, 26077, 26198,
26318, 26437, 26556, 26673, 26789, 26905, 27019, 27132,
27244, 27355, 27466, 27575, 27683, 27790, 27896, 28001,
28105, 28208, 28309, 28410, 28510, 28608, 28706, 28802,
28897, 28992, 29085, 29177, 29268, 29358, 29446, 29534,
29621, 29706, 29790, 29873, 29955, 30036, 30116, 30195,
30272, 30349, 30424, 30498, 30571, 30643, 30713, 30783,
30851, 30918, 30984, 31049, 31113, 31175, 31236, 31297,
31356, 31413, 31470, 31525, 31580, 31633, 31684, 31735,
31785, 31833, 31880, 31926, 31970, 32014, 32056, 32097,
32137, 32176, 32213, 32249, 32284, 32318, 32350, 32382,
32412, 32441, 32468, 32495, 32520, 32544, 32567, 32588,
32609, 32628, 32646, 32662, 32678, 32692, 32705, 32717,
32727, 32736, 32744, 32751, 32757, 32761, 32764, 32766,
32767, 32766, 32764, 32761, 32757, 32751, 32744, 32736,
32727, 32717, 32705, 32692, 32678, 32662, 32646, 32628,
32609, 32588, 32567, 32544, 32520, 32495, 32468, 32441,
32412, 32382, 32350, 32318, 32284, 32249, 32213, 32176,
32137, 32097, 32056, 32014, 31970, 31926, 31880, 31833,
31785, 31735, 31684, 31633, 31580, 31525, 31470, 31413,
31356, 31297, 31236, 31175, 31113, 31049, 30984, 30918,
30851, 30783, 30713, 30643, 30571, 30498, 30424, 30349,
30272, 30195, 30116, 30036, 29955, 29873, 29790, 29706,
29621, 29534, 29446, 29358, 29268, 29177, 29085, 28992,
28897, 28802, 28706, 28608, 28510, 28410, 28309, 28208,
28105, 28001, 27896, 27790, 27683, 27575, 27466, 27355,
27244, 27132, 27019, 26905, 26789, 26673, 26556, 26437,
26318, 26198, 26077, 25954, 25831, 25707, 25582, 25456,
25329, 25201, 25072, 24942, 24811, 24679, 24546, 24413,
24278, 24143, 24006, 23869, 23731, 23592, 23452, 23311,
23169, 23027, 22883, 22739, 22594, 22448, 22301, 22153,
22004, 21855, 21705, 21554, 21402, 21249, 21096, 20942,
20787, 20631, 20474, 20317, 20159, 20000, 19840, 19680,
19519, 19357, 19194, 19031, 18867, 18702, 18537, 18371,
18204, 18036, 17868, 17699, 17530, 17360, 17189, 17017,
16845, 16672, 16499, 16325, 16150, 15975, 15799, 15623,
15446, 15268, 15090, 14911, 14732, 14552, 14372, 14191,
14009, 13827, 13645, 13462, 13278, 13094, 12909, 12724,
12539, 12353, 12166, 11980, 11792, 11604, 11416, 11227,
11038, 10849, 10659, 10469, 10278, 10087, 9895, 9703,
9511, 9319, 9126, 8932, 8739, 8545, 8351, 8156,
7961, 7766, 7571, 7375, 7179, 6982, 6786, 6589,
6392, 6195, 5997, 5799, 5601, 5403, 5205, 5006,
4807, 4608, 4409, 4210, 4011, 3811, 3611, 3411,
3211, 3011, 2811, 2610, 2410, 2209, 2009, 1808,
1607, 1406, 1206, 1005, 804, 603, 402, 201,
0, -201, -402, -603, -804, -1005, -1206, -1406,
-1607, -1808, -2009, -2209, -2410, -2610, -2811, -3011,
-3211, -3411, -3611, -3811, -4011, -4210, -4409, -4608,
-4807, -5006, -5205, -5403, -5601, -5799, -5997, -6195,
-6392, -6589, -6786, -6982, -7179, -7375, -7571, -7766,
-7961, -8156, -8351, -8545, -8739, -8932, -9126, -9319,
-9511, -9703, -9895, -10087, -10278, -10469, -10659, -10849,
-11038, -11227, -11416, -11604, -11792, -11980, -12166, -12353,
-12539, -12724, -12909, -13094, -13278, -13462, -13645, -13827,
-14009, -14191, -14372, -14552, -14732, -14911, -15090, -15268,
-15446, -15623, -15799, -15975, -16150, -16325, -16499, -16672,
-16845, -17017, -17189, -17360, -17530, -17699, -17868, -18036,
-18204, -18371, -18537, -18702, -18867, -19031, -19194, -19357,
-19519, -19680, -19840, -20000, -20159, -20317, -20474, -20631,
-20787, -20942, -21096, -21249, -21402, -21554, -21705, -21855,
-22004, -22153, -22301, -22448, -22594, -22739, -22883, -23027,
-23169, -23311, -23452, -23592, -23731, -23869, -24006, -24143,
-24278, -24413, -24546, -24679, -24811, -24942, -25072, -25201,
-25329, -25456, -25582, -25707, -25831, -25954, -26077, -26198,
-26318, -26437, -26556, -26673, -26789, -26905, -27019, -27132,
-27244, -27355, -27466, -27575, -27683, -27790, -27896, -28001,
-28105, -28208, -28309, -28410, -28510, -28608, -28706, -28802,
-28897, -28992, -29085, -29177, -29268, -29358, -29446, -29534,
-29621, -29706, -29790, -29873, -29955, -30036, -30116, -30195,
-30272, -30349, -30424, -30498, -30571, -30643, -30713, -30783,
-30851, -30918, -30984, -31049, -31113, -31175, -31236, -31297,
-31356, -31413, -31470, -31525, -31580, -31633, -31684, -31735,
-31785, -31833, -31880, -31926, -31970, -32014, -32056, -32097,
-32137, -32176, -32213, -32249, -32284, -32318, -32350, -32382,
-32412, -32441, -32468, -32495, -32520, -32544, -32567, -32588,
-32609, -32628, -32646, -32662, -32678, -32692, -32705, -32717,
-32727, -32736, -32744, -32751, -32757, -32761, -32764, -32766,
-32767, -32766, -32764, -32761, -32757, -32751, -32744, -32736,
-32727, -32717, -32705, -32692, -32678, -32662, -32646, -32628,
-32609, -32588, -32567, -32544, -32520, -32495, -32468, -32441,
-32412, -32382, -32350, -32318, -32284, -32249, -32213, -32176,
-32137, -32097, -32056, -32014, -31970, -31926, -31880, -31833,
-31785, -31735, -31684, -31633, -31580, -31525, -31470, -31413,
-31356, -31297, -31236, -31175, -31113, -31049, -30984, -30918,
-30851, -30783, -30713, -30643, -30571, -30498, -30424, -30349,
-30272, -30195, -30116, -30036, -29955, -29873, -29790, -29706,
-29621, -29534, -29446, -29358, -29268, -29177, -29085, -28992,
-28897, -28802, -28706, -28608, -28510, -28410, -28309, -28208,
-28105, -28001, -27896, -27790, -27683, -27575, -27466, -27355,
-27244, -27132, -27019, -26905, -26789, -26673, -26556, -26437,
-26318, -26198, -26077, -25954, -25831, -25707, -25582, -25456,
-25329, -25201, -25072, -24942, -24811, -24679, -24546, -24413,
-24278, -24143, -24006, -23869, -23731, -23592, -23452, -23311,
-23169, -23027, -22883, -22739, -22594, -22448, -22301, -22153,
-22004, -21855, -21705, -21554, -21402, -21249, -21096, -20942,
-20787, -20631, -20474, -20317, -20159, -20000, -19840, -19680,
-19519, -19357, -19194, -19031, -18867, -18702, -18537, -18371,
-18204, -18036, -17868, -17699, -17530, -17360, -17189, -17017,
-16845, -16672, -16499, -16325, -16150, -15975, -15799, -15623,
-15446, -15268, -15090, -14911, -14732, -14552, -14372, -14191,
-14009, -13827, -13645, -13462, -13278, -13094, -12909, -12724,
-12539, -12353, -12166, -11980, -11792, -11604, -11416, -11227,
-11038, -10849, -10659, -10469, -10278, -10087, -9895, -9703,
-9511, -9319, -9126, -8932, -8739, -8545, -8351, -8156,
-7961, -7766, -7571, -7375, -7179, -6982, -6786, -6589,
-6392, -6195, -5997, -5799, -5601, -5403, -5205, -5006,
-4807, -4608, -4409, -4210, -4011, -3811, -3611, -3411,
-3211, -3011, -2811, -2610, -2410, -2209, -2009, -1808,
-1607, -1406, -1206, -1005, -804, -603, -402, -201
};
#endif // WEBRTC_COMMON_AUDIO_SIGNAL_PROCESSING_COMPLEX_FFT_TABLES_H_

82
webrtc/common_audio/signal_processing/copy_set_operations.c Normal file
View File

@ -0,0 +1,82 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the implementation of functions
* WebRtcSpl_MemSetW16()
* WebRtcSpl_MemSetW32()
* WebRtcSpl_MemCpyReversedOrder()
* WebRtcSpl_CopyFromEndW16()
* WebRtcSpl_ZerosArrayW16()
* WebRtcSpl_ZerosArrayW32()
*
* The description header can be found in signal_processing_library.h
*
*/
#include <string.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_MemSetW16(int16_t *ptr, int16_t set_value, size_t length)
{
size_t j;
int16_t *arrptr = ptr;
for (j = length; j > 0; j--)
{
*arrptr++ = set_value;
}
}
void WebRtcSpl_MemSetW32(int32_t *ptr, int32_t set_value, size_t length)
{
size_t j;
int32_t *arrptr = ptr;
for (j = length; j > 0; j--)
{
*arrptr++ = set_value;
}
}
void WebRtcSpl_MemCpyReversedOrder(int16_t* dest,
int16_t* source,
size_t length)
{
size_t j;
int16_t* destPtr = dest;
int16_t* sourcePtr = source;
for (j = 0; j < length; j++)
{
*destPtr-- = *sourcePtr++;
}
}
void WebRtcSpl_CopyFromEndW16(const int16_t *vector_in,
size_t length,
size_t samples,
int16_t *vector_out)
{
// Copy the last <samples> of the input vector to vector_out
WEBRTC_SPL_MEMCPY_W16(vector_out, &vector_in[length - samples], samples);
}
void WebRtcSpl_ZerosArrayW16(int16_t *vector, size_t length)
{
WebRtcSpl_MemSetW16(vector, 0, length);
}
void WebRtcSpl_ZerosArrayW32(int32_t *vector, size_t length)
{
WebRtcSpl_MemSetW32(vector, 0, length);
}

30
webrtc/common_audio/signal_processing/cross_correlation.c Normal file
View File

@ -0,0 +1,30 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
/* C version of WebRtcSpl_CrossCorrelation() for generic platforms. */
void WebRtcSpl_CrossCorrelationC(int32_t* cross_correlation,
const int16_t* seq1,
const int16_t* seq2,
size_t dim_seq,
size_t dim_cross_correlation,
int right_shifts,
int step_seq2) {
size_t i = 0, j = 0;
for (i = 0; i < dim_cross_correlation; i++) {
int32_t corr = 0;
for (j = 0; j < dim_seq; j++)
corr += (seq1[j] * seq2[j]) >> right_shifts;
seq2 += step_seq2;
*cross_correlation++ = corr;
}
}

104
webrtc/common_audio/signal_processing/cross_correlation_mips.c Normal file
View File

@ -0,0 +1,104 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_CrossCorrelation_mips(int32_t* cross_correlation,
const int16_t* seq1,
const int16_t* seq2,
size_t dim_seq,
size_t dim_cross_correlation,
int right_shifts,
int step_seq2) {
int32_t t0 = 0, t1 = 0, t2 = 0, t3 = 0, sum = 0;
int16_t *pseq2 = NULL;
int16_t *pseq1 = NULL;
int16_t *pseq1_0 = (int16_t*)&seq1[0];
int16_t *pseq2_0 = (int16_t*)&seq2[0];
int k = 0;
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"sll %[step_seq2], %[step_seq2], 1 \n\t"
"andi %[t0], %[dim_seq], 1 \n\t"
"bgtz %[t0], 3f \n\t"
" nop \n\t"
"1: \n\t"
"move %[pseq1], %[pseq1_0] \n\t"
"move %[pseq2], %[pseq2_0] \n\t"
"sra %[k], %[dim_seq], 1 \n\t"
"addiu %[dim_cc], %[dim_cc], -1 \n\t"
"xor %[sum], %[sum], %[sum] \n\t"
"2: \n\t"
"lh %[t0], 0(%[pseq1]) \n\t"
"lh %[t1], 0(%[pseq2]) \n\t"
"lh %[t2], 2(%[pseq1]) \n\t"
"lh %[t3], 2(%[pseq2]) \n\t"
"mul %[t0], %[t0], %[t1] \n\t"
"addiu %[k], %[k], -1 \n\t"
"mul %[t2], %[t2], %[t3] \n\t"
"addiu %[pseq1], %[pseq1], 4 \n\t"
"addiu %[pseq2], %[pseq2], 4 \n\t"
"srav %[t0], %[t0], %[right_shifts] \n\t"
"addu %[sum], %[sum], %[t0] \n\t"
"srav %[t2], %[t2], %[right_shifts] \n\t"
"bgtz %[k], 2b \n\t"
" addu %[sum], %[sum], %[t2] \n\t"
"addu %[pseq2_0], %[pseq2_0], %[step_seq2] \n\t"
"sw %[sum], 0(%[cc]) \n\t"
"bgtz %[dim_cc], 1b \n\t"
" addiu %[cc], %[cc], 4 \n\t"
"b 6f \n\t"
" nop \n\t"
"3: \n\t"
"move %[pseq1], %[pseq1_0] \n\t"
"move %[pseq2], %[pseq2_0] \n\t"
"sra %[k], %[dim_seq], 1 \n\t"
"addiu %[dim_cc], %[dim_cc], -1 \n\t"
"beqz %[k], 5f \n\t"
" xor %[sum], %[sum], %[sum] \n\t"
"4: \n\t"
"lh %[t0], 0(%[pseq1]) \n\t"
"lh %[t1], 0(%[pseq2]) \n\t"
"lh %[t2], 2(%[pseq1]) \n\t"
"lh %[t3], 2(%[pseq2]) \n\t"
"mul %[t0], %[t0], %[t1] \n\t"
"addiu %[k], %[k], -1 \n\t"
"mul %[t2], %[t2], %[t3] \n\t"
"addiu %[pseq1], %[pseq1], 4 \n\t"
"addiu %[pseq2], %[pseq2], 4 \n\t"
"srav %[t0], %[t0], %[right_shifts] \n\t"
"addu %[sum], %[sum], %[t0] \n\t"
"srav %[t2], %[t2], %[right_shifts] \n\t"
"bgtz %[k], 4b \n\t"
" addu %[sum], %[sum], %[t2] \n\t"
"5: \n\t"
"lh %[t0], 0(%[pseq1]) \n\t"
"lh %[t1], 0(%[pseq2]) \n\t"
"mul %[t0], %[t0], %[t1] \n\t"
"srav %[t0], %[t0], %[right_shifts] \n\t"
"addu %[sum], %[sum], %[t0] \n\t"
"addu %[pseq2_0], %[pseq2_0], %[step_seq2] \n\t"
"sw %[sum], 0(%[cc]) \n\t"
"bgtz %[dim_cc], 3b \n\t"
" addiu %[cc], %[cc], 4 \n\t"
"6: \n\t"
".set pop \n\t"
: [step_seq2] "+r" (step_seq2), [t0] "=&r" (t0), [t1] "=&r" (t1),
[t2] "=&r" (t2), [t3] "=&r" (t3), [pseq1] "=&r" (pseq1),
[pseq2] "=&r" (pseq2), [pseq1_0] "+r" (pseq1_0), [pseq2_0] "+r" (pseq2_0),
[k] "=&r" (k), [dim_cc] "+r" (dim_cross_correlation), [sum] "=&r" (sum),
[cc] "+r" (cross_correlation)
: [dim_seq] "r" (dim_seq), [right_shifts] "r" (right_shifts)
: "hi", "lo", "memory"
);
}

87
webrtc/common_audio/signal_processing/cross_correlation_neon.c Normal file
View File

@ -0,0 +1,87 @@
/*
* Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include <arm_neon.h>
static inline void DotProductWithScaleNeon(int32_t* cross_correlation,
const int16_t* vector1,
const int16_t* vector2,
size_t length,
int scaling) {
size_t i = 0;
size_t len1 = length >> 3;
size_t len2 = length & 7;
int64x2_t sum0 = vdupq_n_s64(0);
int64x2_t sum1 = vdupq_n_s64(0);
for (i = len1; i > 0; i -= 1) {
int16x8_t seq1_16x8 = vld1q_s16(vector1);
int16x8_t seq2_16x8 = vld1q_s16(vector2);
#if defined(WEBRTC_ARCH_ARM64)
int32x4_t tmp0 = vmull_s16(vget_low_s16(seq1_16x8),
vget_low_s16(seq2_16x8));
int32x4_t tmp1 = vmull_high_s16(seq1_16x8, seq2_16x8);
#else
int32x4_t tmp0 = vmull_s16(vget_low_s16(seq1_16x8),
vget_low_s16(seq2_16x8));
int32x4_t tmp1 = vmull_s16(vget_high_s16(seq1_16x8),
vget_high_s16(seq2_16x8));
#endif
sum0 = vpadalq_s32(sum0, tmp0);
sum1 = vpadalq_s32(sum1, tmp1);
vector1 += 8;
vector2 += 8;
}
// Calculate the rest of the samples.
int64_t sum_res = 0;
for (i = len2; i > 0; i -= 1) {
sum_res += WEBRTC_SPL_MUL_16_16(*vector1, *vector2);
vector1++;
vector2++;
}
sum0 = vaddq_s64(sum0, sum1);
#if defined(WEBRTC_ARCH_ARM64)
int64_t sum2 = vaddvq_s64(sum0);
*cross_correlation = (int32_t)((sum2 + sum_res) >> scaling);
#else
int64x1_t shift = vdup_n_s64(-scaling);
int64x1_t sum2 = vadd_s64(vget_low_s64(sum0), vget_high_s64(sum0));
sum2 = vadd_s64(sum2, vdup_n_s64(sum_res));
sum2 = vshl_s64(sum2, shift);
vst1_lane_s32(cross_correlation, vreinterpret_s32_s64(sum2), 0);
#endif
}
/* NEON version of WebRtcSpl_CrossCorrelation() for ARM32/64 platforms. */
void WebRtcSpl_CrossCorrelationNeon(int32_t* cross_correlation,
const int16_t* seq1,
const int16_t* seq2,
size_t dim_seq,
size_t dim_cross_correlation,
int right_shifts,
int step_seq2) {
size_t i = 0;
for (i = 0; i < dim_cross_correlation; i++) {
const int16_t* seq1_ptr = seq1;
const int16_t* seq2_ptr = seq2 + (step_seq2 * i);
DotProductWithScaleNeon(cross_correlation,
seq1_ptr,
seq2_ptr,
dim_seq,
right_shifts);
cross_correlation++;
}
}

138
webrtc/common_audio/signal_processing/division_operations.c Normal file
View File

@ -0,0 +1,138 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains implementations of the divisions
* WebRtcSpl_DivU32U16()
* WebRtcSpl_DivW32W16()
* WebRtcSpl_DivW32W16ResW16()
* WebRtcSpl_DivResultInQ31()
* WebRtcSpl_DivW32HiLow()
*
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
uint32_t WebRtcSpl_DivU32U16(uint32_t num, uint16_t den)
{
// Guard against division with 0
if (den != 0)
{
return (uint32_t)(num / den);
} else
{
return (uint32_t)0xFFFFFFFF;
}
}
int32_t WebRtcSpl_DivW32W16(int32_t num, int16_t den)
{
// Guard against division with 0
if (den != 0)
{
return (int32_t)(num / den);
} else
{
return (int32_t)0x7FFFFFFF;
}
}
int16_t WebRtcSpl_DivW32W16ResW16(int32_t num, int16_t den)
{
// Guard against division with 0
if (den != 0)
{
return (int16_t)(num / den);
} else
{
return (int16_t)0x7FFF;
}
}
int32_t WebRtcSpl_DivResultInQ31(int32_t num, int32_t den)
{
int32_t L_num = num;
int32_t L_den = den;
int32_t div = 0;
int k = 31;
int change_sign = 0;
if (num == 0)
return 0;
if (num < 0)
{
change_sign++;
L_num = -num;
}
if (den < 0)
{
change_sign++;
L_den = -den;
}
while (k--)
{
div <<= 1;
L_num <<= 1;
if (L_num >= L_den)
{
L_num -= L_den;
div++;
}
}
if (change_sign == 1)
{
div = -div;
}
return div;
}
int32_t WebRtcSpl_DivW32HiLow(int32_t num, int16_t den_hi, int16_t den_low)
{
int16_t approx, tmp_hi, tmp_low, num_hi, num_low;
int32_t tmpW32;
approx = (int16_t)WebRtcSpl_DivW32W16((int32_t)0x1FFFFFFF, den_hi);
// result in Q14 (Note: 3FFFFFFF = 0.5 in Q30)
// tmpW32 = 1/den = approx * (2.0 - den * approx) (in Q30)
tmpW32 = (den_hi * approx << 1) + ((den_low * approx >> 15) << 1);
// tmpW32 = den * approx
tmpW32 = (int32_t)0x7fffffffL - tmpW32; // result in Q30 (tmpW32 = 2.0-(den*approx))
// Store tmpW32 in hi and low format
tmp_hi = (int16_t)(tmpW32 >> 16);
tmp_low = (int16_t)((tmpW32 - ((int32_t)tmp_hi << 16)) >> 1);
// tmpW32 = 1/den in Q29
tmpW32 = (tmp_hi * approx + (tmp_low * approx >> 15)) << 1;
// 1/den in hi and low format
tmp_hi = (int16_t)(tmpW32 >> 16);
tmp_low = (int16_t)((tmpW32 - ((int32_t)tmp_hi << 16)) >> 1);
// Store num in hi and low format
num_hi = (int16_t)(num >> 16);
num_low = (int16_t)((num - ((int32_t)num_hi << 16)) >> 1);
// num * (1/den) by 32 bit multiplication (result in Q28)
tmpW32 = num_hi * tmp_hi + (num_hi * tmp_low >> 15) +
(num_low * tmp_hi >> 15);
// Put result in Q31 (convert from Q28)
tmpW32 = WEBRTC_SPL_LSHIFT_W32(tmpW32, 3);
return tmpW32;
}

32
webrtc/common_audio/signal_processing/dot_product_with_scale.c Normal file
View File

@ -0,0 +1,32 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
int32_t WebRtcSpl_DotProductWithScale(const int16_t* vector1,
const int16_t* vector2,
size_t length,
int scaling) {
int32_t sum = 0;
size_t i = 0;
/* Unroll the loop to improve performance. */
for (i = 0; i + 3 < length; i += 4) {
sum += (vector1[i + 0] * vector2[i + 0]) >> scaling;
sum += (vector1[i + 1] * vector2[i + 1]) >> scaling;
sum += (vector1[i + 2] * vector2[i + 2]) >> scaling;
sum += (vector1[i + 3] * vector2[i + 3]) >> scaling;
}
for (; i < length; i++) {
sum += (vector1[i] * vector2[i]) >> scaling;
}
return sum;
}

48
webrtc/common_audio/signal_processing/downsample_fast.c Normal file
View File

@ -0,0 +1,48 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// TODO(Bjornv): Change the function parameter order to WebRTC code style.
// C version of WebRtcSpl_DownsampleFast() for generic platforms.
int WebRtcSpl_DownsampleFastC(const int16_t* data_in,
size_t data_in_length,
int16_t* data_out,
size_t data_out_length,
const int16_t* __restrict coefficients,
size_t coefficients_length,
int factor,
size_t delay) {
size_t i = 0;
size_t j = 0;
int32_t out_s32 = 0;
size_t endpos = delay + factor * (data_out_length - 1) + 1;
// Return error if any of the running conditions doesn't meet.
if (data_out_length == 0 || coefficients_length == 0
|| data_in_length < endpos) {
return -1;
}
for (i = delay; i < endpos; i += factor) {
out_s32 = 2048; // Round value, 0.5 in Q12.
for (j = 0; j < coefficients_length; j++) {
out_s32 += coefficients[j] * data_in[i - j]; // Q12.
}
out_s32 >>= 12; // Q0.
// Saturate and store the output.
*data_out++ = WebRtcSpl_SatW32ToW16(out_s32);
}
return 0;
}

169
webrtc/common_audio/signal_processing/downsample_fast_mips.c Normal file
View File

@ -0,0 +1,169 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// Version of WebRtcSpl_DownsampleFast() for MIPS platforms.
int WebRtcSpl_DownsampleFast_mips(const int16_t* data_in,
size_t data_in_length,
int16_t* data_out,
size_t data_out_length,
const int16_t* __restrict coefficients,
size_t coefficients_length,
int factor,
size_t delay) {
int i;
int j;
int k;
int32_t out_s32 = 0;
size_t endpos = delay + factor * (data_out_length - 1) + 1;
int32_t tmp1, tmp2, tmp3, tmp4, factor_2;
int16_t* p_coefficients;
int16_t* p_data_in;
int16_t* p_data_in_0 = (int16_t*)&data_in[delay];
int16_t* p_coefficients_0 = (int16_t*)&coefficients[0];
#if !defined(MIPS_DSP_R1_LE)
int32_t max_16 = 0x7FFF;
int32_t min_16 = 0xFFFF8000;
#endif // #if !defined(MIPS_DSP_R1_LE)
// Return error if any of the running conditions doesn't meet.
if (data_out_length == 0 || coefficients_length == 0
|| data_in_length < endpos) {
return -1;
}
#if defined(MIPS_DSP_R2_LE)
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"subu %[i], %[endpos], %[delay] \n\t"
"sll %[factor_2], %[factor], 1 \n\t"
"1: \n\t"
"move %[p_data_in], %[p_data_in_0] \n\t"
"mult $zero, $zero \n\t"
"move %[p_coefs], %[p_coefs_0] \n\t"
"sra %[j], %[coef_length], 2 \n\t"
"beq %[j], $zero, 3f \n\t"
" andi %[k], %[coef_length], 3 \n\t"
"2: \n\t"
"lwl %[tmp1], 1(%[p_data_in]) \n\t"
"lwl %[tmp2], 3(%[p_coefs]) \n\t"
"lwl %[tmp3], -3(%[p_data_in]) \n\t"
"lwl %[tmp4], 7(%[p_coefs]) \n\t"
"lwr %[tmp1], -2(%[p_data_in]) \n\t"
"lwr %[tmp2], 0(%[p_coefs]) \n\t"
"lwr %[tmp3], -6(%[p_data_in]) \n\t"
"lwr %[tmp4], 4(%[p_coefs]) \n\t"
"packrl.ph %[tmp1], %[tmp1], %[tmp1] \n\t"
"packrl.ph %[tmp3], %[tmp3], %[tmp3] \n\t"
"dpa.w.ph $ac0, %[tmp1], %[tmp2] \n\t"
"dpa.w.ph $ac0, %[tmp3], %[tmp4] \n\t"
"addiu %[j], %[j], -1 \n\t"
"addiu %[p_data_in], %[p_data_in], -8 \n\t"
"bgtz %[j], 2b \n\t"
" addiu %[p_coefs], %[p_coefs], 8 \n\t"
"3: \n\t"
"beq %[k], $zero, 5f \n\t"
" nop \n\t"
"4: \n\t"
"lhu %[tmp1], 0(%[p_data_in]) \n\t"
"lhu %[tmp2], 0(%[p_coefs]) \n\t"
"addiu %[p_data_in], %[p_data_in], -2 \n\t"
"addiu %[k], %[k], -1 \n\t"
"dpa.w.ph $ac0, %[tmp1], %[tmp2] \n\t"
"bgtz %[k], 4b \n\t"
" addiu %[p_coefs], %[p_coefs], 2 \n\t"
"5: \n\t"
"extr_r.w %[out_s32], $ac0, 12 \n\t"
"addu %[p_data_in_0], %[p_data_in_0], %[factor_2] \n\t"
"subu %[i], %[i], %[factor] \n\t"
"shll_s.w %[out_s32], %[out_s32], 16 \n\t"
"sra %[out_s32], %[out_s32], 16 \n\t"
"sh %[out_s32], 0(%[data_out]) \n\t"
"bgtz %[i], 1b \n\t"
" addiu %[data_out], %[data_out], 2 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3),
[tmp4] "=&r" (tmp4), [p_data_in] "=&r" (p_data_in),
[p_data_in_0] "+r" (p_data_in_0), [p_coefs] "=&r" (p_coefficients),
[j] "=&r" (j), [out_s32] "=&r" (out_s32), [factor_2] "=&r" (factor_2),
[i] "=&r" (i), [k] "=&r" (k)
: [coef_length] "r" (coefficients_length), [data_out] "r" (data_out),
[p_coefs_0] "r" (p_coefficients_0), [endpos] "r" (endpos),
[delay] "r" (delay), [factor] "r" (factor)
: "memory", "hi", "lo"
);
#else // #if defined(MIPS_DSP_R2_LE)
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"sll %[factor_2], %[factor], 1 \n\t"
"subu %[i], %[endpos], %[delay] \n\t"
"1: \n\t"
"move %[p_data_in], %[p_data_in_0] \n\t"
"addiu %[out_s32], $zero, 2048 \n\t"
"move %[p_coefs], %[p_coefs_0] \n\t"
"sra %[j], %[coef_length], 1 \n\t"
"beq %[j], $zero, 3f \n\t"
" andi %[k], %[coef_length], 1 \n\t"
"2: \n\t"
"lh %[tmp1], 0(%[p_data_in]) \n\t"
"lh %[tmp2], 0(%[p_coefs]) \n\t"
"lh %[tmp3], -2(%[p_data_in]) \n\t"
"lh %[tmp4], 2(%[p_coefs]) \n\t"
"mul %[tmp1], %[tmp1], %[tmp2] \n\t"
"addiu %[p_coefs], %[p_coefs], 4 \n\t"
"mul %[tmp3], %[tmp3], %[tmp4] \n\t"
"addiu %[j], %[j], -1 \n\t"
"addiu %[p_data_in], %[p_data_in], -4 \n\t"
"addu %[tmp1], %[tmp1], %[tmp3] \n\t"
"bgtz %[j], 2b \n\t"
" addu %[out_s32], %[out_s32], %[tmp1] \n\t"
"3: \n\t"
"beq %[k], $zero, 4f \n\t"
" nop \n\t"
"lh %[tmp1], 0(%[p_data_in]) \n\t"
"lh %[tmp2], 0(%[p_coefs]) \n\t"
"mul %[tmp1], %[tmp1], %[tmp2] \n\t"
"addu %[out_s32], %[out_s32], %[tmp1] \n\t"
"4: \n\t"
"sra %[out_s32], %[out_s32], 12 \n\t"
"addu %[p_data_in_0], %[p_data_in_0], %[factor_2] \n\t"
#if defined(MIPS_DSP_R1_LE)
"shll_s.w %[out_s32], %[out_s32], 16 \n\t"
"sra %[out_s32], %[out_s32], 16 \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"slt %[tmp1], %[max_16], %[out_s32] \n\t"
"movn %[out_s32], %[max_16], %[tmp1] \n\t"
"slt %[tmp1], %[out_s32], %[min_16] \n\t"
"movn %[out_s32], %[min_16], %[tmp1] \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"subu %[i], %[i], %[factor] \n\t"
"sh %[out_s32], 0(%[data_out]) \n\t"
"bgtz %[i], 1b \n\t"
" addiu %[data_out], %[data_out], 2 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3),
[tmp4] "=&r" (tmp4), [p_data_in] "=&r" (p_data_in), [k] "=&r" (k),
[p_data_in_0] "+r" (p_data_in_0), [p_coefs] "=&r" (p_coefficients),
[j] "=&r" (j), [out_s32] "=&r" (out_s32), [factor_2] "=&r" (factor_2),
[i] "=&r" (i)
: [coef_length] "r" (coefficients_length), [data_out] "r" (data_out),
[p_coefs_0] "r" (p_coefficients_0), [endpos] "r" (endpos),
#if !defined(MIPS_DSP_R1_LE)
[max_16] "r" (max_16), [min_16] "r" (min_16),
#endif // #if !defined(MIPS_DSP_R1_LE)
[delay] "r" (delay), [factor] "r" (factor)
: "memory", "hi", "lo"
);
#endif // #if defined(MIPS_DSP_R2_LE)
return 0;
}

217
webrtc/common_audio/signal_processing/downsample_fast_neon.c Normal file
View File

@ -0,0 +1,217 @@
/*
* Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include <arm_neon.h>
// NEON intrinsics version of WebRtcSpl_DownsampleFast()
// for ARM 32-bit/64-bit platforms.
int WebRtcSpl_DownsampleFastNeon(const int16_t* data_in,
size_t data_in_length,
int16_t* data_out,
size_t data_out_length,
const int16_t* __restrict coefficients,
size_t coefficients_length,
int factor,
size_t delay) {
size_t i = 0;
size_t j = 0;
int32_t out_s32 = 0;
size_t endpos = delay + factor * (data_out_length - 1) + 1;
size_t res = data_out_length & 0x7;
size_t endpos1 = endpos - factor * res;
// Return error if any of the running conditions doesn't meet.
if (data_out_length == 0 || coefficients_length == 0
|| data_in_length < endpos) {
return -1;
}
// First part, unroll the loop 8 times, with 3 subcases
// (factor == 2, 4, others).
switch (factor) {
case 2: {
for (i = delay; i < endpos1; i += 16) {
// Round value, 0.5 in Q12.
int32x4_t out32x4_0 = vdupq_n_s32(2048);
int32x4_t out32x4_1 = vdupq_n_s32(2048);
#if defined(WEBRTC_ARCH_ARM64)
// Unroll the loop 2 times.
for (j = 0; j < coefficients_length - 1; j += 2) {
int32x2_t coeff32 = vld1_dup_s32((int32_t*)&coefficients[j]);
int16x4_t coeff16x4 = vreinterpret_s16_s32(coeff32);
int16x8x2_t in16x8x2 = vld2q_s16(&data_in[i - j - 1]);
// Mul and accumulate low 64-bit data.
int16x4_t in16x4_0 = vget_low_s16(in16x8x2.val[0]);
int16x4_t in16x4_1 = vget_low_s16(in16x8x2.val[1]);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_0, coeff16x4, 1);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_1, coeff16x4, 0);
// Mul and accumulate high 64-bit data.
// TODO: vget_high_s16 need extra cost on ARM64. This could be
// replaced by vmlal_high_lane_s16. But for the interface of
// vmlal_high_lane_s16, there is a bug in gcc 4.9.
// This issue need to be tracked in the future.
int16x4_t in16x4_2 = vget_high_s16(in16x8x2.val[0]);
int16x4_t in16x4_3 = vget_high_s16(in16x8x2.val[1]);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_2, coeff16x4, 1);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_3, coeff16x4, 0);
}
for (; j < coefficients_length; j++) {
int16x4_t coeff16x4 = vld1_dup_s16(&coefficients[j]);
int16x8x2_t in16x8x2 = vld2q_s16(&data_in[i - j]);
// Mul and accumulate low 64-bit data.
int16x4_t in16x4_0 = vget_low_s16(in16x8x2.val[0]);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_0, coeff16x4, 0);
// Mul and accumulate high 64-bit data.
// TODO: vget_high_s16 need extra cost on ARM64. This could be
// replaced by vmlal_high_lane_s16. But for the interface of
// vmlal_high_lane_s16, there is a bug in gcc 4.9.
// This issue need to be tracked in the future.
int16x4_t in16x4_1 = vget_high_s16(in16x8x2.val[0]);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_1, coeff16x4, 0);
}
#else
// On ARMv7, the loop unrolling 2 times results in performance
// regression.
for (j = 0; j < coefficients_length; j++) {
int16x4_t coeff16x4 = vld1_dup_s16(&coefficients[j]);
int16x8x2_t in16x8x2 = vld2q_s16(&data_in[i - j]);
// Mul and accumulate.
int16x4_t in16x4_0 = vget_low_s16(in16x8x2.val[0]);
int16x4_t in16x4_1 = vget_high_s16(in16x8x2.val[0]);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_0, coeff16x4, 0);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_1, coeff16x4, 0);
}
#endif
// Saturate and store the output.
int16x4_t out16x4_0 = vqshrn_n_s32(out32x4_0, 12);
int16x4_t out16x4_1 = vqshrn_n_s32(out32x4_1, 12);
vst1q_s16(data_out, vcombine_s16(out16x4_0, out16x4_1));
data_out += 8;
}
break;
}
case 4: {
for (i = delay; i < endpos1; i += 32) {
// Round value, 0.5 in Q12.
int32x4_t out32x4_0 = vdupq_n_s32(2048);
int32x4_t out32x4_1 = vdupq_n_s32(2048);
// Unroll the loop 4 times.
for (j = 0; j < coefficients_length - 3; j += 4) {
int16x4_t coeff16x4 = vld1_s16(&coefficients[j]);
int16x8x4_t in16x8x4 = vld4q_s16(&data_in[i - j - 3]);
// Mul and accumulate low 64-bit data.
int16x4_t in16x4_0 = vget_low_s16(in16x8x4.val[0]);
int16x4_t in16x4_2 = vget_low_s16(in16x8x4.val[1]);
int16x4_t in16x4_4 = vget_low_s16(in16x8x4.val[2]);
int16x4_t in16x4_6 = vget_low_s16(in16x8x4.val[3]);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_0, coeff16x4, 3);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_2, coeff16x4, 2);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_4, coeff16x4, 1);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_6, coeff16x4, 0);
// Mul and accumulate high 64-bit data.
// TODO: vget_high_s16 need extra cost on ARM64. This could be
// replaced by vmlal_high_lane_s16. But for the interface of
// vmlal_high_lane_s16, there is a bug in gcc 4.9.
// This issue need to be tracked in the future.
int16x4_t in16x4_1 = vget_high_s16(in16x8x4.val[0]);
int16x4_t in16x4_3 = vget_high_s16(in16x8x4.val[1]);
int16x4_t in16x4_5 = vget_high_s16(in16x8x4.val[2]);
int16x4_t in16x4_7 = vget_high_s16(in16x8x4.val[3]);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_1, coeff16x4, 3);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_3, coeff16x4, 2);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_5, coeff16x4, 1);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_7, coeff16x4, 0);
}
for (; j < coefficients_length; j++) {
int16x4_t coeff16x4 = vld1_dup_s16(&coefficients[j]);
int16x8x4_t in16x8x4 = vld4q_s16(&data_in[i - j]);
// Mul and accumulate low 64-bit data.
int16x4_t in16x4_0 = vget_low_s16(in16x8x4.val[0]);
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_0, coeff16x4, 0);
// Mul and accumulate high 64-bit data.
// TODO: vget_high_s16 need extra cost on ARM64. This could be
// replaced by vmlal_high_lane_s16. But for the interface of
// vmlal_high_lane_s16, there is a bug in gcc 4.9.
// This issue need to be tracked in the future.
int16x4_t in16x4_1 = vget_high_s16(in16x8x4.val[0]);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_1, coeff16x4, 0);
}
// Saturate and store the output.
int16x4_t out16x4_0 = vqshrn_n_s32(out32x4_0, 12);
int16x4_t out16x4_1 = vqshrn_n_s32(out32x4_1, 12);
vst1q_s16(data_out, vcombine_s16(out16x4_0, out16x4_1));
data_out += 8;
}
break;
}
default: {
for (i = delay; i < endpos1; i += factor * 8) {
// Round value, 0.5 in Q12.
int32x4_t out32x4_0 = vdupq_n_s32(2048);
int32x4_t out32x4_1 = vdupq_n_s32(2048);
for (j = 0; j < coefficients_length; j++) {
int16x4_t coeff16x4 = vld1_dup_s16(&coefficients[j]);
int16x4_t in16x4_0 = vld1_dup_s16(&data_in[i - j]);
in16x4_0 = vld1_lane_s16(&data_in[i + factor - j], in16x4_0, 1);
in16x4_0 = vld1_lane_s16(&data_in[i + factor * 2 - j], in16x4_0, 2);
in16x4_0 = vld1_lane_s16(&data_in[i + factor * 3 - j], in16x4_0, 3);
int16x4_t in16x4_1 = vld1_dup_s16(&data_in[i + factor * 4 - j]);
in16x4_1 = vld1_lane_s16(&data_in[i + factor * 5 - j], in16x4_1, 1);
in16x4_1 = vld1_lane_s16(&data_in[i + factor * 6 - j], in16x4_1, 2);
in16x4_1 = vld1_lane_s16(&data_in[i + factor * 7 - j], in16x4_1, 3);
// Mul and accumulate.
out32x4_0 = vmlal_lane_s16(out32x4_0, in16x4_0, coeff16x4, 0);
out32x4_1 = vmlal_lane_s16(out32x4_1, in16x4_1, coeff16x4, 0);
}
// Saturate and store the output.
int16x4_t out16x4_0 = vqshrn_n_s32(out32x4_0, 12);
int16x4_t out16x4_1 = vqshrn_n_s32(out32x4_1, 12);
vst1q_s16(data_out, vcombine_s16(out16x4_0, out16x4_1));
data_out += 8;
}
break;
}
}
// Second part, do the rest iterations (if any).
for (; i < endpos; i += factor) {
out_s32 = 2048; // Round value, 0.5 in Q12.
for (j = 0; j < coefficients_length; j++) {
out_s32 = WebRtc_MulAccumW16(coefficients[j], data_in[i - j], out_s32);
}
// Saturate and store the output.
out_s32 >>= 12;
*data_out++ = WebRtcSpl_SatW32ToW16(out_s32);
}
return 0;
}

39
webrtc/common_audio/signal_processing/energy.c Normal file
View File

@ -0,0 +1,39 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_Energy().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
int32_t WebRtcSpl_Energy(int16_t* vector,
size_t vector_length,
int* scale_factor)
{
int32_t en = 0;
size_t i;
int scaling =
WebRtcSpl_GetScalingSquare(vector, vector_length, vector_length);
size_t looptimes = vector_length;
int16_t *vectorptr = vector;
for (i = 0; i < looptimes; i++)
{
en += (*vectorptr * *vectorptr) >> scaling;
vectorptr++;
}
*scale_factor = scaling;
return en;
}

89
webrtc/common_audio/signal_processing/filter_ar.c Normal file
View File

@ -0,0 +1,89 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_FilterAR().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
size_t WebRtcSpl_FilterAR(const int16_t* a,
size_t a_length,
const int16_t* x,
size_t x_length,
int16_t* state,
size_t state_length,
int16_t* state_low,
size_t state_low_length,
int16_t* filtered,
int16_t* filtered_low,
size_t filtered_low_length)
{
int32_t o;
int32_t oLOW;
size_t i, j, stop;
const int16_t* x_ptr = &x[0];
int16_t* filteredFINAL_ptr = filtered;
int16_t* filteredFINAL_LOW_ptr = filtered_low;
for (i = 0; i < x_length; i++)
{
// Calculate filtered[i] and filtered_low[i]
const int16_t* a_ptr = &a[1];
int16_t* filtered_ptr = &filtered[i - 1];
int16_t* filtered_low_ptr = &filtered_low[i - 1];
int16_t* state_ptr = &state[state_length - 1];
int16_t* state_low_ptr = &state_low[state_length - 1];
o = (int32_t)(*x_ptr++) << 12;
oLOW = (int32_t)0;
stop = (i < a_length) ? i + 1 : a_length;
for (j = 1; j < stop; j++)
{
o -= *a_ptr * *filtered_ptr--;
oLOW -= *a_ptr++ * *filtered_low_ptr--;
}
for (j = i + 1; j < a_length; j++)
{
o -= *a_ptr * *state_ptr--;
oLOW -= *a_ptr++ * *state_low_ptr--;
}
o += (oLOW >> 12);
*filteredFINAL_ptr = (int16_t)((o + (int32_t)2048) >> 12);
*filteredFINAL_LOW_ptr++ = (int16_t)(o - ((int32_t)(*filteredFINAL_ptr++)
<< 12));
}
// Save the filter state
if (x_length >= state_length)
{
WebRtcSpl_CopyFromEndW16(filtered, x_length, a_length - 1, state);
WebRtcSpl_CopyFromEndW16(filtered_low, x_length, a_length - 1, state_low);
} else
{
for (i = 0; i < state_length - x_length; i++)
{
state[i] = state[i + x_length];
state_low[i] = state_low[i + x_length];
}
for (i = 0; i < x_length; i++)
{
state[state_length - x_length + i] = filtered[i];
state[state_length - x_length + i] = filtered_low[i];
}
}
return x_length;
}

42
webrtc/common_audio/signal_processing/filter_ar_fast_q12.c Normal file
View File

@ -0,0 +1,42 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <assert.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// TODO(bjornv): Change the return type to report errors.
void WebRtcSpl_FilterARFastQ12(const int16_t* data_in,
int16_t* data_out,
const int16_t* __restrict coefficients,
size_t coefficients_length,
size_t data_length) {
size_t i = 0;
size_t j = 0;
assert(data_length > 0);
assert(coefficients_length > 1);
for (i = 0; i < data_length; i++) {
int32_t output = 0;
int32_t sum = 0;
for (j = coefficients_length - 1; j > 0; j--) {
sum += coefficients[j] * data_out[i - j];
}
output = coefficients[0] * data_in[i];
output -= sum;
// Saturate and store the output.
output = WEBRTC_SPL_SAT(134215679, output, -134217728);
data_out[i] = (int16_t)((output + 2048) >> 12);
}
}

218
webrtc/common_audio/signal_processing/filter_ar_fast_q12_armv7.S Normal file
View File

@ -0,0 +1,218 @@
@
@ Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
@
@ Use of this source code is governed by a BSD-style license
@ that can be found in the LICENSE file in the root of the source
@ tree. An additional intellectual property rights grant can be found
@ in the file PATENTS. All contributing project authors may
@ be found in the AUTHORS file in the root of the source tree.
@
@ This file contains the function WebRtcSpl_FilterARFastQ12(), optimized for
@ ARMv7 platform. The description header can be found in
@ signal_processing_library.h
@
@ Output is bit-exact with the generic C code as in filter_ar_fast_q12.c, and
@ the reference C code at end of this file.
@ Assumptions:
@ (1) data_length > 0
@ (2) coefficients_length > 1
@ Register usage:
@
@ r0: &data_in[i]
@ r1: &data_out[i], for result ouput
@ r2: &coefficients[0]
@ r3: coefficients_length
@ r4: Iteration counter for the outer loop.
@ r5: data_out[j] as multiplication inputs
@ r6: Calculated value for output data_out[]; interation counter for inner loop
@ r7: Partial sum of a filtering multiplication results
@ r8: Partial sum of a filtering multiplication results
@ r9: &data_out[], for filtering input; data_in[i]
@ r10: coefficients[j]
@ r11: Scratch
@ r12: &coefficients[j]
#include "webrtc/system_wrappers/interface/asm_defines.h"
GLOBAL_FUNCTION WebRtcSpl_FilterARFastQ12
.align 2
DEFINE_FUNCTION WebRtcSpl_FilterARFastQ12
push {r4-r11}
ldrsh r12, [sp, #32] @ data_length
subs r4, r12, #1
beq ODD_LENGTH @ jump if data_length == 1
LOOP_LENGTH:
add r12, r2, r3, lsl #1
sub r12, #4 @ &coefficients[coefficients_length - 2]
sub r9, r1, r3, lsl #1
add r9, #2 @ &data_out[i - coefficients_length + 1]
ldr r5, [r9], #4 @ data_out[i - coefficients_length + {1,2}]
mov r7, #0 @ sum1
mov r8, #0 @ sum2
subs r6, r3, #3 @ Iteration counter for inner loop.
beq ODD_A_LENGTH @ branch if coefficients_length == 3
blt POST_LOOP_A_LENGTH @ branch if coefficients_length == 2
LOOP_A_LENGTH:
ldr r10, [r12], #-4 @ coefficients[j - 1], coefficients[j]
subs r6, #2
smlatt r8, r10, r5, r8 @ sum2 += coefficients[j] * data_out[i - j + 1];
smlatb r7, r10, r5, r7 @ sum1 += coefficients[j] * data_out[i - j];
smlabt r7, r10, r5, r7 @ coefficients[j - 1] * data_out[i - j + 1];
ldr r5, [r9], #4 @ data_out[i - j + 2], data_out[i - j + 3]
smlabb r8, r10, r5, r8 @ coefficients[j - 1] * data_out[i - j + 2];
bgt LOOP_A_LENGTH
blt POST_LOOP_A_LENGTH
ODD_A_LENGTH:
ldrsh r10, [r12, #2] @ Filter coefficients coefficients[2]
sub r12, #2 @ &coefficients[0]
smlabb r7, r10, r5, r7 @ sum1 += coefficients[2] * data_out[i - 2];
smlabt r8, r10, r5, r8 @ sum2 += coefficients[2] * data_out[i - 1];
ldr r5, [r9, #-2] @ data_out[i - 1], data_out[i]
POST_LOOP_A_LENGTH:
ldr r10, [r12] @ coefficients[0], coefficients[1]
smlatb r7, r10, r5, r7 @ sum1 += coefficients[1] * data_out[i - 1];
ldr r9, [r0], #4 @ data_in[i], data_in[i + 1]
smulbb r6, r10, r9 @ output1 = coefficients[0] * data_in[i];
sub r6, r7 @ output1 -= sum1;
sbfx r11, r6, #12, #16
ssat r7, #16, r6, asr #12
cmp r7, r11
addeq r6, r6, #2048
ssat r6, #16, r6, asr #12
strh r6, [r1], #2 @ Store data_out[i]
smlatb r8, r10, r6, r8 @ sum2 += coefficients[1] * data_out[i];
smulbt r6, r10, r9 @ output2 = coefficients[0] * data_in[i + 1];
sub r6, r8 @ output1 -= sum1;
sbfx r11, r6, #12, #16
ssat r7, #16, r6, asr #12
cmp r7, r11
addeq r6, r6, #2048
ssat r6, #16, r6, asr #12
strh r6, [r1], #2 @ Store data_out[i + 1]
subs r4, #2
bgt LOOP_LENGTH
blt END @ For even data_length, it's done. Jump to END.
@ Process i = data_length -1, for the case of an odd length.
ODD_LENGTH:
add r12, r2, r3, lsl #1
sub r12, #4 @ &coefficients[coefficients_length - 2]
sub r9, r1, r3, lsl #1
add r9, #2 @ &data_out[i - coefficients_length + 1]
mov r7, #0 @ sum1
mov r8, #0 @ sum1
subs r6, r3, #2 @ inner loop counter
beq EVEN_A_LENGTH @ branch if coefficients_length == 2
LOOP2_A_LENGTH:
ldr r10, [r12], #-4 @ coefficients[j - 1], coefficients[j]
ldr r5, [r9], #4 @ data_out[i - j], data_out[i - j + 1]
subs r6, #2
smlatb r7, r10, r5, r7 @ sum1 += coefficients[j] * data_out[i - j];
smlabt r8, r10, r5, r8 @ coefficients[j - 1] * data_out[i - j + 1];
bgt LOOP2_A_LENGTH
addlt r12, #2
blt POST_LOOP2_A_LENGTH
EVEN_A_LENGTH:
ldrsh r10, [r12, #2] @ Filter coefficients coefficients[1]
ldrsh r5, [r9] @ data_out[i - 1]
smlabb r7, r10, r5, r7 @ sum1 += coefficients[1] * data_out[i - 1];
POST_LOOP2_A_LENGTH:
ldrsh r10, [r12] @ Filter coefficients coefficients[0]
ldrsh r9, [r0] @ data_in[i]
smulbb r6, r10, r9 @ output1 = coefficients[0] * data_in[i];
sub r6, r7 @ output1 -= sum1;
sub r6, r8 @ output1 -= sum1;
sbfx r8, r6, #12, #16
ssat r7, #16, r6, asr #12
cmp r7, r8
addeq r6, r6, #2048
ssat r6, #16, r6, asr #12
strh r6, [r1] @ Store the data_out[i]
END:
pop {r4-r11}
bx lr
@Reference C code:
@
@void WebRtcSpl_FilterARFastQ12(int16_t* data_in,
@ int16_t* data_out,
@ int16_t* __restrict coefficients,
@ size_t coefficients_length,
@ size_t data_length) {
@ size_t i = 0;
@ size_t j = 0;
@
@ assert(data_length > 0);
@ assert(coefficients_length > 1);
@
@ for (i = 0; i < data_length - 1; i += 2) {
@ int32_t output1 = 0;
@ int32_t sum1 = 0;
@ int32_t output2 = 0;
@ int32_t sum2 = 0;
@
@ for (j = coefficients_length - 1; j > 2; j -= 2) {
@ sum1 += coefficients[j] * data_out[i - j];
@ sum1 += coefficients[j - 1] * data_out[i - j + 1];
@ sum2 += coefficients[j] * data_out[i - j + 1];
@ sum2 += coefficients[j - 1] * data_out[i - j + 2];
@ }
@
@ if (j == 2) {
@ sum1 += coefficients[2] * data_out[i - 2];
@ sum2 += coefficients[2] * data_out[i - 1];
@ }
@
@ sum1 += coefficients[1] * data_out[i - 1];
@ output1 = coefficients[0] * data_in[i];
@ output1 -= sum1;
@ // Saturate and store the output.
@ output1 = WEBRTC_SPL_SAT(134215679, output1, -134217728);
@ data_out[i] = (int16_t)((output1 + 2048) >> 12);
@
@ sum2 += coefficients[1] * data_out[i];
@ output2 = coefficients[0] * data_in[i + 1];
@ output2 -= sum2;
@ // Saturate and store the output.
@ output2 = WEBRTC_SPL_SAT(134215679, output2, -134217728);
@ data_out[i + 1] = (int16_t)((output2 + 2048) >> 12);
@ }
@
@ if (i == data_length - 1) {
@ int32_t output1 = 0;
@ int32_t sum1 = 0;
@
@ for (j = coefficients_length - 1; j > 1; j -= 2) {
@ sum1 += coefficients[j] * data_out[i - j];
@ sum1 += coefficients[j - 1] * data_out[i - j + 1];
@ }
@
@ if (j == 1) {
@ sum1 += coefficients[1] * data_out[i - 1];
@ }
@
@ output1 = coefficients[0] * data_in[i];
@ output1 -= sum1;
@ // Saturate and store the output.
@ output1 = WEBRTC_SPL_SAT(134215679, output1, -134217728);
@ data_out[i] = (int16_t)((output1 + 2048) >> 12);
@ }
@}

140
webrtc/common_audio/signal_processing/filter_ar_fast_q12_mips.c Normal file
View File

@ -0,0 +1,140 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <assert.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_FilterARFastQ12(const int16_t* data_in,
int16_t* data_out,
const int16_t* __restrict coefficients,
size_t coefficients_length,
size_t data_length) {
int r0, r1, r2, r3;
int coef0, offset;
int i, j, k;
int coefptr, outptr, tmpout, inptr;
#if !defined(MIPS_DSP_R1_LE)
int max16 = 0x7FFF;
int min16 = 0xFFFF8000;
#endif // #if !defined(MIPS_DSP_R1_LE)
assert(data_length > 0);
assert(coefficients_length > 1);
__asm __volatile (
".set push \n\t"
".set noreorder \n\t"
"addiu %[i], %[data_length], 0 \n\t"
"lh %[coef0], 0(%[coefficients]) \n\t"
"addiu %[j], %[coefficients_length], -1 \n\t"
"andi %[k], %[j], 1 \n\t"
"sll %[offset], %[j], 1 \n\t"
"subu %[outptr], %[data_out], %[offset] \n\t"
"addiu %[inptr], %[data_in], 0 \n\t"
"bgtz %[k], 3f \n\t"
" addu %[coefptr], %[coefficients], %[offset] \n\t"
"1: \n\t"
"lh %[r0], 0(%[inptr]) \n\t"
"addiu %[i], %[i], -1 \n\t"
"addiu %[tmpout], %[outptr], 0 \n\t"
"mult %[r0], %[coef0] \n\t"
"2: \n\t"
"lh %[r0], 0(%[tmpout]) \n\t"
"lh %[r1], 0(%[coefptr]) \n\t"
"lh %[r2], 2(%[tmpout]) \n\t"
"lh %[r3], -2(%[coefptr]) \n\t"
"addiu %[tmpout], %[tmpout], 4 \n\t"
"msub %[r0], %[r1] \n\t"
"msub %[r2], %[r3] \n\t"
"addiu %[j], %[j], -2 \n\t"
"bgtz %[j], 2b \n\t"
" addiu %[coefptr], %[coefptr], -4 \n\t"
#if defined(MIPS_DSP_R1_LE)
"extr_r.w %[r0], $ac0, 12 \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"mflo %[r0] \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"addu %[coefptr], %[coefficients], %[offset] \n\t"
"addiu %[inptr], %[inptr], 2 \n\t"
"addiu %[j], %[coefficients_length], -1 \n\t"
#if defined(MIPS_DSP_R1_LE)
"shll_s.w %[r0], %[r0], 16 \n\t"
"sra %[r0], %[r0], 16 \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"addiu %[r0], %[r0], 2048 \n\t"
"sra %[r0], %[r0], 12 \n\t"
"slt %[r1], %[max16], %[r0] \n\t"
"movn %[r0], %[max16], %[r1] \n\t"
"slt %[r1], %[r0], %[min16] \n\t"
"movn %[r0], %[min16], %[r1] \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"sh %[r0], 0(%[tmpout]) \n\t"
"bgtz %[i], 1b \n\t"
" addiu %[outptr], %[outptr], 2 \n\t"
"b 5f \n\t"
" nop \n\t"
"3: \n\t"
"lh %[r0], 0(%[inptr]) \n\t"
"addiu %[i], %[i], -1 \n\t"
"addiu %[tmpout], %[outptr], 0 \n\t"
"mult %[r0], %[coef0] \n\t"
"4: \n\t"
"lh %[r0], 0(%[tmpout]) \n\t"
"lh %[r1], 0(%[coefptr]) \n\t"
"lh %[r2], 2(%[tmpout]) \n\t"
"lh %[r3], -2(%[coefptr]) \n\t"
"addiu %[tmpout], %[tmpout], 4 \n\t"
"msub %[r0], %[r1] \n\t"
"msub %[r2], %[r3] \n\t"
"addiu %[j], %[j], -2 \n\t"
"bgtz %[j], 4b \n\t"
" addiu %[coefptr], %[coefptr], -4 \n\t"
"lh %[r0], 0(%[tmpout]) \n\t"
"lh %[r1], 0(%[coefptr]) \n\t"
"msub %[r0], %[r1] \n\t"
#if defined(MIPS_DSP_R1_LE)
"extr_r.w %[r0], $ac0, 12 \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"mflo %[r0] \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"addu %[coefptr], %[coefficients], %[offset] \n\t"
"addiu %[inptr], %[inptr], 2 \n\t"
"addiu %[j], %[coefficients_length], -1 \n\t"
#if defined(MIPS_DSP_R1_LE)
"shll_s.w %[r0], %[r0], 16 \n\t"
"sra %[r0], %[r0], 16 \n\t"
#else // #if defined(MIPS_DSP_R1_LE)
"addiu %[r0], %[r0], 2048 \n\t"
"sra %[r0], %[r0], 12 \n\t"
"slt %[r1], %[max16], %[r0] \n\t"
"movn %[r0], %[max16], %[r1] \n\t"
"slt %[r1], %[r0], %[min16] \n\t"
"movn %[r0], %[min16], %[r1] \n\t"
#endif // #if defined(MIPS_DSP_R1_LE)
"sh %[r0], 2(%[tmpout]) \n\t"
"bgtz %[i], 3b \n\t"
" addiu %[outptr], %[outptr], 2 \n\t"
"5: \n\t"
".set pop \n\t"
: [i] "=&r" (i), [j] "=&r" (j), [k] "=&r" (k), [r0] "=&r" (r0),
[r1] "=&r" (r1), [r2] "=&r" (r2), [r3] "=&r" (r3),
[coef0] "=&r" (coef0), [offset] "=&r" (offset),
[outptr] "=&r" (outptr), [inptr] "=&r" (inptr),
[coefptr] "=&r" (coefptr), [tmpout] "=&r" (tmpout)
: [coefficients] "r" (coefficients), [data_length] "r" (data_length),
[coefficients_length] "r" (coefficients_length),
#if !defined(MIPS_DSP_R1_LE)
[max16] "r" (max16), [min16] "r" (min16),
#endif
[data_out] "r" (data_out), [data_in] "r" (data_in)
: "hi", "lo", "memory"
);
}

45
webrtc/common_audio/signal_processing/filter_ma_fast_q12.c Normal file
View File

@ -0,0 +1,45 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_FilterMAFastQ12().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_FilterMAFastQ12(const int16_t* in_ptr,
int16_t* out_ptr,
const int16_t* B,
size_t B_length,
size_t length)
{
size_t i, j;
for (i = 0; i < length; i++)
{
int32_t o = 0;
for (j = 0; j < B_length; j++)
{
o += B[j] * in_ptr[i - j];
}
// If output is higher than 32768, saturate it. Same with negative side
// 2^27 = 134217728, which corresponds to 32768 in Q12
// Saturate the output
o = WEBRTC_SPL_SAT((int32_t)134215679, o, (int32_t)-134217728);
*out_ptr++ = (int16_t)((o + (int32_t)2048) >> 12);
}
return;
}

77
webrtc/common_audio/signal_processing/get_hanning_window.c Normal file
View File

@ -0,0 +1,77 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_GetHanningWindow().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// Hanning table with 256 entries
static const int16_t kHanningTable[] = {
1, 2, 6, 10, 15, 22, 30, 39,
50, 62, 75, 89, 104, 121, 138, 157,
178, 199, 222, 246, 271, 297, 324, 353,
383, 413, 446, 479, 513, 549, 586, 624,
663, 703, 744, 787, 830, 875, 920, 967,
1015, 1064, 1114, 1165, 1218, 1271, 1325, 1381,
1437, 1494, 1553, 1612, 1673, 1734, 1796, 1859,
1924, 1989, 2055, 2122, 2190, 2259, 2329, 2399,
2471, 2543, 2617, 2691, 2765, 2841, 2918, 2995,
3073, 3152, 3232, 3312, 3393, 3475, 3558, 3641,
3725, 3809, 3895, 3980, 4067, 4154, 4242, 4330,
4419, 4509, 4599, 4689, 4781, 4872, 4964, 5057,
5150, 5244, 5338, 5432, 5527, 5622, 5718, 5814,
5910, 6007, 6104, 6202, 6299, 6397, 6495, 6594,
6693, 6791, 6891, 6990, 7090, 7189, 7289, 7389,
7489, 7589, 7690, 7790, 7890, 7991, 8091, 8192,
8293, 8393, 8494, 8594, 8694, 8795, 8895, 8995,
9095, 9195, 9294, 9394, 9493, 9593, 9691, 9790,
9889, 9987, 10085, 10182, 10280, 10377, 10474, 10570,
10666, 10762, 10857, 10952, 11046, 11140, 11234, 11327,
11420, 11512, 11603, 11695, 11785, 11875, 11965, 12054,
12142, 12230, 12317, 12404, 12489, 12575, 12659, 12743,
12826, 12909, 12991, 13072, 13152, 13232, 13311, 13389,
13466, 13543, 13619, 13693, 13767, 13841, 13913, 13985,
14055, 14125, 14194, 14262, 14329, 14395, 14460, 14525,
14588, 14650, 14711, 14772, 14831, 14890, 14947, 15003,
15059, 15113, 15166, 15219, 15270, 15320, 15369, 15417,
15464, 15509, 15554, 15597, 15640, 15681, 15721, 15760,
15798, 15835, 15871, 15905, 15938, 15971, 16001, 16031,
16060, 16087, 16113, 16138, 16162, 16185, 16206, 16227,
16246, 16263, 16280, 16295, 16309, 16322, 16334, 16345,
16354, 16362, 16369, 16374, 16378, 16382, 16383, 16384
};
void WebRtcSpl_GetHanningWindow(int16_t *v, size_t size)
{
size_t jj;
int16_t *vptr1;
int32_t index;
int32_t factor = ((int32_t)0x40000000);
factor = WebRtcSpl_DivW32W16(factor, (int16_t)size);
if (size < 513)
index = (int32_t)-0x200000;
else
index = (int32_t)-0x100000;
vptr1 = v;
for (jj = 0; jj < size; jj++)
{
index += factor;
(*vptr1++) = kHanningTable[index >> 22];
}
}

46
webrtc/common_audio/signal_processing/get_scaling_square.c Normal file
View File

@ -0,0 +1,46 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_GetScalingSquare().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
int16_t WebRtcSpl_GetScalingSquare(int16_t* in_vector,
size_t in_vector_length,
size_t times)
{
int16_t nbits = WebRtcSpl_GetSizeInBits((uint32_t)times);
size_t i;
int16_t smax = -1;
int16_t sabs;
int16_t *sptr = in_vector;
int16_t t;
size_t looptimes = in_vector_length;
for (i = looptimes; i > 0; i--)
{
sabs = (*sptr > 0 ? *sptr++ : -*sptr++);
smax = (sabs > smax ? sabs : smax);
}
t = WebRtcSpl_NormW32(WEBRTC_SPL_MUL(smax, smax));
if (smax == 0)
{
return 0; // Since norm(0) returns 0
} else
{
return (t > nbits) ? 0 : nbits - t;
}
}

90
webrtc/common_audio/signal_processing/ilbc_specific_functions.c Normal file
View File

@ -0,0 +1,90 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains implementations of the iLBC specific functions
* WebRtcSpl_ReverseOrderMultArrayElements()
* WebRtcSpl_ElementwiseVectorMult()
* WebRtcSpl_AddVectorsAndShift()
* WebRtcSpl_AddAffineVectorToVector()
* WebRtcSpl_AffineTransformVector()
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_ReverseOrderMultArrayElements(int16_t *out, const int16_t *in,
const int16_t *win,
size_t vector_length,
int16_t right_shifts)
{
size_t i;
int16_t *outptr = out;
const int16_t *inptr = in;
const int16_t *winptr = win;
for (i = 0; i < vector_length; i++)
{
*outptr++ = (int16_t)((*inptr++ * *winptr--) >> right_shifts);
}
}
void WebRtcSpl_ElementwiseVectorMult(int16_t *out, const int16_t *in,
const int16_t *win, size_t vector_length,
int16_t right_shifts)
{
size_t i;
int16_t *outptr = out;
const int16_t *inptr = in;
const int16_t *winptr = win;
for (i = 0; i < vector_length; i++)
{
*outptr++ = (int16_t)((*inptr++ * *winptr++) >> right_shifts);
}
}
void WebRtcSpl_AddVectorsAndShift(int16_t *out, const int16_t *in1,
const int16_t *in2, size_t vector_length,
int16_t right_shifts)
{
size_t i;
int16_t *outptr = out;
const int16_t *in1ptr = in1;
const int16_t *in2ptr = in2;
for (i = vector_length; i > 0; i--)
{
(*outptr++) = (int16_t)(((*in1ptr++) + (*in2ptr++)) >> right_shifts);
}
}
void WebRtcSpl_AddAffineVectorToVector(int16_t *out, int16_t *in,
int16_t gain, int32_t add_constant,
int16_t right_shifts,
size_t vector_length)
{
size_t i;
for (i = 0; i < vector_length; i++)
{
out[i] += (int16_t)((in[i] * gain + add_constant) >> right_shifts);
}
}
void WebRtcSpl_AffineTransformVector(int16_t *out, int16_t *in,
int16_t gain, int32_t add_constant,
int16_t right_shifts, size_t vector_length)
{
size_t i;
for (i = 0; i < vector_length; i++)
{
out[i] = (int16_t)((in[i] * gain + add_constant) >> right_shifts);
}
}

97
webrtc/common_audio/signal_processing/include/real_fft.h Normal file
View File

@ -0,0 +1,97 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_COMMON_AUDIO_SIGNAL_PROCESSING_INCLUDE_REAL_FFT_H_
#define WEBRTC_COMMON_AUDIO_SIGNAL_PROCESSING_INCLUDE_REAL_FFT_H_
#include "webrtc/typedefs.h"
// For ComplexFFT(), the maximum fft order is 10;
// for OpenMax FFT in ARM, it is 12;
// WebRTC APM uses orders of only 7 and 8.
enum {kMaxFFTOrder = 10};
struct RealFFT;
#ifdef __cplusplus
extern "C" {
#endif
struct RealFFT* WebRtcSpl_CreateRealFFT(int order);
void WebRtcSpl_FreeRealFFT(struct RealFFT* self);
// Compute an FFT for a real-valued signal of length of 2^order,
// where 1 < order <= MAX_FFT_ORDER. Transform length is determined by the
// specification structure, which must be initialized prior to calling the FFT
// function with WebRtcSpl_CreateRealFFT().
// The relationship between the input and output sequences can
// be expressed in terms of the DFT, i.e.:
// x[n] = (2^(-scalefactor)/N) . SUM[k=0,...,N-1] X[k].e^(jnk.2.pi/N)
// n=0,1,2,...N-1
// N=2^order.
// The conjugate-symmetric output sequence is represented using a CCS vector,
// which is of length N+2, and is organized as follows:
// Index: 0 1 2 3 4 5 . . . N-2 N-1 N N+1
// Component: R0 0 R1 I1 R2 I2 . . . R[N/2-1] I[N/2-1] R[N/2] 0
// where R[n] and I[n], respectively, denote the real and imaginary components
// for FFT bin 'n'. Bins are numbered from 0 to N/2, where N is the FFT length.
// Bin index 0 corresponds to the DC component, and bin index N/2 corresponds to
// the foldover frequency.
//
// Input Arguments:
// self - pointer to preallocated and initialized FFT specification structure.
// real_data_in - the input signal. For an ARM Neon platform, it must be
// aligned on a 32-byte boundary.
//
// Output Arguments:
// complex_data_out - the output complex signal with (2^order + 2) 16-bit
// elements. For an ARM Neon platform, it must be different
// from real_data_in, and aligned on a 32-byte boundary.
//
// Return Value:
// 0 - FFT calculation is successful.
// -1 - Error with bad arguments (NULL pointers).
int WebRtcSpl_RealForwardFFT(struct RealFFT* self,
const int16_t* real_data_in,
int16_t* complex_data_out);
// Compute the inverse FFT for a conjugate-symmetric input sequence of length of
// 2^order, where 1 < order <= MAX_FFT_ORDER. Transform length is determined by
// the specification structure, which must be initialized prior to calling the
// FFT function with WebRtcSpl_CreateRealFFT().
// For a transform of length M, the input sequence is represented using a packed
// CCS vector of length M+2, which is explained in the comments for
// WebRtcSpl_RealForwardFFTC above.
//
// Input Arguments:
// self - pointer to preallocated and initialized FFT specification structure.
// complex_data_in - the input complex signal with (2^order + 2) 16-bit
// elements. For an ARM Neon platform, it must be aligned on
// a 32-byte boundary.
//
// Output Arguments:
// real_data_out - the output real signal. For an ARM Neon platform, it must
// be different to complex_data_in, and aligned on a 32-byte
// boundary.
//
// Return Value:
// 0 or a positive number - a value that the elements in the |real_data_out|
// should be shifted left with in order to get
// correct physical values.
// -1 - Error with bad arguments (NULL pointers).
int WebRtcSpl_RealInverseFFT(struct RealFFT* self,
const int16_t* complex_data_in,
int16_t* real_data_out);
#ifdef __cplusplus
}
#endif
#endif // WEBRTC_COMMON_AUDIO_SIGNAL_PROCESSING_INCLUDE_REAL_FFT_H_

1645
webrtc/common_audio/signal_processing/include/signal_processing_library.h Normal file
View File

File diff suppressed because it is too large Load Diff

173
webrtc/common_audio/signal_processing/include/spl_inl.h Normal file
View File

@ -0,0 +1,173 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
// This header file includes the inline functions in
// the fix point signal processing library.
#ifndef WEBRTC_SPL_SPL_INL_H_
#define WEBRTC_SPL_SPL_INL_H_
#ifdef WEBRTC_ARCH_ARM_V7
#include "webrtc/common_audio/signal_processing/include/spl_inl_armv7.h"
#else
#if defined(MIPS32_LE)
#include "webrtc/common_audio/signal_processing/include/spl_inl_mips.h"
#endif
#if !defined(MIPS_DSP_R1_LE)
static __inline int16_t WebRtcSpl_SatW32ToW16(int32_t value32) {
int16_t out16 = (int16_t) value32;
if (value32 > 32767)
out16 = 32767;
else if (value32 < -32768)
out16 = -32768;
return out16;
}
static __inline int32_t WebRtcSpl_AddSatW32(int32_t l_var1, int32_t l_var2) {
int32_t l_sum;
// Perform long addition
l_sum = l_var1 + l_var2;
if (l_var1 < 0) { // Check for underflow.
if ((l_var2 < 0) && (l_sum >= 0)) {
l_sum = (int32_t)0x80000000;
}
} else { // Check for overflow.
if ((l_var2 > 0) && (l_sum < 0)) {
l_sum = (int32_t)0x7FFFFFFF;
}
}
return l_sum;
}
static __inline int32_t WebRtcSpl_SubSatW32(int32_t l_var1, int32_t l_var2) {
int32_t l_diff;
// Perform subtraction.
l_diff = l_var1 - l_var2;
if (l_var1 < 0) { // Check for underflow.
if ((l_var2 > 0) && (l_diff > 0)) {
l_diff = (int32_t)0x80000000;
}
} else { // Check for overflow.
if ((l_var2 < 0) && (l_diff < 0)) {
l_diff = (int32_t)0x7FFFFFFF;
}
}
return l_diff;
}
static __inline int16_t WebRtcSpl_AddSatW16(int16_t a, int16_t b) {
return WebRtcSpl_SatW32ToW16((int32_t) a + (int32_t) b);
}
static __inline int16_t WebRtcSpl_SubSatW16(int16_t var1, int16_t var2) {
return WebRtcSpl_SatW32ToW16((int32_t) var1 - (int32_t) var2);
}
#endif // #if !defined(MIPS_DSP_R1_LE)
#if !defined(MIPS32_LE)
static __inline int16_t WebRtcSpl_GetSizeInBits(uint32_t n) {
int16_t bits;
if (0xFFFF0000 & n) {
bits = 16;
} else {
bits = 0;
}
if (0x0000FF00 & (n >> bits)) bits += 8;
if (0x000000F0 & (n >> bits)) bits += 4;
if (0x0000000C & (n >> bits)) bits += 2;
if (0x00000002 & (n >> bits)) bits += 1;
if (0x00000001 & (n >> bits)) bits += 1;
return bits;
}
static __inline int16_t WebRtcSpl_NormW32(int32_t a) {
int16_t zeros;
if (a == 0) {
return 0;
}
else if (a < 0) {
a = ~a;
}
if (!(0xFFFF8000 & a)) {
zeros = 16;
} else {
zeros = 0;
}
if (!(0xFF800000 & (a << zeros))) zeros += 8;
if (!(0xF8000000 & (a << zeros))) zeros += 4;
if (!(0xE0000000 & (a << zeros))) zeros += 2;
if (!(0xC0000000 & (a << zeros))) zeros += 1;
return zeros;
}
static __inline int16_t WebRtcSpl_NormU32(uint32_t a) {
int16_t zeros;
if (a == 0) return 0;
if (!(0xFFFF0000 & a)) {
zeros = 16;
} else {
zeros = 0;
}
if (!(0xFF000000 & (a << zeros))) zeros += 8;
if (!(0xF0000000 & (a << zeros))) zeros += 4;
if (!(0xC0000000 & (a << zeros))) zeros += 2;
if (!(0x80000000 & (a << zeros))) zeros += 1;
return zeros;
}
static __inline int16_t WebRtcSpl_NormW16(int16_t a) {
int16_t zeros;
if (a == 0) {
return 0;
}
else if (a < 0) {
a = ~a;
}
if (!(0xFF80 & a)) {
zeros = 8;
} else {
zeros = 0;
}
if (!(0xF800 & (a << zeros))) zeros += 4;
if (!(0xE000 & (a << zeros))) zeros += 2;
if (!(0xC000 & (a << zeros))) zeros += 1;
return zeros;
}
static __inline int32_t WebRtc_MulAccumW16(int16_t a, int16_t b, int32_t c) {
return (a * b + c);
}
#endif // #if !defined(MIPS32_LE)
#endif // WEBRTC_ARCH_ARM_V7
#endif // WEBRTC_SPL_SPL_INL_H_

136
webrtc/common_audio/signal_processing/include/spl_inl_armv7.h Normal file
View File

@ -0,0 +1,136 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/* This header file includes the inline functions for ARM processors in
* the fix point signal processing library.
*/
#ifndef WEBRTC_SPL_SPL_INL_ARMV7_H_
#define WEBRTC_SPL_SPL_INL_ARMV7_H_
/* TODO(kma): Replace some assembly code with GCC intrinsics
* (e.g. __builtin_clz).
*/
/* This function produces result that is not bit exact with that by the generic
* C version in some cases, although the former is at least as accurate as the
* later.
*/
static __inline int32_t WEBRTC_SPL_MUL_16_32_RSFT16(int16_t a, int32_t b) {
int32_t tmp = 0;
__asm __volatile ("smulwb %0, %1, %2":"=r"(tmp):"r"(b), "r"(a));
return tmp;
}
static __inline int32_t WEBRTC_SPL_MUL_16_16(int16_t a, int16_t b) {
int32_t tmp = 0;
__asm __volatile ("smulbb %0, %1, %2":"=r"(tmp):"r"(a), "r"(b));
return tmp;
}
// TODO(kma): add unit test.
static __inline int32_t WebRtc_MulAccumW16(int16_t a, int16_t b, int32_t c) {
int32_t tmp = 0;
__asm __volatile ("smlabb %0, %1, %2, %3":"=r"(tmp):"r"(a), "r"(b), "r"(c));
return tmp;
}
static __inline int16_t WebRtcSpl_AddSatW16(int16_t a, int16_t b) {
int32_t s_sum = 0;
__asm __volatile ("qadd16 %0, %1, %2":"=r"(s_sum):"r"(a), "r"(b));
return (int16_t) s_sum;
}
static __inline int32_t WebRtcSpl_AddSatW32(int32_t l_var1, int32_t l_var2) {
int32_t l_sum = 0;
__asm __volatile ("qadd %0, %1, %2":"=r"(l_sum):"r"(l_var1), "r"(l_var2));
return l_sum;
}
static __inline int32_t WebRtcSpl_SubSatW32(int32_t l_var1, int32_t l_var2) {
int32_t l_sub = 0;
__asm __volatile ("qsub %0, %1, %2":"=r"(l_sub):"r"(l_var1), "r"(l_var2));
return l_sub;
}
static __inline int16_t WebRtcSpl_SubSatW16(int16_t var1, int16_t var2) {
int32_t s_sub = 0;
__asm __volatile ("qsub16 %0, %1, %2":"=r"(s_sub):"r"(var1), "r"(var2));
return (int16_t)s_sub;
}
static __inline int16_t WebRtcSpl_GetSizeInBits(uint32_t n) {
int32_t tmp = 0;
__asm __volatile ("clz %0, %1":"=r"(tmp):"r"(n));
return (int16_t)(32 - tmp);
}
static __inline int16_t WebRtcSpl_NormW32(int32_t a) {
int32_t tmp = 0;
if (a == 0) {
return 0;
}
else if (a < 0) {
a ^= 0xFFFFFFFF;
}
__asm __volatile ("clz %0, %1":"=r"(tmp):"r"(a));
return (int16_t)(tmp - 1);
}
static __inline int16_t WebRtcSpl_NormU32(uint32_t a) {
int tmp = 0;
if (a == 0) return 0;
__asm __volatile ("clz %0, %1":"=r"(tmp):"r"(a));
return (int16_t)tmp;
}
static __inline int16_t WebRtcSpl_NormW16(int16_t a) {
int32_t tmp = 0;
int32_t a_32 = a;
if (a_32 == 0) {
return 0;
}
else if (a_32 < 0) {
a_32 ^= 0xFFFFFFFF;
}
__asm __volatile ("clz %0, %1":"=r"(tmp):"r"(a_32));
return (int16_t)(tmp - 17);
}
// TODO(kma): add unit test.
static __inline int16_t WebRtcSpl_SatW32ToW16(int32_t value32) {
int32_t out = 0;
__asm __volatile ("ssat %0, #16, %1" : "=r"(out) : "r"(value32));
return (int16_t)out;
}
#endif // WEBRTC_SPL_SPL_INL_ARMV7_H_

225
webrtc/common_audio/signal_processing/include/spl_inl_mips.h Normal file
View File

@ -0,0 +1,225 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
// This header file includes the inline functions in
// the fix point signal processing library.
#ifndef WEBRTC_SPL_SPL_INL_MIPS_H_
#define WEBRTC_SPL_SPL_INL_MIPS_H_
static __inline int32_t WEBRTC_SPL_MUL_16_16(int32_t a,
int32_t b) {
int32_t value32 = 0;
int32_t a1 = 0, b1 = 0;
__asm __volatile(
#if defined(MIPS32_R2_LE)
"seh %[a1], %[a] \n\t"
"seh %[b1], %[b] \n\t"
#else
"sll %[a1], %[a], 16 \n\t"
"sll %[b1], %[b], 16 \n\t"
"sra %[a1], %[a1], 16 \n\t"
"sra %[b1], %[b1], 16 \n\t"
#endif
"mul %[value32], %[a1], %[b1] \n\t"
: [value32] "=r" (value32), [a1] "=&r" (a1), [b1] "=&r" (b1)
: [a] "r" (a), [b] "r" (b)
: "hi", "lo"
);
return value32;
}
static __inline int32_t WEBRTC_SPL_MUL_16_32_RSFT16(int16_t a,
int32_t b) {
int32_t value32 = 0, b1 = 0, b2 = 0;
int32_t a1 = 0;
__asm __volatile(
#if defined(MIPS32_R2_LE)
"seh %[a1], %[a] \n\t"
#else
"sll %[a1], %[a], 16 \n\t"
"sra %[a1], %[a1], 16 \n\t"
#endif
"andi %[b2], %[b], 0xFFFF \n\t"
"sra %[b1], %[b], 16 \n\t"
"sra %[b2], %[b2], 1 \n\t"
"mul %[value32], %[a1], %[b1] \n\t"
"mul %[b2], %[a1], %[b2] \n\t"
"addiu %[b2], %[b2], 0x4000 \n\t"
"sra %[b2], %[b2], 15 \n\t"
"addu %[value32], %[value32], %[b2] \n\t"
: [value32] "=&r" (value32), [b1] "=&r" (b1), [b2] "=&r" (b2),
[a1] "=&r" (a1)
: [a] "r" (a), [b] "r" (b)
: "hi", "lo"
);
return value32;
}
#if defined(MIPS_DSP_R1_LE)
static __inline int16_t WebRtcSpl_SatW32ToW16(int32_t value32) {
__asm __volatile(
"shll_s.w %[value32], %[value32], 16 \n\t"
"sra %[value32], %[value32], 16 \n\t"
: [value32] "+r" (value32)
:
);
int16_t out16 = (int16_t)value32;
return out16;
}
static __inline int16_t WebRtcSpl_AddSatW16(int16_t a, int16_t b) {
int32_t value32 = 0;
__asm __volatile(
"addq_s.ph %[value32], %[a], %[b] \n\t"
: [value32] "=r" (value32)
: [a] "r" (a), [b] "r" (b)
);
return (int16_t)value32;
}
static __inline int32_t WebRtcSpl_AddSatW32(int32_t l_var1, int32_t l_var2) {
int32_t l_sum;
__asm __volatile(
"addq_s.w %[l_sum], %[l_var1], %[l_var2] \n\t"
: [l_sum] "=r" (l_sum)
: [l_var1] "r" (l_var1), [l_var2] "r" (l_var2)
);
return l_sum;
}
static __inline int16_t WebRtcSpl_SubSatW16(int16_t var1, int16_t var2) {
int32_t value32;
__asm __volatile(
"subq_s.ph %[value32], %[var1], %[var2] \n\t"
: [value32] "=r" (value32)
: [var1] "r" (var1), [var2] "r" (var2)
);
return (int16_t)value32;
}
static __inline int32_t WebRtcSpl_SubSatW32(int32_t l_var1, int32_t l_var2) {
int32_t l_diff;
__asm __volatile(
"subq_s.w %[l_diff], %[l_var1], %[l_var2] \n\t"
: [l_diff] "=r" (l_diff)
: [l_var1] "r" (l_var1), [l_var2] "r" (l_var2)
);
return l_diff;
}
#endif
static __inline int16_t WebRtcSpl_GetSizeInBits(uint32_t n) {
int bits = 0;
int i32 = 32;
__asm __volatile(
"clz %[bits], %[n] \n\t"
"subu %[bits], %[i32], %[bits] \n\t"
: [bits] "=&r" (bits)
: [n] "r" (n), [i32] "r" (i32)
);
return (int16_t)bits;
}
static __inline int16_t WebRtcSpl_NormW32(int32_t a) {
int zeros = 0;
__asm __volatile(
".set push \n\t"
".set noreorder \n\t"
"bnez %[a], 1f \n\t"
" sra %[zeros], %[a], 31 \n\t"
"b 2f \n\t"
" move %[zeros], $zero \n\t"
"1: \n\t"
"xor %[zeros], %[a], %[zeros] \n\t"
"clz %[zeros], %[zeros] \n\t"
"addiu %[zeros], %[zeros], -1 \n\t"
"2: \n\t"
".set pop \n\t"
: [zeros]"=&r"(zeros)
: [a] "r" (a)
);
return (int16_t)zeros;
}
static __inline int16_t WebRtcSpl_NormU32(uint32_t a) {
int zeros = 0;
__asm __volatile(
"clz %[zeros], %[a] \n\t"
: [zeros] "=r" (zeros)
: [a] "r" (a)
);
return (int16_t)(zeros & 0x1f);
}
static __inline int16_t WebRtcSpl_NormW16(int16_t a) {
int zeros = 0;
int a0 = a << 16;
__asm __volatile(
".set push \n\t"
".set noreorder \n\t"
"bnez %[a0], 1f \n\t"
" sra %[zeros], %[a0], 31 \n\t"
"b 2f \n\t"
" move %[zeros], $zero \n\t"
"1: \n\t"
"xor %[zeros], %[a0], %[zeros] \n\t"
"clz %[zeros], %[zeros] \n\t"
"addiu %[zeros], %[zeros], -1 \n\t"
"2: \n\t"
".set pop \n\t"
: [zeros]"=&r"(zeros)
: [a0] "r" (a0)
);
return (int16_t)zeros;
}
static __inline int32_t WebRtc_MulAccumW16(int16_t a,
int16_t b,
int32_t c) {
int32_t res = 0, c1 = 0;
__asm __volatile(
#if defined(MIPS32_R2_LE)
"seh %[a], %[a] \n\t"
"seh %[b], %[b] \n\t"
#else
"sll %[a], %[a], 16 \n\t"
"sll %[b], %[b], 16 \n\t"
"sra %[a], %[a], 16 \n\t"
"sra %[b], %[b], 16 \n\t"
#endif
"mul %[res], %[a], %[b] \n\t"
"addu %[c1], %[c], %[res] \n\t"
: [c1] "=r" (c1), [res] "=&r" (res)
: [a] "r" (a), [b] "r" (b), [c] "r" (c)
: "hi", "lo"
);
return (c1);
}
#endif // WEBRTC_SPL_SPL_INL_MIPS_H_

246
webrtc/common_audio/signal_processing/levinson_durbin.c Normal file
View File

@ -0,0 +1,246 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_LevinsonDurbin().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#define SPL_LEVINSON_MAXORDER 20
int16_t WebRtcSpl_LevinsonDurbin(const int32_t* R, int16_t* A, int16_t* K,
size_t order)
{
size_t i, j;
// Auto-correlation coefficients in high precision
int16_t R_hi[SPL_LEVINSON_MAXORDER + 1], R_low[SPL_LEVINSON_MAXORDER + 1];
// LPC coefficients in high precision
int16_t A_hi[SPL_LEVINSON_MAXORDER + 1], A_low[SPL_LEVINSON_MAXORDER + 1];
// LPC coefficients for next iteration
int16_t A_upd_hi[SPL_LEVINSON_MAXORDER + 1], A_upd_low[SPL_LEVINSON_MAXORDER + 1];
// Reflection coefficient in high precision
int16_t K_hi, K_low;
// Prediction gain Alpha in high precision and with scale factor
int16_t Alpha_hi, Alpha_low, Alpha_exp;
int16_t tmp_hi, tmp_low;
int32_t temp1W32, temp2W32, temp3W32;
int16_t norm;
// Normalize the autocorrelation R[0]...R[order+1]
norm = WebRtcSpl_NormW32(R[0]);
for (i = 0; i <= order; ++i)
{
temp1W32 = WEBRTC_SPL_LSHIFT_W32(R[i], norm);
// Put R in hi and low format
R_hi[i] = (int16_t)(temp1W32 >> 16);
R_low[i] = (int16_t)((temp1W32 - ((int32_t)R_hi[i] << 16)) >> 1);
}
// K = A[1] = -R[1] / R[0]
temp2W32 = WEBRTC_SPL_LSHIFT_W32((int32_t)R_hi[1],16)
+ WEBRTC_SPL_LSHIFT_W32((int32_t)R_low[1],1); // R[1] in Q31
temp3W32 = WEBRTC_SPL_ABS_W32(temp2W32); // abs R[1]
temp1W32 = WebRtcSpl_DivW32HiLow(temp3W32, R_hi[0], R_low[0]); // abs(R[1])/R[0] in Q31
// Put back the sign on R[1]
if (temp2W32 > 0)
{
temp1W32 = -temp1W32;
}
// Put K in hi and low format
K_hi = (int16_t)(temp1W32 >> 16);
K_low = (int16_t)((temp1W32 - ((int32_t)K_hi << 16)) >> 1);
// Store first reflection coefficient
K[0] = K_hi;
temp1W32 >>= 4; // A[1] in Q27.
// Put A[1] in hi and low format
A_hi[1] = (int16_t)(temp1W32 >> 16);
A_low[1] = (int16_t)((temp1W32 - ((int32_t)A_hi[1] << 16)) >> 1);
// Alpha = R[0] * (1-K^2)
temp1W32 = ((K_hi * K_low >> 14) + K_hi * K_hi) << 1; // = k^2 in Q31
temp1W32 = WEBRTC_SPL_ABS_W32(temp1W32); // Guard against <0
temp1W32 = (int32_t)0x7fffffffL - temp1W32; // temp1W32 = (1 - K[0]*K[0]) in Q31
// Store temp1W32 = 1 - K[0]*K[0] on hi and low format
tmp_hi = (int16_t)(temp1W32 >> 16);
tmp_low = (int16_t)((temp1W32 - ((int32_t)tmp_hi << 16)) >> 1);
// Calculate Alpha in Q31
temp1W32 = (R_hi[0] * tmp_hi + (R_hi[0] * tmp_low >> 15) +
(R_low[0] * tmp_hi >> 15)) << 1;
// Normalize Alpha and put it in hi and low format
Alpha_exp = WebRtcSpl_NormW32(temp1W32);
temp1W32 = WEBRTC_SPL_LSHIFT_W32(temp1W32, Alpha_exp);
Alpha_hi = (int16_t)(temp1W32 >> 16);
Alpha_low = (int16_t)((temp1W32 - ((int32_t)Alpha_hi << 16)) >> 1);
// Perform the iterative calculations in the Levinson-Durbin algorithm
for (i = 2; i <= order; i++)
{
/* ----
temp1W32 = R[i] + > R[j]*A[i-j]
/
----
j=1..i-1
*/
temp1W32 = 0;
for (j = 1; j < i; j++)
{
// temp1W32 is in Q31
temp1W32 += (R_hi[j] * A_hi[i - j] << 1) +
(((R_hi[j] * A_low[i - j] >> 15) +
(R_low[j] * A_hi[i - j] >> 15)) << 1);
}
temp1W32 = WEBRTC_SPL_LSHIFT_W32(temp1W32, 4);
temp1W32 += (WEBRTC_SPL_LSHIFT_W32((int32_t)R_hi[i], 16)
+ WEBRTC_SPL_LSHIFT_W32((int32_t)R_low[i], 1));
// K = -temp1W32 / Alpha
temp2W32 = WEBRTC_SPL_ABS_W32(temp1W32); // abs(temp1W32)
temp3W32 = WebRtcSpl_DivW32HiLow(temp2W32, Alpha_hi, Alpha_low); // abs(temp1W32)/Alpha
// Put the sign of temp1W32 back again
if (temp1W32 > 0)
{
temp3W32 = -temp3W32;
}
// Use the Alpha shifts from earlier to de-normalize
norm = WebRtcSpl_NormW32(temp3W32);
if ((Alpha_exp <= norm) || (temp3W32 == 0))
{
temp3W32 = WEBRTC_SPL_LSHIFT_W32(temp3W32, Alpha_exp);
} else
{
if (temp3W32 > 0)
{
temp3W32 = (int32_t)0x7fffffffL;
} else
{
temp3W32 = (int32_t)0x80000000L;
}
}
// Put K on hi and low format
K_hi = (int16_t)(temp3W32 >> 16);
K_low = (int16_t)((temp3W32 - ((int32_t)K_hi << 16)) >> 1);
// Store Reflection coefficient in Q15
K[i - 1] = K_hi;
// Test for unstable filter.
// If unstable return 0 and let the user decide what to do in that case
if ((int32_t)WEBRTC_SPL_ABS_W16(K_hi) > (int32_t)32750)
{
return 0; // Unstable filter
}
/*
Compute updated LPC coefficient: Anew[i]
Anew[j]= A[j] + K*A[i-j] for j=1..i-1
Anew[i]= K
*/
for (j = 1; j < i; j++)
{
// temp1W32 = A[j] in Q27
temp1W32 = WEBRTC_SPL_LSHIFT_W32((int32_t)A_hi[j],16)
+ WEBRTC_SPL_LSHIFT_W32((int32_t)A_low[j],1);
// temp1W32 += K*A[i-j] in Q27
temp1W32 += (K_hi * A_hi[i - j] + (K_hi * A_low[i - j] >> 15) +
(K_low * A_hi[i - j] >> 15)) << 1;
// Put Anew in hi and low format
A_upd_hi[j] = (int16_t)(temp1W32 >> 16);
A_upd_low[j] = (int16_t)(
(temp1W32 - ((int32_t)A_upd_hi[j] << 16)) >> 1);
}
// temp3W32 = K in Q27 (Convert from Q31 to Q27)
temp3W32 >>= 4;
// Store Anew in hi and low format
A_upd_hi[i] = (int16_t)(temp3W32 >> 16);
A_upd_low[i] = (int16_t)(
(temp3W32 - ((int32_t)A_upd_hi[i] << 16)) >> 1);
// Alpha = Alpha * (1-K^2)
temp1W32 = ((K_hi * K_low >> 14) + K_hi * K_hi) << 1; // K*K in Q31
temp1W32 = WEBRTC_SPL_ABS_W32(temp1W32); // Guard against <0
temp1W32 = (int32_t)0x7fffffffL - temp1W32; // 1 - K*K in Q31
// Convert 1- K^2 in hi and low format
tmp_hi = (int16_t)(temp1W32 >> 16);
tmp_low = (int16_t)((temp1W32 - ((int32_t)tmp_hi << 16)) >> 1);
// Calculate Alpha = Alpha * (1-K^2) in Q31
temp1W32 = (Alpha_hi * tmp_hi + (Alpha_hi * tmp_low >> 15) +
(Alpha_low * tmp_hi >> 15)) << 1;
// Normalize Alpha and store it on hi and low format
norm = WebRtcSpl_NormW32(temp1W32);
temp1W32 = WEBRTC_SPL_LSHIFT_W32(temp1W32, norm);
Alpha_hi = (int16_t)(temp1W32 >> 16);
Alpha_low = (int16_t)((temp1W32 - ((int32_t)Alpha_hi << 16)) >> 1);
// Update the total normalization of Alpha
Alpha_exp = Alpha_exp + norm;
// Update A[]
for (j = 1; j <= i; j++)
{
A_hi[j] = A_upd_hi[j];
A_low[j] = A_upd_low[j];
}
}
/*
Set A[0] to 1.0 and store the A[i] i=1...order in Q12
(Convert from Q27 and use rounding)
*/
A[0] = 4096;
for (i = 1; i <= order; i++)
{
// temp1W32 in Q27
temp1W32 = WEBRTC_SPL_LSHIFT_W32((int32_t)A_hi[i], 16)
+ WEBRTC_SPL_LSHIFT_W32((int32_t)A_low[i], 1);
// Round and store upper word
A[i] = (int16_t)(((temp1W32 << 1) + 32768) >> 16);
}
return 1; // Stable filters
}

56
webrtc/common_audio/signal_processing/lpc_to_refl_coef.c Normal file
View File

@ -0,0 +1,56 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_LpcToReflCoef().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#define SPL_LPC_TO_REFL_COEF_MAX_AR_MODEL_ORDER 50
void WebRtcSpl_LpcToReflCoef(int16_t* a16, int use_order, int16_t* k16)
{
int m, k;
int32_t tmp32[SPL_LPC_TO_REFL_COEF_MAX_AR_MODEL_ORDER];
int32_t tmp_inv_denom32;
int16_t tmp_inv_denom16;
k16[use_order - 1] = a16[use_order] << 3; // Q12<<3 => Q15
for (m = use_order - 1; m > 0; m--)
{
// (1 - k^2) in Q30
tmp_inv_denom32 = 1073741823 - k16[m] * k16[m];
// (1 - k^2) in Q15
tmp_inv_denom16 = (int16_t)(tmp_inv_denom32 >> 15);
for (k = 1; k <= m; k++)
{
// tmp[k] = (a[k] - RC[m] * a[m-k+1]) / (1.0 - RC[m]*RC[m]);
// [Q12<<16 - (Q15*Q12)<<1] = [Q28 - Q28] = Q28
tmp32[k] = (a16[k] << 16) - (k16[m] * a16[m - k + 1] << 1);
tmp32[k] = WebRtcSpl_DivW32W16(tmp32[k], tmp_inv_denom16); //Q28/Q15 = Q13
}
for (k = 1; k < m; k++)
{
a16[k] = (int16_t)(tmp32[k] >> 1); // Q13>>1 => Q12
}
tmp32[m] = WEBRTC_SPL_SAT(8191, tmp32[m], -8191);
k16[m - 1] = (int16_t)WEBRTC_SPL_LSHIFT_W32(tmp32[m], 2); //Q13<<2 => Q15
}
return;
}

224
webrtc/common_audio/signal_processing/min_max_operations.c Normal file
View File

@ -0,0 +1,224 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the implementation of functions
* WebRtcSpl_MaxAbsValueW16C()
* WebRtcSpl_MaxAbsValueW32C()
* WebRtcSpl_MaxValueW16C()
* WebRtcSpl_MaxValueW32C()
* WebRtcSpl_MinValueW16C()
* WebRtcSpl_MinValueW32C()
* WebRtcSpl_MaxAbsIndexW16()
* WebRtcSpl_MaxIndexW16()
* WebRtcSpl_MaxIndexW32()
* WebRtcSpl_MinIndexW16()
* WebRtcSpl_MinIndexW32()
*
*/
#include <assert.h>
#include <stdlib.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// TODO(bjorn/kma): Consolidate function pairs (e.g. combine
// WebRtcSpl_MaxAbsValueW16C and WebRtcSpl_MaxAbsIndexW16 into a single one.)
// TODO(kma): Move the next six functions into min_max_operations_c.c.
// Maximum absolute value of word16 vector. C version for generic platforms.
int16_t WebRtcSpl_MaxAbsValueW16C(const int16_t* vector, size_t length) {
size_t i = 0;
int absolute = 0, maximum = 0;
assert(length > 0);
for (i = 0; i < length; i++) {
absolute = abs((int)vector[i]);
if (absolute > maximum) {
maximum = absolute;
}
}
// Guard the case for abs(-32768).
if (maximum > WEBRTC_SPL_WORD16_MAX) {
maximum = WEBRTC_SPL_WORD16_MAX;
}
return (int16_t)maximum;
}
// Maximum absolute value of word32 vector. C version for generic platforms.
int32_t WebRtcSpl_MaxAbsValueW32C(const int32_t* vector, size_t length) {
// Use uint32_t for the local variables, to accommodate the return value
// of abs(0x80000000), which is 0x80000000.
uint32_t absolute = 0, maximum = 0;
size_t i = 0;
assert(length > 0);
for (i = 0; i < length; i++) {
absolute = abs((int)vector[i]);
if (absolute > maximum) {
maximum = absolute;
}
}
maximum = WEBRTC_SPL_MIN(maximum, WEBRTC_SPL_WORD32_MAX);
return (int32_t)maximum;
}
// Maximum value of word16 vector. C version for generic platforms.
int16_t WebRtcSpl_MaxValueW16C(const int16_t* vector, size_t length) {
int16_t maximum = WEBRTC_SPL_WORD16_MIN;
size_t i = 0;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] > maximum)
maximum = vector[i];
}
return maximum;
}
// Maximum value of word32 vector. C version for generic platforms.
int32_t WebRtcSpl_MaxValueW32C(const int32_t* vector, size_t length) {
int32_t maximum = WEBRTC_SPL_WORD32_MIN;
size_t i = 0;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] > maximum)
maximum = vector[i];
}
return maximum;
}
// Minimum value of word16 vector. C version for generic platforms.
int16_t WebRtcSpl_MinValueW16C(const int16_t* vector, size_t length) {
int16_t minimum = WEBRTC_SPL_WORD16_MAX;
size_t i = 0;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] < minimum)
minimum = vector[i];
}
return minimum;
}
// Minimum value of word32 vector. C version for generic platforms.
int32_t WebRtcSpl_MinValueW32C(const int32_t* vector, size_t length) {
int32_t minimum = WEBRTC_SPL_WORD32_MAX;
size_t i = 0;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] < minimum)
minimum = vector[i];
}
return minimum;
}
// Index of maximum absolute value in a word16 vector.
size_t WebRtcSpl_MaxAbsIndexW16(const int16_t* vector, size_t length) {
// Use type int for local variables, to accomodate the value of abs(-32768).
size_t i = 0, index = 0;
int absolute = 0, maximum = 0;
assert(length > 0);
for (i = 0; i < length; i++) {
absolute = abs((int)vector[i]);
if (absolute > maximum) {
maximum = absolute;
index = i;
}
}
return index;
}
// Index of maximum value in a word16 vector.
size_t WebRtcSpl_MaxIndexW16(const int16_t* vector, size_t length) {
size_t i = 0, index = 0;
int16_t maximum = WEBRTC_SPL_WORD16_MIN;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] > maximum) {
maximum = vector[i];
index = i;
}
}
return index;
}
// Index of maximum value in a word32 vector.
size_t WebRtcSpl_MaxIndexW32(const int32_t* vector, size_t length) {
size_t i = 0, index = 0;
int32_t maximum = WEBRTC_SPL_WORD32_MIN;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] > maximum) {
maximum = vector[i];
index = i;
}
}
return index;
}
// Index of minimum value in a word16 vector.
size_t WebRtcSpl_MinIndexW16(const int16_t* vector, size_t length) {
size_t i = 0, index = 0;
int16_t minimum = WEBRTC_SPL_WORD16_MAX;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] < minimum) {
minimum = vector[i];
index = i;
}
}
return index;
}
// Index of minimum value in a word32 vector.
size_t WebRtcSpl_MinIndexW32(const int32_t* vector, size_t length) {
size_t i = 0, index = 0;
int32_t minimum = WEBRTC_SPL_WORD32_MAX;
assert(length > 0);
for (i = 0; i < length; i++) {
if (vector[i] < minimum) {
minimum = vector[i];
index = i;
}
}
return index;
}

376
webrtc/common_audio/signal_processing/min_max_operations_mips.c Normal file
View File

@ -0,0 +1,376 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the implementation of function
* WebRtcSpl_MaxAbsValueW16()
*
* The description header can be found in signal_processing_library.h.
*
*/
#include <assert.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// Maximum absolute value of word16 vector.
int16_t WebRtcSpl_MaxAbsValueW16_mips(const int16_t* vector, size_t length) {
int32_t totMax = 0;
int32_t tmp32_0, tmp32_1, tmp32_2, tmp32_3;
size_t i, loop_size;
assert(length > 0);
#if defined(MIPS_DSP_R1)
const int32_t* tmpvec32 = (int32_t*)vector;
loop_size = length >> 4;
for (i = 0; i < loop_size; i++) {
__asm__ volatile (
"lw %[tmp32_0], 0(%[tmpvec32]) \n\t"
"lw %[tmp32_1], 4(%[tmpvec32]) \n\t"
"lw %[tmp32_2], 8(%[tmpvec32]) \n\t"
"lw %[tmp32_3], 12(%[tmpvec32]) \n\t"
"absq_s.ph %[tmp32_0], %[tmp32_0] \n\t"
"absq_s.ph %[tmp32_1], %[tmp32_1] \n\t"
"cmp.lt.ph %[totMax], %[tmp32_0] \n\t"
"pick.ph %[totMax], %[tmp32_0], %[totMax] \n\t"
"lw %[tmp32_0], 16(%[tmpvec32]) \n\t"
"absq_s.ph %[tmp32_2], %[tmp32_2] \n\t"
"cmp.lt.ph %[totMax], %[tmp32_1] \n\t"
"pick.ph %[totMax], %[tmp32_1], %[totMax] \n\t"
"lw %[tmp32_1], 20(%[tmpvec32]) \n\t"
"absq_s.ph %[tmp32_3], %[tmp32_3] \n\t"
"cmp.lt.ph %[totMax], %[tmp32_2] \n\t"
"pick.ph %[totMax], %[tmp32_2], %[totMax] \n\t"
"lw %[tmp32_2], 24(%[tmpvec32]) \n\t"
"cmp.lt.ph %[totMax], %[tmp32_3] \n\t"
"pick.ph %[totMax], %[tmp32_3], %[totMax] \n\t"
"lw %[tmp32_3], 28(%[tmpvec32]) \n\t"
"absq_s.ph %[tmp32_0], %[tmp32_0] \n\t"
"absq_s.ph %[tmp32_1], %[tmp32_1] \n\t"
"cmp.lt.ph %[totMax], %[tmp32_0] \n\t"
"pick.ph %[totMax], %[tmp32_0], %[totMax] \n\t"
"absq_s.ph %[tmp32_2], %[tmp32_2] \n\t"
"cmp.lt.ph %[totMax], %[tmp32_1] \n\t"
"pick.ph %[totMax], %[tmp32_1], %[totMax] \n\t"
"absq_s.ph %[tmp32_3], %[tmp32_3] \n\t"
"cmp.lt.ph %[totMax], %[tmp32_2] \n\t"
"pick.ph %[totMax], %[tmp32_2], %[totMax] \n\t"
"cmp.lt.ph %[totMax], %[tmp32_3] \n\t"
"pick.ph %[totMax], %[tmp32_3], %[totMax] \n\t"
"addiu %[tmpvec32], %[tmpvec32], 32 \n\t"
: [tmp32_0] "=&r" (tmp32_0), [tmp32_1] "=&r" (tmp32_1),
[tmp32_2] "=&r" (tmp32_2), [tmp32_3] "=&r" (tmp32_3),
[totMax] "+r" (totMax), [tmpvec32] "+r" (tmpvec32)
:
: "memory"
);
}
__asm__ volatile (
"rotr %[tmp32_0], %[totMax], 16 \n\t"
"cmp.lt.ph %[totMax], %[tmp32_0] \n\t"
"pick.ph %[totMax], %[tmp32_0], %[totMax] \n\t"
"packrl.ph %[totMax], $0, %[totMax] \n\t"
: [tmp32_0] "=&r" (tmp32_0), [totMax] "+r" (totMax)
:
);
loop_size = length & 0xf;
for (i = 0; i < loop_size; i++) {
__asm__ volatile (
"lh %[tmp32_0], 0(%[tmpvec32]) \n\t"
"addiu %[tmpvec32], %[tmpvec32], 2 \n\t"
"absq_s.w %[tmp32_0], %[tmp32_0] \n\t"
"slt %[tmp32_1], %[totMax], %[tmp32_0] \n\t"
"movn %[totMax], %[tmp32_0], %[tmp32_1] \n\t"
: [tmp32_0] "=&r" (tmp32_0), [tmp32_1] "=&r" (tmp32_1),
[tmpvec32] "+r" (tmpvec32), [totMax] "+r" (totMax)
:
: "memory"
);
}
#else // #if defined(MIPS_DSP_R1)
int32_t v16MaxMax = WEBRTC_SPL_WORD16_MAX;
int32_t r, r1, r2, r3;
const int16_t* tmpvector = vector;
loop_size = length >> 4;
for (i = 0; i < loop_size; i++) {
__asm__ volatile (
"lh %[tmp32_0], 0(%[tmpvector]) \n\t"
"lh %[tmp32_1], 2(%[tmpvector]) \n\t"
"lh %[tmp32_2], 4(%[tmpvector]) \n\t"
"lh %[tmp32_3], 6(%[tmpvector]) \n\t"
"abs %[tmp32_0], %[tmp32_0] \n\t"
"abs %[tmp32_1], %[tmp32_1] \n\t"
"abs %[tmp32_2], %[tmp32_2] \n\t"
"abs %[tmp32_3], %[tmp32_3] \n\t"
"slt %[r], %[totMax], %[tmp32_0] \n\t"
"movn %[totMax], %[tmp32_0], %[r] \n\t"
"slt %[r1], %[totMax], %[tmp32_1] \n\t"
"movn %[totMax], %[tmp32_1], %[r1] \n\t"
"slt %[r2], %[totMax], %[tmp32_2] \n\t"
"movn %[totMax], %[tmp32_2], %[r2] \n\t"
"slt %[r3], %[totMax], %[tmp32_3] \n\t"
"movn %[totMax], %[tmp32_3], %[r3] \n\t"
"lh %[tmp32_0], 8(%[tmpvector]) \n\t"
"lh %[tmp32_1], 10(%[tmpvector]) \n\t"
"lh %[tmp32_2], 12(%[tmpvector]) \n\t"
"lh %[tmp32_3], 14(%[tmpvector]) \n\t"
"abs %[tmp32_0], %[tmp32_0] \n\t"
"abs %[tmp32_1], %[tmp32_1] \n\t"
"abs %[tmp32_2], %[tmp32_2] \n\t"
"abs %[tmp32_3], %[tmp32_3] \n\t"
"slt %[r], %[totMax], %[tmp32_0] \n\t"
"movn %[totMax], %[tmp32_0], %[r] \n\t"
"slt %[r1], %[totMax], %[tmp32_1] \n\t"
"movn %[totMax], %[tmp32_1], %[r1] \n\t"
"slt %[r2], %[totMax], %[tmp32_2] \n\t"
"movn %[totMax], %[tmp32_2], %[r2] \n\t"
"slt %[r3], %[totMax], %[tmp32_3] \n\t"
"movn %[totMax], %[tmp32_3], %[r3] \n\t"
"lh %[tmp32_0], 16(%[tmpvector]) \n\t"
"lh %[tmp32_1], 18(%[tmpvector]) \n\t"
"lh %[tmp32_2], 20(%[tmpvector]) \n\t"
"lh %[tmp32_3], 22(%[tmpvector]) \n\t"
"abs %[tmp32_0], %[tmp32_0] \n\t"
"abs %[tmp32_1], %[tmp32_1] \n\t"
"abs %[tmp32_2], %[tmp32_2] \n\t"
"abs %[tmp32_3], %[tmp32_3] \n\t"
"slt %[r], %[totMax], %[tmp32_0] \n\t"
"movn %[totMax], %[tmp32_0], %[r] \n\t"
"slt %[r1], %[totMax], %[tmp32_1] \n\t"
"movn %[totMax], %[tmp32_1], %[r1] \n\t"
"slt %[r2], %[totMax], %[tmp32_2] \n\t"
"movn %[totMax], %[tmp32_2], %[r2] \n\t"
"slt %[r3], %[totMax], %[tmp32_3] \n\t"
"movn %[totMax], %[tmp32_3], %[r3] \n\t"
"lh %[tmp32_0], 24(%[tmpvector]) \n\t"
"lh %[tmp32_1], 26(%[tmpvector]) \n\t"
"lh %[tmp32_2], 28(%[tmpvector]) \n\t"
"lh %[tmp32_3], 30(%[tmpvector]) \n\t"
"abs %[tmp32_0], %[tmp32_0] \n\t"
"abs %[tmp32_1], %[tmp32_1] \n\t"
"abs %[tmp32_2], %[tmp32_2] \n\t"
"abs %[tmp32_3], %[tmp32_3] \n\t"
"slt %[r], %[totMax], %[tmp32_0] \n\t"
"movn %[totMax], %[tmp32_0], %[r] \n\t"
"slt %[r1], %[totMax], %[tmp32_1] \n\t"
"movn %[totMax], %[tmp32_1], %[r1] \n\t"
"slt %[r2], %[totMax], %[tmp32_2] \n\t"
"movn %[totMax], %[tmp32_2], %[r2] \n\t"
"slt %[r3], %[totMax], %[tmp32_3] \n\t"
"movn %[totMax], %[tmp32_3], %[r3] \n\t"
"addiu %[tmpvector], %[tmpvector], 32 \n\t"
: [tmp32_0] "=&r" (tmp32_0), [tmp32_1] "=&r" (tmp32_1),
[tmp32_2] "=&r" (tmp32_2), [tmp32_3] "=&r" (tmp32_3),
[totMax] "+r" (totMax), [r] "=&r" (r), [tmpvector] "+r" (tmpvector),
[r1] "=&r" (r1), [r2] "=&r" (r2), [r3] "=&r" (r3)
:
: "memory"
);
}
loop_size = length & 0xf;
for (i = 0; i < loop_size; i++) {
__asm__ volatile (
"lh %[tmp32_0], 0(%[tmpvector]) \n\t"
"addiu %[tmpvector], %[tmpvector], 2 \n\t"
"abs %[tmp32_0], %[tmp32_0] \n\t"
"slt %[tmp32_1], %[totMax], %[tmp32_0] \n\t"
"movn %[totMax], %[tmp32_0], %[tmp32_1] \n\t"
: [tmp32_0] "=&r" (tmp32_0), [tmp32_1] "=&r" (tmp32_1),
[tmpvector] "+r" (tmpvector), [totMax] "+r" (totMax)
:
: "memory"
);
}
__asm__ volatile (
"slt %[r], %[v16MaxMax], %[totMax] \n\t"
"movn %[totMax], %[v16MaxMax], %[r] \n\t"
: [totMax] "+r" (totMax), [r] "=&r" (r)
: [v16MaxMax] "r" (v16MaxMax)
);
#endif // #if defined(MIPS_DSP_R1)
return (int16_t)totMax;
}
#if defined(MIPS_DSP_R1_LE)
// Maximum absolute value of word32 vector. Version for MIPS platform.
int32_t WebRtcSpl_MaxAbsValueW32_mips(const int32_t* vector, size_t length) {
// Use uint32_t for the local variables, to accommodate the return value
// of abs(0x80000000), which is 0x80000000.
uint32_t absolute = 0, maximum = 0;
int tmp1 = 0, max_value = 0x7fffffff;
assert(length > 0);
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"1: \n\t"
"lw %[absolute], 0(%[vector]) \n\t"
"absq_s.w %[absolute], %[absolute] \n\t"
"addiu %[length], %[length], -1 \n\t"
"slt %[tmp1], %[maximum], %[absolute] \n\t"
"movn %[maximum], %[absolute], %[tmp1] \n\t"
"bgtz %[length], 1b \n\t"
" addiu %[vector], %[vector], 4 \n\t"
"slt %[tmp1], %[max_value], %[maximum] \n\t"
"movn %[maximum], %[max_value], %[tmp1] \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [maximum] "+r" (maximum), [absolute] "+r" (absolute)
: [vector] "r" (vector), [length] "r" (length), [max_value] "r" (max_value)
: "memory"
);
return (int32_t)maximum;
}
#endif // #if defined(MIPS_DSP_R1_LE)
// Maximum value of word16 vector. Version for MIPS platform.
int16_t WebRtcSpl_MaxValueW16_mips(const int16_t* vector, size_t length) {
int16_t maximum = WEBRTC_SPL_WORD16_MIN;
int tmp1;
int16_t value;
assert(length > 0);
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"1: \n\t"
"lh %[value], 0(%[vector]) \n\t"
"addiu %[length], %[length], -1 \n\t"
"slt %[tmp1], %[maximum], %[value] \n\t"
"movn %[maximum], %[value], %[tmp1] \n\t"
"bgtz %[length], 1b \n\t"
" addiu %[vector], %[vector], 2 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [maximum] "+r" (maximum), [value] "=&r" (value)
: [vector] "r" (vector), [length] "r" (length)
: "memory"
);
return maximum;
}
// Maximum value of word32 vector. Version for MIPS platform.
int32_t WebRtcSpl_MaxValueW32_mips(const int32_t* vector, size_t length) {
int32_t maximum = WEBRTC_SPL_WORD32_MIN;
int tmp1, value;
assert(length > 0);
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"1: \n\t"
"lw %[value], 0(%[vector]) \n\t"
"addiu %[length], %[length], -1 \n\t"
"slt %[tmp1], %[maximum], %[value] \n\t"
"movn %[maximum], %[value], %[tmp1] \n\t"
"bgtz %[length], 1b \n\t"
" addiu %[vector], %[vector], 4 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [maximum] "+r" (maximum), [value] "=&r" (value)
: [vector] "r" (vector), [length] "r" (length)
: "memory"
);
return maximum;
}
// Minimum value of word16 vector. Version for MIPS platform.
int16_t WebRtcSpl_MinValueW16_mips(const int16_t* vector, size_t length) {
int16_t minimum = WEBRTC_SPL_WORD16_MAX;
int tmp1;
int16_t value;
assert(length > 0);
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"1: \n\t"
"lh %[value], 0(%[vector]) \n\t"
"addiu %[length], %[length], -1 \n\t"
"slt %[tmp1], %[value], %[minimum] \n\t"
"movn %[minimum], %[value], %[tmp1] \n\t"
"bgtz %[length], 1b \n\t"
" addiu %[vector], %[vector], 2 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [minimum] "+r" (minimum), [value] "=&r" (value)
: [vector] "r" (vector), [length] "r" (length)
: "memory"
);
return minimum;
}
// Minimum value of word32 vector. Version for MIPS platform.
int32_t WebRtcSpl_MinValueW32_mips(const int32_t* vector, size_t length) {
int32_t minimum = WEBRTC_SPL_WORD32_MAX;
int tmp1, value;
assert(length > 0);
__asm__ volatile (
".set push \n\t"
".set noreorder \n\t"
"1: \n\t"
"lw %[value], 0(%[vector]) \n\t"
"addiu %[length], %[length], -1 \n\t"
"slt %[tmp1], %[value], %[minimum] \n\t"
"movn %[minimum], %[value], %[tmp1] \n\t"
"bgtz %[length], 1b \n\t"
" addiu %[vector], %[vector], 4 \n\t"
".set pop \n\t"
: [tmp1] "=&r" (tmp1), [minimum] "+r" (minimum), [value] "=&r" (value)
: [vector] "r" (vector), [length] "r" (length)
: "memory"
);
return minimum;
}

283
webrtc/common_audio/signal_processing/min_max_operations_neon.c Normal file
View File

@ -0,0 +1,283 @@
/*
* Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <arm_neon.h>
#include <assert.h>
#include <stdlib.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// Maximum absolute value of word16 vector. C version for generic platforms.
int16_t WebRtcSpl_MaxAbsValueW16Neon(const int16_t* vector, size_t length) {
int absolute = 0, maximum = 0;
assert(length > 0);
const int16_t* p_start = vector;
size_t rest = length & 7;
const int16_t* p_end = vector + length - rest;
int16x8_t v;
uint16x8_t max_qv;
max_qv = vdupq_n_u16(0);
while (p_start < p_end) {
v = vld1q_s16(p_start);
// Note vabs doesn't change the value of -32768.
v = vabsq_s16(v);
// Use u16 so we don't lose the value -32768.
max_qv = vmaxq_u16(max_qv, vreinterpretq_u16_s16(v));
p_start += 8;
}
#ifdef WEBRTC_ARCH_ARM64
maximum = (int)vmaxvq_u16(max_qv);
#else
uint16x4_t max_dv;
max_dv = vmax_u16(vget_low_u16(max_qv), vget_high_u16(max_qv));
max_dv = vpmax_u16(max_dv, max_dv);
max_dv = vpmax_u16(max_dv, max_dv);
maximum = (int)vget_lane_u16(max_dv, 0);
#endif
p_end = vector + length;
while (p_start < p_end) {
absolute = abs((int)(*p_start));
if (absolute > maximum) {
maximum = absolute;
}
p_start++;
}
// Guard the case for abs(-32768).
if (maximum > WEBRTC_SPL_WORD16_MAX) {
maximum = WEBRTC_SPL_WORD16_MAX;
}
return (int16_t)maximum;
}
// Maximum absolute value of word32 vector. NEON intrinsics version for
// ARM 32-bit/64-bit platforms.
int32_t WebRtcSpl_MaxAbsValueW32Neon(const int32_t* vector, size_t length) {
// Use uint32_t for the local variables, to accommodate the return value
// of abs(0x80000000), which is 0x80000000.
uint32_t absolute = 0, maximum = 0;
size_t i = 0;
size_t residual = length & 0x7;
assert(length > 0);
const int32_t* p_start = vector;
uint32x4_t max32x4_0 = vdupq_n_u32(0);
uint32x4_t max32x4_1 = vdupq_n_u32(0);
// First part, unroll the loop 8 times.
for (i = 0; i < length - residual; i += 8) {
int32x4_t in32x4_0 = vld1q_s32(p_start);
p_start += 4;
int32x4_t in32x4_1 = vld1q_s32(p_start);
p_start += 4;
in32x4_0 = vabsq_s32(in32x4_0);
in32x4_1 = vabsq_s32(in32x4_1);
// vabs doesn't change the value of 0x80000000.
// Use u32 so we don't lose the value 0x80000000.
max32x4_0 = vmaxq_u32(max32x4_0, vreinterpretq_u32_s32(in32x4_0));
max32x4_1 = vmaxq_u32(max32x4_1, vreinterpretq_u32_s32(in32x4_1));
}
uint32x4_t max32x4 = vmaxq_u32(max32x4_0, max32x4_1);
#if defined(WEBRTC_ARCH_ARM64)
maximum = vmaxvq_u32(max32x4);
#else
uint32x2_t max32x2 = vmax_u32(vget_low_u32(max32x4), vget_high_u32(max32x4));
max32x2 = vpmax_u32(max32x2, max32x2);
maximum = vget_lane_u32(max32x2, 0);
#endif
// Second part, do the remaining iterations (if any).
for (i = residual; i > 0; i--) {
absolute = abs((int)(*p_start));
if (absolute > maximum) {
maximum = absolute;
}
p_start++;
}
// Guard against the case for 0x80000000.
maximum = WEBRTC_SPL_MIN(maximum, WEBRTC_SPL_WORD32_MAX);
return (int32_t)maximum;
}
// Maximum value of word16 vector. NEON intrinsics version for
// ARM 32-bit/64-bit platforms.
int16_t WebRtcSpl_MaxValueW16Neon(const int16_t* vector, size_t length) {
int16_t maximum = WEBRTC_SPL_WORD16_MIN;
size_t i = 0;
size_t residual = length & 0x7;
assert(length > 0);
const int16_t* p_start = vector;
int16x8_t max16x8 = vdupq_n_s16(WEBRTC_SPL_WORD16_MIN);
// First part, unroll the loop 8 times.
for (i = 0; i < length - residual; i += 8) {
int16x8_t in16x8 = vld1q_s16(p_start);
max16x8 = vmaxq_s16(max16x8, in16x8);
p_start += 8;
}
#if defined(WEBRTC_ARCH_ARM64)
maximum = vmaxvq_s16(max16x8);
#else
int16x4_t max16x4 = vmax_s16(vget_low_s16(max16x8), vget_high_s16(max16x8));
max16x4 = vpmax_s16(max16x4, max16x4);
max16x4 = vpmax_s16(max16x4, max16x4);
maximum = vget_lane_s16(max16x4, 0);
#endif
// Second part, do the remaining iterations (if any).
for (i = residual; i > 0; i--) {
if (*p_start > maximum)
maximum = *p_start;
p_start++;
}
return maximum;
}
// Maximum value of word32 vector. NEON intrinsics version for
// ARM 32-bit/64-bit platforms.
int32_t WebRtcSpl_MaxValueW32Neon(const int32_t* vector, size_t length) {
int32_t maximum = WEBRTC_SPL_WORD32_MIN;
size_t i = 0;
size_t residual = length & 0x7;
assert(length > 0);
const int32_t* p_start = vector;
int32x4_t max32x4_0 = vdupq_n_s32(WEBRTC_SPL_WORD32_MIN);
int32x4_t max32x4_1 = vdupq_n_s32(WEBRTC_SPL_WORD32_MIN);
// First part, unroll the loop 8 times.
for (i = 0; i < length - residual; i += 8) {
int32x4_t in32x4_0 = vld1q_s32(p_start);
p_start += 4;
int32x4_t in32x4_1 = vld1q_s32(p_start);
p_start += 4;
max32x4_0 = vmaxq_s32(max32x4_0, in32x4_0);
max32x4_1 = vmaxq_s32(max32x4_1, in32x4_1);
}
int32x4_t max32x4 = vmaxq_s32(max32x4_0, max32x4_1);
#if defined(WEBRTC_ARCH_ARM64)
maximum = vmaxvq_s32(max32x4);
#else
int32x2_t max32x2 = vmax_s32(vget_low_s32(max32x4), vget_high_s32(max32x4));
max32x2 = vpmax_s32(max32x2, max32x2);
maximum = vget_lane_s32(max32x2, 0);
#endif
// Second part, do the remaining iterations (if any).
for (i = residual; i > 0; i--) {
if (*p_start > maximum)
maximum = *p_start;
p_start++;
}
return maximum;
}
// Minimum value of word16 vector. NEON intrinsics version for
// ARM 32-bit/64-bit platforms.
int16_t WebRtcSpl_MinValueW16Neon(const int16_t* vector, size_t length) {
int16_t minimum = WEBRTC_SPL_WORD16_MAX;
size_t i = 0;
size_t residual = length & 0x7;
assert(length > 0);
const int16_t* p_start = vector;
int16x8_t min16x8 = vdupq_n_s16(WEBRTC_SPL_WORD16_MAX);
// First part, unroll the loop 8 times.
for (i = 0; i < length - residual; i += 8) {
int16x8_t in16x8 = vld1q_s16(p_start);
min16x8 = vminq_s16(min16x8, in16x8);
p_start += 8;
}
#if defined(WEBRTC_ARCH_ARM64)
minimum = vminvq_s16(min16x8);
#else
int16x4_t min16x4 = vmin_s16(vget_low_s16(min16x8), vget_high_s16(min16x8));
min16x4 = vpmin_s16(min16x4, min16x4);
min16x4 = vpmin_s16(min16x4, min16x4);
minimum = vget_lane_s16(min16x4, 0);
#endif
// Second part, do the remaining iterations (if any).
for (i = residual; i > 0; i--) {
if (*p_start < minimum)
minimum = *p_start;
p_start++;
}
return minimum;
}
// Minimum value of word32 vector. NEON intrinsics version for
// ARM 32-bit/64-bit platforms.
int32_t WebRtcSpl_MinValueW32Neon(const int32_t* vector, size_t length) {
int32_t minimum = WEBRTC_SPL_WORD32_MAX;
size_t i = 0;
size_t residual = length & 0x7;
assert(length > 0);
const int32_t* p_start = vector;
int32x4_t min32x4_0 = vdupq_n_s32(WEBRTC_SPL_WORD32_MAX);
int32x4_t min32x4_1 = vdupq_n_s32(WEBRTC_SPL_WORD32_MAX);
// First part, unroll the loop 8 times.
for (i = 0; i < length - residual; i += 8) {
int32x4_t in32x4_0 = vld1q_s32(p_start);
p_start += 4;
int32x4_t in32x4_1 = vld1q_s32(p_start);
p_start += 4;
min32x4_0 = vminq_s32(min32x4_0, in32x4_0);
min32x4_1 = vminq_s32(min32x4_1, in32x4_1);
}
int32x4_t min32x4 = vminq_s32(min32x4_0, min32x4_1);
#if defined(WEBRTC_ARCH_ARM64)
minimum = vminvq_s32(min32x4);
#else
int32x2_t min32x2 = vmin_s32(vget_low_s32(min32x4), vget_high_s32(min32x4));
min32x2 = vpmin_s32(min32x2, min32x2);
minimum = vget_lane_s32(min32x2, 0);
#endif
// Second part, do the remaining iterations (if any).
for (i = residual; i > 0; i--) {
if (*p_start < minimum)
minimum = *p_start;
p_start++;
}
return minimum;
}

115
webrtc/common_audio/signal_processing/randomization_functions.c Normal file
View File

@ -0,0 +1,115 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains implementations of the randomization functions
* WebRtcSpl_RandU()
* WebRtcSpl_RandN()
* WebRtcSpl_RandUArray()
*
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
static const uint32_t kMaxSeedUsed = 0x80000000;
static const int16_t kRandNTable[] = {
9178, -7260, 40, 10189, 4894, -3531, -13779, 14764,
-4008, -8884, -8990, 1008, 7368, 5184, 3251, -5817,
-9786, 5963, 1770, 8066, -7135, 10772, -2298, 1361,
6484, 2241, -8633, 792, 199, -3344, 6553, -10079,
-15040, 95, 11608, -12469, 14161, -4176, 2476, 6403,
13685, -16005, 6646, 2239, 10916, -3004, -602, -3141,
2142, 14144, -5829, 5305, 8209, 4713, 2697, -5112,
16092, -1210, -2891, -6631, -5360, -11878, -6781, -2739,
-6392, 536, 10923, 10872, 5059, -4748, -7770, 5477,
38, -1025, -2892, 1638, 6304, 14375, -11028, 1553,
-1565, 10762, -393, 4040, 5257, 12310, 6554, -4799,
4899, -6354, 1603, -1048, -2220, 8247, -186, -8944,
-12004, 2332, 4801, -4933, 6371, 131, 8614, -5927,
-8287, -22760, 4033, -15162, 3385, 3246, 3153, -5250,
3766, 784, 6494, -62, 3531, -1582, 15572, 662,
-3952, -330, -3196, 669, 7236, -2678, -6569, 23319,
-8645, -741, 14830, -15976, 4903, 315, -11342, 10311,
1858, -7777, 2145, 5436, 5677, -113, -10033, 826,
-1353, 17210, 7768, 986, -1471, 8291, -4982, 8207,
-14911, -6255, -2449, -11881, -7059, -11703, -4338, 8025,
7538, -2823, -12490, 9470, -1613, -2529, -10092, -7807,
9480, 6970, -12844, 5123, 3532, 4816, 4803, -8455,
-5045, 14032, -4378, -1643, 5756, -11041, -2732, -16618,
-6430, -18375, -3320, 6098, 5131, -4269, -8840, 2482,
-7048, 1547, -21890, -6505, -7414, -424, -11722, 7955,
1653, -17299, 1823, 473, -9232, 3337, 1111, 873,
4018, -8982, 9889, 3531, -11763, -3799, 7373, -4539,
3231, 7054, -8537, 7616, 6244, 16635, 447, -2915,
13967, 705, -2669, -1520, -1771, -16188, 5956, 5117,
6371, -9936, -1448, 2480, 5128, 7550, -8130, 5236,
8213, -6443, 7707, -1950, -13811, 7218, 7031, -3883,
67, 5731, -2874, 13480, -3743, 9298, -3280, 3552,
-4425, -18, -3785, -9988, -5357, 5477, -11794, 2117,
1416, -9935, 3376, 802, -5079, -8243, 12652, 66,
3653, -2368, 6781, -21895, -7227, 2487, 7839, -385,
6646, -7016, -4658, 5531, -1705, 834, 129, 3694,
-1343, 2238, -22640, -6417, -11139, 11301, -2945, -3494,
-5626, 185, -3615, -2041, -7972, -3106, -60, -23497,
-1566, 17064, 3519, 2518, 304, -6805, -10269, 2105,
1936, -426, -736, -8122, -1467, 4238, -6939, -13309,
360, 7402, -7970, 12576, 3287, 12194, -6289, -16006,
9171, 4042, -9193, 9123, -2512, 6388, -4734, -8739,
1028, -5406, -1696, 5889, -666, -4736, 4971, 3565,
9362, -6292, 3876, -3652, -19666, 7523, -4061, 391,
-11773, 7502, -3763, 4929, -9478, 13278, 2805, 4496,
7814, 16419, 12455, -14773, 2127, -2746, 3763, 4847,
3698, 6978, 4751, -6957, -3581, -45, 6252, 1513,
-4797, -7925, 11270, 16188, -2359, -5269, 9376, -10777,
7262, 20031, -6515, -2208, -5353, 8085, -1341, -1303,
7333, 5576, 3625, 5763, -7931, 9833, -3371, -10305,
6534, -13539, -9971, 997, 8464, -4064, -1495, 1857,
13624, 5458, 9490, -11086, -4524, 12022, -550, -198,
408, -8455, -7068, 10289, 9712, -3366, 9028, -7621,
-5243, 2362, 6909, 4672, -4933, -1799, 4709, -4563,
-62, -566, 1624, -7010, 14730, -17791, -3697, -2344,
-1741, 7099, -9509, -6855, -1989, 3495, -2289, 2031,
12784, 891, 14189, -3963, -5683, 421, -12575, 1724,
-12682, -5970, -8169, 3143, -1824, -5488, -5130, 8536,
12799, 794, 5738, 3459, -11689, -258, -3738, -3775,
-8742, 2333, 8312, -9383, 10331, 13119, 8398, 10644,
-19433, -6446, -16277, -11793, 16284, 9345, 15222, 15834,
2009, -7349, 130, -14547, 338, -5998, 3337, 21492,
2406, 7703, -951, 11196, -564, 3406, 2217, 4806,
2374, -5797, 11839, 8940, -11874, 18213, 2855, 10492
};
static uint32_t IncreaseSeed(uint32_t* seed) {
seed[0] = (seed[0] * ((int32_t)69069) + 1) & (kMaxSeedUsed - 1);
return seed[0];
}
int16_t WebRtcSpl_RandU(uint32_t* seed) {
return (int16_t)(IncreaseSeed(seed) >> 16);
}
int16_t WebRtcSpl_RandN(uint32_t* seed) {
return kRandNTable[IncreaseSeed(seed) >> 23];
}
// Creates an array of uniformly distributed variables.
int16_t WebRtcSpl_RandUArray(int16_t* vector,
int16_t vector_length,
uint32_t* seed) {
int i;
for (i = 0; i < vector_length; i++) {
vector[i] = WebRtcSpl_RandU(seed);
}
return vector_length;
}

102
webrtc/common_audio/signal_processing/real_fft.c Normal file
View File

@ -0,0 +1,102 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/signal_processing/include/real_fft.h"
#include <stdlib.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
struct RealFFT {
int order;
};
struct RealFFT* WebRtcSpl_CreateRealFFT(int order) {
struct RealFFT* self = NULL;
if (order > kMaxFFTOrder || order < 0) {
return NULL;
}
self = malloc(sizeof(struct RealFFT));
if (self == NULL) {
return NULL;
}
self->order = order;
return self;
}
void WebRtcSpl_FreeRealFFT(struct RealFFT* self) {
if (self != NULL) {
free(self);
}
}
// The C version FFT functions (i.e. WebRtcSpl_RealForwardFFT and
// WebRtcSpl_RealInverseFFT) are real-valued FFT wrappers for complex-valued
// FFT implementation in SPL.
int WebRtcSpl_RealForwardFFT(struct RealFFT* self,
const int16_t* real_data_in,
int16_t* complex_data_out) {
int i = 0;
int j = 0;
int result = 0;
int n = 1 << self->order;
// The complex-value FFT implementation needs a buffer to hold 2^order
// 16-bit COMPLEX numbers, for both time and frequency data.
int16_t complex_buffer[2 << kMaxFFTOrder];
// Insert zeros to the imaginary parts for complex forward FFT input.
for (i = 0, j = 0; i < n; i += 1, j += 2) {
complex_buffer[j] = real_data_in[i];
complex_buffer[j + 1] = 0;
};
WebRtcSpl_ComplexBitReverse(complex_buffer, self->order);
result = WebRtcSpl_ComplexFFT(complex_buffer, self->order, 1);
// For real FFT output, use only the first N + 2 elements from
// complex forward FFT.
memcpy(complex_data_out, complex_buffer, sizeof(int16_t) * (n + 2));
return result;
}
int WebRtcSpl_RealInverseFFT(struct RealFFT* self,
const int16_t* complex_data_in,
int16_t* real_data_out) {
int i = 0;
int j = 0;
int result = 0;
int n = 1 << self->order;
// Create the buffer specific to complex-valued FFT implementation.
int16_t complex_buffer[2 << kMaxFFTOrder];
// For n-point FFT, first copy the first n + 2 elements into complex
// FFT, then construct the remaining n - 2 elements by real FFT's
// conjugate-symmetric properties.
memcpy(complex_buffer, complex_data_in, sizeof(int16_t) * (n + 2));
for (i = n + 2; i < 2 * n; i += 2) {
complex_buffer[i] = complex_data_in[2 * n - i];
complex_buffer[i + 1] = -complex_data_in[2 * n - i + 1];
}
WebRtcSpl_ComplexBitReverse(complex_buffer, self->order);
result = WebRtcSpl_ComplexIFFT(complex_buffer, self->order, 1);
// Strip out the imaginary parts of the complex inverse FFT output.
for (i = 0, j = 0; i < n; i += 1, j += 2) {
real_data_out[i] = complex_buffer[j];
}
return result;
}

59
webrtc/common_audio/signal_processing/refl_coef_to_lpc.c Normal file
View File

@ -0,0 +1,59 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_ReflCoefToLpc().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_ReflCoefToLpc(const int16_t *k, int use_order, int16_t *a)
{
int16_t any[WEBRTC_SPL_MAX_LPC_ORDER + 1];
int16_t *aptr, *aptr2, *anyptr;
const int16_t *kptr;
int m, i;
kptr = k;
*a = 4096; // i.e., (Word16_MAX >> 3)+1.
*any = *a;
a[1] = *k >> 3;
for (m = 1; m < use_order; m++)
{
kptr++;
aptr = a;
aptr++;
aptr2 = &a[m];
anyptr = any;
anyptr++;
any[m + 1] = *kptr >> 3;
for (i = 0; i < m; i++)
{
*anyptr = *aptr + (int16_t)((*aptr2 * *kptr) >> 15);
anyptr++;
aptr++;
aptr2--;
}
aptr = a;
anyptr = any;
for (i = 0; i < (m + 2); i++)
{
*aptr = *anyptr;
aptr++;
anyptr++;
}
}
}

505
webrtc/common_audio/signal_processing/resample.c Normal file
View File

@ -0,0 +1,505 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the resampling functions for 22 kHz.
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/common_audio/signal_processing/resample_by_2_internal.h"
// Declaration of internally used functions
static void WebRtcSpl_32khzTo22khzIntToShort(const int32_t *In, int16_t *Out,
int32_t K);
void WebRtcSpl_32khzTo22khzIntToInt(const int32_t *In, int32_t *Out,
int32_t K);
// interpolation coefficients
static const int16_t kCoefficients32To22[5][9] = {
{127, -712, 2359, -6333, 23456, 16775, -3695, 945, -154},
{-39, 230, -830, 2785, 32366, -2324, 760, -218, 38},
{117, -663, 2222, -6133, 26634, 13070, -3174, 831, -137},
{-77, 457, -1677, 5958, 31175, -4136, 1405, -408, 71},
{ 98, -560, 1900, -5406, 29240, 9423, -2480, 663, -110}
};
//////////////////////
// 22 kHz -> 16 kHz //
//////////////////////
// number of subblocks; options: 1, 2, 4, 5, 10
#define SUB_BLOCKS_22_16 5
// 22 -> 16 resampler
void WebRtcSpl_Resample22khzTo16khz(const int16_t* in, int16_t* out,
WebRtcSpl_State22khzTo16khz* state, int32_t* tmpmem)
{
int k;
// process two blocks of 10/SUB_BLOCKS_22_16 ms (to reduce temp buffer size)
for (k = 0; k < SUB_BLOCKS_22_16; k++)
{
///// 22 --> 44 /////
// int16_t in[220/SUB_BLOCKS_22_16]
// int32_t out[440/SUB_BLOCKS_22_16]
/////
WebRtcSpl_UpBy2ShortToInt(in, 220 / SUB_BLOCKS_22_16, tmpmem + 16, state->S_22_44);
///// 44 --> 32 /////
// int32_t in[440/SUB_BLOCKS_22_16]
// int32_t out[320/SUB_BLOCKS_22_16]
/////
// copy state to and from input array
tmpmem[8] = state->S_44_32[0];
tmpmem[9] = state->S_44_32[1];
tmpmem[10] = state->S_44_32[2];
tmpmem[11] = state->S_44_32[3];
tmpmem[12] = state->S_44_32[4];
tmpmem[13] = state->S_44_32[5];
tmpmem[14] = state->S_44_32[6];
tmpmem[15] = state->S_44_32[7];
state->S_44_32[0] = tmpmem[440 / SUB_BLOCKS_22_16 + 8];
state->S_44_32[1] = tmpmem[440 / SUB_BLOCKS_22_16 + 9];
state->S_44_32[2] = tmpmem[440 / SUB_BLOCKS_22_16 + 10];
state->S_44_32[3] = tmpmem[440 / SUB_BLOCKS_22_16 + 11];
state->S_44_32[4] = tmpmem[440 / SUB_BLOCKS_22_16 + 12];
state->S_44_32[5] = tmpmem[440 / SUB_BLOCKS_22_16 + 13];
state->S_44_32[6] = tmpmem[440 / SUB_BLOCKS_22_16 + 14];
state->S_44_32[7] = tmpmem[440 / SUB_BLOCKS_22_16 + 15];
WebRtcSpl_Resample44khzTo32khz(tmpmem + 8, tmpmem, 40 / SUB_BLOCKS_22_16);
///// 32 --> 16 /////
// int32_t in[320/SUB_BLOCKS_22_16]
// int32_t out[160/SUB_BLOCKS_22_16]
/////
WebRtcSpl_DownBy2IntToShort(tmpmem, 320 / SUB_BLOCKS_22_16, out, state->S_32_16);
// move input/output pointers 10/SUB_BLOCKS_22_16 ms seconds ahead
in += 220 / SUB_BLOCKS_22_16;
out += 160 / SUB_BLOCKS_22_16;
}
}
// initialize state of 22 -> 16 resampler
void WebRtcSpl_ResetResample22khzTo16khz(WebRtcSpl_State22khzTo16khz* state)
{
int k;
for (k = 0; k < 8; k++)
{
state->S_22_44[k] = 0;
state->S_44_32[k] = 0;
state->S_32_16[k] = 0;
}
}
//////////////////////
// 16 kHz -> 22 kHz //
//////////////////////
// number of subblocks; options: 1, 2, 4, 5, 10
#define SUB_BLOCKS_16_22 4
// 16 -> 22 resampler
void WebRtcSpl_Resample16khzTo22khz(const int16_t* in, int16_t* out,
WebRtcSpl_State16khzTo22khz* state, int32_t* tmpmem)
{
int k;
// process two blocks of 10/SUB_BLOCKS_16_22 ms (to reduce temp buffer size)
for (k = 0; k < SUB_BLOCKS_16_22; k++)
{
///// 16 --> 32 /////
// int16_t in[160/SUB_BLOCKS_16_22]
// int32_t out[320/SUB_BLOCKS_16_22]
/////
WebRtcSpl_UpBy2ShortToInt(in, 160 / SUB_BLOCKS_16_22, tmpmem + 8, state->S_16_32);
///// 32 --> 22 /////
// int32_t in[320/SUB_BLOCKS_16_22]
// int32_t out[220/SUB_BLOCKS_16_22]
/////
// copy state to and from input array
tmpmem[0] = state->S_32_22[0];
tmpmem[1] = state->S_32_22[1];
tmpmem[2] = state->S_32_22[2];
tmpmem[3] = state->S_32_22[3];
tmpmem[4] = state->S_32_22[4];
tmpmem[5] = state->S_32_22[5];
tmpmem[6] = state->S_32_22[6];
tmpmem[7] = state->S_32_22[7];
state->S_32_22[0] = tmpmem[320 / SUB_BLOCKS_16_22];
state->S_32_22[1] = tmpmem[320 / SUB_BLOCKS_16_22 + 1];
state->S_32_22[2] = tmpmem[320 / SUB_BLOCKS_16_22 + 2];
state->S_32_22[3] = tmpmem[320 / SUB_BLOCKS_16_22 + 3];
state->S_32_22[4] = tmpmem[320 / SUB_BLOCKS_16_22 + 4];
state->S_32_22[5] = tmpmem[320 / SUB_BLOCKS_16_22 + 5];
state->S_32_22[6] = tmpmem[320 / SUB_BLOCKS_16_22 + 6];
state->S_32_22[7] = tmpmem[320 / SUB_BLOCKS_16_22 + 7];
WebRtcSpl_32khzTo22khzIntToShort(tmpmem, out, 20 / SUB_BLOCKS_16_22);
// move input/output pointers 10/SUB_BLOCKS_16_22 ms seconds ahead
in += 160 / SUB_BLOCKS_16_22;
out += 220 / SUB_BLOCKS_16_22;
}
}
// initialize state of 16 -> 22 resampler
void WebRtcSpl_ResetResample16khzTo22khz(WebRtcSpl_State16khzTo22khz* state)
{
int k;
for (k = 0; k < 8; k++)
{
state->S_16_32[k] = 0;
state->S_32_22[k] = 0;
}
}
//////////////////////
// 22 kHz -> 8 kHz //
//////////////////////
// number of subblocks; options: 1, 2, 5, 10
#define SUB_BLOCKS_22_8 2
// 22 -> 8 resampler
void WebRtcSpl_Resample22khzTo8khz(const int16_t* in, int16_t* out,
WebRtcSpl_State22khzTo8khz* state, int32_t* tmpmem)
{
int k;
// process two blocks of 10/SUB_BLOCKS_22_8 ms (to reduce temp buffer size)
for (k = 0; k < SUB_BLOCKS_22_8; k++)
{
///// 22 --> 22 lowpass /////
// int16_t in[220/SUB_BLOCKS_22_8]
// int32_t out[220/SUB_BLOCKS_22_8]
/////
WebRtcSpl_LPBy2ShortToInt(in, 220 / SUB_BLOCKS_22_8, tmpmem + 16, state->S_22_22);
///// 22 --> 16 /////
// int32_t in[220/SUB_BLOCKS_22_8]
// int32_t out[160/SUB_BLOCKS_22_8]
/////
// copy state to and from input array
tmpmem[8] = state->S_22_16[0];
tmpmem[9] = state->S_22_16[1];
tmpmem[10] = state->S_22_16[2];
tmpmem[11] = state->S_22_16[3];
tmpmem[12] = state->S_22_16[4];
tmpmem[13] = state->S_22_16[5];
tmpmem[14] = state->S_22_16[6];
tmpmem[15] = state->S_22_16[7];
state->S_22_16[0] = tmpmem[220 / SUB_BLOCKS_22_8 + 8];
state->S_22_16[1] = tmpmem[220 / SUB_BLOCKS_22_8 + 9];
state->S_22_16[2] = tmpmem[220 / SUB_BLOCKS_22_8 + 10];
state->S_22_16[3] = tmpmem[220 / SUB_BLOCKS_22_8 + 11];
state->S_22_16[4] = tmpmem[220 / SUB_BLOCKS_22_8 + 12];
state->S_22_16[5] = tmpmem[220 / SUB_BLOCKS_22_8 + 13];
state->S_22_16[6] = tmpmem[220 / SUB_BLOCKS_22_8 + 14];
state->S_22_16[7] = tmpmem[220 / SUB_BLOCKS_22_8 + 15];
WebRtcSpl_Resample44khzTo32khz(tmpmem + 8, tmpmem, 20 / SUB_BLOCKS_22_8);
///// 16 --> 8 /////
// int32_t in[160/SUB_BLOCKS_22_8]
// int32_t out[80/SUB_BLOCKS_22_8]
/////
WebRtcSpl_DownBy2IntToShort(tmpmem, 160 / SUB_BLOCKS_22_8, out, state->S_16_8);
// move input/output pointers 10/SUB_BLOCKS_22_8 ms seconds ahead
in += 220 / SUB_BLOCKS_22_8;
out += 80 / SUB_BLOCKS_22_8;
}
}
// initialize state of 22 -> 8 resampler
void WebRtcSpl_ResetResample22khzTo8khz(WebRtcSpl_State22khzTo8khz* state)
{
int k;
for (k = 0; k < 8; k++)
{
state->S_22_22[k] = 0;
state->S_22_22[k + 8] = 0;
state->S_22_16[k] = 0;
state->S_16_8[k] = 0;
}
}
//////////////////////
// 8 kHz -> 22 kHz //
//////////////////////
// number of subblocks; options: 1, 2, 5, 10
#define SUB_BLOCKS_8_22 2
// 8 -> 22 resampler
void WebRtcSpl_Resample8khzTo22khz(const int16_t* in, int16_t* out,
WebRtcSpl_State8khzTo22khz* state, int32_t* tmpmem)
{
int k;
// process two blocks of 10/SUB_BLOCKS_8_22 ms (to reduce temp buffer size)
for (k = 0; k < SUB_BLOCKS_8_22; k++)
{
///// 8 --> 16 /////
// int16_t in[80/SUB_BLOCKS_8_22]
// int32_t out[160/SUB_BLOCKS_8_22]
/////
WebRtcSpl_UpBy2ShortToInt(in, 80 / SUB_BLOCKS_8_22, tmpmem + 18, state->S_8_16);
///// 16 --> 11 /////
// int32_t in[160/SUB_BLOCKS_8_22]
// int32_t out[110/SUB_BLOCKS_8_22]
/////
// copy state to and from input array
tmpmem[10] = state->S_16_11[0];
tmpmem[11] = state->S_16_11[1];
tmpmem[12] = state->S_16_11[2];
tmpmem[13] = state->S_16_11[3];
tmpmem[14] = state->S_16_11[4];
tmpmem[15] = state->S_16_11[5];
tmpmem[16] = state->S_16_11[6];
tmpmem[17] = state->S_16_11[7];
state->S_16_11[0] = tmpmem[160 / SUB_BLOCKS_8_22 + 10];
state->S_16_11[1] = tmpmem[160 / SUB_BLOCKS_8_22 + 11];
state->S_16_11[2] = tmpmem[160 / SUB_BLOCKS_8_22 + 12];
state->S_16_11[3] = tmpmem[160 / SUB_BLOCKS_8_22 + 13];
state->S_16_11[4] = tmpmem[160 / SUB_BLOCKS_8_22 + 14];
state->S_16_11[5] = tmpmem[160 / SUB_BLOCKS_8_22 + 15];
state->S_16_11[6] = tmpmem[160 / SUB_BLOCKS_8_22 + 16];
state->S_16_11[7] = tmpmem[160 / SUB_BLOCKS_8_22 + 17];
WebRtcSpl_32khzTo22khzIntToInt(tmpmem + 10, tmpmem, 10 / SUB_BLOCKS_8_22);
///// 11 --> 22 /////
// int32_t in[110/SUB_BLOCKS_8_22]
// int16_t out[220/SUB_BLOCKS_8_22]
/////
WebRtcSpl_UpBy2IntToShort(tmpmem, 110 / SUB_BLOCKS_8_22, out, state->S_11_22);
// move input/output pointers 10/SUB_BLOCKS_8_22 ms seconds ahead
in += 80 / SUB_BLOCKS_8_22;
out += 220 / SUB_BLOCKS_8_22;
}
}
// initialize state of 8 -> 22 resampler
void WebRtcSpl_ResetResample8khzTo22khz(WebRtcSpl_State8khzTo22khz* state)
{
int k;
for (k = 0; k < 8; k++)
{
state->S_8_16[k] = 0;
state->S_16_11[k] = 0;
state->S_11_22[k] = 0;
}
}
// compute two inner-products and store them to output array
static void WebRtcSpl_DotProdIntToInt(const int32_t* in1, const int32_t* in2,
const int16_t* coef_ptr, int32_t* out1,
int32_t* out2)
{
int32_t tmp1 = 16384;
int32_t tmp2 = 16384;
int16_t coef;
coef = coef_ptr[0];
tmp1 += coef * in1[0];
tmp2 += coef * in2[-0];
coef = coef_ptr[1];
tmp1 += coef * in1[1];
tmp2 += coef * in2[-1];
coef = coef_ptr[2];
tmp1 += coef * in1[2];
tmp2 += coef * in2[-2];
coef = coef_ptr[3];
tmp1 += coef * in1[3];
tmp2 += coef * in2[-3];
coef = coef_ptr[4];
tmp1 += coef * in1[4];
tmp2 += coef * in2[-4];
coef = coef_ptr[5];
tmp1 += coef * in1[5];
tmp2 += coef * in2[-5];
coef = coef_ptr[6];
tmp1 += coef * in1[6];
tmp2 += coef * in2[-6];
coef = coef_ptr[7];
tmp1 += coef * in1[7];
tmp2 += coef * in2[-7];
coef = coef_ptr[8];
*out1 = tmp1 + coef * in1[8];
*out2 = tmp2 + coef * in2[-8];
}
// compute two inner-products and store them to output array
static void WebRtcSpl_DotProdIntToShort(const int32_t* in1, const int32_t* in2,
const int16_t* coef_ptr, int16_t* out1,
int16_t* out2)
{
int32_t tmp1 = 16384;
int32_t tmp2 = 16384;
int16_t coef;
coef = coef_ptr[0];
tmp1 += coef * in1[0];
tmp2 += coef * in2[-0];
coef = coef_ptr[1];
tmp1 += coef * in1[1];
tmp2 += coef * in2[-1];
coef = coef_ptr[2];
tmp1 += coef * in1[2];
tmp2 += coef * in2[-2];
coef = coef_ptr[3];
tmp1 += coef * in1[3];
tmp2 += coef * in2[-3];
coef = coef_ptr[4];
tmp1 += coef * in1[4];
tmp2 += coef * in2[-4];
coef = coef_ptr[5];
tmp1 += coef * in1[5];
tmp2 += coef * in2[-5];
coef = coef_ptr[6];
tmp1 += coef * in1[6];
tmp2 += coef * in2[-6];
coef = coef_ptr[7];
tmp1 += coef * in1[7];
tmp2 += coef * in2[-7];
coef = coef_ptr[8];
tmp1 += coef * in1[8];
tmp2 += coef * in2[-8];
// scale down, round and saturate
tmp1 >>= 15;
if (tmp1 > (int32_t)0x00007FFF)
tmp1 = 0x00007FFF;
if (tmp1 < (int32_t)0xFFFF8000)
tmp1 = 0xFFFF8000;
tmp2 >>= 15;
if (tmp2 > (int32_t)0x00007FFF)
tmp2 = 0x00007FFF;
if (tmp2 < (int32_t)0xFFFF8000)
tmp2 = 0xFFFF8000;
*out1 = (int16_t)tmp1;
*out2 = (int16_t)tmp2;
}
// Resampling ratio: 11/16
// input: int32_t (normalized, not saturated) :: size 16 * K
// output: int32_t (shifted 15 positions to the left, + offset 16384) :: size 11 * K
// K: Number of blocks
void WebRtcSpl_32khzTo22khzIntToInt(const int32_t* In,
int32_t* Out,
int32_t K)
{
/////////////////////////////////////////////////////////////
// Filter operation:
//
// Perform resampling (16 input samples -> 11 output samples);
// process in sub blocks of size 16 samples.
int32_t m;
for (m = 0; m < K; m++)
{
// first output sample
Out[0] = ((int32_t)In[3] << 15) + (1 << 14);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToInt(&In[0], &In[22], kCoefficients32To22[0], &Out[1], &Out[10]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToInt(&In[2], &In[20], kCoefficients32To22[1], &Out[2], &Out[9]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToInt(&In[3], &In[19], kCoefficients32To22[2], &Out[3], &Out[8]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToInt(&In[5], &In[17], kCoefficients32To22[3], &Out[4], &Out[7]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToInt(&In[6], &In[16], kCoefficients32To22[4], &Out[5], &Out[6]);
// update pointers
In += 16;
Out += 11;
}
}
// Resampling ratio: 11/16
// input: int32_t (normalized, not saturated) :: size 16 * K
// output: int16_t (saturated) :: size 11 * K
// K: Number of blocks
void WebRtcSpl_32khzTo22khzIntToShort(const int32_t *In,
int16_t *Out,
int32_t K)
{
/////////////////////////////////////////////////////////////
// Filter operation:
//
// Perform resampling (16 input samples -> 11 output samples);
// process in sub blocks of size 16 samples.
int32_t tmp;
int32_t m;
for (m = 0; m < K; m++)
{
// first output sample
tmp = In[3];
if (tmp > (int32_t)0x00007FFF)
tmp = 0x00007FFF;
if (tmp < (int32_t)0xFFFF8000)
tmp = 0xFFFF8000;
Out[0] = (int16_t)tmp;
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToShort(&In[0], &In[22], kCoefficients32To22[0], &Out[1], &Out[10]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToShort(&In[2], &In[20], kCoefficients32To22[1], &Out[2], &Out[9]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToShort(&In[3], &In[19], kCoefficients32To22[2], &Out[3], &Out[8]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToShort(&In[5], &In[17], kCoefficients32To22[3], &Out[4], &Out[7]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_DotProdIntToShort(&In[6], &In[16], kCoefficients32To22[4], &Out[5], &Out[6]);
// update pointers
In += 16;
Out += 11;
}
}

186
webrtc/common_audio/signal_processing/resample_48khz.c Normal file
View File

@ -0,0 +1,186 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains resampling functions between 48 kHz and nb/wb.
* The description header can be found in signal_processing_library.h
*
*/
#include <string.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/common_audio/signal_processing/resample_by_2_internal.h"
////////////////////////////
///// 48 kHz -> 16 kHz /////
////////////////////////////
// 48 -> 16 resampler
void WebRtcSpl_Resample48khzTo16khz(const int16_t* in, int16_t* out,
WebRtcSpl_State48khzTo16khz* state, int32_t* tmpmem)
{
///// 48 --> 48(LP) /////
// int16_t in[480]
// int32_t out[480]
/////
WebRtcSpl_LPBy2ShortToInt(in, 480, tmpmem + 16, state->S_48_48);
///// 48 --> 32 /////
// int32_t in[480]
// int32_t out[320]
/////
// copy state to and from input array
memcpy(tmpmem + 8, state->S_48_32, 8 * sizeof(int32_t));
memcpy(state->S_48_32, tmpmem + 488, 8 * sizeof(int32_t));
WebRtcSpl_Resample48khzTo32khz(tmpmem + 8, tmpmem, 160);
///// 32 --> 16 /////
// int32_t in[320]
// int16_t out[160]
/////
WebRtcSpl_DownBy2IntToShort(tmpmem, 320, out, state->S_32_16);
}
// initialize state of 48 -> 16 resampler
void WebRtcSpl_ResetResample48khzTo16khz(WebRtcSpl_State48khzTo16khz* state)
{
memset(state->S_48_48, 0, 16 * sizeof(int32_t));
memset(state->S_48_32, 0, 8 * sizeof(int32_t));
memset(state->S_32_16, 0, 8 * sizeof(int32_t));
}
////////////////////////////
///// 16 kHz -> 48 kHz /////
////////////////////////////
// 16 -> 48 resampler
void WebRtcSpl_Resample16khzTo48khz(const int16_t* in, int16_t* out,
WebRtcSpl_State16khzTo48khz* state, int32_t* tmpmem)
{
///// 16 --> 32 /////
// int16_t in[160]
// int32_t out[320]
/////
WebRtcSpl_UpBy2ShortToInt(in, 160, tmpmem + 16, state->S_16_32);
///// 32 --> 24 /////
// int32_t in[320]
// int32_t out[240]
// copy state to and from input array
/////
memcpy(tmpmem + 8, state->S_32_24, 8 * sizeof(int32_t));
memcpy(state->S_32_24, tmpmem + 328, 8 * sizeof(int32_t));
WebRtcSpl_Resample32khzTo24khz(tmpmem + 8, tmpmem, 80);
///// 24 --> 48 /////
// int32_t in[240]
// int16_t out[480]
/////
WebRtcSpl_UpBy2IntToShort(tmpmem, 240, out, state->S_24_48);
}
// initialize state of 16 -> 48 resampler
void WebRtcSpl_ResetResample16khzTo48khz(WebRtcSpl_State16khzTo48khz* state)
{
memset(state->S_16_32, 0, 8 * sizeof(int32_t));
memset(state->S_32_24, 0, 8 * sizeof(int32_t));
memset(state->S_24_48, 0, 8 * sizeof(int32_t));
}
////////////////////////////
///// 48 kHz -> 8 kHz /////
////////////////////////////
// 48 -> 8 resampler
void WebRtcSpl_Resample48khzTo8khz(const int16_t* in, int16_t* out,
WebRtcSpl_State48khzTo8khz* state, int32_t* tmpmem)
{
///// 48 --> 24 /////
// int16_t in[480]
// int32_t out[240]
/////
WebRtcSpl_DownBy2ShortToInt(in, 480, tmpmem + 256, state->S_48_24);
///// 24 --> 24(LP) /////
// int32_t in[240]
// int32_t out[240]
/////
WebRtcSpl_LPBy2IntToInt(tmpmem + 256, 240, tmpmem + 16, state->S_24_24);
///// 24 --> 16 /////
// int32_t in[240]
// int32_t out[160]
/////
// copy state to and from input array
memcpy(tmpmem + 8, state->S_24_16, 8 * sizeof(int32_t));
memcpy(state->S_24_16, tmpmem + 248, 8 * sizeof(int32_t));
WebRtcSpl_Resample48khzTo32khz(tmpmem + 8, tmpmem, 80);
///// 16 --> 8 /////
// int32_t in[160]
// int16_t out[80]
/////
WebRtcSpl_DownBy2IntToShort(tmpmem, 160, out, state->S_16_8);
}
// initialize state of 48 -> 8 resampler
void WebRtcSpl_ResetResample48khzTo8khz(WebRtcSpl_State48khzTo8khz* state)
{
memset(state->S_48_24, 0, 8 * sizeof(int32_t));
memset(state->S_24_24, 0, 16 * sizeof(int32_t));
memset(state->S_24_16, 0, 8 * sizeof(int32_t));
memset(state->S_16_8, 0, 8 * sizeof(int32_t));
}
////////////////////////////
///// 8 kHz -> 48 kHz /////
////////////////////////////
// 8 -> 48 resampler
void WebRtcSpl_Resample8khzTo48khz(const int16_t* in, int16_t* out,
WebRtcSpl_State8khzTo48khz* state, int32_t* tmpmem)
{
///// 8 --> 16 /////
// int16_t in[80]
// int32_t out[160]
/////
WebRtcSpl_UpBy2ShortToInt(in, 80, tmpmem + 264, state->S_8_16);
///// 16 --> 12 /////
// int32_t in[160]
// int32_t out[120]
/////
// copy state to and from input array
memcpy(tmpmem + 256, state->S_16_12, 8 * sizeof(int32_t));
memcpy(state->S_16_12, tmpmem + 416, 8 * sizeof(int32_t));
WebRtcSpl_Resample32khzTo24khz(tmpmem + 256, tmpmem + 240, 40);
///// 12 --> 24 /////
// int32_t in[120]
// int16_t out[240]
/////
WebRtcSpl_UpBy2IntToInt(tmpmem + 240, 120, tmpmem, state->S_12_24);
///// 24 --> 48 /////
// int32_t in[240]
// int16_t out[480]
/////
WebRtcSpl_UpBy2IntToShort(tmpmem, 240, out, state->S_24_48);
}
// initialize state of 8 -> 48 resampler
void WebRtcSpl_ResetResample8khzTo48khz(WebRtcSpl_State8khzTo48khz* state)
{
memset(state->S_8_16, 0, 8 * sizeof(int32_t));
memset(state->S_16_12, 0, 8 * sizeof(int32_t));
memset(state->S_12_24, 0, 8 * sizeof(int32_t));
memset(state->S_24_48, 0, 8 * sizeof(int32_t));
}

183
webrtc/common_audio/signal_processing/resample_by_2.c Normal file
View File

@ -0,0 +1,183 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the resampling by two functions.
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#ifdef WEBRTC_ARCH_ARM_V7
// allpass filter coefficients.
static const uint32_t kResampleAllpass1[3] = {3284, 24441, 49528 << 15};
static const uint32_t kResampleAllpass2[3] =
{12199, 37471 << 15, 60255 << 15};
// Multiply two 32-bit values and accumulate to another input value.
// Return: state + ((diff * tbl_value) >> 16)
static __inline int32_t MUL_ACCUM_1(int32_t tbl_value,
int32_t diff,
int32_t state) {
int32_t result;
__asm __volatile ("smlawb %0, %1, %2, %3": "=r"(result): "r"(diff),
"r"(tbl_value), "r"(state));
return result;
}
// Multiply two 32-bit values and accumulate to another input value.
// Return: Return: state + (((diff << 1) * tbl_value) >> 32)
//
// The reason to introduce this function is that, in case we can't use smlawb
// instruction (in MUL_ACCUM_1) due to input value range, we can still use
// smmla to save some cycles.
static __inline int32_t MUL_ACCUM_2(int32_t tbl_value,
int32_t diff,
int32_t state) {
int32_t result;
__asm __volatile ("smmla %0, %1, %2, %3": "=r"(result): "r"(diff << 1),
"r"(tbl_value), "r"(state));
return result;
}
#else
// allpass filter coefficients.
static const uint16_t kResampleAllpass1[3] = {3284, 24441, 49528};
static const uint16_t kResampleAllpass2[3] = {12199, 37471, 60255};
// Multiply a 32-bit value with a 16-bit value and accumulate to another input:
#define MUL_ACCUM_1(a, b, c) WEBRTC_SPL_SCALEDIFF32(a, b, c)
#define MUL_ACCUM_2(a, b, c) WEBRTC_SPL_SCALEDIFF32(a, b, c)
#endif // WEBRTC_ARCH_ARM_V7
// decimator
#if !defined(MIPS32_LE)
void WebRtcSpl_DownsampleBy2(const int16_t* in, size_t len,
int16_t* out, int32_t* filtState) {
int32_t tmp1, tmp2, diff, in32, out32;
size_t i;
register int32_t state0 = filtState[0];
register int32_t state1 = filtState[1];
register int32_t state2 = filtState[2];
register int32_t state3 = filtState[3];
register int32_t state4 = filtState[4];
register int32_t state5 = filtState[5];
register int32_t state6 = filtState[6];
register int32_t state7 = filtState[7];
for (i = (len >> 1); i > 0; i--) {
// lower allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state1;
tmp1 = MUL_ACCUM_1(kResampleAllpass2[0], diff, state0);
state0 = in32;
diff = tmp1 - state2;
tmp2 = MUL_ACCUM_2(kResampleAllpass2[1], diff, state1);
state1 = tmp1;
diff = tmp2 - state3;
state3 = MUL_ACCUM_2(kResampleAllpass2[2], diff, state2);
state2 = tmp2;
// upper allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state5;
tmp1 = MUL_ACCUM_1(kResampleAllpass1[0], diff, state4);
state4 = in32;
diff = tmp1 - state6;
tmp2 = MUL_ACCUM_1(kResampleAllpass1[1], diff, state5);
state5 = tmp1;
diff = tmp2 - state7;
state7 = MUL_ACCUM_2(kResampleAllpass1[2], diff, state6);
state6 = tmp2;
// add two allpass outputs, divide by two and round
out32 = (state3 + state7 + 1024) >> 11;
// limit amplitude to prevent wrap-around, and write to output array
*out++ = WebRtcSpl_SatW32ToW16(out32);
}
filtState[0] = state0;
filtState[1] = state1;
filtState[2] = state2;
filtState[3] = state3;
filtState[4] = state4;
filtState[5] = state5;
filtState[6] = state6;
filtState[7] = state7;
}
#endif // #if defined(MIPS32_LE)
void WebRtcSpl_UpsampleBy2(const int16_t* in, size_t len,
int16_t* out, int32_t* filtState) {
int32_t tmp1, tmp2, diff, in32, out32;
size_t i;
register int32_t state0 = filtState[0];
register int32_t state1 = filtState[1];
register int32_t state2 = filtState[2];
register int32_t state3 = filtState[3];
register int32_t state4 = filtState[4];
register int32_t state5 = filtState[5];
register int32_t state6 = filtState[6];
register int32_t state7 = filtState[7];
for (i = len; i > 0; i--) {
// lower allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state1;
tmp1 = MUL_ACCUM_1(kResampleAllpass1[0], diff, state0);
state0 = in32;
diff = tmp1 - state2;
tmp2 = MUL_ACCUM_1(kResampleAllpass1[1], diff, state1);
state1 = tmp1;
diff = tmp2 - state3;
state3 = MUL_ACCUM_2(kResampleAllpass1[2], diff, state2);
state2 = tmp2;
// round; limit amplitude to prevent wrap-around; write to output array
out32 = (state3 + 512) >> 10;
*out++ = WebRtcSpl_SatW32ToW16(out32);
// upper allpass filter
diff = in32 - state5;
tmp1 = MUL_ACCUM_1(kResampleAllpass2[0], diff, state4);
state4 = in32;
diff = tmp1 - state6;
tmp2 = MUL_ACCUM_2(kResampleAllpass2[1], diff, state5);
state5 = tmp1;
diff = tmp2 - state7;
state7 = MUL_ACCUM_2(kResampleAllpass2[2], diff, state6);
state6 = tmp2;
// round; limit amplitude to prevent wrap-around; write to output array
out32 = (state7 + 512) >> 10;
*out++ = WebRtcSpl_SatW32ToW16(out32);
}
filtState[0] = state0;
filtState[1] = state1;
filtState[2] = state2;
filtState[3] = state3;
filtState[4] = state4;
filtState[5] = state5;
filtState[6] = state6;
filtState[7] = state7;
}

679
webrtc/common_audio/signal_processing/resample_by_2_internal.c Normal file
View File

@ -0,0 +1,679 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This header file contains some internal resampling functions.
*
*/
#include "webrtc/common_audio/signal_processing/resample_by_2_internal.h"
// allpass filter coefficients.
static const int16_t kResampleAllpass[2][3] = {
{821, 6110, 12382},
{3050, 9368, 15063}
};
//
// decimator
// input: int32_t (shifted 15 positions to the left, + offset 16384) OVERWRITTEN!
// output: int16_t (saturated) (of length len/2)
// state: filter state array; length = 8
void WebRtcSpl_DownBy2IntToShort(int32_t *in, int32_t len, int16_t *out,
int32_t *state)
{
int32_t tmp0, tmp1, diff;
int32_t i;
len >>= 1;
// lower allpass filter (operates on even input samples)
for (i = 0; i < len; i++)
{
tmp0 = in[i << 1];
diff = tmp0 - state[1];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[0] + diff * kResampleAllpass[1][0];
state[0] = tmp0;
diff = tmp1 - state[2];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[1] + diff * kResampleAllpass[1][1];
state[1] = tmp1;
diff = tmp0 - state[3];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[3] = state[2] + diff * kResampleAllpass[1][2];
state[2] = tmp0;
// divide by two and store temporarily
in[i << 1] = (state[3] >> 1);
}
in++;
// upper allpass filter (operates on odd input samples)
for (i = 0; i < len; i++)
{
tmp0 = in[i << 1];
diff = tmp0 - state[5];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[4] + diff * kResampleAllpass[0][0];
state[4] = tmp0;
diff = tmp1 - state[6];
// scale down and round
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[5] + diff * kResampleAllpass[0][1];
state[5] = tmp1;
diff = tmp0 - state[7];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[7] = state[6] + diff * kResampleAllpass[0][2];
state[6] = tmp0;
// divide by two and store temporarily
in[i << 1] = (state[7] >> 1);
}
in--;
// combine allpass outputs
for (i = 0; i < len; i += 2)
{
// divide by two, add both allpass outputs and round
tmp0 = (in[i << 1] + in[(i << 1) + 1]) >> 15;
tmp1 = (in[(i << 1) + 2] + in[(i << 1) + 3]) >> 15;
if (tmp0 > (int32_t)0x00007FFF)
tmp0 = 0x00007FFF;
if (tmp0 < (int32_t)0xFFFF8000)
tmp0 = 0xFFFF8000;
out[i] = (int16_t)tmp0;
if (tmp1 > (int32_t)0x00007FFF)
tmp1 = 0x00007FFF;
if (tmp1 < (int32_t)0xFFFF8000)
tmp1 = 0xFFFF8000;
out[i + 1] = (int16_t)tmp1;
}
}
//
// decimator
// input: int16_t
// output: int32_t (shifted 15 positions to the left, + offset 16384) (of length len/2)
// state: filter state array; length = 8
void WebRtcSpl_DownBy2ShortToInt(const int16_t *in,
int32_t len,
int32_t *out,
int32_t *state)
{
int32_t tmp0, tmp1, diff;
int32_t i;
len >>= 1;
// lower allpass filter (operates on even input samples)
for (i = 0; i < len; i++)
{
tmp0 = ((int32_t)in[i << 1] << 15) + (1 << 14);
diff = tmp0 - state[1];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[0] + diff * kResampleAllpass[1][0];
state[0] = tmp0;
diff = tmp1 - state[2];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[1] + diff * kResampleAllpass[1][1];
state[1] = tmp1;
diff = tmp0 - state[3];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[3] = state[2] + diff * kResampleAllpass[1][2];
state[2] = tmp0;
// divide by two and store temporarily
out[i] = (state[3] >> 1);
}
in++;
// upper allpass filter (operates on odd input samples)
for (i = 0; i < len; i++)
{
tmp0 = ((int32_t)in[i << 1] << 15) + (1 << 14);
diff = tmp0 - state[5];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[4] + diff * kResampleAllpass[0][0];
state[4] = tmp0;
diff = tmp1 - state[6];
// scale down and round
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[5] + diff * kResampleAllpass[0][1];
state[5] = tmp1;
diff = tmp0 - state[7];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[7] = state[6] + diff * kResampleAllpass[0][2];
state[6] = tmp0;
// divide by two and store temporarily
out[i] += (state[7] >> 1);
}
in--;
}
//
// interpolator
// input: int16_t
// output: int32_t (normalized, not saturated) (of length len*2)
// state: filter state array; length = 8
void WebRtcSpl_UpBy2ShortToInt(const int16_t *in, int32_t len, int32_t *out,
int32_t *state)
{
int32_t tmp0, tmp1, diff;
int32_t i;
// upper allpass filter (generates odd output samples)
for (i = 0; i < len; i++)
{
tmp0 = ((int32_t)in[i] << 15) + (1 << 14);
diff = tmp0 - state[5];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[4] + diff * kResampleAllpass[0][0];
state[4] = tmp0;
diff = tmp1 - state[6];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[5] + diff * kResampleAllpass[0][1];
state[5] = tmp1;
diff = tmp0 - state[7];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[7] = state[6] + diff * kResampleAllpass[0][2];
state[6] = tmp0;
// scale down, round and store
out[i << 1] = state[7] >> 15;
}
out++;
// lower allpass filter (generates even output samples)
for (i = 0; i < len; i++)
{
tmp0 = ((int32_t)in[i] << 15) + (1 << 14);
diff = tmp0 - state[1];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[0] + diff * kResampleAllpass[1][0];
state[0] = tmp0;
diff = tmp1 - state[2];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[1] + diff * kResampleAllpass[1][1];
state[1] = tmp1;
diff = tmp0 - state[3];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[3] = state[2] + diff * kResampleAllpass[1][2];
state[2] = tmp0;
// scale down, round and store
out[i << 1] = state[3] >> 15;
}
}
//
// interpolator
// input: int32_t (shifted 15 positions to the left, + offset 16384)
// output: int32_t (shifted 15 positions to the left, + offset 16384) (of length len*2)
// state: filter state array; length = 8
void WebRtcSpl_UpBy2IntToInt(const int32_t *in, int32_t len, int32_t *out,
int32_t *state)
{
int32_t tmp0, tmp1, diff;
int32_t i;
// upper allpass filter (generates odd output samples)
for (i = 0; i < len; i++)
{
tmp0 = in[i];
diff = tmp0 - state[5];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[4] + diff * kResampleAllpass[0][0];
state[4] = tmp0;
diff = tmp1 - state[6];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[5] + diff * kResampleAllpass[0][1];
state[5] = tmp1;
diff = tmp0 - state[7];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[7] = state[6] + diff * kResampleAllpass[0][2];
state[6] = tmp0;
// scale down, round and store
out[i << 1] = state[7];
}
out++;
// lower allpass filter (generates even output samples)
for (i = 0; i < len; i++)
{
tmp0 = in[i];
diff = tmp0 - state[1];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[0] + diff * kResampleAllpass[1][0];
state[0] = tmp0;
diff = tmp1 - state[2];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[1] + diff * kResampleAllpass[1][1];
state[1] = tmp1;
diff = tmp0 - state[3];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[3] = state[2] + diff * kResampleAllpass[1][2];
state[2] = tmp0;
// scale down, round and store
out[i << 1] = state[3];
}
}
//
// interpolator
// input: int32_t (shifted 15 positions to the left, + offset 16384)
// output: int16_t (saturated) (of length len*2)
// state: filter state array; length = 8
void WebRtcSpl_UpBy2IntToShort(const int32_t *in, int32_t len, int16_t *out,
int32_t *state)
{
int32_t tmp0, tmp1, diff;
int32_t i;
// upper allpass filter (generates odd output samples)
for (i = 0; i < len; i++)
{
tmp0 = in[i];
diff = tmp0 - state[5];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[4] + diff * kResampleAllpass[0][0];
state[4] = tmp0;
diff = tmp1 - state[6];
// scale down and round
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[5] + diff * kResampleAllpass[0][1];
state[5] = tmp1;
diff = tmp0 - state[7];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[7] = state[6] + diff * kResampleAllpass[0][2];
state[6] = tmp0;
// scale down, saturate and store
tmp1 = state[7] >> 15;
if (tmp1 > (int32_t)0x00007FFF)
tmp1 = 0x00007FFF;
if (tmp1 < (int32_t)0xFFFF8000)
tmp1 = 0xFFFF8000;
out[i << 1] = (int16_t)tmp1;
}
out++;
// lower allpass filter (generates even output samples)
for (i = 0; i < len; i++)
{
tmp0 = in[i];
diff = tmp0 - state[1];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[0] + diff * kResampleAllpass[1][0];
state[0] = tmp0;
diff = tmp1 - state[2];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[1] + diff * kResampleAllpass[1][1];
state[1] = tmp1;
diff = tmp0 - state[3];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[3] = state[2] + diff * kResampleAllpass[1][2];
state[2] = tmp0;
// scale down, saturate and store
tmp1 = state[3] >> 15;
if (tmp1 > (int32_t)0x00007FFF)
tmp1 = 0x00007FFF;
if (tmp1 < (int32_t)0xFFFF8000)
tmp1 = 0xFFFF8000;
out[i << 1] = (int16_t)tmp1;
}
}
// lowpass filter
// input: int16_t
// output: int32_t (normalized, not saturated)
// state: filter state array; length = 8
void WebRtcSpl_LPBy2ShortToInt(const int16_t* in, int32_t len, int32_t* out,
int32_t* state)
{
int32_t tmp0, tmp1, diff;
int32_t i;
len >>= 1;
// lower allpass filter: odd input -> even output samples
in++;
// initial state of polyphase delay element
tmp0 = state[12];
for (i = 0; i < len; i++)
{
diff = tmp0 - state[1];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[0] + diff * kResampleAllpass[1][0];
state[0] = tmp0;
diff = tmp1 - state[2];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[1] + diff * kResampleAllpass[1][1];
state[1] = tmp1;
diff = tmp0 - state[3];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[3] = state[2] + diff * kResampleAllpass[1][2];
state[2] = tmp0;
// scale down, round and store
out[i << 1] = state[3] >> 1;
tmp0 = ((int32_t)in[i << 1] << 15) + (1 << 14);
}
in--;
// upper allpass filter: even input -> even output samples
for (i = 0; i < len; i++)
{
tmp0 = ((int32_t)in[i << 1] << 15) + (1 << 14);
diff = tmp0 - state[5];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[4] + diff * kResampleAllpass[0][0];
state[4] = tmp0;
diff = tmp1 - state[6];
// scale down and round
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[5] + diff * kResampleAllpass[0][1];
state[5] = tmp1;
diff = tmp0 - state[7];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[7] = state[6] + diff * kResampleAllpass[0][2];
state[6] = tmp0;
// average the two allpass outputs, scale down and store
out[i << 1] = (out[i << 1] + (state[7] >> 1)) >> 15;
}
// switch to odd output samples
out++;
// lower allpass filter: even input -> odd output samples
for (i = 0; i < len; i++)
{
tmp0 = ((int32_t)in[i << 1] << 15) + (1 << 14);
diff = tmp0 - state[9];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[8] + diff * kResampleAllpass[1][0];
state[8] = tmp0;
diff = tmp1 - state[10];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[9] + diff * kResampleAllpass[1][1];
state[9] = tmp1;
diff = tmp0 - state[11];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[11] = state[10] + diff * kResampleAllpass[1][2];
state[10] = tmp0;
// scale down, round and store
out[i << 1] = state[11] >> 1;
}
// upper allpass filter: odd input -> odd output samples
in++;
for (i = 0; i < len; i++)
{
tmp0 = ((int32_t)in[i << 1] << 15) + (1 << 14);
diff = tmp0 - state[13];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[12] + diff * kResampleAllpass[0][0];
state[12] = tmp0;
diff = tmp1 - state[14];
// scale down and round
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[13] + diff * kResampleAllpass[0][1];
state[13] = tmp1;
diff = tmp0 - state[15];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[15] = state[14] + diff * kResampleAllpass[0][2];
state[14] = tmp0;
// average the two allpass outputs, scale down and store
out[i << 1] = (out[i << 1] + (state[15] >> 1)) >> 15;
}
}
// lowpass filter
// input: int32_t (shifted 15 positions to the left, + offset 16384)
// output: int32_t (normalized, not saturated)
// state: filter state array; length = 8
void WebRtcSpl_LPBy2IntToInt(const int32_t* in, int32_t len, int32_t* out,
int32_t* state)
{
int32_t tmp0, tmp1, diff;
int32_t i;
len >>= 1;
// lower allpass filter: odd input -> even output samples
in++;
// initial state of polyphase delay element
tmp0 = state[12];
for (i = 0; i < len; i++)
{
diff = tmp0 - state[1];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[0] + diff * kResampleAllpass[1][0];
state[0] = tmp0;
diff = tmp1 - state[2];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[1] + diff * kResampleAllpass[1][1];
state[1] = tmp1;
diff = tmp0 - state[3];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[3] = state[2] + diff * kResampleAllpass[1][2];
state[2] = tmp0;
// scale down, round and store
out[i << 1] = state[3] >> 1;
tmp0 = in[i << 1];
}
in--;
// upper allpass filter: even input -> even output samples
for (i = 0; i < len; i++)
{
tmp0 = in[i << 1];
diff = tmp0 - state[5];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[4] + diff * kResampleAllpass[0][0];
state[4] = tmp0;
diff = tmp1 - state[6];
// scale down and round
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[5] + diff * kResampleAllpass[0][1];
state[5] = tmp1;
diff = tmp0 - state[7];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[7] = state[6] + diff * kResampleAllpass[0][2];
state[6] = tmp0;
// average the two allpass outputs, scale down and store
out[i << 1] = (out[i << 1] + (state[7] >> 1)) >> 15;
}
// switch to odd output samples
out++;
// lower allpass filter: even input -> odd output samples
for (i = 0; i < len; i++)
{
tmp0 = in[i << 1];
diff = tmp0 - state[9];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[8] + diff * kResampleAllpass[1][0];
state[8] = tmp0;
diff = tmp1 - state[10];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[9] + diff * kResampleAllpass[1][1];
state[9] = tmp1;
diff = tmp0 - state[11];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[11] = state[10] + diff * kResampleAllpass[1][2];
state[10] = tmp0;
// scale down, round and store
out[i << 1] = state[11] >> 1;
}
// upper allpass filter: odd input -> odd output samples
in++;
for (i = 0; i < len; i++)
{
tmp0 = in[i << 1];
diff = tmp0 - state[13];
// scale down and round
diff = (diff + (1 << 13)) >> 14;
tmp1 = state[12] + diff * kResampleAllpass[0][0];
state[12] = tmp0;
diff = tmp1 - state[14];
// scale down and round
diff = diff >> 14;
if (diff < 0)
diff += 1;
tmp0 = state[13] + diff * kResampleAllpass[0][1];
state[13] = tmp1;
diff = tmp0 - state[15];
// scale down and truncate
diff = diff >> 14;
if (diff < 0)
diff += 1;
state[15] = state[14] + diff * kResampleAllpass[0][2];
state[14] = tmp0;
// average the two allpass outputs, scale down and store
out[i << 1] = (out[i << 1] + (state[15] >> 1)) >> 15;
}
}

47
webrtc/common_audio/signal_processing/resample_by_2_internal.h Normal file
View File

@ -0,0 +1,47 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This header file contains some internal resampling functions.
*
*/
#ifndef WEBRTC_SPL_RESAMPLE_BY_2_INTERNAL_H_
#define WEBRTC_SPL_RESAMPLE_BY_2_INTERNAL_H_
#include "webrtc/typedefs.h"
/*******************************************************************
* resample_by_2_fast.c
* Functions for internal use in the other resample functions
******************************************************************/
void WebRtcSpl_DownBy2IntToShort(int32_t *in, int32_t len, int16_t *out,
int32_t *state);
void WebRtcSpl_DownBy2ShortToInt(const int16_t *in, int32_t len,
int32_t *out, int32_t *state);
void WebRtcSpl_UpBy2ShortToInt(const int16_t *in, int32_t len,
int32_t *out, int32_t *state);
void WebRtcSpl_UpBy2IntToInt(const int32_t *in, int32_t len, int32_t *out,
int32_t *state);
void WebRtcSpl_UpBy2IntToShort(const int32_t *in, int32_t len,
int16_t *out, int32_t *state);
void WebRtcSpl_LPBy2ShortToInt(const int16_t* in, int32_t len,
int32_t* out, int32_t* state);
void WebRtcSpl_LPBy2IntToInt(const int32_t* in, int32_t len, int32_t* out,
int32_t* state);
#endif // WEBRTC_SPL_RESAMPLE_BY_2_INTERNAL_H_

290
webrtc/common_audio/signal_processing/resample_by_2_mips.c Normal file
View File

@ -0,0 +1,290 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the resampling by two functions.
* The description header can be found in signal_processing_library.h
*
*/
#if defined(MIPS32_LE)
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// allpass filter coefficients.
static const uint16_t kResampleAllpass1[3] = {3284, 24441, 49528};
static const uint16_t kResampleAllpass2[3] = {12199, 37471, 60255};
// Multiply a 32-bit value with a 16-bit value and accumulate to another input:
#define MUL_ACCUM_1(a, b, c) WEBRTC_SPL_SCALEDIFF32(a, b, c)
#define MUL_ACCUM_2(a, b, c) WEBRTC_SPL_SCALEDIFF32(a, b, c)
// decimator
void WebRtcSpl_DownsampleBy2(const int16_t* in,
size_t len,
int16_t* out,
int32_t* filtState) {
int32_t out32;
size_t i, len1;
register int32_t state0 = filtState[0];
register int32_t state1 = filtState[1];
register int32_t state2 = filtState[2];
register int32_t state3 = filtState[3];
register int32_t state4 = filtState[4];
register int32_t state5 = filtState[5];
register int32_t state6 = filtState[6];
register int32_t state7 = filtState[7];
#if defined(MIPS_DSP_R2_LE)
int32_t k1Res0, k1Res1, k1Res2, k2Res0, k2Res1, k2Res2;
k1Res0= 3284;
k1Res1= 24441;
k1Res2= 49528;
k2Res0= 12199;
k2Res1= 37471;
k2Res2= 60255;
len1 = (len >> 1);
const int32_t* inw = (int32_t*)in;
int32_t tmp11, tmp12, tmp21, tmp22;
int32_t in322, in321;
int32_t diff1, diff2;
for (i = len1; i > 0; i--) {
__asm__ volatile (
"lh %[in321], 0(%[inw]) \n\t"
"lh %[in322], 2(%[inw]) \n\t"
"sll %[in321], %[in321], 10 \n\t"
"sll %[in322], %[in322], 10 \n\t"
"addiu %[inw], %[inw], 4 \n\t"
"subu %[diff1], %[in321], %[state1] \n\t"
"subu %[diff2], %[in322], %[state5] \n\t"
: [in322] "=&r" (in322), [in321] "=&r" (in321),
[diff1] "=&r" (diff1), [diff2] "=r" (diff2), [inw] "+r" (inw)
: [state1] "r" (state1), [state5] "r" (state5)
: "memory"
);
__asm__ volatile (
"mult $ac0, %[diff1], %[k2Res0] \n\t"
"mult $ac1, %[diff2], %[k1Res0] \n\t"
"extr.w %[tmp11], $ac0, 16 \n\t"
"extr.w %[tmp12], $ac1, 16 \n\t"
"addu %[tmp11], %[state0], %[tmp11] \n\t"
"addu %[tmp12], %[state4], %[tmp12] \n\t"
"addiu %[state0], %[in321], 0 \n\t"
"addiu %[state4], %[in322], 0 \n\t"
"subu %[diff1], %[tmp11], %[state2] \n\t"
"subu %[diff2], %[tmp12], %[state6] \n\t"
"mult $ac0, %[diff1], %[k2Res1] \n\t"
"mult $ac1, %[diff2], %[k1Res1] \n\t"
"extr.w %[tmp21], $ac0, 16 \n\t"
"extr.w %[tmp22], $ac1, 16 \n\t"
"addu %[tmp21], %[state1], %[tmp21] \n\t"
"addu %[tmp22], %[state5], %[tmp22] \n\t"
"addiu %[state1], %[tmp11], 0 \n\t"
"addiu %[state5], %[tmp12], 0 \n\t"
: [tmp22] "=r" (tmp22), [tmp21] "=&r" (tmp21),
[tmp11] "=&r" (tmp11), [state0] "+r" (state0),
[state1] "+r" (state1),
[state2] "+r" (state2),
[state4] "+r" (state4), [tmp12] "=&r" (tmp12),
[state6] "+r" (state6), [state5] "+r" (state5)
: [k1Res1] "r" (k1Res1), [k2Res1] "r" (k2Res1), [k2Res0] "r" (k2Res0),
[diff2] "r" (diff2), [diff1] "r" (diff1), [in322] "r" (in322),
[in321] "r" (in321), [k1Res0] "r" (k1Res0)
: "hi", "lo", "$ac1hi", "$ac1lo"
);
// upper allpass filter
__asm__ volatile (
"subu %[diff1], %[tmp21], %[state3] \n\t"
"subu %[diff2], %[tmp22], %[state7] \n\t"
"mult $ac0, %[diff1], %[k2Res2] \n\t"
"mult $ac1, %[diff2], %[k1Res2] \n\t"
"extr.w %[state3], $ac0, 16 \n\t"
"extr.w %[state7], $ac1, 16 \n\t"
"addu %[state3], %[state2], %[state3] \n\t"
"addu %[state7], %[state6], %[state7] \n\t"
"addiu %[state2], %[tmp21], 0 \n\t"
"addiu %[state6], %[tmp22], 0 \n\t"
// add two allpass outputs, divide by two and round
"addu %[out32], %[state3], %[state7] \n\t"
"addiu %[out32], %[out32], 1024 \n\t"
"sra %[out32], %[out32], 11 \n\t"
: [state3] "+r" (state3), [state6] "+r" (state6),
[state2] "+r" (state2), [diff2] "=&r" (diff2),
[out32] "=r" (out32), [diff1] "=&r" (diff1), [state7] "+r" (state7)
: [tmp22] "r" (tmp22), [tmp21] "r" (tmp21),
[k1Res2] "r" (k1Res2), [k2Res2] "r" (k2Res2)
: "hi", "lo", "$ac1hi", "$ac1lo"
);
// limit amplitude to prevent wrap-around, and write to output array
*out++ = WebRtcSpl_SatW32ToW16(out32);
}
#else // #if defined(MIPS_DSP_R2_LE)
int32_t tmp1, tmp2, diff;
int32_t in32;
len1 = (len >> 1)/4;
for (i = len1; i > 0; i--) {
// lower allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state1;
tmp1 = MUL_ACCUM_1(kResampleAllpass2[0], diff, state0);
state0 = in32;
diff = tmp1 - state2;
tmp2 = MUL_ACCUM_2(kResampleAllpass2[1], diff, state1);
state1 = tmp1;
diff = tmp2 - state3;
state3 = MUL_ACCUM_2(kResampleAllpass2[2], diff, state2);
state2 = tmp2;
// upper allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state5;
tmp1 = MUL_ACCUM_1(kResampleAllpass1[0], diff, state4);
state4 = in32;
diff = tmp1 - state6;
tmp2 = MUL_ACCUM_1(kResampleAllpass1[1], diff, state5);
state5 = tmp1;
diff = tmp2 - state7;
state7 = MUL_ACCUM_2(kResampleAllpass1[2], diff, state6);
state6 = tmp2;
// add two allpass outputs, divide by two and round
out32 = (state3 + state7 + 1024) >> 11;
// limit amplitude to prevent wrap-around, and write to output array
*out++ = WebRtcSpl_SatW32ToW16(out32);
// lower allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state1;
tmp1 = MUL_ACCUM_1(kResampleAllpass2[0], diff, state0);
state0 = in32;
diff = tmp1 - state2;
tmp2 = MUL_ACCUM_2(kResampleAllpass2[1], diff, state1);
state1 = tmp1;
diff = tmp2 - state3;
state3 = MUL_ACCUM_2(kResampleAllpass2[2], diff, state2);
state2 = tmp2;
// upper allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state5;
tmp1 = MUL_ACCUM_1(kResampleAllpass1[0], diff, state4);
state4 = in32;
diff = tmp1 - state6;
tmp2 = MUL_ACCUM_1(kResampleAllpass1[1], diff, state5);
state5 = tmp1;
diff = tmp2 - state7;
state7 = MUL_ACCUM_2(kResampleAllpass1[2], diff, state6);
state6 = tmp2;
// add two allpass outputs, divide by two and round
out32 = (state3 + state7 + 1024) >> 11;
// limit amplitude to prevent wrap-around, and write to output array
*out++ = WebRtcSpl_SatW32ToW16(out32);
// lower allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state1;
tmp1 = MUL_ACCUM_1(kResampleAllpass2[0], diff, state0);
state0 = in32;
diff = tmp1 - state2;
tmp2 = MUL_ACCUM_2(kResampleAllpass2[1], diff, state1);
state1 = tmp1;
diff = tmp2 - state3;
state3 = MUL_ACCUM_2(kResampleAllpass2[2], diff, state2);
state2 = tmp2;
// upper allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state5;
tmp1 = MUL_ACCUM_1(kResampleAllpass1[0], diff, state4);
state4 = in32;
diff = tmp1 - state6;
tmp2 = MUL_ACCUM_1(kResampleAllpass1[1], diff, state5);
state5 = tmp1;
diff = tmp2 - state7;
state7 = MUL_ACCUM_2(kResampleAllpass1[2], diff, state6);
state6 = tmp2;
// add two allpass outputs, divide by two and round
out32 = (state3 + state7 + 1024) >> 11;
// limit amplitude to prevent wrap-around, and write to output array
*out++ = WebRtcSpl_SatW32ToW16(out32);
// lower allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state1;
tmp1 = MUL_ACCUM_1(kResampleAllpass2[0], diff, state0);
state0 = in32;
diff = tmp1 - state2;
tmp2 = MUL_ACCUM_2(kResampleAllpass2[1], diff, state1);
state1 = tmp1;
diff = tmp2 - state3;
state3 = MUL_ACCUM_2(kResampleAllpass2[2], diff, state2);
state2 = tmp2;
// upper allpass filter
in32 = (int32_t)(*in++) << 10;
diff = in32 - state5;
tmp1 = MUL_ACCUM_1(kResampleAllpass1[0], diff, state4);
state4 = in32;
diff = tmp1 - state6;
tmp2 = MUL_ACCUM_1(kResampleAllpass1[1], diff, state5);
state5 = tmp1;
diff = tmp2 - state7;
state7 = MUL_ACCUM_2(kResampleAllpass1[2], diff, state6);
state6 = tmp2;
// add two allpass outputs, divide by two and round
out32 = (state3 + state7 + 1024) >> 11;
// limit amplitude to prevent wrap-around, and write to output array
*out++ = WebRtcSpl_SatW32ToW16(out32);
}
#endif // #if defined(MIPS_DSP_R2_LE)
__asm__ volatile (
"sw %[state0], 0(%[filtState]) \n\t"
"sw %[state1], 4(%[filtState]) \n\t"
"sw %[state2], 8(%[filtState]) \n\t"
"sw %[state3], 12(%[filtState]) \n\t"
"sw %[state4], 16(%[filtState]) \n\t"
"sw %[state5], 20(%[filtState]) \n\t"
"sw %[state6], 24(%[filtState]) \n\t"
"sw %[state7], 28(%[filtState]) \n\t"
:
: [state0] "r" (state0), [state1] "r" (state1), [state2] "r" (state2),
[state3] "r" (state3), [state4] "r" (state4), [state5] "r" (state5),
[state6] "r" (state6), [state7] "r" (state7), [filtState] "r" (filtState)
: "memory"
);
}
#endif // #if defined(MIPS32_LE)

239
webrtc/common_audio/signal_processing/resample_fractional.c Normal file
View File

@ -0,0 +1,239 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the resampling functions between 48, 44, 32 and 24 kHz.
* The description headers can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
// interpolation coefficients
static const int16_t kCoefficients48To32[2][8] = {
{778, -2050, 1087, 23285, 12903, -3783, 441, 222},
{222, 441, -3783, 12903, 23285, 1087, -2050, 778}
};
static const int16_t kCoefficients32To24[3][8] = {
{767, -2362, 2434, 24406, 10620, -3838, 721, 90},
{386, -381, -2646, 19062, 19062, -2646, -381, 386},
{90, 721, -3838, 10620, 24406, 2434, -2362, 767}
};
static const int16_t kCoefficients44To32[4][9] = {
{117, -669, 2245, -6183, 26267, 13529, -3245, 845, -138},
{-101, 612, -2283, 8532, 29790, -5138, 1789, -524, 91},
{50, -292, 1016, -3064, 32010, 3933, -1147, 315, -53},
{-156, 974, -3863, 18603, 21691, -6246, 2353, -712, 126}
};
// Resampling ratio: 2/3
// input: int32_t (normalized, not saturated) :: size 3 * K
// output: int32_t (shifted 15 positions to the left, + offset 16384) :: size 2 * K
// K: number of blocks
void WebRtcSpl_Resample48khzTo32khz(const int32_t *In, int32_t *Out, size_t K)
{
/////////////////////////////////////////////////////////////
// Filter operation:
//
// Perform resampling (3 input samples -> 2 output samples);
// process in sub blocks of size 3 samples.
int32_t tmp;
size_t m;
for (m = 0; m < K; m++)
{
tmp = 1 << 14;
tmp += kCoefficients48To32[0][0] * In[0];
tmp += kCoefficients48To32[0][1] * In[1];
tmp += kCoefficients48To32[0][2] * In[2];
tmp += kCoefficients48To32[0][3] * In[3];
tmp += kCoefficients48To32[0][4] * In[4];
tmp += kCoefficients48To32[0][5] * In[5];
tmp += kCoefficients48To32[0][6] * In[6];
tmp += kCoefficients48To32[0][7] * In[7];
Out[0] = tmp;
tmp = 1 << 14;
tmp += kCoefficients48To32[1][0] * In[1];
tmp += kCoefficients48To32[1][1] * In[2];
tmp += kCoefficients48To32[1][2] * In[3];
tmp += kCoefficients48To32[1][3] * In[4];
tmp += kCoefficients48To32[1][4] * In[5];
tmp += kCoefficients48To32[1][5] * In[6];
tmp += kCoefficients48To32[1][6] * In[7];
tmp += kCoefficients48To32[1][7] * In[8];
Out[1] = tmp;
// update pointers
In += 3;
Out += 2;
}
}
// Resampling ratio: 3/4
// input: int32_t (normalized, not saturated) :: size 4 * K
// output: int32_t (shifted 15 positions to the left, + offset 16384) :: size 3 * K
// K: number of blocks
void WebRtcSpl_Resample32khzTo24khz(const int32_t *In, int32_t *Out, size_t K)
{
/////////////////////////////////////////////////////////////
// Filter operation:
//
// Perform resampling (4 input samples -> 3 output samples);
// process in sub blocks of size 4 samples.
size_t m;
int32_t tmp;
for (m = 0; m < K; m++)
{
tmp = 1 << 14;
tmp += kCoefficients32To24[0][0] * In[0];
tmp += kCoefficients32To24[0][1] * In[1];
tmp += kCoefficients32To24[0][2] * In[2];
tmp += kCoefficients32To24[0][3] * In[3];
tmp += kCoefficients32To24[0][4] * In[4];
tmp += kCoefficients32To24[0][5] * In[5];
tmp += kCoefficients32To24[0][6] * In[6];
tmp += kCoefficients32To24[0][7] * In[7];
Out[0] = tmp;
tmp = 1 << 14;
tmp += kCoefficients32To24[1][0] * In[1];
tmp += kCoefficients32To24[1][1] * In[2];
tmp += kCoefficients32To24[1][2] * In[3];
tmp += kCoefficients32To24[1][3] * In[4];
tmp += kCoefficients32To24[1][4] * In[5];
tmp += kCoefficients32To24[1][5] * In[6];
tmp += kCoefficients32To24[1][6] * In[7];
tmp += kCoefficients32To24[1][7] * In[8];
Out[1] = tmp;
tmp = 1 << 14;
tmp += kCoefficients32To24[2][0] * In[2];
tmp += kCoefficients32To24[2][1] * In[3];
tmp += kCoefficients32To24[2][2] * In[4];
tmp += kCoefficients32To24[2][3] * In[5];
tmp += kCoefficients32To24[2][4] * In[6];
tmp += kCoefficients32To24[2][5] * In[7];
tmp += kCoefficients32To24[2][6] * In[8];
tmp += kCoefficients32To24[2][7] * In[9];
Out[2] = tmp;
// update pointers
In += 4;
Out += 3;
}
}
//
// fractional resampling filters
// Fout = 11/16 * Fin
// Fout = 8/11 * Fin
//
// compute two inner-products and store them to output array
static void WebRtcSpl_ResampDotProduct(const int32_t *in1, const int32_t *in2,
const int16_t *coef_ptr, int32_t *out1,
int32_t *out2)
{
int32_t tmp1 = 16384;
int32_t tmp2 = 16384;
int16_t coef;
coef = coef_ptr[0];
tmp1 += coef * in1[0];
tmp2 += coef * in2[-0];
coef = coef_ptr[1];
tmp1 += coef * in1[1];
tmp2 += coef * in2[-1];
coef = coef_ptr[2];
tmp1 += coef * in1[2];
tmp2 += coef * in2[-2];
coef = coef_ptr[3];
tmp1 += coef * in1[3];
tmp2 += coef * in2[-3];
coef = coef_ptr[4];
tmp1 += coef * in1[4];
tmp2 += coef * in2[-4];
coef = coef_ptr[5];
tmp1 += coef * in1[5];
tmp2 += coef * in2[-5];
coef = coef_ptr[6];
tmp1 += coef * in1[6];
tmp2 += coef * in2[-6];
coef = coef_ptr[7];
tmp1 += coef * in1[7];
tmp2 += coef * in2[-7];
coef = coef_ptr[8];
*out1 = tmp1 + coef * in1[8];
*out2 = tmp2 + coef * in2[-8];
}
// Resampling ratio: 8/11
// input: int32_t (normalized, not saturated) :: size 11 * K
// output: int32_t (shifted 15 positions to the left, + offset 16384) :: size 8 * K
// K: number of blocks
void WebRtcSpl_Resample44khzTo32khz(const int32_t *In, int32_t *Out, size_t K)
{
/////////////////////////////////////////////////////////////
// Filter operation:
//
// Perform resampling (11 input samples -> 8 output samples);
// process in sub blocks of size 11 samples.
int32_t tmp;
size_t m;
for (m = 0; m < K; m++)
{
tmp = 1 << 14;
// first output sample
Out[0] = ((int32_t)In[3] << 15) + tmp;
// sum and accumulate filter coefficients and input samples
tmp += kCoefficients44To32[3][0] * In[5];
tmp += kCoefficients44To32[3][1] * In[6];
tmp += kCoefficients44To32[3][2] * In[7];
tmp += kCoefficients44To32[3][3] * In[8];
tmp += kCoefficients44To32[3][4] * In[9];
tmp += kCoefficients44To32[3][5] * In[10];
tmp += kCoefficients44To32[3][6] * In[11];
tmp += kCoefficients44To32[3][7] * In[12];
tmp += kCoefficients44To32[3][8] * In[13];
Out[4] = tmp;
// sum and accumulate filter coefficients and input samples
WebRtcSpl_ResampDotProduct(&In[0], &In[17], kCoefficients44To32[0], &Out[1], &Out[7]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_ResampDotProduct(&In[2], &In[15], kCoefficients44To32[1], &Out[2], &Out[6]);
// sum and accumulate filter coefficients and input samples
WebRtcSpl_ResampDotProduct(&In[3], &In[14], kCoefficients44To32[2], &Out[3], &Out[5]);
// update pointers
In += 11;
Out += 8;
}
}

140
webrtc/common_audio/signal_processing/spl_init.c Normal file
View File

@ -0,0 +1,140 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/* The global function contained in this file initializes SPL function
* pointers, currently only for ARM platforms.
*
* Some code came from common/rtcd.c in the WebM project.
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/system_wrappers/interface/cpu_features_wrapper.h"
/* Declare function pointers. */
MaxAbsValueW16 WebRtcSpl_MaxAbsValueW16;
MaxAbsValueW32 WebRtcSpl_MaxAbsValueW32;
MaxValueW16 WebRtcSpl_MaxValueW16;
MaxValueW32 WebRtcSpl_MaxValueW32;
MinValueW16 WebRtcSpl_MinValueW16;
MinValueW32 WebRtcSpl_MinValueW32;
CrossCorrelation WebRtcSpl_CrossCorrelation;
DownsampleFast WebRtcSpl_DownsampleFast;
ScaleAndAddVectorsWithRound WebRtcSpl_ScaleAndAddVectorsWithRound;
#if (defined(WEBRTC_DETECT_NEON) || !defined(WEBRTC_HAS_NEON)) && \
!defined(MIPS32_LE)
/* Initialize function pointers to the generic C version. */
static void InitPointersToC() {
WebRtcSpl_MaxAbsValueW16 = WebRtcSpl_MaxAbsValueW16C;
WebRtcSpl_MaxAbsValueW32 = WebRtcSpl_MaxAbsValueW32C;
WebRtcSpl_MaxValueW16 = WebRtcSpl_MaxValueW16C;
WebRtcSpl_MaxValueW32 = WebRtcSpl_MaxValueW32C;
WebRtcSpl_MinValueW16 = WebRtcSpl_MinValueW16C;
WebRtcSpl_MinValueW32 = WebRtcSpl_MinValueW32C;
WebRtcSpl_CrossCorrelation = WebRtcSpl_CrossCorrelationC;
WebRtcSpl_DownsampleFast = WebRtcSpl_DownsampleFastC;
WebRtcSpl_ScaleAndAddVectorsWithRound =
WebRtcSpl_ScaleAndAddVectorsWithRoundC;
}
#endif
#if defined(WEBRTC_DETECT_NEON) || defined(WEBRTC_HAS_NEON)
/* Initialize function pointers to the Neon version. */
static void InitPointersToNeon() {
WebRtcSpl_MaxAbsValueW16 = WebRtcSpl_MaxAbsValueW16Neon;
WebRtcSpl_MaxAbsValueW32 = WebRtcSpl_MaxAbsValueW32Neon;
WebRtcSpl_MaxValueW16 = WebRtcSpl_MaxValueW16Neon;
WebRtcSpl_MaxValueW32 = WebRtcSpl_MaxValueW32Neon;
WebRtcSpl_MinValueW16 = WebRtcSpl_MinValueW16Neon;
WebRtcSpl_MinValueW32 = WebRtcSpl_MinValueW32Neon;
WebRtcSpl_CrossCorrelation = WebRtcSpl_CrossCorrelationNeon;
WebRtcSpl_DownsampleFast = WebRtcSpl_DownsampleFastNeon;
WebRtcSpl_ScaleAndAddVectorsWithRound =
WebRtcSpl_ScaleAndAddVectorsWithRoundC;
}
#endif
#if defined(MIPS32_LE)
/* Initialize function pointers to the MIPS version. */
static void InitPointersToMIPS() {
WebRtcSpl_MaxAbsValueW16 = WebRtcSpl_MaxAbsValueW16_mips;
WebRtcSpl_MaxValueW16 = WebRtcSpl_MaxValueW16_mips;
WebRtcSpl_MaxValueW32 = WebRtcSpl_MaxValueW32_mips;
WebRtcSpl_MinValueW16 = WebRtcSpl_MinValueW16_mips;
WebRtcSpl_MinValueW32 = WebRtcSpl_MinValueW32_mips;
WebRtcSpl_CrossCorrelation = WebRtcSpl_CrossCorrelation_mips;
WebRtcSpl_DownsampleFast = WebRtcSpl_DownsampleFast_mips;
#if defined(MIPS_DSP_R1_LE)
WebRtcSpl_MaxAbsValueW32 = WebRtcSpl_MaxAbsValueW32_mips;
WebRtcSpl_ScaleAndAddVectorsWithRound =
WebRtcSpl_ScaleAndAddVectorsWithRound_mips;
#else
WebRtcSpl_MaxAbsValueW32 = WebRtcSpl_MaxAbsValueW32C;
WebRtcSpl_ScaleAndAddVectorsWithRound =
WebRtcSpl_ScaleAndAddVectorsWithRoundC;
#endif
}
#endif
static void InitFunctionPointers(void) {
#if defined(WEBRTC_DETECT_NEON)
if ((WebRtc_GetCPUFeaturesARM() & kCPUFeatureNEON) != 0) {
InitPointersToNeon();
} else {
InitPointersToC();
}
#elif defined(WEBRTC_HAS_NEON)
InitPointersToNeon();
#elif defined(MIPS32_LE)
InitPointersToMIPS();
#else
InitPointersToC();
#endif /* WEBRTC_DETECT_NEON */
}
#if defined(WEBRTC_POSIX)
#include <pthread.h>
static void once(void (*func)(void)) {
static pthread_once_t lock = PTHREAD_ONCE_INIT;
pthread_once(&lock, func);
}
#elif defined(_WIN32)
#include <windows.h>
static void once(void (*func)(void)) {
/* Didn't use InitializeCriticalSection() since there's no race-free context
* in which to execute it.
*
* TODO(kma): Change to different implementation (e.g.
* InterlockedCompareExchangePointer) to avoid issues similar to
* http://code.google.com/p/webm/issues/detail?id=467.
*/
static CRITICAL_SECTION lock = {(void *)((size_t)-1), -1, 0, 0, 0, 0};
static int done = 0;
EnterCriticalSection(&lock);
if (!done) {
func();
done = 1;
}
LeaveCriticalSection(&lock);
}
/* There's no fallback version as an #else block here to ensure thread safety.
* In case of neither pthread for WEBRTC_POSIX nor _WIN32 is present, build
* system should pick it up.
*/
#endif /* WEBRTC_POSIX */
void WebRtcSpl_Init() {
once(InitFunctionPointers);
}

184
webrtc/common_audio/signal_processing/spl_sqrt.c Normal file
View File

@ -0,0 +1,184 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_Sqrt().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include <assert.h>
int32_t WebRtcSpl_SqrtLocal(int32_t in);
int32_t WebRtcSpl_SqrtLocal(int32_t in)
{
int16_t x_half, t16;
int32_t A, B, x2;
/* The following block performs:
y=in/2
x=y-2^30
x_half=x/2^31
t = 1 + (x_half) - 0.5*((x_half)^2) + 0.5*((x_half)^3) - 0.625*((x_half)^4)
+ 0.875*((x_half)^5)
*/
B = in / 2;
B = B - ((int32_t)0x40000000); // B = in/2 - 1/2
x_half = (int16_t)(B >> 16); // x_half = x/2 = (in-1)/2
B = B + ((int32_t)0x40000000); // B = 1 + x/2
B = B + ((int32_t)0x40000000); // Add 0.5 twice (since 1.0 does not exist in Q31)
x2 = ((int32_t)x_half) * ((int32_t)x_half) * 2; // A = (x/2)^2
A = -x2; // A = -(x/2)^2
B = B + (A >> 1); // B = 1 + x/2 - 0.5*(x/2)^2
A >>= 16;
A = A * A * 2; // A = (x/2)^4
t16 = (int16_t)(A >> 16);
B += -20480 * t16 * 2; // B = B - 0.625*A
// After this, B = 1 + x/2 - 0.5*(x/2)^2 - 0.625*(x/2)^4
A = x_half * t16 * 2; // A = (x/2)^5
t16 = (int16_t)(A >> 16);
B += 28672 * t16 * 2; // B = B + 0.875*A
// After this, B = 1 + x/2 - 0.5*(x/2)^2 - 0.625*(x/2)^4 + 0.875*(x/2)^5
t16 = (int16_t)(x2 >> 16);
A = x_half * t16 * 2; // A = x/2^3
B = B + (A >> 1); // B = B + 0.5*A
// After this, B = 1 + x/2 - 0.5*(x/2)^2 + 0.5*(x/2)^3 - 0.625*(x/2)^4 + 0.875*(x/2)^5
B = B + ((int32_t)32768); // Round off bit
return B;
}
int32_t WebRtcSpl_Sqrt(int32_t value)
{
/*
Algorithm:
Six term Taylor Series is used here to compute the square root of a number
y^0.5 = (1+x)^0.5 where x = y-1
= 1+(x/2)-0.5*((x/2)^2+0.5*((x/2)^3-0.625*((x/2)^4+0.875*((x/2)^5)
0.5 <= x < 1
Example of how the algorithm works, with ut=sqrt(in), and
with in=73632 and ut=271 (even shift value case):
in=73632
y= in/131072
x=y-1
t = 1 + (x/2) - 0.5*((x/2)^2) + 0.5*((x/2)^3) - 0.625*((x/2)^4) + 0.875*((x/2)^5)
ut=t*(1/sqrt(2))*512
or:
in=73632
in2=73632*2^14
y= in2/2^31
x=y-1
t = 1 + (x/2) - 0.5*((x/2)^2) + 0.5*((x/2)^3) - 0.625*((x/2)^4) + 0.875*((x/2)^5)
ut=t*(1/sqrt(2))
ut2=ut*2^9
which gives:
in = 73632
in2 = 1206386688
y = 0.56176757812500
x = -0.43823242187500
t = 0.74973506527313
ut = 0.53014274874797
ut2 = 2.714330873589594e+002
or:
in=73632
in2=73632*2^14
y=in2/2
x=y-2^30
x_half=x/2^31
t = 1 + (x_half) - 0.5*((x_half)^2) + 0.5*((x_half)^3) - 0.625*((x_half)^4)
+ 0.875*((x_half)^5)
ut=t*(1/sqrt(2))
ut2=ut*2^9
which gives:
in = 73632
in2 = 1206386688
y = 603193344
x = -470548480
x_half = -0.21911621093750
t = 0.74973506527313
ut = 0.53014274874797
ut2 = 2.714330873589594e+002
*/
int16_t x_norm, nshift, t16, sh;
int32_t A;
int16_t k_sqrt_2 = 23170; // 1/sqrt2 (==5a82)
A = value;
if (A == 0)
return (int32_t)0; // sqrt(0) = 0
sh = WebRtcSpl_NormW32(A); // # shifts to normalize A
A = WEBRTC_SPL_LSHIFT_W32(A, sh); // Normalize A
if (A < (WEBRTC_SPL_WORD32_MAX - 32767))
{
A = A + ((int32_t)32768); // Round off bit
} else
{
A = WEBRTC_SPL_WORD32_MAX;
}
x_norm = (int16_t)(A >> 16); // x_norm = AH
nshift = (sh / 2);
assert(nshift >= 0);
A = (int32_t)WEBRTC_SPL_LSHIFT_W32((int32_t)x_norm, 16);
A = WEBRTC_SPL_ABS_W32(A); // A = abs(x_norm<<16)
A = WebRtcSpl_SqrtLocal(A); // A = sqrt(A)
if (2 * nshift == sh) {
// Even shift value case
t16 = (int16_t)(A >> 16); // t16 = AH
A = k_sqrt_2 * t16 * 2; // A = 1/sqrt(2)*t16
A = A + ((int32_t)32768); // Round off
A = A & ((int32_t)0x7fff0000); // Round off
A >>= 15; // A = A>>16
} else
{
A >>= 16; // A = A>>16
}
A = A & ((int32_t)0x0000ffff);
A >>= nshift; // De-normalize the result.
return A;
}

77
webrtc/common_audio/signal_processing/spl_sqrt_floor.c Normal file
View File

@ -0,0 +1,77 @@
/*
* Written by Wilco Dijkstra, 1996. The following email exchange establishes the
* license.
*
* From: Wilco Dijkstra <Wilco.Dijkstra@ntlworld.com>
* Date: Fri, Jun 24, 2011 at 3:20 AM
* Subject: Re: sqrt routine
* To: Kevin Ma <kma@google.com>
* Hi Kevin,
* Thanks for asking. Those routines are public domain (originally posted to
* comp.sys.arm a long time ago), so you can use them freely for any purpose.
* Cheers,
* Wilco
*
* ----- Original Message -----
* From: "Kevin Ma" <kma@google.com>
* To: <Wilco.Dijkstra@ntlworld.com>
* Sent: Thursday, June 23, 2011 11:44 PM
* Subject: Fwd: sqrt routine
* Hi Wilco,
* I saw your sqrt routine from several web sites, including
* http://www.finesse.demon.co.uk/steven/sqrt.html.
* Just wonder if there's any copyright information with your Successive
* approximation routines, or if I can freely use it for any purpose.
* Thanks.
* Kevin
*/
// Minor modifications in code style for WebRTC, 2012.
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
/*
* Algorithm:
* Successive approximation of the equation (root + delta) ^ 2 = N
* until delta < 1. If delta < 1 we have the integer part of SQRT (N).
* Use delta = 2^i for i = 15 .. 0.
*
* Output precision is 16 bits. Note for large input values (close to
* 0x7FFFFFFF), bit 15 (the highest bit of the low 16-bit half word)
* contains the MSB information (a non-sign value). Do with caution
* if you need to cast the output to int16_t type.
*
* If the input value is negative, it returns 0.
*/
#define WEBRTC_SPL_SQRT_ITER(N) \
try1 = root + (1 << (N)); \
if (value >= try1 << (N)) \
{ \
value -= try1 << (N); \
root |= 2 << (N); \
}
int32_t WebRtcSpl_SqrtFloor(int32_t value)
{
int32_t root = 0, try1;
WEBRTC_SPL_SQRT_ITER (15);
WEBRTC_SPL_SQRT_ITER (14);
WEBRTC_SPL_SQRT_ITER (13);
WEBRTC_SPL_SQRT_ITER (12);
WEBRTC_SPL_SQRT_ITER (11);
WEBRTC_SPL_SQRT_ITER (10);
WEBRTC_SPL_SQRT_ITER ( 9);
WEBRTC_SPL_SQRT_ITER ( 8);
WEBRTC_SPL_SQRT_ITER ( 7);
WEBRTC_SPL_SQRT_ITER ( 6);
WEBRTC_SPL_SQRT_ITER ( 5);
WEBRTC_SPL_SQRT_ITER ( 4);
WEBRTC_SPL_SQRT_ITER ( 3);
WEBRTC_SPL_SQRT_ITER ( 2);
WEBRTC_SPL_SQRT_ITER ( 1);
WEBRTC_SPL_SQRT_ITER ( 0);
return root >> 1;
}

110
webrtc/common_audio/signal_processing/spl_sqrt_floor_arm.S Normal file
View File

@ -0,0 +1,110 @@
@
@ Written by Wilco Dijkstra, 1996. The following email exchange establishes the
@ license.
@
@ From: Wilco Dijkstra <Wilco.Dijkstra@ntlworld.com>
@ Date: Fri, Jun 24, 2011 at 3:20 AM
@ Subject: Re: sqrt routine
@ To: Kevin Ma <kma@google.com>
@ Hi Kevin,
@ Thanks for asking. Those routines are public domain (originally posted to
@ comp.sys.arm a long time ago), so you can use them freely for any purpose.
@ Cheers,
@ Wilco
@
@ ----- Original Message -----
@ From: "Kevin Ma" <kma@google.com>
@ To: <Wilco.Dijkstra@ntlworld.com>
@ Sent: Thursday, June 23, 2011 11:44 PM
@ Subject: Fwd: sqrt routine
@ Hi Wilco,
@ I saw your sqrt routine from several web sites, including
@ http://www.finesse.demon.co.uk/steven/sqrt.html.
@ Just wonder if there's any copyright information with your Successive
@ approximation routines, or if I can freely use it for any purpose.
@ Thanks.
@ Kevin
@ Minor modifications in code style for WebRTC, 2012.
@ Output is bit-exact with the reference C code in spl_sqrt_floor.c.
@ Input : r0 32 bit unsigned integer
@ Output: r0 = INT (SQRT (r0)), precision is 16 bits
@ Registers touched: r1, r2
#include "webrtc/system_wrappers/interface/asm_defines.h"
GLOBAL_FUNCTION WebRtcSpl_SqrtFloor
.align 2
DEFINE_FUNCTION WebRtcSpl_SqrtFloor
mov r1, #3 << 30
mov r2, #1 << 30
@ unroll for i = 0 .. 15
cmp r0, r2, ror #2 * 0
subhs r0, r0, r2, ror #2 * 0
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 1
subhs r0, r0, r2, ror #2 * 1
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 2
subhs r0, r0, r2, ror #2 * 2
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 3
subhs r0, r0, r2, ror #2 * 3
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 4
subhs r0, r0, r2, ror #2 * 4
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 5
subhs r0, r0, r2, ror #2 * 5
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 6
subhs r0, r0, r2, ror #2 * 6
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 7
subhs r0, r0, r2, ror #2 * 7
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 8
subhs r0, r0, r2, ror #2 * 8
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 9
subhs r0, r0, r2, ror #2 * 9
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 10
subhs r0, r0, r2, ror #2 * 10
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 11
subhs r0, r0, r2, ror #2 * 11
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 12
subhs r0, r0, r2, ror #2 * 12
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 13
subhs r0, r0, r2, ror #2 * 13
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 14
subhs r0, r0, r2, ror #2 * 14
adc r2, r1, r2, lsl #1
cmp r0, r2, ror #2 * 15
subhs r0, r0, r2, ror #2 * 15
adc r2, r1, r2, lsl #1
bic r0, r2, #3 << 30 @ for rounding add: cmp r0, r2 adc r2, #1
bx lr

207
webrtc/common_audio/signal_processing/spl_sqrt_floor_mips.c Normal file
View File

@ -0,0 +1,207 @@
/*
* Written by Wilco Dijkstra, 1996. The following email exchange establishes the
* license.
*
* From: Wilco Dijkstra <Wilco.Dijkstra@ntlworld.com>
* Date: Fri, Jun 24, 2011 at 3:20 AM
* Subject: Re: sqrt routine
* To: Kevin Ma <kma@google.com>
* Hi Kevin,
* Thanks for asking. Those routines are public domain (originally posted to
* comp.sys.arm a long time ago), so you can use them freely for any purpose.
* Cheers,
* Wilco
*
* ----- Original Message -----
* From: "Kevin Ma" <kma@google.com>
* To: <Wilco.Dijkstra@ntlworld.com>
* Sent: Thursday, June 23, 2011 11:44 PM
* Subject: Fwd: sqrt routine
* Hi Wilco,
* I saw your sqrt routine from several web sites, including
* http://www.finesse.demon.co.uk/steven/sqrt.html.
* Just wonder if there's any copyright information with your Successive
* approximation routines, or if I can freely use it for any purpose.
* Thanks.
* Kevin
*/
// Minor modifications in code style for WebRTC, 2012.
// Code optimizations for MIPS, 2013.
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
/*
* Algorithm:
* Successive approximation of the equation (root + delta) ^ 2 = N
* until delta < 1. If delta < 1 we have the integer part of SQRT (N).
* Use delta = 2^i for i = 15 .. 0.
*
* Output precision is 16 bits. Note for large input values (close to
* 0x7FFFFFFF), bit 15 (the highest bit of the low 16-bit half word)
* contains the MSB information (a non-sign value). Do with caution
* if you need to cast the output to int16_t type.
*
* If the input value is negative, it returns 0.
*/
int32_t WebRtcSpl_SqrtFloor(int32_t value)
{
int32_t root = 0, tmp1, tmp2, tmp3, tmp4;
__asm __volatile(
".set push \n\t"
".set noreorder \n\t"
"lui %[tmp1], 0x4000 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"sub %[tmp3], %[value], %[tmp1] \n\t"
"lui %[tmp1], 0x1 \n\t"
"or %[tmp4], %[root], %[tmp1] \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x4000 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 14 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x8000 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x2000 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 13 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x4000 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x1000 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 12 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x2000 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x800 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 11 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x1000 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x400 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 10 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x800 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x200 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 9 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x400 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x100 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 8 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x200 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x80 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 7 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x100 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x40 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 6 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x80 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x20 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 5 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x40 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x10 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 4 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x20 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x8 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 3 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x10 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x4 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 2 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x8 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x2 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"sll %[tmp1], 1 \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"subu %[tmp3], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x4 \n\t"
"movz %[value], %[tmp3], %[tmp2] \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
"addiu %[tmp1], $0, 0x1 \n\t"
"addu %[tmp1], %[tmp1], %[root] \n\t"
"slt %[tmp2], %[value], %[tmp1] \n\t"
"ori %[tmp4], %[root], 0x2 \n\t"
"movz %[root], %[tmp4], %[tmp2] \n\t"
".set pop \n\t"
: [root] "+r" (root), [value] "+r" (value),
[tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2),
[tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4)
:
);
return root >> 1;
}

208
webrtc/common_audio/signal_processing/splitting_filter.c Normal file
View File

@ -0,0 +1,208 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the splitting filter functions.
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include <assert.h>
// Maximum number of samples in a low/high-band frame.
enum
{
kMaxBandFrameLength = 320 // 10 ms at 64 kHz.
};
// QMF filter coefficients in Q16.
static const uint16_t WebRtcSpl_kAllPassFilter1[3] = {6418, 36982, 57261};
static const uint16_t WebRtcSpl_kAllPassFilter2[3] = {21333, 49062, 63010};
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcSpl_AllPassQMF(...)
//
// Allpass filter used by the analysis and synthesis parts of the QMF filter.
//
// Input:
// - in_data : Input data sequence (Q10)
// - data_length : Length of data sequence (>2)
// - filter_coefficients : Filter coefficients (length 3, Q16)
//
// Input & Output:
// - filter_state : Filter state (length 6, Q10).
//
// Output:
// - out_data : Output data sequence (Q10), length equal to
// |data_length|
//
void WebRtcSpl_AllPassQMF(int32_t* in_data, size_t data_length,
int32_t* out_data, const uint16_t* filter_coefficients,
int32_t* filter_state)
{
// The procedure is to filter the input with three first order all pass filters
// (cascade operations).
//
// a_3 + q^-1 a_2 + q^-1 a_1 + q^-1
// y[n] = ----------- ----------- ----------- x[n]
// 1 + a_3q^-1 1 + a_2q^-1 1 + a_1q^-1
//
// The input vector |filter_coefficients| includes these three filter coefficients.
// The filter state contains the in_data state, in_data[-1], followed by
// the out_data state, out_data[-1]. This is repeated for each cascade.
// The first cascade filter will filter the |in_data| and store the output in
// |out_data|. The second will the take the |out_data| as input and make an
// intermediate storage in |in_data|, to save memory. The third, and final, cascade
// filter operation takes the |in_data| (which is the output from the previous cascade
// filter) and store the output in |out_data|.
// Note that the input vector values are changed during the process.
size_t k;
int32_t diff;
// First all-pass cascade; filter from in_data to out_data.
// Let y_i[n] indicate the output of cascade filter i (with filter coefficient a_i) at
// vector position n. Then the final output will be y[n] = y_3[n]
// First loop, use the states stored in memory.
// "diff" should be safe from wrap around since max values are 2^25
// diff = (x[0] - y_1[-1])
diff = WebRtcSpl_SubSatW32(in_data[0], filter_state[1]);
// y_1[0] = x[-1] + a_1 * (x[0] - y_1[-1])
out_data[0] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[0], diff, filter_state[0]);
// For the remaining loops, use previous values.
for (k = 1; k < data_length; k++)
{
// diff = (x[n] - y_1[n-1])
diff = WebRtcSpl_SubSatW32(in_data[k], out_data[k - 1]);
// y_1[n] = x[n-1] + a_1 * (x[n] - y_1[n-1])
out_data[k] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[0], diff, in_data[k - 1]);
}
// Update states.
filter_state[0] = in_data[data_length - 1]; // x[N-1], becomes x[-1] next time
filter_state[1] = out_data[data_length - 1]; // y_1[N-1], becomes y_1[-1] next time
// Second all-pass cascade; filter from out_data to in_data.
// diff = (y_1[0] - y_2[-1])
diff = WebRtcSpl_SubSatW32(out_data[0], filter_state[3]);
// y_2[0] = y_1[-1] + a_2 * (y_1[0] - y_2[-1])
in_data[0] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[1], diff, filter_state[2]);
for (k = 1; k < data_length; k++)
{
// diff = (y_1[n] - y_2[n-1])
diff = WebRtcSpl_SubSatW32(out_data[k], in_data[k - 1]);
// y_2[0] = y_1[-1] + a_2 * (y_1[0] - y_2[-1])
in_data[k] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[1], diff, out_data[k-1]);
}
filter_state[2] = out_data[data_length - 1]; // y_1[N-1], becomes y_1[-1] next time
filter_state[3] = in_data[data_length - 1]; // y_2[N-1], becomes y_2[-1] next time
// Third all-pass cascade; filter from in_data to out_data.
// diff = (y_2[0] - y[-1])
diff = WebRtcSpl_SubSatW32(in_data[0], filter_state[5]);
// y[0] = y_2[-1] + a_3 * (y_2[0] - y[-1])
out_data[0] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[2], diff, filter_state[4]);
for (k = 1; k < data_length; k++)
{
// diff = (y_2[n] - y[n-1])
diff = WebRtcSpl_SubSatW32(in_data[k], out_data[k - 1]);
// y[n] = y_2[n-1] + a_3 * (y_2[n] - y[n-1])
out_data[k] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[2], diff, in_data[k-1]);
}
filter_state[4] = in_data[data_length - 1]; // y_2[N-1], becomes y_2[-1] next time
filter_state[5] = out_data[data_length - 1]; // y[N-1], becomes y[-1] next time
}
void WebRtcSpl_AnalysisQMF(const int16_t* in_data, size_t in_data_length,
int16_t* low_band, int16_t* high_band,
int32_t* filter_state1, int32_t* filter_state2)
{
size_t i;
int16_t k;
int32_t tmp;
int32_t half_in1[kMaxBandFrameLength];
int32_t half_in2[kMaxBandFrameLength];
int32_t filter1[kMaxBandFrameLength];
int32_t filter2[kMaxBandFrameLength];
const size_t band_length = in_data_length / 2;
assert(in_data_length % 2 == 0);
assert(band_length <= kMaxBandFrameLength);
// Split even and odd samples. Also shift them to Q10.
for (i = 0, k = 0; i < band_length; i++, k += 2)
{
half_in2[i] = WEBRTC_SPL_LSHIFT_W32((int32_t)in_data[k], 10);
half_in1[i] = WEBRTC_SPL_LSHIFT_W32((int32_t)in_data[k + 1], 10);
}
// All pass filter even and odd samples, independently.
WebRtcSpl_AllPassQMF(half_in1, band_length, filter1,
WebRtcSpl_kAllPassFilter1, filter_state1);
WebRtcSpl_AllPassQMF(half_in2, band_length, filter2,
WebRtcSpl_kAllPassFilter2, filter_state2);
// Take the sum and difference of filtered version of odd and even
// branches to get upper & lower band.
for (i = 0; i < band_length; i++)
{
tmp = (filter1[i] + filter2[i] + 1024) >> 11;
low_band[i] = WebRtcSpl_SatW32ToW16(tmp);
tmp = (filter1[i] - filter2[i] + 1024) >> 11;
high_band[i] = WebRtcSpl_SatW32ToW16(tmp);
}
}
void WebRtcSpl_SynthesisQMF(const int16_t* low_band, const int16_t* high_band,
size_t band_length, int16_t* out_data,
int32_t* filter_state1, int32_t* filter_state2)
{
int32_t tmp;
int32_t half_in1[kMaxBandFrameLength];
int32_t half_in2[kMaxBandFrameLength];
int32_t filter1[kMaxBandFrameLength];
int32_t filter2[kMaxBandFrameLength];
size_t i;
int16_t k;
assert(band_length <= kMaxBandFrameLength);
// Obtain the sum and difference channels out of upper and lower-band channels.
// Also shift to Q10 domain.
for (i = 0; i < band_length; i++)
{
tmp = (int32_t)low_band[i] + (int32_t)high_band[i];
half_in1[i] = WEBRTC_SPL_LSHIFT_W32(tmp, 10);
tmp = (int32_t)low_band[i] - (int32_t)high_band[i];
half_in2[i] = WEBRTC_SPL_LSHIFT_W32(tmp, 10);
}
// all-pass filter the sum and difference channels
WebRtcSpl_AllPassQMF(half_in1, band_length, filter1,
WebRtcSpl_kAllPassFilter2, filter_state1);
WebRtcSpl_AllPassQMF(half_in2, band_length, filter2,
WebRtcSpl_kAllPassFilter1, filter_state2);
// The filtered signals are even and odd samples of the output. Combine
// them. The signals are Q10 should shift them back to Q0 and take care of
// saturation.
for (i = 0, k = 0; i < band_length; i++)
{
tmp = (filter2[i] + 512) >> 10;
out_data[k++] = WebRtcSpl_SatW32ToW16(tmp);
tmp = (filter1[i] + 512) >> 10;
out_data[k++] = WebRtcSpl_SatW32ToW16(tmp);
}
}

35
webrtc/common_audio/signal_processing/sqrt_of_one_minus_x_squared.c Normal file
View File

@ -0,0 +1,35 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains the function WebRtcSpl_SqrtOfOneMinusXSquared().
* The description header can be found in signal_processing_library.h
*
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_SqrtOfOneMinusXSquared(int16_t *xQ15, size_t vector_length,
int16_t *yQ15)
{
int32_t sq;
size_t m;
int16_t tmp;
for (m = 0; m < vector_length; m++)
{
tmp = xQ15[m];
sq = tmp * tmp; // x^2 in Q30
sq = 1073741823 - sq; // 1-x^2, where 1 ~= 0.99999999906 is 1073741823 in Q30
sq = WebRtcSpl_Sqrt(sq); // sqrt(1-x^2) in Q15
yQ15[m] = (int16_t)sq;
}
}

165
webrtc/common_audio/signal_processing/vector_scaling_operations.c Normal file
View File

@ -0,0 +1,165 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains implementations of the functions
* WebRtcSpl_VectorBitShiftW16()
* WebRtcSpl_VectorBitShiftW32()
* WebRtcSpl_VectorBitShiftW32ToW16()
* WebRtcSpl_ScaleVector()
* WebRtcSpl_ScaleVectorWithSat()
* WebRtcSpl_ScaleAndAddVectors()
* WebRtcSpl_ScaleAndAddVectorsWithRoundC()
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
void WebRtcSpl_VectorBitShiftW16(int16_t *res, size_t length,
const int16_t *in, int16_t right_shifts)
{
size_t i;
if (right_shifts > 0)
{
for (i = length; i > 0; i--)
{
(*res++) = ((*in++) >> right_shifts);
}
} else
{
for (i = length; i > 0; i--)
{
(*res++) = ((*in++) << (-right_shifts));
}
}
}
void WebRtcSpl_VectorBitShiftW32(int32_t *out_vector,
size_t vector_length,
const int32_t *in_vector,
int16_t right_shifts)
{
size_t i;
if (right_shifts > 0)
{
for (i = vector_length; i > 0; i--)
{
(*out_vector++) = ((*in_vector++) >> right_shifts);
}
} else
{
for (i = vector_length; i > 0; i--)
{
(*out_vector++) = ((*in_vector++) << (-right_shifts));
}
}
}
void WebRtcSpl_VectorBitShiftW32ToW16(int16_t* out, size_t length,
const int32_t* in, int right_shifts) {
size_t i;
int32_t tmp_w32;
if (right_shifts >= 0) {
for (i = length; i > 0; i--) {
tmp_w32 = (*in++) >> right_shifts;
(*out++) = WebRtcSpl_SatW32ToW16(tmp_w32);
}
} else {
int left_shifts = -right_shifts;
for (i = length; i > 0; i--) {
tmp_w32 = (*in++) << left_shifts;
(*out++) = WebRtcSpl_SatW32ToW16(tmp_w32);
}
}
}
void WebRtcSpl_ScaleVector(const int16_t *in_vector, int16_t *out_vector,
int16_t gain, size_t in_vector_length,
int16_t right_shifts)
{
// Performs vector operation: out_vector = (gain*in_vector)>>right_shifts
size_t i;
const int16_t *inptr;
int16_t *outptr;
inptr = in_vector;
outptr = out_vector;
for (i = 0; i < in_vector_length; i++)
{
*outptr++ = (int16_t)((*inptr++ * gain) >> right_shifts);
}
}
void WebRtcSpl_ScaleVectorWithSat(const int16_t *in_vector, int16_t *out_vector,
int16_t gain, size_t in_vector_length,
int16_t right_shifts)
{
// Performs vector operation: out_vector = (gain*in_vector)>>right_shifts
size_t i;
const int16_t *inptr;
int16_t *outptr;
inptr = in_vector;
outptr = out_vector;
for (i = 0; i < in_vector_length; i++) {
*outptr++ = WebRtcSpl_SatW32ToW16((*inptr++ * gain) >> right_shifts);
}
}
void WebRtcSpl_ScaleAndAddVectors(const int16_t *in1, int16_t gain1, int shift1,
const int16_t *in2, int16_t gain2, int shift2,
int16_t *out, size_t vector_length)
{
// Performs vector operation: out = (gain1*in1)>>shift1 + (gain2*in2)>>shift2
size_t i;
const int16_t *in1ptr;
const int16_t *in2ptr;
int16_t *outptr;
in1ptr = in1;
in2ptr = in2;
outptr = out;
for (i = 0; i < vector_length; i++)
{
*outptr++ = (int16_t)((gain1 * *in1ptr++) >> shift1) +
(int16_t)((gain2 * *in2ptr++) >> shift2);
}
}
// C version of WebRtcSpl_ScaleAndAddVectorsWithRound() for generic platforms.
int WebRtcSpl_ScaleAndAddVectorsWithRoundC(const int16_t* in_vector1,
int16_t in_vector1_scale,
const int16_t* in_vector2,
int16_t in_vector2_scale,
int right_shifts,
int16_t* out_vector,
size_t length) {
size_t i = 0;
int round_value = (1 << right_shifts) >> 1;
if (in_vector1 == NULL || in_vector2 == NULL || out_vector == NULL ||
length == 0 || right_shifts < 0) {
return -1;
}
for (i = 0; i < length; i++) {
out_vector[i] = (int16_t)((
in_vector1[i] * in_vector1_scale + in_vector2[i] * in_vector2_scale +
round_value) >> right_shifts);
}
return 0;
}

57
webrtc/common_audio/signal_processing/vector_scaling_operations_mips.c Normal file
View File

@ -0,0 +1,57 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file contains implementations of the functions
* WebRtcSpl_ScaleAndAddVectorsWithRound_mips()
*/
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
int WebRtcSpl_ScaleAndAddVectorsWithRound_mips(const int16_t* in_vector1,
int16_t in_vector1_scale,
const int16_t* in_vector2,
int16_t in_vector2_scale,
int right_shifts,
int16_t* out_vector,
size_t length) {
int16_t r0 = 0, r1 = 0;
int16_t *in1 = (int16_t*)in_vector1;
int16_t *in2 = (int16_t*)in_vector2;
int16_t *out = out_vector;
size_t i = 0;
int value32 = 0;
if (in_vector1 == NULL || in_vector2 == NULL || out_vector == NULL ||
length == 0 || right_shifts < 0) {
return -1;
}
for (i = 0; i < length; i++) {
__asm __volatile (
"lh %[r0], 0(%[in1]) \n\t"
"lh %[r1], 0(%[in2]) \n\t"
"mult %[r0], %[in_vector1_scale] \n\t"
"madd %[r1], %[in_vector2_scale] \n\t"
"extrv_r.w %[value32], $ac0, %[right_shifts] \n\t"
"addiu %[in1], %[in1], 2 \n\t"
"addiu %[in2], %[in2], 2 \n\t"
"sh %[value32], 0(%[out]) \n\t"
"addiu %[out], %[out], 2 \n\t"
: [value32] "=&r" (value32), [out] "+r" (out), [in1] "+r" (in1),
[in2] "+r" (in2), [r0] "=&r" (r0), [r1] "=&r" (r1)
: [in_vector1_scale] "r" (in_vector1_scale),
[in_vector2_scale] "r" (in_vector2_scale),
[right_shifts] "r" (right_shifts)
: "hi", "lo", "memory"
);
}
return 0;
}

50
webrtc/common_audio/vad/include/vad.h Normal file
View File

@ -0,0 +1,50 @@
/*
* Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_COMMON_AUDIO_VAD_INCLUDE_VAD_H_
#define WEBRTC_COMMON_AUDIO_VAD_INCLUDE_VAD_H_
#include "webrtc/base/checks.h"
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/common_audio/vad/include/webrtc_vad.h"
#include "webrtc/typedefs.h"
namespace webrtc {
class Vad {
public:
enum Aggressiveness {
kVadNormal = 0,
kVadLowBitrate = 1,
kVadAggressive = 2,
kVadVeryAggressive = 3
};
enum Activity { kPassive = 0, kActive = 1, kError = -1 };
virtual ~Vad() = default;
// Calculates a VAD decision for the given audio frame. Valid sample rates
// are 8000, 16000, and 32000 Hz; the number of samples must be such that the
// frame is 10, 20, or 30 ms long.
virtual Activity VoiceActivity(const int16_t* audio,
size_t num_samples,
int sample_rate_hz) = 0;
// Resets VAD state.
virtual void Reset() = 0;
};
// Returns a Vad instance that's implemented on top of WebRtcVad.
rtc::scoped_ptr<Vad> CreateVad(Vad::Aggressiveness aggressiveness);
} // namespace webrtc
#endif // WEBRTC_COMMON_AUDIO_VAD_INCLUDE_VAD_H_

86
webrtc/common_audio/vad/include/webrtc_vad.h Normal file
View File

@ -0,0 +1,86 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This header file includes the VAD API calls. Specific function calls are given below.
*/
#ifndef WEBRTC_COMMON_AUDIO_VAD_INCLUDE_WEBRTC_VAD_H_ // NOLINT
#define WEBRTC_COMMON_AUDIO_VAD_INCLUDE_WEBRTC_VAD_H_
#include <stddef.h>
#include "webrtc/typedefs.h"
typedef struct WebRtcVadInst VadInst;
#ifdef __cplusplus
extern "C" {
#endif
// Creates an instance to the VAD structure.
VadInst* WebRtcVad_Create();
// Frees the dynamic memory of a specified VAD instance.
//
// - handle [i] : Pointer to VAD instance that should be freed.
void WebRtcVad_Free(VadInst* handle);
// Initializes a VAD instance.
//
// - handle [i/o] : Instance that should be initialized.
//
// returns : 0 - (OK),
// -1 - (NULL pointer or Default mode could not be set).
int WebRtcVad_Init(VadInst* handle);
// Sets the VAD operating mode. A more aggressive (higher mode) VAD is more
// restrictive in reporting speech. Put in other words the probability of being
// speech when the VAD returns 1 is increased with increasing mode. As a
// consequence also the missed detection rate goes up.
//
// - handle [i/o] : VAD instance.
// - mode [i] : Aggressiveness mode (0, 1, 2, or 3).
//
// returns : 0 - (OK),
// -1 - (NULL pointer, mode could not be set or the VAD instance
// has not been initialized).
int WebRtcVad_set_mode(VadInst* handle, int mode);
// Calculates a VAD decision for the |audio_frame|. For valid sampling rates
// frame lengths, see the description of WebRtcVad_ValidRatesAndFrameLengths().
//
// - handle [i/o] : VAD Instance. Needs to be initialized by
// WebRtcVad_Init() before call.
// - fs [i] : Sampling frequency (Hz): 8000, 16000, or 32000
// - audio_frame [i] : Audio frame buffer.
// - frame_length [i] : Length of audio frame buffer in number of samples.
//
// returns : 1 - (Active Voice),
// 0 - (Non-active Voice),
// -1 - (Error)
int WebRtcVad_Process(VadInst* handle, int fs, const int16_t* audio_frame,
size_t frame_length);
// Checks for valid combinations of |rate| and |frame_length|. We support 10,
// 20 and 30 ms frames and the rates 8000, 16000 and 32000 Hz.
//
// - rate [i] : Sampling frequency (Hz).
// - frame_length [i] : Speech frame buffer length in number of samples.
//
// returns : 0 - (valid combination), -1 - (invalid combination)
int WebRtcVad_ValidRateAndFrameLength(int rate, size_t frame_length);
#ifdef __cplusplus
}
#endif
#endif // WEBRTC_COMMON_AUDIO_VAD_INCLUDE_WEBRTC_VAD_H_ // NOLINT

63
webrtc/common_audio/vad/vad.cc Normal file
View File

@ -0,0 +1,63 @@
/*
* Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/vad/include/vad.h"
#include "webrtc/base/checks.h"
namespace webrtc {
namespace {
class VadImpl final : public Vad {
public:
explicit VadImpl(Aggressiveness aggressiveness)
: handle_(nullptr), aggressiveness_(aggressiveness) {
Reset();
}
~VadImpl() override { WebRtcVad_Free(handle_); }
Activity VoiceActivity(const int16_t* audio,
size_t num_samples,
int sample_rate_hz) override {
int ret = WebRtcVad_Process(handle_, sample_rate_hz, audio, num_samples);
switch (ret) {
case 0:
return kPassive;
case 1:
return kActive;
default:
RTC_DCHECK(false) << "WebRtcVad_Process returned an error.";
return kError;
}
}
void Reset() override {
if (handle_)
WebRtcVad_Free(handle_);
handle_ = WebRtcVad_Create();
RTC_CHECK(handle_);
RTC_CHECK_EQ(WebRtcVad_Init(handle_), 0);
RTC_CHECK_EQ(WebRtcVad_set_mode(handle_, aggressiveness_), 0);
}
private:
VadInst* handle_;
Aggressiveness aggressiveness_;
};
} // namespace
rtc::scoped_ptr<Vad> CreateVad(Vad::Aggressiveness aggressiveness) {
return rtc::scoped_ptr<Vad>(new VadImpl(aggressiveness));
}
} // namespace webrtc

676
webrtc/common_audio/vad/vad_core.c Normal file
View File

@ -0,0 +1,676 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/vad/vad_core.h"
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/common_audio/vad/vad_filterbank.h"
#include "webrtc/common_audio/vad/vad_gmm.h"
#include "webrtc/common_audio/vad/vad_sp.h"
#include "webrtc/typedefs.h"
// Spectrum Weighting
static const int16_t kSpectrumWeight[kNumChannels] = { 6, 8, 10, 12, 14, 16 };
static const int16_t kNoiseUpdateConst = 655; // Q15
static const int16_t kSpeechUpdateConst = 6554; // Q15
static const int16_t kBackEta = 154; // Q8
// Minimum difference between the two models, Q5
static const int16_t kMinimumDifference[kNumChannels] = {
544, 544, 576, 576, 576, 576 };
// Upper limit of mean value for speech model, Q7
static const int16_t kMaximumSpeech[kNumChannels] = {
11392, 11392, 11520, 11520, 11520, 11520 };
// Minimum value for mean value
static const int16_t kMinimumMean[kNumGaussians] = { 640, 768 };
// Upper limit of mean value for noise model, Q7
static const int16_t kMaximumNoise[kNumChannels] = {
9216, 9088, 8960, 8832, 8704, 8576 };
// Start values for the Gaussian models, Q7
// Weights for the two Gaussians for the six channels (noise)
static const int16_t kNoiseDataWeights[kTableSize] = {
34, 62, 72, 66, 53, 25, 94, 66, 56, 62, 75, 103 };
// Weights for the two Gaussians for the six channels (speech)
static const int16_t kSpeechDataWeights[kTableSize] = {
48, 82, 45, 87, 50, 47, 80, 46, 83, 41, 78, 81 };
// Means for the two Gaussians for the six channels (noise)
static const int16_t kNoiseDataMeans[kTableSize] = {
6738, 4892, 7065, 6715, 6771, 3369, 7646, 3863, 7820, 7266, 5020, 4362 };
// Means for the two Gaussians for the six channels (speech)
static const int16_t kSpeechDataMeans[kTableSize] = {
8306, 10085, 10078, 11823, 11843, 6309, 9473, 9571, 10879, 7581, 8180, 7483
};
// Stds for the two Gaussians for the six channels (noise)
static const int16_t kNoiseDataStds[kTableSize] = {
378, 1064, 493, 582, 688, 593, 474, 697, 475, 688, 421, 455 };
// Stds for the two Gaussians for the six channels (speech)
static const int16_t kSpeechDataStds[kTableSize] = {
555, 505, 567, 524, 585, 1231, 509, 828, 492, 1540, 1079, 850 };
// Constants used in GmmProbability().
//
// Maximum number of counted speech (VAD = 1) frames in a row.
static const int16_t kMaxSpeechFrames = 6;
// Minimum standard deviation for both speech and noise.
static const int16_t kMinStd = 384;
// Constants in WebRtcVad_InitCore().
// Default aggressiveness mode.
static const short kDefaultMode = 0;
static const int kInitCheck = 42;
// Constants used in WebRtcVad_set_mode_core().
//
// Thresholds for different frame lengths (10 ms, 20 ms and 30 ms).
//
// Mode 0, Quality.
static const int16_t kOverHangMax1Q[3] = { 8, 4, 3 };
static const int16_t kOverHangMax2Q[3] = { 14, 7, 5 };
static const int16_t kLocalThresholdQ[3] = { 24, 21, 24 };
static const int16_t kGlobalThresholdQ[3] = { 57, 48, 57 };
// Mode 1, Low bitrate.
static const int16_t kOverHangMax1LBR[3] = { 8, 4, 3 };
static const int16_t kOverHangMax2LBR[3] = { 14, 7, 5 };
static const int16_t kLocalThresholdLBR[3] = { 37, 32, 37 };
static const int16_t kGlobalThresholdLBR[3] = { 100, 80, 100 };
// Mode 2, Aggressive.
static const int16_t kOverHangMax1AGG[3] = { 6, 3, 2 };
static const int16_t kOverHangMax2AGG[3] = { 9, 5, 3 };
static const int16_t kLocalThresholdAGG[3] = { 82, 78, 82 };
static const int16_t kGlobalThresholdAGG[3] = { 285, 260, 285 };
// Mode 3, Very aggressive.
static const int16_t kOverHangMax1VAG[3] = { 6, 3, 2 };
static const int16_t kOverHangMax2VAG[3] = { 9, 5, 3 };
static const int16_t kLocalThresholdVAG[3] = { 94, 94, 94 };
static const int16_t kGlobalThresholdVAG[3] = { 1100, 1050, 1100 };
// Calculates the weighted average w.r.t. number of Gaussians. The |data| are
// updated with an |offset| before averaging.
//
// - data [i/o] : Data to average.
// - offset [i] : An offset added to |data|.
// - weights [i] : Weights used for averaging.
//
// returns : The weighted average.
static int32_t WeightedAverage(int16_t* data, int16_t offset,
const int16_t* weights) {
int k;
int32_t weighted_average = 0;
for (k = 0; k < kNumGaussians; k++) {
data[k * kNumChannels] += offset;
weighted_average += data[k * kNumChannels] * weights[k * kNumChannels];
}
return weighted_average;
}
// Calculates the probabilities for both speech and background noise using
// Gaussian Mixture Models (GMM). A hypothesis-test is performed to decide which
// type of signal is most probable.
//
// - self [i/o] : Pointer to VAD instance
// - features [i] : Feature vector of length |kNumChannels|
// = log10(energy in frequency band)
// - total_power [i] : Total power in audio frame.
// - frame_length [i] : Number of input samples
//
// - returns : the VAD decision (0 - noise, 1 - speech).
static int16_t GmmProbability(VadInstT* self, int16_t* features,
int16_t total_power, size_t frame_length) {
int channel, k;
int16_t feature_minimum;
int16_t h0, h1;
int16_t log_likelihood_ratio;
int16_t vadflag = 0;
int16_t shifts_h0, shifts_h1;
int16_t tmp_s16, tmp1_s16, tmp2_s16;
int16_t diff;
int gaussian;
int16_t nmk, nmk2, nmk3, smk, smk2, nsk, ssk;
int16_t delt, ndelt;
int16_t maxspe, maxmu;
int16_t deltaN[kTableSize], deltaS[kTableSize];
int16_t ngprvec[kTableSize] = { 0 }; // Conditional probability = 0.
int16_t sgprvec[kTableSize] = { 0 }; // Conditional probability = 0.
int32_t h0_test, h1_test;
int32_t tmp1_s32, tmp2_s32;
int32_t sum_log_likelihood_ratios = 0;
int32_t noise_global_mean, speech_global_mean;
int32_t noise_probability[kNumGaussians], speech_probability[kNumGaussians];
int16_t overhead1, overhead2, individualTest, totalTest;
// Set various thresholds based on frame lengths (80, 160 or 240 samples).
if (frame_length == 80) {
overhead1 = self->over_hang_max_1[0];
overhead2 = self->over_hang_max_2[0];
individualTest = self->individual[0];
totalTest = self->total[0];
} else if (frame_length == 160) {
overhead1 = self->over_hang_max_1[1];
overhead2 = self->over_hang_max_2[1];
individualTest = self->individual[1];
totalTest = self->total[1];
} else {
overhead1 = self->over_hang_max_1[2];
overhead2 = self->over_hang_max_2[2];
individualTest = self->individual[2];
totalTest = self->total[2];
}
if (total_power > kMinEnergy) {
// The signal power of current frame is large enough for processing. The
// processing consists of two parts:
// 1) Calculating the likelihood of speech and thereby a VAD decision.
// 2) Updating the underlying model, w.r.t., the decision made.
// The detection scheme is an LRT with hypothesis
// H0: Noise
// H1: Speech
//
// We combine a global LRT with local tests, for each frequency sub-band,
// here defined as |channel|.
for (channel = 0; channel < kNumChannels; channel++) {
// For each channel we model the probability with a GMM consisting of
// |kNumGaussians|, with different means and standard deviations depending
// on H0 or H1.
h0_test = 0;
h1_test = 0;
for (k = 0; k < kNumGaussians; k++) {
gaussian = channel + k * kNumChannels;
// Probability under H0, that is, probability of frame being noise.
// Value given in Q27 = Q7 * Q20.
tmp1_s32 = WebRtcVad_GaussianProbability(features[channel],
self->noise_means[gaussian],
self->noise_stds[gaussian],
&deltaN[gaussian]);
noise_probability[k] = kNoiseDataWeights[gaussian] * tmp1_s32;
h0_test += noise_probability[k]; // Q27
// Probability under H1, that is, probability of frame being speech.
// Value given in Q27 = Q7 * Q20.
tmp1_s32 = WebRtcVad_GaussianProbability(features[channel],
self->speech_means[gaussian],
self->speech_stds[gaussian],
&deltaS[gaussian]);
speech_probability[k] = kSpeechDataWeights[gaussian] * tmp1_s32;
h1_test += speech_probability[k]; // Q27
}
// Calculate the log likelihood ratio: log2(Pr{X|H1} / Pr{X|H1}).
// Approximation:
// log2(Pr{X|H1} / Pr{X|H1}) = log2(Pr{X|H1}*2^Q) - log2(Pr{X|H1}*2^Q)
// = log2(h1_test) - log2(h0_test)
// = log2(2^(31-shifts_h1)*(1+b1))
// - log2(2^(31-shifts_h0)*(1+b0))
// = shifts_h0 - shifts_h1
// + log2(1+b1) - log2(1+b0)
// ~= shifts_h0 - shifts_h1
//
// Note that b0 and b1 are values less than 1, hence, 0 <= log2(1+b0) < 1.
// Further, b0 and b1 are independent and on the average the two terms
// cancel.
shifts_h0 = WebRtcSpl_NormW32(h0_test);
shifts_h1 = WebRtcSpl_NormW32(h1_test);
if (h0_test == 0) {
shifts_h0 = 31;
}
if (h1_test == 0) {
shifts_h1 = 31;
}
log_likelihood_ratio = shifts_h0 - shifts_h1;
// Update |sum_log_likelihood_ratios| with spectrum weighting. This is
// used for the global VAD decision.
sum_log_likelihood_ratios +=
(int32_t) (log_likelihood_ratio * kSpectrumWeight[channel]);
// Local VAD decision.
if ((log_likelihood_ratio << 2) > individualTest) {
vadflag = 1;
}
// TODO(bjornv): The conditional probabilities below are applied on the
// hard coded number of Gaussians set to two. Find a way to generalize.
// Calculate local noise probabilities used later when updating the GMM.
h0 = (int16_t) (h0_test >> 12); // Q15
if (h0 > 0) {
// High probability of noise. Assign conditional probabilities for each
// Gaussian in the GMM.
tmp1_s32 = (noise_probability[0] & 0xFFFFF000) << 2; // Q29
ngprvec[channel] = (int16_t) WebRtcSpl_DivW32W16(tmp1_s32, h0); // Q14
ngprvec[channel + kNumChannels] = 16384 - ngprvec[channel];
} else {
// Low noise probability. Assign conditional probability 1 to the first
// Gaussian and 0 to the rest (which is already set at initialization).
ngprvec[channel] = 16384;
}
// Calculate local speech probabilities used later when updating the GMM.
h1 = (int16_t) (h1_test >> 12); // Q15
if (h1 > 0) {
// High probability of speech. Assign conditional probabilities for each
// Gaussian in the GMM. Otherwise use the initialized values, i.e., 0.
tmp1_s32 = (speech_probability[0] & 0xFFFFF000) << 2; // Q29
sgprvec[channel] = (int16_t) WebRtcSpl_DivW32W16(tmp1_s32, h1); // Q14
sgprvec[channel + kNumChannels] = 16384 - sgprvec[channel];
}
}
// Make a global VAD decision.
vadflag |= (sum_log_likelihood_ratios >= totalTest);
// Update the model parameters.
maxspe = 12800;
for (channel = 0; channel < kNumChannels; channel++) {
// Get minimum value in past which is used for long term correction in Q4.
feature_minimum = WebRtcVad_FindMinimum(self, features[channel], channel);
// Compute the "global" mean, that is the sum of the two means weighted.
noise_global_mean = WeightedAverage(&self->noise_means[channel], 0,
&kNoiseDataWeights[channel]);
tmp1_s16 = (int16_t) (noise_global_mean >> 6); // Q8
for (k = 0; k < kNumGaussians; k++) {
gaussian = channel + k * kNumChannels;
nmk = self->noise_means[gaussian];
smk = self->speech_means[gaussian];
nsk = self->noise_stds[gaussian];
ssk = self->speech_stds[gaussian];
// Update noise mean vector if the frame consists of noise only.
nmk2 = nmk;
if (!vadflag) {
// deltaN = (x-mu)/sigma^2
// ngprvec[k] = |noise_probability[k]| /
// (|noise_probability[0]| + |noise_probability[1]|)
// (Q14 * Q11 >> 11) = Q14.
delt = (int16_t)((ngprvec[gaussian] * deltaN[gaussian]) >> 11);
// Q7 + (Q14 * Q15 >> 22) = Q7.
nmk2 = nmk + (int16_t)((delt * kNoiseUpdateConst) >> 22);
}
// Long term correction of the noise mean.
// Q8 - Q8 = Q8.
ndelt = (feature_minimum << 4) - tmp1_s16;
// Q7 + (Q8 * Q8) >> 9 = Q7.
nmk3 = nmk2 + (int16_t)((ndelt * kBackEta) >> 9);
// Control that the noise mean does not drift to much.
tmp_s16 = (int16_t) ((k + 5) << 7);
if (nmk3 < tmp_s16) {
nmk3 = tmp_s16;
}
tmp_s16 = (int16_t) ((72 + k - channel) << 7);
if (nmk3 > tmp_s16) {
nmk3 = tmp_s16;
}
self->noise_means[gaussian] = nmk3;
if (vadflag) {
// Update speech mean vector:
// |deltaS| = (x-mu)/sigma^2
// sgprvec[k] = |speech_probability[k]| /
// (|speech_probability[0]| + |speech_probability[1]|)
// (Q14 * Q11) >> 11 = Q14.
delt = (int16_t)((sgprvec[gaussian] * deltaS[gaussian]) >> 11);
// Q14 * Q15 >> 21 = Q8.
tmp_s16 = (int16_t)((delt * kSpeechUpdateConst) >> 21);
// Q7 + (Q8 >> 1) = Q7. With rounding.
smk2 = smk + ((tmp_s16 + 1) >> 1);
// Control that the speech mean does not drift to much.
maxmu = maxspe + 640;
if (smk2 < kMinimumMean[k]) {
smk2 = kMinimumMean[k];
}
if (smk2 > maxmu) {
smk2 = maxmu;
}
self->speech_means[gaussian] = smk2; // Q7.
// (Q7 >> 3) = Q4. With rounding.
tmp_s16 = ((smk + 4) >> 3);
tmp_s16 = features[channel] - tmp_s16; // Q4
// (Q11 * Q4 >> 3) = Q12.
tmp1_s32 = (deltaS[gaussian] * tmp_s16) >> 3;
tmp2_s32 = tmp1_s32 - 4096;
tmp_s16 = sgprvec[gaussian] >> 2;
// (Q14 >> 2) * Q12 = Q24.
tmp1_s32 = tmp_s16 * tmp2_s32;
tmp2_s32 = tmp1_s32 >> 4; // Q20
// 0.1 * Q20 / Q7 = Q13.
if (tmp2_s32 > 0) {
tmp_s16 = (int16_t) WebRtcSpl_DivW32W16(tmp2_s32, ssk * 10);
} else {
tmp_s16 = (int16_t) WebRtcSpl_DivW32W16(-tmp2_s32, ssk * 10);
tmp_s16 = -tmp_s16;
}
// Divide by 4 giving an update factor of 0.025 (= 0.1 / 4).
// Note that division by 4 equals shift by 2, hence,
// (Q13 >> 8) = (Q13 >> 6) / 4 = Q7.
tmp_s16 += 128; // Rounding.
ssk += (tmp_s16 >> 8);
if (ssk < kMinStd) {
ssk = kMinStd;
}
self->speech_stds[gaussian] = ssk;
} else {
// Update GMM variance vectors.
// deltaN * (features[channel] - nmk) - 1
// Q4 - (Q7 >> 3) = Q4.
tmp_s16 = features[channel] - (nmk >> 3);
// (Q11 * Q4 >> 3) = Q12.
tmp1_s32 = (deltaN[gaussian] * tmp_s16) >> 3;
tmp1_s32 -= 4096;
// (Q14 >> 2) * Q12 = Q24.
tmp_s16 = (ngprvec[gaussian] + 2) >> 2;
tmp2_s32 = tmp_s16 * tmp1_s32;
// Q20 * approx 0.001 (2^-10=0.0009766), hence,
// (Q24 >> 14) = (Q24 >> 4) / 2^10 = Q20.
tmp1_s32 = tmp2_s32 >> 14;
// Q20 / Q7 = Q13.
if (tmp1_s32 > 0) {
tmp_s16 = (int16_t) WebRtcSpl_DivW32W16(tmp1_s32, nsk);
} else {
tmp_s16 = (int16_t) WebRtcSpl_DivW32W16(-tmp1_s32, nsk);
tmp_s16 = -tmp_s16;
}
tmp_s16 += 32; // Rounding
nsk += tmp_s16 >> 6; // Q13 >> 6 = Q7.
if (nsk < kMinStd) {
nsk = kMinStd;
}
self->noise_stds[gaussian] = nsk;
}
}
// Separate models if they are too close.
// |noise_global_mean| in Q14 (= Q7 * Q7).
noise_global_mean = WeightedAverage(&self->noise_means[channel], 0,
&kNoiseDataWeights[channel]);
// |speech_global_mean| in Q14 (= Q7 * Q7).
speech_global_mean = WeightedAverage(&self->speech_means[channel], 0,
&kSpeechDataWeights[channel]);
// |diff| = "global" speech mean - "global" noise mean.
// (Q14 >> 9) - (Q14 >> 9) = Q5.
diff = (int16_t) (speech_global_mean >> 9) -
(int16_t) (noise_global_mean >> 9);
if (diff < kMinimumDifference[channel]) {
tmp_s16 = kMinimumDifference[channel] - diff;
// |tmp1_s16| = ~0.8 * (kMinimumDifference - diff) in Q7.
// |tmp2_s16| = ~0.2 * (kMinimumDifference - diff) in Q7.
tmp1_s16 = (int16_t)((13 * tmp_s16) >> 2);
tmp2_s16 = (int16_t)((3 * tmp_s16) >> 2);
// Move Gaussian means for speech model by |tmp1_s16| and update
// |speech_global_mean|. Note that |self->speech_means[channel]| is
// changed after the call.
speech_global_mean = WeightedAverage(&self->speech_means[channel],
tmp1_s16,
&kSpeechDataWeights[channel]);
// Move Gaussian means for noise model by -|tmp2_s16| and update
// |noise_global_mean|. Note that |self->noise_means[channel]| is
// changed after the call.
noise_global_mean = WeightedAverage(&self->noise_means[channel],
-tmp2_s16,
&kNoiseDataWeights[channel]);
}
// Control that the speech & noise means do not drift to much.
maxspe = kMaximumSpeech[channel];
tmp2_s16 = (int16_t) (speech_global_mean >> 7);
if (tmp2_s16 > maxspe) {
// Upper limit of speech model.
tmp2_s16 -= maxspe;
for (k = 0; k < kNumGaussians; k++) {
self->speech_means[channel + k * kNumChannels] -= tmp2_s16;
}
}
tmp2_s16 = (int16_t) (noise_global_mean >> 7);
if (tmp2_s16 > kMaximumNoise[channel]) {
tmp2_s16 -= kMaximumNoise[channel];
for (k = 0; k < kNumGaussians; k++) {
self->noise_means[channel + k * kNumChannels] -= tmp2_s16;
}
}
}
self->frame_counter++;
}
// Smooth with respect to transition hysteresis.
if (!vadflag) {
if (self->over_hang > 0) {
vadflag = 2 + self->over_hang;
self->over_hang--;
}
self->num_of_speech = 0;
} else {
self->num_of_speech++;
if (self->num_of_speech > kMaxSpeechFrames) {
self->num_of_speech = kMaxSpeechFrames;
self->over_hang = overhead2;
} else {
self->over_hang = overhead1;
}
}
return vadflag;
}
// Initialize the VAD. Set aggressiveness mode to default value.
int WebRtcVad_InitCore(VadInstT* self) {
int i;
if (self == NULL) {
return -1;
}
// Initialization of general struct variables.
self->vad = 1; // Speech active (=1).
self->frame_counter = 0;
self->over_hang = 0;
self->num_of_speech = 0;
// Initialization of downsampling filter state.
memset(self->downsampling_filter_states, 0,
sizeof(self->downsampling_filter_states));
// Initialization of 48 to 8 kHz downsampling.
WebRtcSpl_ResetResample48khzTo8khz(&self->state_48_to_8);
// Read initial PDF parameters.
for (i = 0; i < kTableSize; i++) {
self->noise_means[i] = kNoiseDataMeans[i];
self->speech_means[i] = kSpeechDataMeans[i];
self->noise_stds[i] = kNoiseDataStds[i];
self->speech_stds[i] = kSpeechDataStds[i];
}
// Initialize Index and Minimum value vectors.
for (i = 0; i < 16 * kNumChannels; i++) {
self->low_value_vector[i] = 10000;
self->index_vector[i] = 0;
}
// Initialize splitting filter states.
memset(self->upper_state, 0, sizeof(self->upper_state));
memset(self->lower_state, 0, sizeof(self->lower_state));
// Initialize high pass filter states.
memset(self->hp_filter_state, 0, sizeof(self->hp_filter_state));
// Initialize mean value memory, for WebRtcVad_FindMinimum().
for (i = 0; i < kNumChannels; i++) {
self->mean_value[i] = 1600;
}
// Set aggressiveness mode to default (=|kDefaultMode|).
if (WebRtcVad_set_mode_core(self, kDefaultMode) != 0) {
return -1;
}
self->init_flag = kInitCheck;
return 0;
}
// Set aggressiveness mode
int WebRtcVad_set_mode_core(VadInstT* self, int mode) {
int return_value = 0;
switch (mode) {
case 0:
// Quality mode.
memcpy(self->over_hang_max_1, kOverHangMax1Q,
sizeof(self->over_hang_max_1));
memcpy(self->over_hang_max_2, kOverHangMax2Q,
sizeof(self->over_hang_max_2));
memcpy(self->individual, kLocalThresholdQ,
sizeof(self->individual));
memcpy(self->total, kGlobalThresholdQ,
sizeof(self->total));
break;
case 1:
// Low bitrate mode.
memcpy(self->over_hang_max_1, kOverHangMax1LBR,
sizeof(self->over_hang_max_1));
memcpy(self->over_hang_max_2, kOverHangMax2LBR,
sizeof(self->over_hang_max_2));
memcpy(self->individual, kLocalThresholdLBR,
sizeof(self->individual));
memcpy(self->total, kGlobalThresholdLBR,
sizeof(self->total));
break;
case 2:
// Aggressive mode.
memcpy(self->over_hang_max_1, kOverHangMax1AGG,
sizeof(self->over_hang_max_1));
memcpy(self->over_hang_max_2, kOverHangMax2AGG,
sizeof(self->over_hang_max_2));
memcpy(self->individual, kLocalThresholdAGG,
sizeof(self->individual));
memcpy(self->total, kGlobalThresholdAGG,
sizeof(self->total));
break;
case 3:
// Very aggressive mode.
memcpy(self->over_hang_max_1, kOverHangMax1VAG,
sizeof(self->over_hang_max_1));
memcpy(self->over_hang_max_2, kOverHangMax2VAG,
sizeof(self->over_hang_max_2));
memcpy(self->individual, kLocalThresholdVAG,
sizeof(self->individual));
memcpy(self->total, kGlobalThresholdVAG,
sizeof(self->total));
break;
default:
return_value = -1;
break;
}
return return_value;
}
// Calculate VAD decision by first extracting feature values and then calculate
// probability for both speech and background noise.
int WebRtcVad_CalcVad48khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length) {
int vad;
size_t i;
int16_t speech_nb[240]; // 30 ms in 8 kHz.
// |tmp_mem| is a temporary memory used by resample function, length is
// frame length in 10 ms (480 samples) + 256 extra.
int32_t tmp_mem[480 + 256] = { 0 };
const size_t kFrameLen10ms48khz = 480;
const size_t kFrameLen10ms8khz = 80;
size_t num_10ms_frames = frame_length / kFrameLen10ms48khz;
for (i = 0; i < num_10ms_frames; i++) {
WebRtcSpl_Resample48khzTo8khz(speech_frame,
&speech_nb[i * kFrameLen10ms8khz],
&inst->state_48_to_8,
tmp_mem);
}
// Do VAD on an 8 kHz signal
vad = WebRtcVad_CalcVad8khz(inst, speech_nb, frame_length / 6);
return vad;
}
int WebRtcVad_CalcVad32khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length)
{
size_t len;
int vad;
int16_t speechWB[480]; // Downsampled speech frame: 960 samples (30ms in SWB)
int16_t speechNB[240]; // Downsampled speech frame: 480 samples (30ms in WB)
// Downsample signal 32->16->8 before doing VAD
WebRtcVad_Downsampling(speech_frame, speechWB, &(inst->downsampling_filter_states[2]),
frame_length);
len = frame_length / 2;
WebRtcVad_Downsampling(speechWB, speechNB, inst->downsampling_filter_states, len);
len /= 2;
// Do VAD on an 8 kHz signal
vad = WebRtcVad_CalcVad8khz(inst, speechNB, len);
return vad;
}
int WebRtcVad_CalcVad16khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length)
{
size_t len;
int vad;
int16_t speechNB[240]; // Downsampled speech frame: 480 samples (30ms in WB)
// Wideband: Downsample signal before doing VAD
WebRtcVad_Downsampling(speech_frame, speechNB, inst->downsampling_filter_states,
frame_length);
len = frame_length / 2;
vad = WebRtcVad_CalcVad8khz(inst, speechNB, len);
return vad;
}
int WebRtcVad_CalcVad8khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length)
{
int16_t feature_vector[kNumChannels], total_power;
// Get power in the bands
total_power = WebRtcVad_CalculateFeatures(inst, speech_frame, frame_length,
feature_vector);
// Make a VAD
inst->vad = GmmProbability(inst, feature_vector, total_power, frame_length);
return inst->vad;
}

115
webrtc/common_audio/vad/vad_core.h Normal file
View File

@ -0,0 +1,115 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This header file includes the descriptions of the core VAD calls.
*/
#ifndef WEBRTC_COMMON_AUDIO_VAD_VAD_CORE_H_
#define WEBRTC_COMMON_AUDIO_VAD_VAD_CORE_H_
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/typedefs.h"
enum { kNumChannels = 6 }; // Number of frequency bands (named channels).
enum { kNumGaussians = 2 }; // Number of Gaussians per channel in the GMM.
enum { kTableSize = kNumChannels * kNumGaussians };
enum { kMinEnergy = 10 }; // Minimum energy required to trigger audio signal.
typedef struct VadInstT_
{
int vad;
int32_t downsampling_filter_states[4];
WebRtcSpl_State48khzTo8khz state_48_to_8;
int16_t noise_means[kTableSize];
int16_t speech_means[kTableSize];
int16_t noise_stds[kTableSize];
int16_t speech_stds[kTableSize];
// TODO(bjornv): Change to |frame_count|.
int32_t frame_counter;
int16_t over_hang; // Over Hang
int16_t num_of_speech;
// TODO(bjornv): Change to |age_vector|.
int16_t index_vector[16 * kNumChannels];
int16_t low_value_vector[16 * kNumChannels];
// TODO(bjornv): Change to |median|.
int16_t mean_value[kNumChannels];
int16_t upper_state[5];
int16_t lower_state[5];
int16_t hp_filter_state[4];
int16_t over_hang_max_1[3];
int16_t over_hang_max_2[3];
int16_t individual[3];
int16_t total[3];
int init_flag;
} VadInstT;
// Initializes the core VAD component. The default aggressiveness mode is
// controlled by |kDefaultMode| in vad_core.c.
//
// - self [i/o] : Instance that should be initialized
//
// returns : 0 (OK), -1 (NULL pointer in or if the default mode can't be
// set)
int WebRtcVad_InitCore(VadInstT* self);
/****************************************************************************
* WebRtcVad_set_mode_core(...)
*
* This function changes the VAD settings
*
* Input:
* - inst : VAD instance
* - mode : Aggressiveness degree
* 0 (High quality) - 3 (Highly aggressive)
*
* Output:
* - inst : Changed instance
*
* Return value : 0 - Ok
* -1 - Error
*/
int WebRtcVad_set_mode_core(VadInstT* self, int mode);
/****************************************************************************
* WebRtcVad_CalcVad48khz(...)
* WebRtcVad_CalcVad32khz(...)
* WebRtcVad_CalcVad16khz(...)
* WebRtcVad_CalcVad8khz(...)
*
* Calculate probability for active speech and make VAD decision.
*
* Input:
* - inst : Instance that should be initialized
* - speech_frame : Input speech frame
* - frame_length : Number of input samples
*
* Output:
* - inst : Updated filter states etc.
*
* Return value : VAD decision
* 0 - No active speech
* 1-6 - Active speech
*/
int WebRtcVad_CalcVad48khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length);
int WebRtcVad_CalcVad32khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length);
int WebRtcVad_CalcVad16khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length);
int WebRtcVad_CalcVad8khz(VadInstT* inst, const int16_t* speech_frame,
size_t frame_length);
#endif // WEBRTC_COMMON_AUDIO_VAD_VAD_CORE_H_

95
webrtc/common_audio/vad/vad_defines.h Normal file
View File

@ -0,0 +1,95 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This header file includes the macros used in VAD.
*/
#ifndef WEBRTC_VAD_DEFINES_H_
#define WEBRTC_VAD_DEFINES_H_
#define NUM_CHANNELS 6 // Eight frequency bands
#define NUM_MODELS 2 // Number of Gaussian models
#define NUM_TABLE_VALUES NUM_CHANNELS * NUM_MODELS
#define MIN_ENERGY 10
#define ALPHA1 6553 // 0.2 in Q15
#define ALPHA2 32439 // 0.99 in Q15
#define NSP_MAX 6 // Maximum number of VAD=1 frames in a row counted
#define MIN_STD 384 // Minimum standard deviation
// Mode 0, Quality thresholds - Different thresholds for the different frame lengths
#define INDIVIDUAL_10MS_Q 24
#define INDIVIDUAL_20MS_Q 21 // (log10(2)*66)<<2 ~=16
#define INDIVIDUAL_30MS_Q 24
#define TOTAL_10MS_Q 57
#define TOTAL_20MS_Q 48
#define TOTAL_30MS_Q 57
#define OHMAX1_10MS_Q 8 // Max Overhang 1
#define OHMAX2_10MS_Q 14 // Max Overhang 2
#define OHMAX1_20MS_Q 4 // Max Overhang 1
#define OHMAX2_20MS_Q 7 // Max Overhang 2
#define OHMAX1_30MS_Q 3
#define OHMAX2_30MS_Q 5
// Mode 1, Low bitrate thresholds - Different thresholds for the different frame lengths
#define INDIVIDUAL_10MS_LBR 37
#define INDIVIDUAL_20MS_LBR 32
#define INDIVIDUAL_30MS_LBR 37
#define TOTAL_10MS_LBR 100
#define TOTAL_20MS_LBR 80
#define TOTAL_30MS_LBR 100
#define OHMAX1_10MS_LBR 8 // Max Overhang 1
#define OHMAX2_10MS_LBR 14 // Max Overhang 2
#define OHMAX1_20MS_LBR 4
#define OHMAX2_20MS_LBR 7
#define OHMAX1_30MS_LBR 3
#define OHMAX2_30MS_LBR 5
// Mode 2, Very aggressive thresholds - Different thresholds for the different frame lengths
#define INDIVIDUAL_10MS_AGG 82
#define INDIVIDUAL_20MS_AGG 78
#define INDIVIDUAL_30MS_AGG 82
#define TOTAL_10MS_AGG 285 //580
#define TOTAL_20MS_AGG 260
#define TOTAL_30MS_AGG 285
#define OHMAX1_10MS_AGG 6 // Max Overhang 1
#define OHMAX2_10MS_AGG 9 // Max Overhang 2
#define OHMAX1_20MS_AGG 3
#define OHMAX2_20MS_AGG 5
#define OHMAX1_30MS_AGG 2
#define OHMAX2_30MS_AGG 3
// Mode 3, Super aggressive thresholds - Different thresholds for the different frame lengths
#define INDIVIDUAL_10MS_VAG 94
#define INDIVIDUAL_20MS_VAG 94
#define INDIVIDUAL_30MS_VAG 94
#define TOTAL_10MS_VAG 1100 //1700
#define TOTAL_20MS_VAG 1050
#define TOTAL_30MS_VAG 1100
#define OHMAX1_10MS_VAG 6 // Max Overhang 1
#define OHMAX2_10MS_VAG 9 // Max Overhang 2
#define OHMAX1_20MS_VAG 3
#define OHMAX2_20MS_VAG 5
#define OHMAX1_30MS_VAG 2
#define OHMAX2_30MS_VAG 3
#endif // WEBRTC_VAD_DEFINES_H_

331
webrtc/common_audio/vad/vad_filterbank.c Normal file
View File

@ -0,0 +1,331 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/vad/vad_filterbank.h"
#include <assert.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/typedefs.h"
// Constants used in LogOfEnergy().
static const int16_t kLogConst = 24660; // 160*log10(2) in Q9.
static const int16_t kLogEnergyIntPart = 14336; // 14 in Q10
// Coefficients used by HighPassFilter, Q14.
static const int16_t kHpZeroCoefs[3] = { 6631, -13262, 6631 };
static const int16_t kHpPoleCoefs[3] = { 16384, -7756, 5620 };
// Allpass filter coefficients, upper and lower, in Q15.
// Upper: 0.64, Lower: 0.17
static const int16_t kAllPassCoefsQ15[2] = { 20972, 5571 };
// Adjustment for division with two in SplitFilter.
static const int16_t kOffsetVector[6] = { 368, 368, 272, 176, 176, 176 };
// High pass filtering, with a cut-off frequency at 80 Hz, if the |data_in| is
// sampled at 500 Hz.
//
// - data_in [i] : Input audio data sampled at 500 Hz.
// - data_length [i] : Length of input and output data.
// - filter_state [i/o] : State of the filter.
// - data_out [o] : Output audio data in the frequency interval
// 80 - 250 Hz.
static void HighPassFilter(const int16_t* data_in, size_t data_length,
int16_t* filter_state, int16_t* data_out) {
size_t i;
const int16_t* in_ptr = data_in;
int16_t* out_ptr = data_out;
int32_t tmp32 = 0;
// The sum of the absolute values of the impulse response:
// The zero/pole-filter has a max amplification of a single sample of: 1.4546
// Impulse response: 0.4047 -0.6179 -0.0266 0.1993 0.1035 -0.0194
// The all-zero section has a max amplification of a single sample of: 1.6189
// Impulse response: 0.4047 -0.8094 0.4047 0 0 0
// The all-pole section has a max amplification of a single sample of: 1.9931
// Impulse response: 1.0000 0.4734 -0.1189 -0.2187 -0.0627 0.04532
for (i = 0; i < data_length; i++) {
// All-zero section (filter coefficients in Q14).
tmp32 = kHpZeroCoefs[0] * *in_ptr;
tmp32 += kHpZeroCoefs[1] * filter_state[0];
tmp32 += kHpZeroCoefs[2] * filter_state[1];
filter_state[1] = filter_state[0];
filter_state[0] = *in_ptr++;
// All-pole section (filter coefficients in Q14).
tmp32 -= kHpPoleCoefs[1] * filter_state[2];
tmp32 -= kHpPoleCoefs[2] * filter_state[3];
filter_state[3] = filter_state[2];
filter_state[2] = (int16_t) (tmp32 >> 14);
*out_ptr++ = filter_state[2];
}
}
// All pass filtering of |data_in|, used before splitting the signal into two
// frequency bands (low pass vs high pass).
// Note that |data_in| and |data_out| can NOT correspond to the same address.
//
// - data_in [i] : Input audio signal given in Q0.
// - data_length [i] : Length of input and output data.
// - filter_coefficient [i] : Given in Q15.
// - filter_state [i/o] : State of the filter given in Q(-1).
// - data_out [o] : Output audio signal given in Q(-1).
static void AllPassFilter(const int16_t* data_in, size_t data_length,
int16_t filter_coefficient, int16_t* filter_state,
int16_t* data_out) {
// The filter can only cause overflow (in the w16 output variable)
// if more than 4 consecutive input numbers are of maximum value and
// has the the same sign as the impulse responses first taps.
// First 6 taps of the impulse response:
// 0.6399 0.5905 -0.3779 0.2418 -0.1547 0.0990
size_t i;
int16_t tmp16 = 0;
int32_t tmp32 = 0;
int32_t state32 = ((int32_t) (*filter_state) << 16); // Q15
for (i = 0; i < data_length; i++) {
tmp32 = state32 + filter_coefficient * *data_in;
tmp16 = (int16_t) (tmp32 >> 16); // Q(-1)
*data_out++ = tmp16;
state32 = (*data_in << 14) - filter_coefficient * tmp16; // Q14
state32 <<= 1; // Q15.
data_in += 2;
}
*filter_state = (int16_t) (state32 >> 16); // Q(-1)
}
// Splits |data_in| into |hp_data_out| and |lp_data_out| corresponding to
// an upper (high pass) part and a lower (low pass) part respectively.
//
// - data_in [i] : Input audio data to be split into two frequency bands.
// - data_length [i] : Length of |data_in|.
// - upper_state [i/o] : State of the upper filter, given in Q(-1).
// - lower_state [i/o] : State of the lower filter, given in Q(-1).
// - hp_data_out [o] : Output audio data of the upper half of the spectrum.
// The length is |data_length| / 2.
// - lp_data_out [o] : Output audio data of the lower half of the spectrum.
// The length is |data_length| / 2.
static void SplitFilter(const int16_t* data_in, size_t data_length,
int16_t* upper_state, int16_t* lower_state,
int16_t* hp_data_out, int16_t* lp_data_out) {
size_t i;
size_t half_length = data_length >> 1; // Downsampling by 2.
int16_t tmp_out;
// All-pass filtering upper branch.
AllPassFilter(&data_in[0], half_length, kAllPassCoefsQ15[0], upper_state,
hp_data_out);
// All-pass filtering lower branch.
AllPassFilter(&data_in[1], half_length, kAllPassCoefsQ15[1], lower_state,
lp_data_out);
// Make LP and HP signals.
for (i = 0; i < half_length; i++) {
tmp_out = *hp_data_out;
*hp_data_out++ -= *lp_data_out;
*lp_data_out++ += tmp_out;
}
}
// Calculates the energy of |data_in| in dB, and also updates an overall
// |total_energy| if necessary.
//
// - data_in [i] : Input audio data for energy calculation.
// - data_length [i] : Length of input data.
// - offset [i] : Offset value added to |log_energy|.
// - total_energy [i/o] : An external energy updated with the energy of
// |data_in|.
// NOTE: |total_energy| is only updated if
// |total_energy| <= |kMinEnergy|.
// - log_energy [o] : 10 * log10("energy of |data_in|") given in Q4.
static void LogOfEnergy(const int16_t* data_in, size_t data_length,
int16_t offset, int16_t* total_energy,
int16_t* log_energy) {
// |tot_rshifts| accumulates the number of right shifts performed on |energy|.
int tot_rshifts = 0;
// The |energy| will be normalized to 15 bits. We use unsigned integer because
// we eventually will mask out the fractional part.
uint32_t energy = 0;
assert(data_in != NULL);
assert(data_length > 0);
energy = (uint32_t) WebRtcSpl_Energy((int16_t*) data_in, data_length,
&tot_rshifts);
if (energy != 0) {
// By construction, normalizing to 15 bits is equivalent with 17 leading
// zeros of an unsigned 32 bit value.
int normalizing_rshifts = 17 - WebRtcSpl_NormU32(energy);
// In a 15 bit representation the leading bit is 2^14. log2(2^14) in Q10 is
// (14 << 10), which is what we initialize |log2_energy| with. For a more
// detailed derivations, see below.
int16_t log2_energy = kLogEnergyIntPart;
tot_rshifts += normalizing_rshifts;
// Normalize |energy| to 15 bits.
// |tot_rshifts| is now the total number of right shifts performed on
// |energy| after normalization. This means that |energy| is in
// Q(-tot_rshifts).
if (normalizing_rshifts < 0) {
energy <<= -normalizing_rshifts;
} else {
energy >>= normalizing_rshifts;
}
// Calculate the energy of |data_in| in dB, in Q4.
//
// 10 * log10("true energy") in Q4 = 2^4 * 10 * log10("true energy") =
// 160 * log10(|energy| * 2^|tot_rshifts|) =
// 160 * log10(2) * log2(|energy| * 2^|tot_rshifts|) =
// 160 * log10(2) * (log2(|energy|) + log2(2^|tot_rshifts|)) =
// (160 * log10(2)) * (log2(|energy|) + |tot_rshifts|) =
// |kLogConst| * (|log2_energy| + |tot_rshifts|)
//
// We know by construction that |energy| is normalized to 15 bits. Hence,
// |energy| = 2^14 + frac_Q15, where frac_Q15 is a fractional part in Q15.
// Further, we'd like |log2_energy| in Q10
// log2(|energy|) in Q10 = 2^10 * log2(2^14 + frac_Q15) =
// 2^10 * log2(2^14 * (1 + frac_Q15 * 2^-14)) =
// 2^10 * (14 + log2(1 + frac_Q15 * 2^-14)) ~=
// (14 << 10) + 2^10 * (frac_Q15 * 2^-14) =
// (14 << 10) + (frac_Q15 * 2^-4) = (14 << 10) + (frac_Q15 >> 4)
//
// Note that frac_Q15 = (|energy| & 0x00003FFF)
// Calculate and add the fractional part to |log2_energy|.
log2_energy += (int16_t) ((energy & 0x00003FFF) >> 4);
// |kLogConst| is in Q9, |log2_energy| in Q10 and |tot_rshifts| in Q0.
// Note that we in our derivation above have accounted for an output in Q4.
*log_energy = (int16_t)(((kLogConst * log2_energy) >> 19) +
((tot_rshifts * kLogConst) >> 9));
if (*log_energy < 0) {
*log_energy = 0;
}
} else {
*log_energy = offset;
return;
}
*log_energy += offset;
// Update the approximate |total_energy| with the energy of |data_in|, if
// |total_energy| has not exceeded |kMinEnergy|. |total_energy| is used as an
// energy indicator in WebRtcVad_GmmProbability() in vad_core.c.
if (*total_energy <= kMinEnergy) {
if (tot_rshifts >= 0) {
// We know by construction that the |energy| > |kMinEnergy| in Q0, so add
// an arbitrary value such that |total_energy| exceeds |kMinEnergy|.
*total_energy += kMinEnergy + 1;
} else {
// By construction |energy| is represented by 15 bits, hence any number of
// right shifted |energy| will fit in an int16_t. In addition, adding the
// value to |total_energy| is wrap around safe as long as
// |kMinEnergy| < 8192.
*total_energy += (int16_t) (energy >> -tot_rshifts); // Q0.
}
}
}
int16_t WebRtcVad_CalculateFeatures(VadInstT* self, const int16_t* data_in,
size_t data_length, int16_t* features) {
int16_t total_energy = 0;
// We expect |data_length| to be 80, 160 or 240 samples, which corresponds to
// 10, 20 or 30 ms in 8 kHz. Therefore, the intermediate downsampled data will
// have at most 120 samples after the first split and at most 60 samples after
// the second split.
int16_t hp_120[120], lp_120[120];
int16_t hp_60[60], lp_60[60];
const size_t half_data_length = data_length >> 1;
size_t length = half_data_length; // |data_length| / 2, corresponds to
// bandwidth = 2000 Hz after downsampling.
// Initialize variables for the first SplitFilter().
int frequency_band = 0;
const int16_t* in_ptr = data_in; // [0 - 4000] Hz.
int16_t* hp_out_ptr = hp_120; // [2000 - 4000] Hz.
int16_t* lp_out_ptr = lp_120; // [0 - 2000] Hz.
assert(data_length <= 240);
assert(4 < kNumChannels - 1); // Checking maximum |frequency_band|.
// Split at 2000 Hz and downsample.
SplitFilter(in_ptr, data_length, &self->upper_state[frequency_band],
&self->lower_state[frequency_band], hp_out_ptr, lp_out_ptr);
// For the upper band (2000 Hz - 4000 Hz) split at 3000 Hz and downsample.
frequency_band = 1;
in_ptr = hp_120; // [2000 - 4000] Hz.
hp_out_ptr = hp_60; // [3000 - 4000] Hz.
lp_out_ptr = lp_60; // [2000 - 3000] Hz.
SplitFilter(in_ptr, length, &self->upper_state[frequency_band],
&self->lower_state[frequency_band], hp_out_ptr, lp_out_ptr);
// Energy in 3000 Hz - 4000 Hz.
length >>= 1; // |data_length| / 4 <=> bandwidth = 1000 Hz.
LogOfEnergy(hp_60, length, kOffsetVector[5], &total_energy, &features[5]);
// Energy in 2000 Hz - 3000 Hz.
LogOfEnergy(lp_60, length, kOffsetVector[4], &total_energy, &features[4]);
// For the lower band (0 Hz - 2000 Hz) split at 1000 Hz and downsample.
frequency_band = 2;
in_ptr = lp_120; // [0 - 2000] Hz.
hp_out_ptr = hp_60; // [1000 - 2000] Hz.
lp_out_ptr = lp_60; // [0 - 1000] Hz.
length = half_data_length; // |data_length| / 2 <=> bandwidth = 2000 Hz.
SplitFilter(in_ptr, length, &self->upper_state[frequency_band],
&self->lower_state[frequency_band], hp_out_ptr, lp_out_ptr);
// Energy in 1000 Hz - 2000 Hz.
length >>= 1; // |data_length| / 4 <=> bandwidth = 1000 Hz.
LogOfEnergy(hp_60, length, kOffsetVector[3], &total_energy, &features[3]);
// For the lower band (0 Hz - 1000 Hz) split at 500 Hz and downsample.
frequency_band = 3;
in_ptr = lp_60; // [0 - 1000] Hz.
hp_out_ptr = hp_120; // [500 - 1000] Hz.
lp_out_ptr = lp_120; // [0 - 500] Hz.
SplitFilter(in_ptr, length, &self->upper_state[frequency_band],
&self->lower_state[frequency_band], hp_out_ptr, lp_out_ptr);
// Energy in 500 Hz - 1000 Hz.
length >>= 1; // |data_length| / 8 <=> bandwidth = 500 Hz.
LogOfEnergy(hp_120, length, kOffsetVector[2], &total_energy, &features[2]);
// For the lower band (0 Hz - 500 Hz) split at 250 Hz and downsample.
frequency_band = 4;
in_ptr = lp_120; // [0 - 500] Hz.
hp_out_ptr = hp_60; // [250 - 500] Hz.
lp_out_ptr = lp_60; // [0 - 250] Hz.
SplitFilter(in_ptr, length, &self->upper_state[frequency_band],
&self->lower_state[frequency_band], hp_out_ptr, lp_out_ptr);
// Energy in 250 Hz - 500 Hz.
length >>= 1; // |data_length| / 16 <=> bandwidth = 250 Hz.
LogOfEnergy(hp_60, length, kOffsetVector[1], &total_energy, &features[1]);
// Remove 0 Hz - 80 Hz, by high pass filtering the lower band.
HighPassFilter(lp_60, length, self->hp_filter_state, hp_120);
// Energy in 80 Hz - 250 Hz.
LogOfEnergy(hp_120, length, kOffsetVector[0], &total_energy, &features[0]);
return total_energy;
}

44
webrtc/common_audio/vad/vad_filterbank.h Normal file
View File

@ -0,0 +1,44 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* This file includes feature calculating functionality used in vad_core.c.
*/
#ifndef WEBRTC_COMMON_AUDIO_VAD_VAD_FILTERBANK_H_
#define WEBRTC_COMMON_AUDIO_VAD_VAD_FILTERBANK_H_
#include "webrtc/common_audio/vad/vad_core.h"
#include "webrtc/typedefs.h"
// Takes |data_length| samples of |data_in| and calculates the logarithm of the
// energy of each of the |kNumChannels| = 6 frequency bands used by the VAD:
// 80 Hz - 250 Hz
// 250 Hz - 500 Hz
// 500 Hz - 1000 Hz
// 1000 Hz - 2000 Hz
// 2000 Hz - 3000 Hz
// 3000 Hz - 4000 Hz
//
// The values are given in Q4 and written to |features|. Further, an approximate
// overall energy is returned. The return value is used in
// WebRtcVad_GmmProbability() as a signal indicator, hence it is arbitrary above
// the threshold |kMinEnergy|.
//
// - self [i/o] : State information of the VAD.
// - data_in [i] : Input audio data, for feature extraction.
// - data_length [i] : Audio data size, in number of samples.
// - features [o] : 10 * log10(energy in each frequency band), Q4.
// - returns : Total energy of the signal (NOTE! This value is not
// exact. It is only used in a comparison.)
int16_t WebRtcVad_CalculateFeatures(VadInstT* self, const int16_t* data_in,
size_t data_length, int16_t* features);
#endif // WEBRTC_COMMON_AUDIO_VAD_VAD_FILTERBANK_H_

83
webrtc/common_audio/vad/vad_gmm.c Normal file
View File

@ -0,0 +1,83 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/vad/vad_gmm.h"
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/typedefs.h"
static const int32_t kCompVar = 22005;
static const int16_t kLog2Exp = 5909; // log2(exp(1)) in Q12.
// For a normal distribution, the probability of |input| is calculated and
// returned (in Q20). The formula for normal distributed probability is
//
// 1 / s * exp(-(x - m)^2 / (2 * s^2))
//
// where the parameters are given in the following Q domains:
// m = |mean| (Q7)
// s = |std| (Q7)
// x = |input| (Q4)
// in addition to the probability we output |delta| (in Q11) used when updating
// the noise/speech model.
int32_t WebRtcVad_GaussianProbability(int16_t input,
int16_t mean,
int16_t std,
int16_t* delta) {
int16_t tmp16, inv_std, inv_std2, exp_value = 0;
int32_t tmp32;
// Calculate |inv_std| = 1 / s, in Q10.
// 131072 = 1 in Q17, and (|std| >> 1) is for rounding instead of truncation.
// Q-domain: Q17 / Q7 = Q10.
tmp32 = (int32_t) 131072 + (int32_t) (std >> 1);
inv_std = (int16_t) WebRtcSpl_DivW32W16(tmp32, std);
// Calculate |inv_std2| = 1 / s^2, in Q14.
tmp16 = (inv_std >> 2); // Q10 -> Q8.
// Q-domain: (Q8 * Q8) >> 2 = Q14.
inv_std2 = (int16_t)((tmp16 * tmp16) >> 2);
// TODO(bjornv): Investigate if changing to
// inv_std2 = (int16_t)((inv_std * inv_std) >> 6);
// gives better accuracy.
tmp16 = (input << 3); // Q4 -> Q7
tmp16 = tmp16 - mean; // Q7 - Q7 = Q7
// To be used later, when updating noise/speech model.
// |delta| = (x - m) / s^2, in Q11.
// Q-domain: (Q14 * Q7) >> 10 = Q11.
*delta = (int16_t)((inv_std2 * tmp16) >> 10);
// Calculate the exponent |tmp32| = (x - m)^2 / (2 * s^2), in Q10. Replacing
// division by two with one shift.
// Q-domain: (Q11 * Q7) >> 8 = Q10.
tmp32 = (*delta * tmp16) >> 9;
// If the exponent is small enough to give a non-zero probability we calculate
// |exp_value| ~= exp(-(x - m)^2 / (2 * s^2))
// ~= exp2(-log2(exp(1)) * |tmp32|).
if (tmp32 < kCompVar) {
// Calculate |tmp16| = log2(exp(1)) * |tmp32|, in Q10.
// Q-domain: (Q12 * Q10) >> 12 = Q10.
tmp16 = (int16_t)((kLog2Exp * tmp32) >> 12);
tmp16 = -tmp16;
exp_value = (0x0400 | (tmp16 & 0x03FF));
tmp16 ^= 0xFFFF;
tmp16 >>= 10;
tmp16 += 1;
// Get |exp_value| = exp(-|tmp32|) in Q10.
exp_value >>= tmp16;
}
// Calculate and return (1 / s) * exp(-(x - m)^2 / (2 * s^2)), in Q20.
// Q-domain: Q10 * Q10 = Q20.
return inv_std * exp_value;
}

39
webrtc/common_audio/vad/vad_gmm.h Normal file
View File

@ -0,0 +1,39 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
// Gaussian probability calculations internally used in vad_core.c.
#ifndef WEBRTC_COMMON_AUDIO_VAD_VAD_GMM_H_
#define WEBRTC_COMMON_AUDIO_VAD_VAD_GMM_H_
#include "webrtc/typedefs.h"
// Calculates the probability for |input|, given that |input| comes from a
// normal distribution with mean and standard deviation (|mean|, |std|).
//
// Inputs:
// - input : input sample in Q4.
// - mean : mean input in the statistical model, Q7.
// - std : standard deviation, Q7.
//
// Output:
//
// - delta : input used when updating the model, Q11.
// |delta| = (|input| - |mean|) / |std|^2.
//
// Return:
// (probability for |input|) =
// 1 / |std| * exp(-(|input| - |mean|)^2 / (2 * |std|^2));
int32_t WebRtcVad_GaussianProbability(int16_t input,
int16_t mean,
int16_t std,
int16_t* delta);
#endif // WEBRTC_COMMON_AUDIO_VAD_VAD_GMM_H_

178
webrtc/common_audio/vad/vad_sp.c Normal file
View File

@ -0,0 +1,178 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/vad/vad_sp.h"
#include <assert.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/common_audio/vad/vad_core.h"
#include "webrtc/typedefs.h"
// Allpass filter coefficients, upper and lower, in Q13.
// Upper: 0.64, Lower: 0.17.
static const int16_t kAllPassCoefsQ13[2] = { 5243, 1392 }; // Q13.
static const int16_t kSmoothingDown = 6553; // 0.2 in Q15.
static const int16_t kSmoothingUp = 32439; // 0.99 in Q15.
// TODO(bjornv): Move this function to vad_filterbank.c.
// Downsampling filter based on splitting filter and allpass functions.
void WebRtcVad_Downsampling(const int16_t* signal_in,
int16_t* signal_out,
int32_t* filter_state,
size_t in_length) {
int16_t tmp16_1 = 0, tmp16_2 = 0;
int32_t tmp32_1 = filter_state[0];
int32_t tmp32_2 = filter_state[1];
size_t n = 0;
// Downsampling by 2 gives half length.
size_t half_length = (in_length >> 1);
// Filter coefficients in Q13, filter state in Q0.
for (n = 0; n < half_length; n++) {
// All-pass filtering upper branch.
tmp16_1 = (int16_t) ((tmp32_1 >> 1) +
((kAllPassCoefsQ13[0] * *signal_in) >> 14));
*signal_out = tmp16_1;
tmp32_1 = (int32_t)(*signal_in++) - ((kAllPassCoefsQ13[0] * tmp16_1) >> 12);
// All-pass filtering lower branch.
tmp16_2 = (int16_t) ((tmp32_2 >> 1) +
((kAllPassCoefsQ13[1] * *signal_in) >> 14));
*signal_out++ += tmp16_2;
tmp32_2 = (int32_t)(*signal_in++) - ((kAllPassCoefsQ13[1] * tmp16_2) >> 12);
}
// Store the filter states.
filter_state[0] = tmp32_1;
filter_state[1] = tmp32_2;
}
// Inserts |feature_value| into |low_value_vector|, if it is one of the 16
// smallest values the last 100 frames. Then calculates and returns the median
// of the five smallest values.
int16_t WebRtcVad_FindMinimum(VadInstT* self,
int16_t feature_value,
int channel) {
int i = 0, j = 0;
int position = -1;
// Offset to beginning of the 16 minimum values in memory.
const int offset = (channel << 4);
int16_t current_median = 1600;
int16_t alpha = 0;
int32_t tmp32 = 0;
// Pointer to memory for the 16 minimum values and the age of each value of
// the |channel|.
int16_t* age = &self->index_vector[offset];
int16_t* smallest_values = &self->low_value_vector[offset];
assert(channel < kNumChannels);
// Each value in |smallest_values| is getting 1 loop older. Update |age|, and
// remove old values.
for (i = 0; i < 16; i++) {
if (age[i] != 100) {
age[i]++;
} else {
// Too old value. Remove from memory and shift larger values downwards.
for (j = i; j < 16; j++) {
smallest_values[j] = smallest_values[j + 1];
age[j] = age[j + 1];
}
age[15] = 101;
smallest_values[15] = 10000;
}
}
// Check if |feature_value| is smaller than any of the values in
// |smallest_values|. If so, find the |position| where to insert the new value
// (|feature_value|).
if (feature_value < smallest_values[7]) {
if (feature_value < smallest_values[3]) {
if (feature_value < smallest_values[1]) {
if (feature_value < smallest_values[0]) {
position = 0;
} else {
position = 1;
}
} else if (feature_value < smallest_values[2]) {
position = 2;
} else {
position = 3;
}
} else if (feature_value < smallest_values[5]) {
if (feature_value < smallest_values[4]) {
position = 4;
} else {
position = 5;
}
} else if (feature_value < smallest_values[6]) {
position = 6;
} else {
position = 7;
}
} else if (feature_value < smallest_values[15]) {
if (feature_value < smallest_values[11]) {
if (feature_value < smallest_values[9]) {
if (feature_value < smallest_values[8]) {
position = 8;
} else {
position = 9;
}
} else if (feature_value < smallest_values[10]) {
position = 10;
} else {
position = 11;
}
} else if (feature_value < smallest_values[13]) {
if (feature_value < smallest_values[12]) {
position = 12;
} else {
position = 13;
}
} else if (feature_value < smallest_values[14]) {
position = 14;
} else {
position = 15;
}
}
// If we have detected a new small value, insert it at the correct position
// and shift larger values up.
if (position > -1) {
for (i = 15; i > position; i--) {
smallest_values[i] = smallest_values[i - 1];
age[i] = age[i - 1];
}
smallest_values[position] = feature_value;
age[position] = 1;
}
// Get |current_median|.
if (self->frame_counter > 2) {
current_median = smallest_values[2];
} else if (self->frame_counter > 0) {
current_median = smallest_values[0];
}
// Smooth the median value.
if (self->frame_counter > 0) {
if (current_median < self->mean_value[channel]) {
alpha = kSmoothingDown; // 0.2 in Q15.
} else {
alpha = kSmoothingUp; // 0.99 in Q15.
}
}
tmp32 = (alpha + 1) * self->mean_value[channel];
tmp32 += (WEBRTC_SPL_WORD16_MAX - alpha) * current_median;
tmp32 += 16384;
self->mean_value[channel] = (int16_t) (tmp32 >> 15);
return self->mean_value[channel];
}

56
webrtc/common_audio/vad/vad_sp.h Normal file
View File

@ -0,0 +1,56 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
// This file includes specific signal processing tools used in vad_core.c.
#ifndef WEBRTC_COMMON_AUDIO_VAD_VAD_SP_H_
#define WEBRTC_COMMON_AUDIO_VAD_VAD_SP_H_
#include "webrtc/common_audio/vad/vad_core.h"
#include "webrtc/typedefs.h"
// Downsamples the signal by a factor 2, eg. 32->16 or 16->8.
//
// Inputs:
// - signal_in : Input signal.
// - in_length : Length of input signal in samples.
//
// Input & Output:
// - filter_state : Current filter states of the two all-pass filters. The
// |filter_state| is updated after all samples have been
// processed.
//
// Output:
// - signal_out : Downsampled signal (of length |in_length| / 2).
void WebRtcVad_Downsampling(const int16_t* signal_in,
int16_t* signal_out,
int32_t* filter_state,
size_t in_length);
// Updates and returns the smoothed feature minimum. As minimum we use the
// median of the five smallest feature values in a 100 frames long window.
// As long as |handle->frame_counter| is zero, that is, we haven't received any
// "valid" data, FindMinimum() outputs the default value of 1600.
//
// Inputs:
// - feature_value : New feature value to update with.
// - channel : Channel number.
//
// Input & Output:
// - handle : State information of the VAD.
//
// Returns:
// : Smoothed minimum value for a moving window.
int16_t WebRtcVad_FindMinimum(VadInstT* handle,
int16_t feature_value,
int channel);
#endif // WEBRTC_COMMON_AUDIO_VAD_VAD_SP_H_

116
webrtc/common_audio/vad/webrtc_vad.c Normal file
View File

@ -0,0 +1,116 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/common_audio/vad/include/webrtc_vad.h"
#include <stdlib.h>
#include <string.h>
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
#include "webrtc/common_audio/vad/vad_core.h"
#include "webrtc/typedefs.h"
static const int kInitCheck = 42;
static const int kValidRates[] = { 8000, 16000, 32000, 48000 };
static const size_t kRatesSize = sizeof(kValidRates) / sizeof(*kValidRates);
static const int kMaxFrameLengthMs = 30;
VadInst* WebRtcVad_Create() {
VadInstT* self = (VadInstT*)malloc(sizeof(VadInstT));
WebRtcSpl_Init();
self->init_flag = 0;
return (VadInst*)self;
}
void WebRtcVad_Free(VadInst* handle) {
free(handle);
}
// TODO(bjornv): Move WebRtcVad_InitCore() code here.
int WebRtcVad_Init(VadInst* handle) {
// Initialize the core VAD component.
return WebRtcVad_InitCore((VadInstT*) handle);
}
// TODO(bjornv): Move WebRtcVad_set_mode_core() code here.
int WebRtcVad_set_mode(VadInst* handle, int mode) {
VadInstT* self = (VadInstT*) handle;
if (handle == NULL) {
return -1;
}
if (self->init_flag != kInitCheck) {
return -1;
}
return WebRtcVad_set_mode_core(self, mode);
}
int WebRtcVad_Process(VadInst* handle, int fs, const int16_t* audio_frame,
size_t frame_length) {
int vad = -1;
VadInstT* self = (VadInstT*) handle;
if (handle == NULL) {
return -1;
}
if (self->init_flag != kInitCheck) {
return -1;
}
if (audio_frame == NULL) {
return -1;
}
if (WebRtcVad_ValidRateAndFrameLength(fs, frame_length) != 0) {
return -1;
}
if (fs == 48000) {
vad = WebRtcVad_CalcVad48khz(self, audio_frame, frame_length);
} else if (fs == 32000) {
vad = WebRtcVad_CalcVad32khz(self, audio_frame, frame_length);
} else if (fs == 16000) {
vad = WebRtcVad_CalcVad16khz(self, audio_frame, frame_length);
} else if (fs == 8000) {
vad = WebRtcVad_CalcVad8khz(self, audio_frame, frame_length);
}
if (vad > 0) {
vad = 1;
}
return vad;
}
int WebRtcVad_ValidRateAndFrameLength(int rate, size_t frame_length) {
int return_value = -1;
size_t i;
int valid_length_ms;
size_t valid_length;
// We only allow 10, 20 or 30 ms frames. Loop through valid frame rates and
// see if we have a matching pair.
for (i = 0; i < kRatesSize; i++) {
if (kValidRates[i] == rate) {
for (valid_length_ms = 10; valid_length_ms <= kMaxFrameLengthMs;
valid_length_ms += 10) {
valid_length = (size_t)(kValidRates[i] / 1000 * valid_length_ms);
if (frame_length == valid_length) {
return_value = 0;
break;
}
}
break;
}
}
return return_value;
}

903
webrtc/common_types.h Normal file
View File

@ -0,0 +1,903 @@
/*
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_COMMON_TYPES_H_
#define WEBRTC_COMMON_TYPES_H_
#include <stddef.h>
#include <string.h>
#include <string>
#include <vector>
#include "webrtc/typedefs.h"
#if defined(_MSC_VER)
// Disable "new behavior: elements of array will be default initialized"
// warning. Affects OverUseDetectorOptions.
#pragma warning(disable:4351)
#endif
#ifdef WEBRTC_EXPORT
#define WEBRTC_DLLEXPORT _declspec(dllexport)
#elif WEBRTC_DLL
#define WEBRTC_DLLEXPORT _declspec(dllimport)
#else
#define WEBRTC_DLLEXPORT
#endif
#ifndef NULL
#define NULL 0
#endif
#define RTP_PAYLOAD_NAME_SIZE 32
#if defined(WEBRTC_WIN) || defined(WIN32)
// Compares two strings without regard to case.
#define STR_CASE_CMP(s1, s2) ::_stricmp(s1, s2)
// Compares characters of two strings without regard to case.
#define STR_NCASE_CMP(s1, s2, n) ::_strnicmp(s1, s2, n)
#else
#define STR_CASE_CMP(s1, s2) ::strcasecmp(s1, s2)
#define STR_NCASE_CMP(s1, s2, n) ::strncasecmp(s1, s2, n)
#endif
namespace webrtc {
class Config;
class InStream
{
public:
// Reads |length| bytes from file to |buf|. Returns the number of bytes read
// or -1 on error.
virtual int Read(void *buf, size_t len) = 0;
virtual int Rewind();
virtual ~InStream() {}
protected:
InStream() {}
};
class OutStream
{
public:
// Writes |length| bytes from |buf| to file. The actual writing may happen
// some time later. Call Flush() to force a write.
virtual bool Write(const void *buf, size_t len) = 0;
virtual int Rewind();
virtual ~OutStream() {}
protected:
OutStream() {}
};
enum TraceModule
{
kTraceUndefined = 0,
// not a module, triggered from the engine code
kTraceVoice = 0x0001,
// not a module, triggered from the engine code
kTraceVideo = 0x0002,
// not a module, triggered from the utility code
kTraceUtility = 0x0003,
kTraceRtpRtcp = 0x0004,
kTraceTransport = 0x0005,
kTraceSrtp = 0x0006,
kTraceAudioCoding = 0x0007,
kTraceAudioMixerServer = 0x0008,
kTraceAudioMixerClient = 0x0009,
kTraceFile = 0x000a,
kTraceAudioProcessing = 0x000b,
kTraceVideoCoding = 0x0010,
kTraceVideoMixer = 0x0011,
kTraceAudioDevice = 0x0012,
kTraceVideoRenderer = 0x0014,
kTraceVideoCapture = 0x0015,
kTraceRemoteBitrateEstimator = 0x0017,
};
enum TraceLevel
{
kTraceNone = 0x0000, // no trace
kTraceStateInfo = 0x0001,
kTraceWarning = 0x0002,
kTraceError = 0x0004,
kTraceCritical = 0x0008,
kTraceApiCall = 0x0010,
kTraceDefault = 0x00ff,
kTraceModuleCall = 0x0020,
kTraceMemory = 0x0100, // memory info
kTraceTimer = 0x0200, // timing info
kTraceStream = 0x0400, // "continuous" stream of data
// used for debug purposes
kTraceDebug = 0x0800, // debug
kTraceInfo = 0x1000, // debug info
// Non-verbose level used by LS_INFO of logging.h. Do not use directly.
kTraceTerseInfo = 0x2000,
kTraceAll = 0xffff
};
// External Trace API
class TraceCallback {
public:
virtual void Print(TraceLevel level, const char* message, int length) = 0;
protected:
virtual ~TraceCallback() {}
TraceCallback() {}
};
enum FileFormats
{
kFileFormatWavFile = 1,
kFileFormatCompressedFile = 2,
kFileFormatPreencodedFile = 4,
kFileFormatPcm16kHzFile = 7,
kFileFormatPcm8kHzFile = 8,
kFileFormatPcm32kHzFile = 9
};
enum ProcessingTypes
{
kPlaybackPerChannel = 0,
kPlaybackAllChannelsMixed,
kRecordingPerChannel,
kRecordingAllChannelsMixed,
kRecordingPreprocessing
};
enum FrameType
{
kFrameEmpty = 0,
kAudioFrameSpeech = 1,
kAudioFrameCN = 2,
kVideoFrameKey = 3, // independent frame
kVideoFrameDelta = 4, // depends on the previus frame
};
// Statistics for an RTCP channel
struct RtcpStatistics {
RtcpStatistics()
: fraction_lost(0),
cumulative_lost(0),
extended_max_sequence_number(0),
jitter(0) {}
uint8_t fraction_lost;
uint32_t cumulative_lost;
uint32_t extended_max_sequence_number;
uint32_t jitter;
};
class RtcpStatisticsCallback {
public:
virtual ~RtcpStatisticsCallback() {}
virtual void StatisticsUpdated(const RtcpStatistics& statistics,
uint32_t ssrc) = 0;
virtual void CNameChanged(const char* cname, uint32_t ssrc) = 0;
};
// Statistics for RTCP packet types.
struct RtcpPacketTypeCounter {
RtcpPacketTypeCounter()
: first_packet_time_ms(-1),
nack_packets(0),
fir_packets(0),
pli_packets(0),
nack_requests(0),
unique_nack_requests(0) {}
void Add(const RtcpPacketTypeCounter& other) {
nack_packets += other.nack_packets;
fir_packets += other.fir_packets;
pli_packets += other.pli_packets;
nack_requests += other.nack_requests;
unique_nack_requests += other.unique_nack_requests;
if (other.first_packet_time_ms != -1 &&
(other.first_packet_time_ms < first_packet_time_ms ||
first_packet_time_ms == -1)) {
// Use oldest time.
first_packet_time_ms = other.first_packet_time_ms;
}
}
int64_t TimeSinceFirstPacketInMs(int64_t now_ms) const {
return (first_packet_time_ms == -1) ? -1 : (now_ms - first_packet_time_ms);
}
int UniqueNackRequestsInPercent() const {
if (nack_requests == 0) {
return 0;
}
return static_cast<int>(
(unique_nack_requests * 100.0f / nack_requests) + 0.5f);
}
int64_t first_packet_time_ms; // Time when first packet is sent/received.
uint32_t nack_packets; // Number of RTCP NACK packets.
uint32_t fir_packets; // Number of RTCP FIR packets.
uint32_t pli_packets; // Number of RTCP PLI packets.
uint32_t nack_requests; // Number of NACKed RTP packets.
uint32_t unique_nack_requests; // Number of unique NACKed RTP packets.
};
class RtcpPacketTypeCounterObserver {
public:
virtual ~RtcpPacketTypeCounterObserver() {}
virtual void RtcpPacketTypesCounterUpdated(
uint32_t ssrc,
const RtcpPacketTypeCounter& packet_counter) = 0;
};
// Rate statistics for a stream.
struct BitrateStatistics {
BitrateStatistics() : bitrate_bps(0), packet_rate(0), timestamp_ms(0) {}
uint32_t bitrate_bps; // Bitrate in bits per second.
uint32_t packet_rate; // Packet rate in packets per second.
uint64_t timestamp_ms; // Ntp timestamp in ms at time of rate estimation.
};
// Callback, used to notify an observer whenever new rates have been estimated.
class BitrateStatisticsObserver {
public:
virtual ~BitrateStatisticsObserver() {}
virtual void Notify(const BitrateStatistics& total_stats,
const BitrateStatistics& retransmit_stats,
uint32_t ssrc) = 0;
};
struct FrameCounts {
FrameCounts() : key_frames(0), delta_frames(0) {}
int key_frames;
int delta_frames;
};
// Callback, used to notify an observer whenever frame counts have been updated.
class FrameCountObserver {
public:
virtual ~FrameCountObserver() {}
virtual void FrameCountUpdated(const FrameCounts& frame_counts,
uint32_t ssrc) = 0;
};
// Callback, used to notify an observer whenever the send-side delay is updated.
class SendSideDelayObserver {
public:
virtual ~SendSideDelayObserver() {}
virtual void SendSideDelayUpdated(int avg_delay_ms,
int max_delay_ms,
uint32_t ssrc) = 0;
};
// ==================================================================
// Voice specific types
// ==================================================================
// Each codec supported can be described by this structure.
struct CodecInst {
int pltype;
char plname[RTP_PAYLOAD_NAME_SIZE];
int plfreq;
int pacsize;
int channels;
int rate; // bits/sec unlike {start,min,max}Bitrate elsewhere in this file!
bool operator==(const CodecInst& other) const {
return pltype == other.pltype &&
(STR_CASE_CMP(plname, other.plname) == 0) &&
plfreq == other.plfreq &&
pacsize == other.pacsize &&
channels == other.channels &&
rate == other.rate;
}
bool operator!=(const CodecInst& other) const {
return !(*this == other);
}
};
// RTP
enum {kRtpCsrcSize = 15}; // RFC 3550 page 13
enum RTPDirections
{
kRtpIncoming = 0,
kRtpOutgoing
};
enum PayloadFrequencies
{
kFreq8000Hz = 8000,
kFreq16000Hz = 16000,
kFreq32000Hz = 32000
};
enum VadModes // degree of bandwidth reduction
{
kVadConventional = 0, // lowest reduction
kVadAggressiveLow,
kVadAggressiveMid,
kVadAggressiveHigh // highest reduction
};
struct NetworkStatistics // NETEQ statistics
{
// current jitter buffer size in ms
uint16_t currentBufferSize;
// preferred (optimal) buffer size in ms
uint16_t preferredBufferSize;
// adding extra delay due to "peaky jitter"
bool jitterPeaksFound;
// Loss rate (network + late); fraction between 0 and 1, scaled to Q14.
uint16_t currentPacketLossRate;
// Late loss rate; fraction between 0 and 1, scaled to Q14.
uint16_t currentDiscardRate;
// fraction (of original stream) of synthesized audio inserted through
// expansion (in Q14)
uint16_t currentExpandRate;
// fraction (of original stream) of synthesized speech inserted through
// expansion (in Q14)
uint16_t currentSpeechExpandRate;
// fraction of synthesized speech inserted through pre-emptive expansion
// (in Q14)
uint16_t currentPreemptiveRate;
// fraction of data removed through acceleration (in Q14)
uint16_t currentAccelerateRate;
// fraction of data coming from secondary decoding (in Q14)
uint16_t currentSecondaryDecodedRate;
// clock-drift in parts-per-million (negative or positive)
int32_t clockDriftPPM;
// average packet waiting time in the jitter buffer (ms)
int meanWaitingTimeMs;
// median packet waiting time in the jitter buffer (ms)
int medianWaitingTimeMs;
// min packet waiting time in the jitter buffer (ms)
int minWaitingTimeMs;
// max packet waiting time in the jitter buffer (ms)
int maxWaitingTimeMs;
// added samples in off mode due to packet loss
size_t addedSamples;
};
// Statistics for calls to AudioCodingModule::PlayoutData10Ms().
struct AudioDecodingCallStats {
AudioDecodingCallStats()
: calls_to_silence_generator(0),
calls_to_neteq(0),
decoded_normal(0),
decoded_plc(0),
decoded_cng(0),
decoded_plc_cng(0) {}
int calls_to_silence_generator; // Number of calls where silence generated,
// and NetEq was disengaged from decoding.
int calls_to_neteq; // Number of calls to NetEq.
int decoded_normal; // Number of calls where audio RTP packet decoded.
int decoded_plc; // Number of calls resulted in PLC.
int decoded_cng; // Number of calls where comfort noise generated due to DTX.
int decoded_plc_cng; // Number of calls resulted where PLC faded to CNG.
};
typedef struct
{
int min; // minumum
int max; // maximum
int average; // average
} StatVal;
typedef struct // All levels are reported in dBm0
{
StatVal speech_rx; // long-term speech levels on receiving side
StatVal speech_tx; // long-term speech levels on transmitting side
StatVal noise_rx; // long-term noise/silence levels on receiving side
StatVal noise_tx; // long-term noise/silence levels on transmitting side
} LevelStatistics;
typedef struct // All levels are reported in dB
{
StatVal erl; // Echo Return Loss
StatVal erle; // Echo Return Loss Enhancement
StatVal rerl; // RERL = ERL + ERLE
// Echo suppression inside EC at the point just before its NLP
StatVal a_nlp;
} EchoStatistics;
enum NsModes // type of Noise Suppression
{
kNsUnchanged = 0, // previously set mode
kNsDefault, // platform default
kNsConference, // conferencing default
kNsLowSuppression, // lowest suppression
kNsModerateSuppression,
kNsHighSuppression,
kNsVeryHighSuppression, // highest suppression
};
enum AgcModes // type of Automatic Gain Control
{
kAgcUnchanged = 0, // previously set mode
kAgcDefault, // platform default
// adaptive mode for use when analog volume control exists (e.g. for
// PC softphone)
kAgcAdaptiveAnalog,
// scaling takes place in the digital domain (e.g. for conference servers
// and embedded devices)
kAgcAdaptiveDigital,
// can be used on embedded devices where the capture signal level
// is predictable
kAgcFixedDigital
};
// EC modes
enum EcModes // type of Echo Control
{
kEcUnchanged = 0, // previously set mode
kEcDefault, // platform default
kEcConference, // conferencing default (aggressive AEC)
kEcAec, // Acoustic Echo Cancellation
kEcAecm, // AEC mobile
};
// AECM modes
enum AecmModes // mode of AECM
{
kAecmQuietEarpieceOrHeadset = 0,
// Quiet earpiece or headset use
kAecmEarpiece, // most earpiece use
kAecmLoudEarpiece, // Loud earpiece or quiet speakerphone use
kAecmSpeakerphone, // most speakerphone use (default)
kAecmLoudSpeakerphone // Loud speakerphone
};
// AGC configuration
typedef struct
{
unsigned short targetLeveldBOv;
unsigned short digitalCompressionGaindB;
bool limiterEnable;
} AgcConfig; // AGC configuration parameters
enum StereoChannel
{
kStereoLeft = 0,
kStereoRight,
kStereoBoth
};
// Audio device layers
enum AudioLayers
{
kAudioPlatformDefault = 0,
kAudioWindowsWave = 1,
kAudioWindowsCore = 2,
kAudioLinuxAlsa = 3,
kAudioLinuxPulse = 4
};
// TODO(henrika): to be removed.
enum NetEqModes // NetEQ playout configurations
{
// Optimized trade-off between low delay and jitter robustness for two-way
// communication.
kNetEqDefault = 0,
// Improved jitter robustness at the cost of increased delay. Can be
// used in one-way communication.
kNetEqStreaming = 1,
// Optimzed for decodability of fax signals rather than for perceived audio
// quality.
kNetEqFax = 2,
// Minimal buffer management. Inserts zeros for lost packets and during
// buffer increases.
kNetEqOff = 3,
};
// TODO(henrika): to be removed.
enum OnHoldModes // On Hold direction
{
kHoldSendAndPlay = 0, // Put both sending and playing in on-hold state.
kHoldSendOnly, // Put only sending in on-hold state.
kHoldPlayOnly // Put only playing in on-hold state.
};
// TODO(henrika): to be removed.
enum AmrMode
{
kRfc3267BwEfficient = 0,
kRfc3267OctetAligned = 1,
kRfc3267FileStorage = 2,
};
// ==================================================================
// Video specific types
// ==================================================================
// Raw video types
enum RawVideoType
{
kVideoI420 = 0,
kVideoYV12 = 1,
kVideoYUY2 = 2,
kVideoUYVY = 3,
kVideoIYUV = 4,
kVideoARGB = 5,
kVideoRGB24 = 6,
kVideoRGB565 = 7,
kVideoARGB4444 = 8,
kVideoARGB1555 = 9,
kVideoMJPEG = 10,
kVideoNV12 = 11,
kVideoNV21 = 12,
kVideoBGRA = 13,
kVideoUnknown = 99
};
// Video codec
enum { kConfigParameterSize = 128};
enum { kPayloadNameSize = 32};
enum { kMaxSimulcastStreams = 4};
enum { kMaxTemporalStreams = 4};
enum VideoCodecComplexity
{
kComplexityNormal = 0,
kComplexityHigh = 1,
kComplexityHigher = 2,
kComplexityMax = 3
};
enum VideoCodecProfile
{
kProfileBase = 0x00,
kProfileMain = 0x01
};
enum VP8ResilienceMode {
kResilienceOff, // The stream produced by the encoder requires a
// recovery frame (typically a key frame) to be
// decodable after a packet loss.
kResilientStream, // A stream produced by the encoder is resilient to
// packet losses, but packets within a frame subsequent
// to a loss can't be decoded.
kResilientFrames // Same as kResilientStream but with added resilience
// within a frame.
};
// VP8 specific
struct VideoCodecVP8 {
bool pictureLossIndicationOn;
bool feedbackModeOn;
VideoCodecComplexity complexity;
VP8ResilienceMode resilience;
unsigned char numberOfTemporalLayers;
bool denoisingOn;
bool errorConcealmentOn;
bool automaticResizeOn;
bool frameDroppingOn;
int keyFrameInterval;
bool operator==(const VideoCodecVP8& other) const {
return pictureLossIndicationOn == other.pictureLossIndicationOn &&
feedbackModeOn == other.feedbackModeOn &&
complexity == other.complexity &&
resilience == other.resilience &&
numberOfTemporalLayers == other.numberOfTemporalLayers &&
denoisingOn == other.denoisingOn &&
errorConcealmentOn == other.errorConcealmentOn &&
automaticResizeOn == other.automaticResizeOn &&
frameDroppingOn == other.frameDroppingOn &&
keyFrameInterval == other.keyFrameInterval;
}
bool operator!=(const VideoCodecVP8& other) const {
return !(*this == other);
}
};
// VP9 specific.
struct VideoCodecVP9 {
VideoCodecComplexity complexity;
int resilience;
unsigned char numberOfTemporalLayers;
bool denoisingOn;
bool frameDroppingOn;
int keyFrameInterval;
bool adaptiveQpMode;
bool automaticResizeOn;
unsigned char numberOfSpatialLayers;
bool flexibleMode;
};
// H264 specific.
struct VideoCodecH264 {
VideoCodecProfile profile;
bool frameDroppingOn;
int keyFrameInterval;
// These are NULL/0 if not externally negotiated.
const uint8_t* spsData;
size_t spsLen;
const uint8_t* ppsData;
size_t ppsLen;
};
// Video codec types
enum VideoCodecType {
kVideoCodecVP8,
kVideoCodecVP9,
kVideoCodecH264,
kVideoCodecI420,
kVideoCodecRED,
kVideoCodecULPFEC,
kVideoCodecGeneric,
kVideoCodecUnknown
};
union VideoCodecUnion {
VideoCodecVP8 VP8;
VideoCodecVP9 VP9;
VideoCodecH264 H264;
};
// Simulcast is when the same stream is encoded multiple times with different
// settings such as resolution.
struct SimulcastStream {
unsigned short width;
unsigned short height;
unsigned char numberOfTemporalLayers;
unsigned int maxBitrate; // kilobits/sec.
unsigned int targetBitrate; // kilobits/sec.
unsigned int minBitrate; // kilobits/sec.
unsigned int qpMax; // minimum quality
bool operator==(const SimulcastStream& other) const {
return width == other.width &&
height == other.height &&
numberOfTemporalLayers == other.numberOfTemporalLayers &&
maxBitrate == other.maxBitrate &&
targetBitrate == other.targetBitrate &&
minBitrate == other.minBitrate &&
qpMax == other.qpMax;
}
bool operator!=(const SimulcastStream& other) const {
return !(*this == other);
}
};
enum VideoCodecMode {
kRealtimeVideo,
kScreensharing
};
// Common video codec properties
struct VideoCodec {
VideoCodecType codecType;
char plName[kPayloadNameSize];
unsigned char plType;
unsigned short width;
unsigned short height;
unsigned int startBitrate; // kilobits/sec.
unsigned int maxBitrate; // kilobits/sec.
unsigned int minBitrate; // kilobits/sec.
unsigned int targetBitrate; // kilobits/sec.
unsigned char maxFramerate;
VideoCodecUnion codecSpecific;
unsigned int qpMax;
unsigned char numberOfSimulcastStreams;
SimulcastStream simulcastStream[kMaxSimulcastStreams];
VideoCodecMode mode;
// When using an external encoder/decoder this allows to pass
// extra options without requiring webrtc to be aware of them.
Config* extra_options;
bool operator==(const VideoCodec& other) const {
bool ret = codecType == other.codecType &&
(STR_CASE_CMP(plName, other.plName) == 0) &&
plType == other.plType &&
width == other.width &&
height == other.height &&
startBitrate == other.startBitrate &&
maxBitrate == other.maxBitrate &&
minBitrate == other.minBitrate &&
targetBitrate == other.targetBitrate &&
maxFramerate == other.maxFramerate &&
qpMax == other.qpMax &&
numberOfSimulcastStreams == other.numberOfSimulcastStreams &&
mode == other.mode;
if (ret && codecType == kVideoCodecVP8) {
ret &= (codecSpecific.VP8 == other.codecSpecific.VP8);
}
for (unsigned char i = 0; i < other.numberOfSimulcastStreams && ret; ++i) {
ret &= (simulcastStream[i] == other.simulcastStream[i]);
}
return ret;
}
bool operator!=(const VideoCodec& other) const {
return !(*this == other);
}
};
// Bandwidth over-use detector options. These are used to drive
// experimentation with bandwidth estimation parameters.
// See modules/remote_bitrate_estimator/overuse_detector.h
struct OverUseDetectorOptions {
OverUseDetectorOptions()
: initial_slope(8.0/512.0),
initial_offset(0),
initial_e(),
initial_process_noise(),
initial_avg_noise(0.0),
initial_var_noise(50) {
initial_e[0][0] = 100;
initial_e[1][1] = 1e-1;
initial_e[0][1] = initial_e[1][0] = 0;
initial_process_noise[0] = 1e-13;
initial_process_noise[1] = 1e-2;
}
double initial_slope;
double initial_offset;
double initial_e[2][2];
double initial_process_noise[2];
double initial_avg_noise;
double initial_var_noise;
};
// This structure will have the information about when packet is actually
// received by socket.
struct PacketTime {
PacketTime() : timestamp(-1), not_before(-1) {}
PacketTime(int64_t timestamp, int64_t not_before)
: timestamp(timestamp), not_before(not_before) {
}
int64_t timestamp; // Receive time after socket delivers the data.
int64_t not_before; // Earliest possible time the data could have arrived,
// indicating the potential error in the |timestamp|
// value,in case the system is busy.
// For example, the time of the last select() call.
// If unknown, this value will be set to zero.
};
struct RTPHeaderExtension {
RTPHeaderExtension();
bool hasTransmissionTimeOffset;
int32_t transmissionTimeOffset;
bool hasAbsoluteSendTime;
uint32_t absoluteSendTime;
bool hasTransportSequenceNumber;
uint16_t transportSequenceNumber;
// Audio Level includes both level in dBov and voiced/unvoiced bit. See:
// https://datatracker.ietf.org/doc/draft-lennox-avt-rtp-audio-level-exthdr/
bool hasAudioLevel;
bool voiceActivity;
uint8_t audioLevel;
// For Coordination of Video Orientation. See
// http://www.etsi.org/deliver/etsi_ts/126100_126199/126114/12.07.00_60/
// ts_126114v120700p.pdf
bool hasVideoRotation;
uint8_t videoRotation;
};
struct RTPHeader {
RTPHeader();
bool markerBit;
uint8_t payloadType;
uint16_t sequenceNumber;
uint32_t timestamp;
uint32_t ssrc;
uint8_t numCSRCs;
uint32_t arrOfCSRCs[kRtpCsrcSize];
size_t paddingLength;
size_t headerLength;
int payload_type_frequency;
RTPHeaderExtension extension;
};
struct RtpPacketCounter {
RtpPacketCounter()
: header_bytes(0),
payload_bytes(0),
padding_bytes(0),
packets(0) {}
void Add(const RtpPacketCounter& other) {
header_bytes += other.header_bytes;
payload_bytes += other.payload_bytes;
padding_bytes += other.padding_bytes;
packets += other.packets;
}
void AddPacket(size_t packet_length, const RTPHeader& header) {
++packets;
header_bytes += header.headerLength;
padding_bytes += header.paddingLength;
payload_bytes +=
packet_length - (header.headerLength + header.paddingLength);
}
size_t TotalBytes() const {
return header_bytes + payload_bytes + padding_bytes;
}
size_t header_bytes; // Number of bytes used by RTP headers.
size_t payload_bytes; // Payload bytes, excluding RTP headers and padding.
size_t padding_bytes; // Number of padding bytes.
uint32_t packets; // Number of packets.
};
// Data usage statistics for a (rtp) stream.
struct StreamDataCounters {
StreamDataCounters();
void Add(const StreamDataCounters& other) {
transmitted.Add(other.transmitted);
retransmitted.Add(other.retransmitted);
fec.Add(other.fec);
if (other.first_packet_time_ms != -1 &&
(other.first_packet_time_ms < first_packet_time_ms ||
first_packet_time_ms == -1)) {
// Use oldest time.
first_packet_time_ms = other.first_packet_time_ms;
}
}
int64_t TimeSinceFirstPacketInMs(int64_t now_ms) const {
return (first_packet_time_ms == -1) ? -1 : (now_ms - first_packet_time_ms);
}
// Returns the number of bytes corresponding to the actual media payload (i.e.
// RTP headers, padding, retransmissions and fec packets are excluded).
// Note this function does not have meaning for an RTX stream.
size_t MediaPayloadBytes() const {
return transmitted.payload_bytes - retransmitted.payload_bytes -
fec.payload_bytes;
}
int64_t first_packet_time_ms; // Time when first packet is sent/received.
RtpPacketCounter transmitted; // Number of transmitted packets/bytes.
RtpPacketCounter retransmitted; // Number of retransmitted packets/bytes.
RtpPacketCounter fec; // Number of redundancy packets/bytes.
};
// Callback, called whenever byte/packet counts have been updated.
class StreamDataCountersCallback {
public:
virtual ~StreamDataCountersCallback() {}
virtual void DataCountersUpdated(const StreamDataCounters& counters,
uint32_t ssrc) = 0;
};
// RTCP mode to use. Compound mode is described by RFC 4585 and reduced-size
// RTCP mode is described by RFC 5506.
enum class RtcpMode { kOff, kCompound, kReducedSize };
} // namespace webrtc
#endif // WEBRTC_COMMON_TYPES_H_

1
webrtc/modules/Makefile.am Normal file
View File

@ -0,0 +1 @@
SUBDIRS = audio_processing

59
webrtc/modules/audio_processing/Makefile.am Normal file
View File

@ -0,0 +1,59 @@
SUBDIRS = utility ns aec aecm agc
lib_LTLIBRARIES = libwebrtc_audio_processing.la
if NS_FIXED
COMMON_CXXFLAGS += -DWEBRTC_NS_FIXED=1
NS_LIB = libns_fix
else
COMMON_CXXFLAGS += -DWEBRTC_NS_FLOAT=1
NS_LIB = libns
endif
webrtcincludedir = $(includedir)/webrtc_audio_processing
webrtcinclude_HEADERS = $(top_srcdir)/src/typedefs.h \
$(top_srcdir)/src/modules/interface/module.h \
interface/audio_processing.h \
$(top_srcdir)/src/common_types.h \
$(top_srcdir)/src/modules/interface/module_common_types.h
libwebrtc_audio_processing_la_SOURCES = interface/audio_processing.h \
audio_buffer.cc \
audio_buffer.h \
audio_processing_impl.cc \
audio_processing_impl.h \
echo_cancellation_impl.cc \
echo_cancellation_impl.h \
echo_control_mobile_impl.cc \
echo_control_mobile_impl.h \
gain_control_impl.cc \
gain_control_impl.h \
high_pass_filter_impl.cc \
high_pass_filter_impl.h \
level_estimator_impl.cc \
level_estimator_impl.h \
noise_suppression_impl.cc \
noise_suppression_impl.h \
splitting_filter.cc \
splitting_filter.h \
processing_component.cc \
processing_component.h \
voice_detection_impl.cc \
voice_detection_impl.h
libwebrtc_audio_processing_la_CXXFLAGS = $(AM_CXXFLAGS) $(COMMON_CXXFLAGS) \
-I$(top_srcdir)/src/common_audio/signal_processing_library/main/interface \
-I$(top_srcdir)/src/common_audio/vad/main/interface \
-I$(top_srcdir)/src/system_wrappers/interface \
-I$(top_srcdir)/src/modules/audio_processing/utility \
-I$(top_srcdir)/src/modules/audio_processing/ns/interface \
-I$(top_srcdir)/src/modules/audio_processing/aec/interface \
-I$(top_srcdir)/src/modules/audio_processing/aecm/interface \
-I$(top_srcdir)/src/modules/audio_processing/agc/interface
libwebrtc_audio_processing_la_LIBADD = $(top_builddir)/src/system_wrappers/libsystem_wrappers.la \
$(top_builddir)/src/common_audio/signal_processing_library/libspl.la \
$(top_builddir)/src/common_audio/vad/libvad.la \
$(top_builddir)/src/modules/audio_processing/utility/libapm_util.la \
$(top_builddir)/src/modules/audio_processing/ns/$(NS_LIB).la \
$(top_builddir)/src/modules/audio_processing/aec/libaec.la \
$(top_builddir)/src/modules/audio_processing/aecm/libaecm.la \
$(top_builddir)/src/modules/audio_processing/agc/libagc.la
libwebrtc_audio_processing_la_LDFLAGS = $(AM_LDFLAGS) -version-info $(LIBWEBRTC_AUDIO_PROCESSING_VERSION_INFO)

2
webrtc/modules/audio_processing/OWNERS Normal file
View File

@ -0,0 +1,2 @@
andrew@webrtc.org
bjornv@webrtc.org

16
webrtc/modules/audio_processing/aec/Makefile.am Normal file
View File

@ -0,0 +1,16 @@
noinst_LTLIBRARIES = libaec.la
libaec_la_SOURCES = interface/echo_cancellation.h \
echo_cancellation.c \
aec_core.h \
aec_core.c \
aec_core_sse2.c \
aec_rdft.h \
aec_rdft.c \
aec_rdft_sse2.c \
resampler.h \
resampler.c
libaec_la_CFLAGS = $(AM_CFLAGS) $(COMMON_CFLAGS) \
-I$(top_srcdir)/src/common_audio/signal_processing_library/main/interface \
-I$(top_srcdir)/src/system_wrappers/interface \
-I$(top_srcdir)/src/modules/audio_processing/utility

40
webrtc/modules/audio_processing/aec/aec.gypi Normal file
View File

@ -0,0 +1,40 @@
# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
{
'targets': [
{
'target_name': 'aec',
'type': '<(library)',
'dependencies': [
'<(webrtc_root)/common_audio/common_audio.gyp:spl',
'apm_util'
],
'include_dirs': [
'interface',
],
'direct_dependent_settings': {
'include_dirs': [
'interface',
],
},
'sources': [
'interface/echo_cancellation.h',
'echo_cancellation.c',
'aec_core.h',
'aec_core.c',
'aec_core_sse2.c',
'aec_rdft.h',
'aec_rdft.c',
'aec_rdft_sse2.c',
'resampler.h',
'resampler.c',
],
},
],
}

1466
webrtc/modules/audio_processing/aec/aec_core.c Normal file
View File

File diff suppressed because it is too large Load Diff

181
webrtc/modules/audio_processing/aec/aec_core.h Normal file
View File

@ -0,0 +1,181 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* Specifies the interface for the AEC core.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_CORE_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_CORE_H_
#include <stdio.h>
#include "signal_processing_library.h"
#include "typedefs.h"
//#define AEC_DEBUG // for recording files
#define FRAME_LEN 80
#define PART_LEN 64 // Length of partition
#define PART_LEN1 (PART_LEN + 1) // Unique fft coefficients
#define PART_LEN2 (PART_LEN * 2) // Length of partition * 2
#define NR_PART 12 // Number of partitions
#define FILT_LEN (PART_LEN * NR_PART) // Filter length
#define FILT_LEN2 (FILT_LEN * 2) // Double filter length
#define FAR_BUF_LEN (FILT_LEN2 * 2)
#define PREF_BAND_SIZE 24
#define BLOCKL_MAX FRAME_LEN
// Maximum delay in fixed point delay estimator, used for logging
enum {kMaxDelay = 100};
typedef float complex_t[2];
// For performance reasons, some arrays of complex numbers are replaced by twice
// as long arrays of float, all the real parts followed by all the imaginary
// ones (complex_t[SIZE] -> float[2][SIZE]). This allows SIMD optimizations and
// is better than two arrays (one for the real parts and one for the imaginary
// parts) as this other way would require two pointers instead of one and cause
// extra register spilling. This also allows the offsets to be calculated at
// compile time.
// Metrics
enum {offsetLevel = -100};
typedef struct {
float sfrsum;
int sfrcounter;
float framelevel;
float frsum;
int frcounter;
float minlevel;
float averagelevel;
} power_level_t;
typedef struct {
float instant;
float average;
float min;
float max;
float sum;
float hisum;
float himean;
int counter;
int hicounter;
} stats_t;
typedef struct {
int farBufWritePos, farBufReadPos;
int knownDelay;
int inSamples, outSamples;
int delayEstCtr;
void *farFrBuf, *nearFrBuf, *outFrBuf;
void *nearFrBufH;
void *outFrBufH;
float xBuf[PART_LEN2]; // farend
float dBuf[PART_LEN2]; // nearend
float eBuf[PART_LEN2]; // error
float dBufH[PART_LEN2]; // nearend
float xPow[PART_LEN1];
float dPow[PART_LEN1];
float dMinPow[PART_LEN1];
float dInitMinPow[PART_LEN1];
float *noisePow;
float xfBuf[2][NR_PART * PART_LEN1]; // farend fft buffer
float wfBuf[2][NR_PART * PART_LEN1]; // filter fft
complex_t sde[PART_LEN1]; // cross-psd of nearend and error
complex_t sxd[PART_LEN1]; // cross-psd of farend and nearend
complex_t xfwBuf[NR_PART * PART_LEN1]; // farend windowed fft buffer
float sx[PART_LEN1], sd[PART_LEN1], se[PART_LEN1]; // far, near and error psd
float hNs[PART_LEN1];
float hNlFbMin, hNlFbLocalMin;
float hNlXdAvgMin;
int hNlNewMin, hNlMinCtr;
float overDrive, overDriveSm;
float targetSupp, minOverDrive;
float outBuf[PART_LEN];
int delayIdx;
short stNearState, echoState;
short divergeState;
int xfBufBlockPos;
short farBuf[FILT_LEN2 * 2];
short mult; // sampling frequency multiple
int sampFreq;
WebRtc_UWord32 seed;
float mu; // stepsize
float errThresh; // error threshold
int noiseEstCtr;
power_level_t farlevel;
power_level_t nearlevel;
power_level_t linoutlevel;
power_level_t nlpoutlevel;
int metricsMode;
int stateCounter;
stats_t erl;
stats_t erle;
stats_t aNlp;
stats_t rerl;
// Quantities to control H band scaling for SWB input
int freq_avg_ic; //initial bin for averaging nlp gain
int flag_Hband_cn; //for comfort noise
float cn_scale_Hband; //scale for comfort noise in H band
int delay_histogram[kMaxDelay];
int delay_logging_enabled;
void* delay_estimator;
#ifdef AEC_DEBUG
FILE *farFile;
FILE *nearFile;
FILE *outFile;
FILE *outLpFile;
#endif
} aec_t;
typedef void (*WebRtcAec_FilterFar_t)(aec_t *aec, float yf[2][PART_LEN1]);
extern WebRtcAec_FilterFar_t WebRtcAec_FilterFar;
typedef void (*WebRtcAec_ScaleErrorSignal_t)(aec_t *aec, float ef[2][PART_LEN1]);
extern WebRtcAec_ScaleErrorSignal_t WebRtcAec_ScaleErrorSignal;
typedef void (*WebRtcAec_FilterAdaptation_t)
(aec_t *aec, float *fft, float ef[2][PART_LEN1]);
extern WebRtcAec_FilterAdaptation_t WebRtcAec_FilterAdaptation;
typedef void (*WebRtcAec_OverdriveAndSuppress_t)
(aec_t *aec, float hNl[PART_LEN1], const float hNlFb, float efw[2][PART_LEN1]);
extern WebRtcAec_OverdriveAndSuppress_t WebRtcAec_OverdriveAndSuppress;
int WebRtcAec_CreateAec(aec_t **aec);
int WebRtcAec_FreeAec(aec_t *aec);
int WebRtcAec_InitAec(aec_t *aec, int sampFreq);
void WebRtcAec_InitAec_SSE2(void);
void WebRtcAec_InitMetrics(aec_t *aec);
void WebRtcAec_ProcessFrame(aec_t *aec, const short *farend,
const short *nearend, const short *nearendH,
short *out, short *outH,
int knownDelay);
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_CORE_H_

417
webrtc/modules/audio_processing/aec/aec_core_sse2.c Normal file
View File

@ -0,0 +1,417 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* The core AEC algorithm, SSE2 version of speed-critical functions.
*/
#include "typedefs.h"
#if defined(WEBRTC_USE_SSE2)
#include <emmintrin.h>
#include <math.h>
#include "aec_core.h"
#include "aec_rdft.h"
__inline static float MulRe(float aRe, float aIm, float bRe, float bIm)
{
return aRe * bRe - aIm * bIm;
}
__inline static float MulIm(float aRe, float aIm, float bRe, float bIm)
{
return aRe * bIm + aIm * bRe;
}
static void FilterFarSSE2(aec_t *aec, float yf[2][PART_LEN1])
{
int i;
for (i = 0; i < NR_PART; i++) {
int j;
int xPos = (i + aec->xfBufBlockPos) * PART_LEN1;
int pos = i * PART_LEN1;
// Check for wrap
if (i + aec->xfBufBlockPos >= NR_PART) {
xPos -= NR_PART*(PART_LEN1);
}
// vectorized code (four at once)
for (j = 0; j + 3 < PART_LEN1; j += 4) {
const __m128 xfBuf_re = _mm_loadu_ps(&aec->xfBuf[0][xPos + j]);
const __m128 xfBuf_im = _mm_loadu_ps(&aec->xfBuf[1][xPos + j]);
const __m128 wfBuf_re = _mm_loadu_ps(&aec->wfBuf[0][pos + j]);
const __m128 wfBuf_im = _mm_loadu_ps(&aec->wfBuf[1][pos + j]);
const __m128 yf_re = _mm_loadu_ps(&yf[0][j]);
const __m128 yf_im = _mm_loadu_ps(&yf[1][j]);
const __m128 a = _mm_mul_ps(xfBuf_re, wfBuf_re);
const __m128 b = _mm_mul_ps(xfBuf_im, wfBuf_im);
const __m128 c = _mm_mul_ps(xfBuf_re, wfBuf_im);
const __m128 d = _mm_mul_ps(xfBuf_im, wfBuf_re);
const __m128 e = _mm_sub_ps(a, b);
const __m128 f = _mm_add_ps(c, d);
const __m128 g = _mm_add_ps(yf_re, e);
const __m128 h = _mm_add_ps(yf_im, f);
_mm_storeu_ps(&yf[0][j], g);
_mm_storeu_ps(&yf[1][j], h);
}
// scalar code for the remaining items.
for (; j < PART_LEN1; j++) {
yf[0][j] += MulRe(aec->xfBuf[0][xPos + j], aec->xfBuf[1][xPos + j],
aec->wfBuf[0][ pos + j], aec->wfBuf[1][ pos + j]);
yf[1][j] += MulIm(aec->xfBuf[0][xPos + j], aec->xfBuf[1][xPos + j],
aec->wfBuf[0][ pos + j], aec->wfBuf[1][ pos + j]);
}
}
}
static void ScaleErrorSignalSSE2(aec_t *aec, float ef[2][PART_LEN1])
{
const __m128 k1e_10f = _mm_set1_ps(1e-10f);
const __m128 kThresh = _mm_set1_ps(aec->errThresh);
const __m128 kMu = _mm_set1_ps(aec->mu);
int i;
// vectorized code (four at once)
for (i = 0; i + 3 < PART_LEN1; i += 4) {
const __m128 xPow = _mm_loadu_ps(&aec->xPow[i]);
const __m128 ef_re_base = _mm_loadu_ps(&ef[0][i]);
const __m128 ef_im_base = _mm_loadu_ps(&ef[1][i]);
const __m128 xPowPlus = _mm_add_ps(xPow, k1e_10f);
__m128 ef_re = _mm_div_ps(ef_re_base, xPowPlus);
__m128 ef_im = _mm_div_ps(ef_im_base, xPowPlus);
const __m128 ef_re2 = _mm_mul_ps(ef_re, ef_re);
const __m128 ef_im2 = _mm_mul_ps(ef_im, ef_im);
const __m128 ef_sum2 = _mm_add_ps(ef_re2, ef_im2);
const __m128 absEf = _mm_sqrt_ps(ef_sum2);
const __m128 bigger = _mm_cmpgt_ps(absEf, kThresh);
__m128 absEfPlus = _mm_add_ps(absEf, k1e_10f);
const __m128 absEfInv = _mm_div_ps(kThresh, absEfPlus);
__m128 ef_re_if = _mm_mul_ps(ef_re, absEfInv);
__m128 ef_im_if = _mm_mul_ps(ef_im, absEfInv);
ef_re_if = _mm_and_ps(bigger, ef_re_if);
ef_im_if = _mm_and_ps(bigger, ef_im_if);
ef_re = _mm_andnot_ps(bigger, ef_re);
ef_im = _mm_andnot_ps(bigger, ef_im);
ef_re = _mm_or_ps(ef_re, ef_re_if);
ef_im = _mm_or_ps(ef_im, ef_im_if);
ef_re = _mm_mul_ps(ef_re, kMu);
ef_im = _mm_mul_ps(ef_im, kMu);
_mm_storeu_ps(&ef[0][i], ef_re);
_mm_storeu_ps(&ef[1][i], ef_im);
}
// scalar code for the remaining items.
for (; i < (PART_LEN1); i++) {
float absEf;
ef[0][i] /= (aec->xPow[i] + 1e-10f);
ef[1][i] /= (aec->xPow[i] + 1e-10f);
absEf = sqrtf(ef[0][i] * ef[0][i] + ef[1][i] * ef[1][i]);
if (absEf > aec->errThresh) {
absEf = aec->errThresh / (absEf + 1e-10f);
ef[0][i] *= absEf;
ef[1][i] *= absEf;
}
// Stepsize factor
ef[0][i] *= aec->mu;
ef[1][i] *= aec->mu;
}
}
static void FilterAdaptationSSE2(aec_t *aec, float *fft, float ef[2][PART_LEN1]) {
int i, j;
for (i = 0; i < NR_PART; i++) {
int xPos = (i + aec->xfBufBlockPos)*(PART_LEN1);
int pos = i * PART_LEN1;
// Check for wrap
if (i + aec->xfBufBlockPos >= NR_PART) {
xPos -= NR_PART * PART_LEN1;
}
// Process the whole array...
for (j = 0; j < PART_LEN; j+= 4) {
// Load xfBuf and ef.
const __m128 xfBuf_re = _mm_loadu_ps(&aec->xfBuf[0][xPos + j]);
const __m128 xfBuf_im = _mm_loadu_ps(&aec->xfBuf[1][xPos + j]);
const __m128 ef_re = _mm_loadu_ps(&ef[0][j]);
const __m128 ef_im = _mm_loadu_ps(&ef[1][j]);
// Calculate the product of conjugate(xfBuf) by ef.
// re(conjugate(a) * b) = aRe * bRe + aIm * bIm
// im(conjugate(a) * b)= aRe * bIm - aIm * bRe
const __m128 a = _mm_mul_ps(xfBuf_re, ef_re);
const __m128 b = _mm_mul_ps(xfBuf_im, ef_im);
const __m128 c = _mm_mul_ps(xfBuf_re, ef_im);
const __m128 d = _mm_mul_ps(xfBuf_im, ef_re);
const __m128 e = _mm_add_ps(a, b);
const __m128 f = _mm_sub_ps(c, d);
// Interleave real and imaginary parts.
const __m128 g = _mm_unpacklo_ps(e, f);
const __m128 h = _mm_unpackhi_ps(e, f);
// Store
_mm_storeu_ps(&fft[2*j + 0], g);
_mm_storeu_ps(&fft[2*j + 4], h);
}
// ... and fixup the first imaginary entry.
fft[1] = MulRe(aec->xfBuf[0][xPos + PART_LEN],
-aec->xfBuf[1][xPos + PART_LEN],
ef[0][PART_LEN], ef[1][PART_LEN]);
aec_rdft_inverse_128(fft);
memset(fft + PART_LEN, 0, sizeof(float)*PART_LEN);
// fft scaling
{
float scale = 2.0f / PART_LEN2;
const __m128 scale_ps = _mm_load_ps1(&scale);
for (j = 0; j < PART_LEN; j+=4) {
const __m128 fft_ps = _mm_loadu_ps(&fft[j]);
const __m128 fft_scale = _mm_mul_ps(fft_ps, scale_ps);
_mm_storeu_ps(&fft[j], fft_scale);
}
}
aec_rdft_forward_128(fft);
{
float wt1 = aec->wfBuf[1][pos];
aec->wfBuf[0][pos + PART_LEN] += fft[1];
for (j = 0; j < PART_LEN; j+= 4) {
__m128 wtBuf_re = _mm_loadu_ps(&aec->wfBuf[0][pos + j]);
__m128 wtBuf_im = _mm_loadu_ps(&aec->wfBuf[1][pos + j]);
const __m128 fft0 = _mm_loadu_ps(&fft[2 * j + 0]);
const __m128 fft4 = _mm_loadu_ps(&fft[2 * j + 4]);
const __m128 fft_re = _mm_shuffle_ps(fft0, fft4, _MM_SHUFFLE(2, 0, 2 ,0));
const __m128 fft_im = _mm_shuffle_ps(fft0, fft4, _MM_SHUFFLE(3, 1, 3 ,1));
wtBuf_re = _mm_add_ps(wtBuf_re, fft_re);
wtBuf_im = _mm_add_ps(wtBuf_im, fft_im);
_mm_storeu_ps(&aec->wfBuf[0][pos + j], wtBuf_re);
_mm_storeu_ps(&aec->wfBuf[1][pos + j], wtBuf_im);
}
aec->wfBuf[1][pos] = wt1;
}
}
}
static __m128 mm_pow_ps(__m128 a, __m128 b)
{
// a^b = exp2(b * log2(a))
// exp2(x) and log2(x) are calculated using polynomial approximations.
__m128 log2_a, b_log2_a, a_exp_b;
// Calculate log2(x), x = a.
{
// To calculate log2(x), we decompose x like this:
// x = y * 2^n
// n is an integer
// y is in the [1.0, 2.0) range
//
// log2(x) = log2(y) + n
// n can be evaluated by playing with float representation.
// log2(y) in a small range can be approximated, this code uses an order
// five polynomial approximation. The coefficients have been
// estimated with the Remez algorithm and the resulting
// polynomial has a maximum relative error of 0.00086%.
// Compute n.
// This is done by masking the exponent, shifting it into the top bit of
// the mantissa, putting eight into the biased exponent (to shift/
// compensate the fact that the exponent has been shifted in the top/
// fractional part and finally getting rid of the implicit leading one
// from the mantissa by substracting it out.
static const ALIGN16_BEG int float_exponent_mask[4] ALIGN16_END =
{0x7F800000, 0x7F800000, 0x7F800000, 0x7F800000};
static const ALIGN16_BEG int eight_biased_exponent[4] ALIGN16_END =
{0x43800000, 0x43800000, 0x43800000, 0x43800000};
static const ALIGN16_BEG int implicit_leading_one[4] ALIGN16_END =
{0x43BF8000, 0x43BF8000, 0x43BF8000, 0x43BF8000};
static const int shift_exponent_into_top_mantissa = 8;
const __m128 two_n = _mm_and_ps(a, *((__m128 *)float_exponent_mask));
const __m128 n_1 = _mm_castsi128_ps(_mm_srli_epi32(_mm_castps_si128(two_n),
shift_exponent_into_top_mantissa));
const __m128 n_0 = _mm_or_ps(n_1, *((__m128 *)eight_biased_exponent));
const __m128 n = _mm_sub_ps(n_0, *((__m128 *)implicit_leading_one));
// Compute y.
static const ALIGN16_BEG int mantissa_mask[4] ALIGN16_END =
{0x007FFFFF, 0x007FFFFF, 0x007FFFFF, 0x007FFFFF};
static const ALIGN16_BEG int zero_biased_exponent_is_one[4] ALIGN16_END =
{0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000};
const __m128 mantissa = _mm_and_ps(a, *((__m128 *)mantissa_mask));
const __m128 y = _mm_or_ps(
mantissa, *((__m128 *)zero_biased_exponent_is_one));
// Approximate log2(y) ~= (y - 1) * pol5(y).
// pol5(y) = C5 * y^5 + C4 * y^4 + C3 * y^3 + C2 * y^2 + C1 * y + C0
static const ALIGN16_BEG float ALIGN16_END C5[4] =
{-3.4436006e-2f, -3.4436006e-2f, -3.4436006e-2f, -3.4436006e-2f};
static const ALIGN16_BEG float ALIGN16_END C4[4] =
{3.1821337e-1f, 3.1821337e-1f, 3.1821337e-1f, 3.1821337e-1f};
static const ALIGN16_BEG float ALIGN16_END C3[4] =
{-1.2315303f, -1.2315303f, -1.2315303f, -1.2315303f};
static const ALIGN16_BEG float ALIGN16_END C2[4] =
{2.5988452f, 2.5988452f, 2.5988452f, 2.5988452f};
static const ALIGN16_BEG float ALIGN16_END C1[4] =
{-3.3241990f, -3.3241990f, -3.3241990f, -3.3241990f};
static const ALIGN16_BEG float ALIGN16_END C0[4] =
{3.1157899f, 3.1157899f, 3.1157899f, 3.1157899f};
const __m128 pol5_y_0 = _mm_mul_ps(y, *((__m128 *)C5));
const __m128 pol5_y_1 = _mm_add_ps(pol5_y_0, *((__m128 *)C4));
const __m128 pol5_y_2 = _mm_mul_ps(pol5_y_1, y);
const __m128 pol5_y_3 = _mm_add_ps(pol5_y_2, *((__m128 *)C3));
const __m128 pol5_y_4 = _mm_mul_ps(pol5_y_3, y);
const __m128 pol5_y_5 = _mm_add_ps(pol5_y_4, *((__m128 *)C2));
const __m128 pol5_y_6 = _mm_mul_ps(pol5_y_5, y);
const __m128 pol5_y_7 = _mm_add_ps(pol5_y_6, *((__m128 *)C1));
const __m128 pol5_y_8 = _mm_mul_ps(pol5_y_7, y);
const __m128 pol5_y = _mm_add_ps(pol5_y_8, *((__m128 *)C0));
const __m128 y_minus_one = _mm_sub_ps(
y, *((__m128 *)zero_biased_exponent_is_one));
const __m128 log2_y = _mm_mul_ps(y_minus_one , pol5_y);
// Combine parts.
log2_a = _mm_add_ps(n, log2_y);
}
// b * log2(a)
b_log2_a = _mm_mul_ps(b, log2_a);
// Calculate exp2(x), x = b * log2(a).
{
// To calculate 2^x, we decompose x like this:
// x = n + y
// n is an integer, the value of x - 0.5 rounded down, therefore
// y is in the [0.5, 1.5) range
//
// 2^x = 2^n * 2^y
// 2^n can be evaluated by playing with float representation.
// 2^y in a small range can be approximated, this code uses an order two
// polynomial approximation. The coefficients have been estimated
// with the Remez algorithm and the resulting polynomial has a
// maximum relative error of 0.17%.
// To avoid over/underflow, we reduce the range of input to ]-127, 129].
static const ALIGN16_BEG float max_input[4] ALIGN16_END =
{129.f, 129.f, 129.f, 129.f};
static const ALIGN16_BEG float min_input[4] ALIGN16_END =
{-126.99999f, -126.99999f, -126.99999f, -126.99999f};
const __m128 x_min = _mm_min_ps(b_log2_a, *((__m128 *)max_input));
const __m128 x_max = _mm_max_ps(x_min, *((__m128 *)min_input));
// Compute n.
static const ALIGN16_BEG float half[4] ALIGN16_END =
{0.5f, 0.5f, 0.5f, 0.5f};
const __m128 x_minus_half = _mm_sub_ps(x_max, *((__m128 *)half));
const __m128i x_minus_half_floor = _mm_cvtps_epi32(x_minus_half);
// Compute 2^n.
static const ALIGN16_BEG int float_exponent_bias[4] ALIGN16_END =
{127, 127, 127, 127};
static const int float_exponent_shift = 23;
const __m128i two_n_exponent = _mm_add_epi32(
x_minus_half_floor, *((__m128i *)float_exponent_bias));
const __m128 two_n = _mm_castsi128_ps(_mm_slli_epi32(
two_n_exponent, float_exponent_shift));
// Compute y.
const __m128 y = _mm_sub_ps(x_max, _mm_cvtepi32_ps(x_minus_half_floor));
// Approximate 2^y ~= C2 * y^2 + C1 * y + C0.
static const ALIGN16_BEG float C2[4] ALIGN16_END =
{3.3718944e-1f, 3.3718944e-1f, 3.3718944e-1f, 3.3718944e-1f};
static const ALIGN16_BEG float C1[4] ALIGN16_END =
{6.5763628e-1f, 6.5763628e-1f, 6.5763628e-1f, 6.5763628e-1f};
static const ALIGN16_BEG float C0[4] ALIGN16_END =
{1.0017247f, 1.0017247f, 1.0017247f, 1.0017247f};
const __m128 exp2_y_0 = _mm_mul_ps(y, *((__m128 *)C2));
const __m128 exp2_y_1 = _mm_add_ps(exp2_y_0, *((__m128 *)C1));
const __m128 exp2_y_2 = _mm_mul_ps(exp2_y_1, y);
const __m128 exp2_y = _mm_add_ps(exp2_y_2, *((__m128 *)C0));
// Combine parts.
a_exp_b = _mm_mul_ps(exp2_y, two_n);
}
return a_exp_b;
}
extern const float WebRtcAec_weightCurve[65];
extern const float WebRtcAec_overDriveCurve[65];
static void OverdriveAndSuppressSSE2(aec_t *aec, float hNl[PART_LEN1],
const float hNlFb,
float efw[2][PART_LEN1]) {
int i;
const __m128 vec_hNlFb = _mm_set1_ps(hNlFb);
const __m128 vec_one = _mm_set1_ps(1.0f);
const __m128 vec_minus_one = _mm_set1_ps(-1.0f);
const __m128 vec_overDriveSm = _mm_set1_ps(aec->overDriveSm);
// vectorized code (four at once)
for (i = 0; i + 3 < PART_LEN1; i+=4) {
// Weight subbands
__m128 vec_hNl = _mm_loadu_ps(&hNl[i]);
const __m128 vec_weightCurve = _mm_loadu_ps(&WebRtcAec_weightCurve[i]);
const __m128 bigger = _mm_cmpgt_ps(vec_hNl, vec_hNlFb);
const __m128 vec_weightCurve_hNlFb = _mm_mul_ps(
vec_weightCurve, vec_hNlFb);
const __m128 vec_one_weightCurve = _mm_sub_ps(vec_one, vec_weightCurve);
const __m128 vec_one_weightCurve_hNl = _mm_mul_ps(
vec_one_weightCurve, vec_hNl);
const __m128 vec_if0 = _mm_andnot_ps(bigger, vec_hNl);
const __m128 vec_if1 = _mm_and_ps(
bigger, _mm_add_ps(vec_weightCurve_hNlFb, vec_one_weightCurve_hNl));
vec_hNl = _mm_or_ps(vec_if0, vec_if1);
{
const __m128 vec_overDriveCurve = _mm_loadu_ps(
&WebRtcAec_overDriveCurve[i]);
const __m128 vec_overDriveSm_overDriveCurve = _mm_mul_ps(
vec_overDriveSm, vec_overDriveCurve);
vec_hNl = mm_pow_ps(vec_hNl, vec_overDriveSm_overDriveCurve);
_mm_storeu_ps(&hNl[i], vec_hNl);
}
// Suppress error signal
{
__m128 vec_efw_re = _mm_loadu_ps(&efw[0][i]);
__m128 vec_efw_im = _mm_loadu_ps(&efw[1][i]);
vec_efw_re = _mm_mul_ps(vec_efw_re, vec_hNl);
vec_efw_im = _mm_mul_ps(vec_efw_im, vec_hNl);
// Ooura fft returns incorrect sign on imaginary component. It matters
// here because we are making an additive change with comfort noise.
vec_efw_im = _mm_mul_ps(vec_efw_im, vec_minus_one);
_mm_storeu_ps(&efw[0][i], vec_efw_re);
_mm_storeu_ps(&efw[1][i], vec_efw_im);
}
}
// scalar code for the remaining items.
for (; i < PART_LEN1; i++) {
// Weight subbands
if (hNl[i] > hNlFb) {
hNl[i] = WebRtcAec_weightCurve[i] * hNlFb +
(1 - WebRtcAec_weightCurve[i]) * hNl[i];
}
hNl[i] = powf(hNl[i], aec->overDriveSm * WebRtcAec_overDriveCurve[i]);
// Suppress error signal
efw[0][i] *= hNl[i];
efw[1][i] *= hNl[i];
// Ooura fft returns incorrect sign on imaginary component. It matters
// here because we are making an additive change with comfort noise.
efw[1][i] *= -1;
}
}
void WebRtcAec_InitAec_SSE2(void) {
WebRtcAec_FilterFar = FilterFarSSE2;
WebRtcAec_ScaleErrorSignal = ScaleErrorSignalSSE2;
WebRtcAec_FilterAdaptation = FilterAdaptationSSE2;
WebRtcAec_OverdriveAndSuppress = OverdriveAndSuppressSSE2;
}
#endif // WEBRTC_USE_SSE2

587
webrtc/modules/audio_processing/aec/aec_rdft.c Normal file
View File

@ -0,0 +1,587 @@
/*
* http://www.kurims.kyoto-u.ac.jp/~ooura/fft.html
* Copyright Takuya OOURA, 1996-2001
*
* You may use, copy, modify and distribute this code for any purpose (include
* commercial use) and without fee. Please refer to this package when you modify
* this code.
*
* Changes by the WebRTC authors:
* - Trivial type modifications.
* - Minimal code subset to do rdft of length 128.
* - Optimizations because of known length.
*
* All changes are covered by the WebRTC license and IP grant:
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "aec_rdft.h"
#include <math.h>
#include "system_wrappers/interface/cpu_features_wrapper.h"
#include "typedefs.h"
// constants shared by all paths (C, SSE2).
float rdft_w[64];
// constants used by the C path.
float rdft_wk3ri_first[32];
float rdft_wk3ri_second[32];
// constants used by SSE2 but initialized in C path.
ALIGN16_BEG float ALIGN16_END rdft_wk1r[32];
ALIGN16_BEG float ALIGN16_END rdft_wk2r[32];
ALIGN16_BEG float ALIGN16_END rdft_wk3r[32];
ALIGN16_BEG float ALIGN16_END rdft_wk1i[32];
ALIGN16_BEG float ALIGN16_END rdft_wk2i[32];
ALIGN16_BEG float ALIGN16_END rdft_wk3i[32];
ALIGN16_BEG float ALIGN16_END cftmdl_wk1r[4];
static int ip[16];
static void bitrv2_32or128(int n, int *ip, float *a) {
// n is 32 or 128
int j, j1, k, k1, m, m2;
float xr, xi, yr, yi;
ip[0] = 0;
{
int l = n;
m = 1;
while ((m << 3) < l) {
l >>= 1;
for (j = 0; j < m; j++) {
ip[m + j] = ip[j] + l;
}
m <<= 1;
}
}
m2 = 2 * m;
for (k = 0; k < m; k++) {
for (j = 0; j < k; j++) {
j1 = 2 * j + ip[k];
k1 = 2 * k + ip[j];
xr = a[j1];
xi = a[j1 + 1];
yr = a[k1];
yi = a[k1 + 1];
a[j1] = yr;
a[j1 + 1] = yi;
a[k1] = xr;
a[k1 + 1] = xi;
j1 += m2;
k1 += 2 * m2;
xr = a[j1];
xi = a[j1 + 1];
yr = a[k1];
yi = a[k1 + 1];
a[j1] = yr;
a[j1 + 1] = yi;
a[k1] = xr;
a[k1 + 1] = xi;
j1 += m2;
k1 -= m2;
xr = a[j1];
xi = a[j1 + 1];
yr = a[k1];
yi = a[k1 + 1];
a[j1] = yr;
a[j1 + 1] = yi;
a[k1] = xr;
a[k1 + 1] = xi;
j1 += m2;
k1 += 2 * m2;
xr = a[j1];
xi = a[j1 + 1];
yr = a[k1];
yi = a[k1 + 1];
a[j1] = yr;
a[j1 + 1] = yi;
a[k1] = xr;
a[k1 + 1] = xi;
}
j1 = 2 * k + m2 + ip[k];
k1 = j1 + m2;
xr = a[j1];
xi = a[j1 + 1];
yr = a[k1];
yi = a[k1 + 1];
a[j1] = yr;
a[j1 + 1] = yi;
a[k1] = xr;
a[k1 + 1] = xi;
}
}
static void makewt_32(void) {
const int nw = 32;
int j, nwh;
float delta, x, y;
ip[0] = nw;
ip[1] = 1;
nwh = nw >> 1;
delta = atanf(1.0f) / nwh;
rdft_w[0] = 1;
rdft_w[1] = 0;
rdft_w[nwh] = cosf(delta * nwh);
rdft_w[nwh + 1] = rdft_w[nwh];
for (j = 2; j < nwh; j += 2) {
x = cosf(delta * j);
y = sinf(delta * j);
rdft_w[j] = x;
rdft_w[j + 1] = y;
rdft_w[nw - j] = y;
rdft_w[nw - j + 1] = x;
}
bitrv2_32or128(nw, ip + 2, rdft_w);
// pre-calculate constants used by cft1st_128 and cftmdl_128...
cftmdl_wk1r[0] = rdft_w[2];
cftmdl_wk1r[1] = rdft_w[2];
cftmdl_wk1r[2] = rdft_w[2];
cftmdl_wk1r[3] = -rdft_w[2];
{
int k1;
for (k1 = 0, j = 0; j < 128; j += 16, k1 += 2) {
const int k2 = 2 * k1;
const float wk2r = rdft_w[k1 + 0];
const float wk2i = rdft_w[k1 + 1];
float wk1r, wk1i;
// ... scalar version.
wk1r = rdft_w[k2 + 0];
wk1i = rdft_w[k2 + 1];
rdft_wk3ri_first[k1 + 0] = wk1r - 2 * wk2i * wk1i;
rdft_wk3ri_first[k1 + 1] = 2 * wk2i * wk1r - wk1i;
wk1r = rdft_w[k2 + 2];
wk1i = rdft_w[k2 + 3];
rdft_wk3ri_second[k1 + 0] = wk1r - 2 * wk2r * wk1i;
rdft_wk3ri_second[k1 + 1] = 2 * wk2r * wk1r - wk1i;
// ... vector version.
rdft_wk1r[k2 + 0] = rdft_w[k2 + 0];
rdft_wk1r[k2 + 1] = rdft_w[k2 + 0];
rdft_wk1r[k2 + 2] = rdft_w[k2 + 2];
rdft_wk1r[k2 + 3] = rdft_w[k2 + 2];
rdft_wk2r[k2 + 0] = rdft_w[k1 + 0];
rdft_wk2r[k2 + 1] = rdft_w[k1 + 0];
rdft_wk2r[k2 + 2] = -rdft_w[k1 + 1];
rdft_wk2r[k2 + 3] = -rdft_w[k1 + 1];
rdft_wk3r[k2 + 0] = rdft_wk3ri_first[k1 + 0];
rdft_wk3r[k2 + 1] = rdft_wk3ri_first[k1 + 0];
rdft_wk3r[k2 + 2] = rdft_wk3ri_second[k1 + 0];
rdft_wk3r[k2 + 3] = rdft_wk3ri_second[k1 + 0];
rdft_wk1i[k2 + 0] = -rdft_w[k2 + 1];
rdft_wk1i[k2 + 1] = rdft_w[k2 + 1];
rdft_wk1i[k2 + 2] = -rdft_w[k2 + 3];
rdft_wk1i[k2 + 3] = rdft_w[k2 + 3];
rdft_wk2i[k2 + 0] = -rdft_w[k1 + 1];
rdft_wk2i[k2 + 1] = rdft_w[k1 + 1];
rdft_wk2i[k2 + 2] = -rdft_w[k1 + 0];
rdft_wk2i[k2 + 3] = rdft_w[k1 + 0];
rdft_wk3i[k2 + 0] = -rdft_wk3ri_first[k1 + 1];
rdft_wk3i[k2 + 1] = rdft_wk3ri_first[k1 + 1];
rdft_wk3i[k2 + 2] = -rdft_wk3ri_second[k1 + 1];
rdft_wk3i[k2 + 3] = rdft_wk3ri_second[k1 + 1];
}
}
}
static void makect_32(void) {
float *c = rdft_w + 32;
const int nc = 32;
int j, nch;
float delta;
ip[1] = nc;
nch = nc >> 1;
delta = atanf(1.0f) / nch;
c[0] = cosf(delta * nch);
c[nch] = 0.5f * c[0];
for (j = 1; j < nch; j++) {
c[j] = 0.5f * cosf(delta * j);
c[nc - j] = 0.5f * sinf(delta * j);
}
}
static void cft1st_128_C(float *a) {
const int n = 128;
int j, k1, k2;
float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;
float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
x0r = a[0] + a[2];
x0i = a[1] + a[3];
x1r = a[0] - a[2];
x1i = a[1] - a[3];
x2r = a[4] + a[6];
x2i = a[5] + a[7];
x3r = a[4] - a[6];
x3i = a[5] - a[7];
a[0] = x0r + x2r;
a[1] = x0i + x2i;
a[4] = x0r - x2r;
a[5] = x0i - x2i;
a[2] = x1r - x3i;
a[3] = x1i + x3r;
a[6] = x1r + x3i;
a[7] = x1i - x3r;
wk1r = rdft_w[2];
x0r = a[8] + a[10];
x0i = a[9] + a[11];
x1r = a[8] - a[10];
x1i = a[9] - a[11];
x2r = a[12] + a[14];
x2i = a[13] + a[15];
x3r = a[12] - a[14];
x3i = a[13] - a[15];
a[8] = x0r + x2r;
a[9] = x0i + x2i;
a[12] = x2i - x0i;
a[13] = x0r - x2r;
x0r = x1r - x3i;
x0i = x1i + x3r;
a[10] = wk1r * (x0r - x0i);
a[11] = wk1r * (x0r + x0i);
x0r = x3i + x1r;
x0i = x3r - x1i;
a[14] = wk1r * (x0i - x0r);
a[15] = wk1r * (x0i + x0r);
k1 = 0;
for (j = 16; j < n; j += 16) {
k1 += 2;
k2 = 2 * k1;
wk2r = rdft_w[k1 + 0];
wk2i = rdft_w[k1 + 1];
wk1r = rdft_w[k2 + 0];
wk1i = rdft_w[k2 + 1];
wk3r = rdft_wk3ri_first[k1 + 0];
wk3i = rdft_wk3ri_first[k1 + 1];
x0r = a[j + 0] + a[j + 2];
x0i = a[j + 1] + a[j + 3];
x1r = a[j + 0] - a[j + 2];
x1i = a[j + 1] - a[j + 3];
x2r = a[j + 4] + a[j + 6];
x2i = a[j + 5] + a[j + 7];
x3r = a[j + 4] - a[j + 6];
x3i = a[j + 5] - a[j + 7];
a[j + 0] = x0r + x2r;
a[j + 1] = x0i + x2i;
x0r -= x2r;
x0i -= x2i;
a[j + 4] = wk2r * x0r - wk2i * x0i;
a[j + 5] = wk2r * x0i + wk2i * x0r;
x0r = x1r - x3i;
x0i = x1i + x3r;
a[j + 2] = wk1r * x0r - wk1i * x0i;
a[j + 3] = wk1r * x0i + wk1i * x0r;
x0r = x1r + x3i;
x0i = x1i - x3r;
a[j + 6] = wk3r * x0r - wk3i * x0i;
a[j + 7] = wk3r * x0i + wk3i * x0r;
wk1r = rdft_w[k2 + 2];
wk1i = rdft_w[k2 + 3];
wk3r = rdft_wk3ri_second[k1 + 0];
wk3i = rdft_wk3ri_second[k1 + 1];
x0r = a[j + 8] + a[j + 10];
x0i = a[j + 9] + a[j + 11];
x1r = a[j + 8] - a[j + 10];
x1i = a[j + 9] - a[j + 11];
x2r = a[j + 12] + a[j + 14];
x2i = a[j + 13] + a[j + 15];
x3r = a[j + 12] - a[j + 14];
x3i = a[j + 13] - a[j + 15];
a[j + 8] = x0r + x2r;
a[j + 9] = x0i + x2i;
x0r -= x2r;
x0i -= x2i;
a[j + 12] = -wk2i * x0r - wk2r * x0i;
a[j + 13] = -wk2i * x0i + wk2r * x0r;
x0r = x1r - x3i;
x0i = x1i + x3r;
a[j + 10] = wk1r * x0r - wk1i * x0i;
a[j + 11] = wk1r * x0i + wk1i * x0r;
x0r = x1r + x3i;
x0i = x1i - x3r;
a[j + 14] = wk3r * x0r - wk3i * x0i;
a[j + 15] = wk3r * x0i + wk3i * x0r;
}
}
static void cftmdl_128_C(float *a) {
const int l = 8;
const int n = 128;
const int m = 32;
int j0, j1, j2, j3, k, k1, k2, m2;
float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;
float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
for (j0 = 0; j0 < l; j0 += 2) {
j1 = j0 + 8;
j2 = j0 + 16;
j3 = j0 + 24;
x0r = a[j0 + 0] + a[j1 + 0];
x0i = a[j0 + 1] + a[j1 + 1];
x1r = a[j0 + 0] - a[j1 + 0];
x1i = a[j0 + 1] - a[j1 + 1];
x2r = a[j2 + 0] + a[j3 + 0];
x2i = a[j2 + 1] + a[j3 + 1];
x3r = a[j2 + 0] - a[j3 + 0];
x3i = a[j2 + 1] - a[j3 + 1];
a[j0 + 0] = x0r + x2r;
a[j0 + 1] = x0i + x2i;
a[j2 + 0] = x0r - x2r;
a[j2 + 1] = x0i - x2i;
a[j1 + 0] = x1r - x3i;
a[j1 + 1] = x1i + x3r;
a[j3 + 0] = x1r + x3i;
a[j3 + 1] = x1i - x3r;
}
wk1r = rdft_w[2];
for (j0 = m; j0 < l + m; j0 += 2) {
j1 = j0 + 8;
j2 = j0 + 16;
j3 = j0 + 24;
x0r = a[j0 + 0] + a[j1 + 0];
x0i = a[j0 + 1] + a[j1 + 1];
x1r = a[j0 + 0] - a[j1 + 0];
x1i = a[j0 + 1] - a[j1 + 1];
x2r = a[j2 + 0] + a[j3 + 0];
x2i = a[j2 + 1] + a[j3 + 1];
x3r = a[j2 + 0] - a[j3 + 0];
x3i = a[j2 + 1] - a[j3 + 1];
a[j0 + 0] = x0r + x2r;
a[j0 + 1] = x0i + x2i;
a[j2 + 0] = x2i - x0i;
a[j2 + 1] = x0r - x2r;
x0r = x1r - x3i;
x0i = x1i + x3r;
a[j1 + 0] = wk1r * (x0r - x0i);
a[j1 + 1] = wk1r * (x0r + x0i);
x0r = x3i + x1r;
x0i = x3r - x1i;
a[j3 + 0] = wk1r * (x0i - x0r);
a[j3 + 1] = wk1r * (x0i + x0r);
}
k1 = 0;
m2 = 2 * m;
for (k = m2; k < n; k += m2) {
k1 += 2;
k2 = 2 * k1;
wk2r = rdft_w[k1 + 0];
wk2i = rdft_w[k1 + 1];
wk1r = rdft_w[k2 + 0];
wk1i = rdft_w[k2 + 1];
wk3r = rdft_wk3ri_first[k1 + 0];
wk3i = rdft_wk3ri_first[k1 + 1];
for (j0 = k; j0 < l + k; j0 += 2) {
j1 = j0 + 8;
j2 = j0 + 16;
j3 = j0 + 24;
x0r = a[j0 + 0] + a[j1 + 0];
x0i = a[j0 + 1] + a[j1 + 1];
x1r = a[j0 + 0] - a[j1 + 0];
x1i = a[j0 + 1] - a[j1 + 1];
x2r = a[j2 + 0] + a[j3 + 0];
x2i = a[j2 + 1] + a[j3 + 1];
x3r = a[j2 + 0] - a[j3 + 0];
x3i = a[j2 + 1] - a[j3 + 1];
a[j0 + 0] = x0r + x2r;
a[j0 + 1] = x0i + x2i;
x0r -= x2r;
x0i -= x2i;
a[j2 + 0] = wk2r * x0r - wk2i * x0i;
a[j2 + 1] = wk2r * x0i + wk2i * x0r;
x0r = x1r - x3i;
x0i = x1i + x3r;
a[j1 + 0] = wk1r * x0r - wk1i * x0i;
a[j1 + 1] = wk1r * x0i + wk1i * x0r;
x0r = x1r + x3i;
x0i = x1i - x3r;
a[j3 + 0] = wk3r * x0r - wk3i * x0i;
a[j3 + 1] = wk3r * x0i + wk3i * x0r;
}
wk1r = rdft_w[k2 + 2];
wk1i = rdft_w[k2 + 3];
wk3r = rdft_wk3ri_second[k1 + 0];
wk3i = rdft_wk3ri_second[k1 + 1];
for (j0 = k + m; j0 < l + (k + m); j0 += 2) {
j1 = j0 + 8;
j2 = j0 + 16;
j3 = j0 + 24;
x0r = a[j0 + 0] + a[j1 + 0];
x0i = a[j0 + 1] + a[j1 + 1];
x1r = a[j0 + 0] - a[j1 + 0];
x1i = a[j0 + 1] - a[j1 + 1];
x2r = a[j2 + 0] + a[j3 + 0];
x2i = a[j2 + 1] + a[j3 + 1];
x3r = a[j2 + 0] - a[j3 + 0];
x3i = a[j2 + 1] - a[j3 + 1];
a[j0 + 0] = x0r + x2r;
a[j0 + 1] = x0i + x2i;
x0r -= x2r;
x0i -= x2i;
a[j2 + 0] = -wk2i * x0r - wk2r * x0i;
a[j2 + 1] = -wk2i * x0i + wk2r * x0r;
x0r = x1r - x3i;
x0i = x1i + x3r;
a[j1 + 0] = wk1r * x0r - wk1i * x0i;
a[j1 + 1] = wk1r * x0i + wk1i * x0r;
x0r = x1r + x3i;
x0i = x1i - x3r;
a[j3 + 0] = wk3r * x0r - wk3i * x0i;
a[j3 + 1] = wk3r * x0i + wk3i * x0r;
}
}
}
static void cftfsub_128(float *a) {
int j, j1, j2, j3, l;
float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
cft1st_128(a);
cftmdl_128(a);
l = 32;
for (j = 0; j < l; j += 2) {
j1 = j + l;
j2 = j1 + l;
j3 = j2 + l;
x0r = a[j] + a[j1];
x0i = a[j + 1] + a[j1 + 1];
x1r = a[j] - a[j1];
x1i = a[j + 1] - a[j1 + 1];
x2r = a[j2] + a[j3];
x2i = a[j2 + 1] + a[j3 + 1];
x3r = a[j2] - a[j3];
x3i = a[j2 + 1] - a[j3 + 1];
a[j] = x0r + x2r;
a[j + 1] = x0i + x2i;
a[j2] = x0r - x2r;
a[j2 + 1] = x0i - x2i;
a[j1] = x1r - x3i;
a[j1 + 1] = x1i + x3r;
a[j3] = x1r + x3i;
a[j3 + 1] = x1i - x3r;
}
}
static void cftbsub_128(float *a) {
int j, j1, j2, j3, l;
float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
cft1st_128(a);
cftmdl_128(a);
l = 32;
for (j = 0; j < l; j += 2) {
j1 = j + l;
j2 = j1 + l;
j3 = j2 + l;
x0r = a[j] + a[j1];
x0i = -a[j + 1] - a[j1 + 1];
x1r = a[j] - a[j1];
x1i = -a[j + 1] + a[j1 + 1];
x2r = a[j2] + a[j3];
x2i = a[j2 + 1] + a[j3 + 1];
x3r = a[j2] - a[j3];
x3i = a[j2 + 1] - a[j3 + 1];
a[j] = x0r + x2r;
a[j + 1] = x0i - x2i;
a[j2] = x0r - x2r;
a[j2 + 1] = x0i + x2i;
a[j1] = x1r - x3i;
a[j1 + 1] = x1i - x3r;
a[j3] = x1r + x3i;
a[j3 + 1] = x1i + x3r;
}
}
static void rftfsub_128_C(float *a) {
const float *c = rdft_w + 32;
int j1, j2, k1, k2;
float wkr, wki, xr, xi, yr, yi;
for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {
k2 = 128 - j2;
k1 = 32 - j1;
wkr = 0.5f - c[k1];
wki = c[j1];
xr = a[j2 + 0] - a[k2 + 0];
xi = a[j2 + 1] + a[k2 + 1];
yr = wkr * xr - wki * xi;
yi = wkr * xi + wki * xr;
a[j2 + 0] -= yr;
a[j2 + 1] -= yi;
a[k2 + 0] += yr;
a[k2 + 1] -= yi;
}
}
static void rftbsub_128_C(float *a) {
const float *c = rdft_w + 32;
int j1, j2, k1, k2;
float wkr, wki, xr, xi, yr, yi;
a[1] = -a[1];
for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {
k2 = 128 - j2;
k1 = 32 - j1;
wkr = 0.5f - c[k1];
wki = c[j1];
xr = a[j2 + 0] - a[k2 + 0];
xi = a[j2 + 1] + a[k2 + 1];
yr = wkr * xr + wki * xi;
yi = wkr * xi - wki * xr;
a[j2 + 0] = a[j2 + 0] - yr;
a[j2 + 1] = yi - a[j2 + 1];
a[k2 + 0] = yr + a[k2 + 0];
a[k2 + 1] = yi - a[k2 + 1];
}
a[65] = -a[65];
}
void aec_rdft_forward_128(float *a) {
const int n = 128;
float xi;
bitrv2_32or128(n, ip + 2, a);
cftfsub_128(a);
rftfsub_128(a);
xi = a[0] - a[1];
a[0] += a[1];
a[1] = xi;
}
void aec_rdft_inverse_128(float *a) {
const int n = 128;
a[1] = 0.5f * (a[0] - a[1]);
a[0] -= a[1];
rftbsub_128(a);
bitrv2_32or128(n, ip + 2, a);
cftbsub_128(a);
}
// code path selection
rft_sub_128_t cft1st_128;
rft_sub_128_t cftmdl_128;
rft_sub_128_t rftfsub_128;
rft_sub_128_t rftbsub_128;
void aec_rdft_init(void) {
cft1st_128 = cft1st_128_C;
cftmdl_128 = cftmdl_128_C;
rftfsub_128 = rftfsub_128_C;
rftbsub_128 = rftbsub_128_C;
if (WebRtc_GetCPUInfo(kSSE2)) {
#if defined(WEBRTC_USE_SSE2)
aec_rdft_init_sse2();
#endif
}
// init library constants.
makewt_32();
makect_32();
}

57
webrtc/modules/audio_processing/aec/aec_rdft.h Normal file
View File

@ -0,0 +1,57 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_
// These intrinsics were unavailable before VS 2008.
// TODO(andrew): move to a common file.
#if defined(_MSC_VER) && _MSC_VER < 1500
#include <emmintrin.h>
static __inline __m128 _mm_castsi128_ps(__m128i a) { return *(__m128*)&a; }
static __inline __m128i _mm_castps_si128(__m128 a) { return *(__m128i*)&a; }
#endif
#ifdef _MSC_VER /* visual c++ */
# define ALIGN16_BEG __declspec(align(16))
# define ALIGN16_END
#else /* gcc or icc */
# define ALIGN16_BEG
# define ALIGN16_END __attribute__((aligned(16)))
#endif
// constants shared by all paths (C, SSE2).
extern float rdft_w[64];
// constants used by the C path.
extern float rdft_wk3ri_first[32];
extern float rdft_wk3ri_second[32];
// constants used by SSE2 but initialized in C path.
extern float rdft_wk1r[32];
extern float rdft_wk2r[32];
extern float rdft_wk3r[32];
extern float rdft_wk1i[32];
extern float rdft_wk2i[32];
extern float rdft_wk3i[32];
extern float cftmdl_wk1r[4];
// code path selection function pointers
typedef void (*rft_sub_128_t)(float *a);
extern rft_sub_128_t rftfsub_128;
extern rft_sub_128_t rftbsub_128;
extern rft_sub_128_t cft1st_128;
extern rft_sub_128_t cftmdl_128;
// entry points
void aec_rdft_init(void);
void aec_rdft_init_sse2(void);
void aec_rdft_forward_128(float *a);
void aec_rdft_inverse_128(float *a);
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_

431
webrtc/modules/audio_processing/aec/aec_rdft_sse2.c Normal file
View File

@ -0,0 +1,431 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "typedefs.h"
#if defined(WEBRTC_USE_SSE2)
#include <emmintrin.h>
#include "aec_rdft.h"
static const ALIGN16_BEG float ALIGN16_END k_swap_sign[4] =
{-1.f, 1.f, -1.f, 1.f};
static void cft1st_128_SSE2(float *a) {
const __m128 mm_swap_sign = _mm_load_ps(k_swap_sign);
int j, k2;
for (k2 = 0, j = 0; j < 128; j += 16, k2 += 4) {
__m128 a00v = _mm_loadu_ps(&a[j + 0]);
__m128 a04v = _mm_loadu_ps(&a[j + 4]);
__m128 a08v = _mm_loadu_ps(&a[j + 8]);
__m128 a12v = _mm_loadu_ps(&a[j + 12]);
__m128 a01v = _mm_shuffle_ps(a00v, a08v, _MM_SHUFFLE(1, 0, 1 ,0));
__m128 a23v = _mm_shuffle_ps(a00v, a08v, _MM_SHUFFLE(3, 2, 3 ,2));
__m128 a45v = _mm_shuffle_ps(a04v, a12v, _MM_SHUFFLE(1, 0, 1 ,0));
__m128 a67v = _mm_shuffle_ps(a04v, a12v, _MM_SHUFFLE(3, 2, 3 ,2));
const __m128 wk1rv = _mm_load_ps(&rdft_wk1r[k2]);
const __m128 wk1iv = _mm_load_ps(&rdft_wk1i[k2]);
const __m128 wk2rv = _mm_load_ps(&rdft_wk2r[k2]);
const __m128 wk2iv = _mm_load_ps(&rdft_wk2i[k2]);
const __m128 wk3rv = _mm_load_ps(&rdft_wk3r[k2]);
const __m128 wk3iv = _mm_load_ps(&rdft_wk3i[k2]);
__m128 x0v = _mm_add_ps(a01v, a23v);
const __m128 x1v = _mm_sub_ps(a01v, a23v);
const __m128 x2v = _mm_add_ps(a45v, a67v);
const __m128 x3v = _mm_sub_ps(a45v, a67v);
__m128 x0w;
a01v = _mm_add_ps(x0v, x2v);
x0v = _mm_sub_ps(x0v, x2v);
x0w = _mm_shuffle_ps(x0v, x0v, _MM_SHUFFLE(2, 3, 0 ,1));
{
const __m128 a45_0v = _mm_mul_ps(wk2rv, x0v);
const __m128 a45_1v = _mm_mul_ps(wk2iv, x0w);
a45v = _mm_add_ps(a45_0v, a45_1v);
}
{
__m128 a23_0v, a23_1v;
const __m128 x3w = _mm_shuffle_ps(x3v, x3v, _MM_SHUFFLE(2, 3, 0 ,1));
const __m128 x3s = _mm_mul_ps(mm_swap_sign, x3w);
x0v = _mm_add_ps(x1v, x3s);
x0w = _mm_shuffle_ps(x0v, x0v, _MM_SHUFFLE(2, 3, 0 ,1));
a23_0v = _mm_mul_ps(wk1rv, x0v);
a23_1v = _mm_mul_ps(wk1iv, x0w);
a23v = _mm_add_ps(a23_0v, a23_1v);
x0v = _mm_sub_ps(x1v, x3s);
x0w = _mm_shuffle_ps(x0v, x0v, _MM_SHUFFLE(2, 3, 0 ,1));
}
{
const __m128 a67_0v = _mm_mul_ps(wk3rv, x0v);
const __m128 a67_1v = _mm_mul_ps(wk3iv, x0w);
a67v = _mm_add_ps(a67_0v, a67_1v);
}
a00v = _mm_shuffle_ps(a01v, a23v, _MM_SHUFFLE(1, 0, 1 ,0));
a04v = _mm_shuffle_ps(a45v, a67v, _MM_SHUFFLE(1, 0, 1 ,0));
a08v = _mm_shuffle_ps(a01v, a23v, _MM_SHUFFLE(3, 2, 3 ,2));
a12v = _mm_shuffle_ps(a45v, a67v, _MM_SHUFFLE(3, 2, 3 ,2));
_mm_storeu_ps(&a[j + 0], a00v);
_mm_storeu_ps(&a[j + 4], a04v);
_mm_storeu_ps(&a[j + 8], a08v);
_mm_storeu_ps(&a[j + 12], a12v);
}
}
static void cftmdl_128_SSE2(float *a) {
const int l = 8;
const __m128 mm_swap_sign = _mm_load_ps(k_swap_sign);
int j0;
__m128 wk1rv = _mm_load_ps(cftmdl_wk1r);
for (j0 = 0; j0 < l; j0 += 2) {
const __m128i a_00 = _mm_loadl_epi64((__m128i*)&a[j0 + 0]);
const __m128i a_08 = _mm_loadl_epi64((__m128i*)&a[j0 + 8]);
const __m128i a_32 = _mm_loadl_epi64((__m128i*)&a[j0 + 32]);
const __m128i a_40 = _mm_loadl_epi64((__m128i*)&a[j0 + 40]);
const __m128 a_00_32 = _mm_shuffle_ps(_mm_castsi128_ps(a_00),
_mm_castsi128_ps(a_32),
_MM_SHUFFLE(1, 0, 1 ,0));
const __m128 a_08_40 = _mm_shuffle_ps(_mm_castsi128_ps(a_08),
_mm_castsi128_ps(a_40),
_MM_SHUFFLE(1, 0, 1 ,0));
__m128 x0r0_0i0_0r1_x0i1 = _mm_add_ps(a_00_32, a_08_40);
const __m128 x1r0_1i0_1r1_x1i1 = _mm_sub_ps(a_00_32, a_08_40);
const __m128i a_16 = _mm_loadl_epi64((__m128i*)&a[j0 + 16]);
const __m128i a_24 = _mm_loadl_epi64((__m128i*)&a[j0 + 24]);
const __m128i a_48 = _mm_loadl_epi64((__m128i*)&a[j0 + 48]);
const __m128i a_56 = _mm_loadl_epi64((__m128i*)&a[j0 + 56]);
const __m128 a_16_48 = _mm_shuffle_ps(_mm_castsi128_ps(a_16),
_mm_castsi128_ps(a_48),
_MM_SHUFFLE(1, 0, 1 ,0));
const __m128 a_24_56 = _mm_shuffle_ps(_mm_castsi128_ps(a_24),
_mm_castsi128_ps(a_56),
_MM_SHUFFLE(1, 0, 1 ,0));
const __m128 x2r0_2i0_2r1_x2i1 = _mm_add_ps(a_16_48, a_24_56);
const __m128 x3r0_3i0_3r1_x3i1 = _mm_sub_ps(a_16_48, a_24_56);
const __m128 xx0 = _mm_add_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
const __m128 xx1 = _mm_sub_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
const __m128 x3i0_3r0_3i1_x3r1 = _mm_castsi128_ps(
_mm_shuffle_epi32(_mm_castps_si128(x3r0_3i0_3r1_x3i1),
_MM_SHUFFLE(2, 3, 0, 1)));
const __m128 x3_swapped = _mm_mul_ps(mm_swap_sign, x3i0_3r0_3i1_x3r1);
const __m128 x1_x3_add = _mm_add_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
const __m128 x1_x3_sub = _mm_sub_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
const __m128 yy0 = _mm_shuffle_ps(x1_x3_add, x1_x3_sub,
_MM_SHUFFLE(2, 2, 2 ,2));
const __m128 yy1 = _mm_shuffle_ps(x1_x3_add, x1_x3_sub,
_MM_SHUFFLE(3, 3, 3 ,3));
const __m128 yy2 = _mm_mul_ps(mm_swap_sign, yy1);
const __m128 yy3 = _mm_add_ps(yy0, yy2);
const __m128 yy4 = _mm_mul_ps(wk1rv, yy3);
_mm_storel_epi64((__m128i*)&a[j0 + 0], _mm_castps_si128(xx0));
_mm_storel_epi64((__m128i*)&a[j0 + 32],
_mm_shuffle_epi32(_mm_castps_si128(xx0),
_MM_SHUFFLE(3, 2, 3, 2)));
_mm_storel_epi64((__m128i*)&a[j0 + 16], _mm_castps_si128(xx1));
_mm_storel_epi64((__m128i*)&a[j0 + 48],
_mm_shuffle_epi32(_mm_castps_si128(xx1),
_MM_SHUFFLE(2, 3, 2, 3)));
a[j0 + 48] = -a[j0 + 48];
_mm_storel_epi64((__m128i*)&a[j0 + 8], _mm_castps_si128(x1_x3_add));
_mm_storel_epi64((__m128i*)&a[j0 + 24], _mm_castps_si128(x1_x3_sub));
_mm_storel_epi64((__m128i*)&a[j0 + 40], _mm_castps_si128(yy4));
_mm_storel_epi64((__m128i*)&a[j0 + 56],
_mm_shuffle_epi32(_mm_castps_si128(yy4),
_MM_SHUFFLE(2, 3, 2, 3)));
}
{
int k = 64;
int k1 = 2;
int k2 = 2 * k1;
const __m128 wk2rv = _mm_load_ps(&rdft_wk2r[k2+0]);
const __m128 wk2iv = _mm_load_ps(&rdft_wk2i[k2+0]);
const __m128 wk1iv = _mm_load_ps(&rdft_wk1i[k2+0]);
const __m128 wk3rv = _mm_load_ps(&rdft_wk3r[k2+0]);
const __m128 wk3iv = _mm_load_ps(&rdft_wk3i[k2+0]);
wk1rv = _mm_load_ps(&rdft_wk1r[k2+0]);
for (j0 = k; j0 < l + k; j0 += 2) {
const __m128i a_00 = _mm_loadl_epi64((__m128i*)&a[j0 + 0]);
const __m128i a_08 = _mm_loadl_epi64((__m128i*)&a[j0 + 8]);
const __m128i a_32 = _mm_loadl_epi64((__m128i*)&a[j0 + 32]);
const __m128i a_40 = _mm_loadl_epi64((__m128i*)&a[j0 + 40]);
const __m128 a_00_32 = _mm_shuffle_ps(_mm_castsi128_ps(a_00),
_mm_castsi128_ps(a_32),
_MM_SHUFFLE(1, 0, 1 ,0));
const __m128 a_08_40 = _mm_shuffle_ps(_mm_castsi128_ps(a_08),
_mm_castsi128_ps(a_40),
_MM_SHUFFLE(1, 0, 1 ,0));
__m128 x0r0_0i0_0r1_x0i1 = _mm_add_ps(a_00_32, a_08_40);
const __m128 x1r0_1i0_1r1_x1i1 = _mm_sub_ps(a_00_32, a_08_40);
const __m128i a_16 = _mm_loadl_epi64((__m128i*)&a[j0 + 16]);
const __m128i a_24 = _mm_loadl_epi64((__m128i*)&a[j0 + 24]);
const __m128i a_48 = _mm_loadl_epi64((__m128i*)&a[j0 + 48]);
const __m128i a_56 = _mm_loadl_epi64((__m128i*)&a[j0 + 56]);
const __m128 a_16_48 = _mm_shuffle_ps(_mm_castsi128_ps(a_16),
_mm_castsi128_ps(a_48),
_MM_SHUFFLE(1, 0, 1 ,0));
const __m128 a_24_56 = _mm_shuffle_ps(_mm_castsi128_ps(a_24),
_mm_castsi128_ps(a_56),
_MM_SHUFFLE(1, 0, 1 ,0));
const __m128 x2r0_2i0_2r1_x2i1 = _mm_add_ps(a_16_48, a_24_56);
const __m128 x3r0_3i0_3r1_x3i1 = _mm_sub_ps(a_16_48, a_24_56);
const __m128 xx = _mm_add_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
const __m128 xx1 = _mm_sub_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
const __m128 xx2 = _mm_mul_ps(xx1 , wk2rv);
const __m128 xx3 = _mm_mul_ps(wk2iv,
_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(xx1),
_MM_SHUFFLE(2, 3, 0, 1))));
const __m128 xx4 = _mm_add_ps(xx2, xx3);
const __m128 x3i0_3r0_3i1_x3r1 = _mm_castsi128_ps(
_mm_shuffle_epi32(_mm_castps_si128(x3r0_3i0_3r1_x3i1),
_MM_SHUFFLE(2, 3, 0, 1)));
const __m128 x3_swapped = _mm_mul_ps(mm_swap_sign, x3i0_3r0_3i1_x3r1);
const __m128 x1_x3_add = _mm_add_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
const __m128 x1_x3_sub = _mm_sub_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
const __m128 xx10 = _mm_mul_ps(x1_x3_add, wk1rv);
const __m128 xx11 = _mm_mul_ps(wk1iv,
_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(x1_x3_add),
_MM_SHUFFLE(2, 3, 0, 1))));
const __m128 xx12 = _mm_add_ps(xx10, xx11);
const __m128 xx20 = _mm_mul_ps(x1_x3_sub, wk3rv);
const __m128 xx21 = _mm_mul_ps(wk3iv,
_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(x1_x3_sub),
_MM_SHUFFLE(2, 3, 0, 1))));
const __m128 xx22 = _mm_add_ps(xx20, xx21);
_mm_storel_epi64((__m128i*)&a[j0 + 0], _mm_castps_si128(xx));
_mm_storel_epi64((__m128i*)&a[j0 + 32],
_mm_shuffle_epi32(_mm_castps_si128(xx),
_MM_SHUFFLE(3, 2, 3, 2)));
_mm_storel_epi64((__m128i*)&a[j0 + 16], _mm_castps_si128(xx4));
_mm_storel_epi64((__m128i*)&a[j0 + 48],
_mm_shuffle_epi32(_mm_castps_si128(xx4),
_MM_SHUFFLE(3, 2, 3, 2)));
_mm_storel_epi64((__m128i*)&a[j0 + 8], _mm_castps_si128(xx12));
_mm_storel_epi64((__m128i*)&a[j0 + 40],
_mm_shuffle_epi32(_mm_castps_si128(xx12),
_MM_SHUFFLE(3, 2, 3, 2)));
_mm_storel_epi64((__m128i*)&a[j0 + 24], _mm_castps_si128(xx22));
_mm_storel_epi64((__m128i*)&a[j0 + 56],
_mm_shuffle_epi32(_mm_castps_si128(xx22),
_MM_SHUFFLE(3, 2, 3, 2)));
}
}
}
static void rftfsub_128_SSE2(float *a) {
const float *c = rdft_w + 32;
int j1, j2, k1, k2;
float wkr, wki, xr, xi, yr, yi;
static const ALIGN16_BEG float ALIGN16_END k_half[4] =
{0.5f, 0.5f, 0.5f, 0.5f};
const __m128 mm_half = _mm_load_ps(k_half);
// Vectorized code (four at once).
// Note: commented number are indexes for the first iteration of the loop.
for (j1 = 1, j2 = 2; j2 + 7 < 64; j1 += 4, j2 += 8) {
// Load 'wk'.
const __m128 c_j1 = _mm_loadu_ps(&c[ j1]); // 1, 2, 3, 4,
const __m128 c_k1 = _mm_loadu_ps(&c[29 - j1]); // 28, 29, 30, 31,
const __m128 wkrt = _mm_sub_ps(mm_half, c_k1); // 28, 29, 30, 31,
const __m128 wkr_ =
_mm_shuffle_ps(wkrt, wkrt, _MM_SHUFFLE(0, 1, 2, 3)); // 31, 30, 29, 28,
const __m128 wki_ = c_j1; // 1, 2, 3, 4,
// Load and shuffle 'a'.
const __m128 a_j2_0 = _mm_loadu_ps(&a[0 + j2]); // 2, 3, 4, 5,
const __m128 a_j2_4 = _mm_loadu_ps(&a[4 + j2]); // 6, 7, 8, 9,
const __m128 a_k2_0 = _mm_loadu_ps(&a[122 - j2]); // 120, 121, 122, 123,
const __m128 a_k2_4 = _mm_loadu_ps(&a[126 - j2]); // 124, 125, 126, 127,
const __m128 a_j2_p0 = _mm_shuffle_ps(a_j2_0, a_j2_4,
_MM_SHUFFLE(2, 0, 2 ,0)); // 2, 4, 6, 8,
const __m128 a_j2_p1 = _mm_shuffle_ps(a_j2_0, a_j2_4,
_MM_SHUFFLE(3, 1, 3 ,1)); // 3, 5, 7, 9,
const __m128 a_k2_p0 = _mm_shuffle_ps(a_k2_4, a_k2_0,
_MM_SHUFFLE(0, 2, 0 ,2)); // 126, 124, 122, 120,
const __m128 a_k2_p1 = _mm_shuffle_ps(a_k2_4, a_k2_0,
_MM_SHUFFLE(1, 3, 1 ,3)); // 127, 125, 123, 121,
// Calculate 'x'.
const __m128 xr_ = _mm_sub_ps(a_j2_p0, a_k2_p0);
// 2-126, 4-124, 6-122, 8-120,
const __m128 xi_ = _mm_add_ps(a_j2_p1, a_k2_p1);
// 3-127, 5-125, 7-123, 9-121,
// Calculate product into 'y'.
// yr = wkr * xr - wki * xi;
// yi = wkr * xi + wki * xr;
const __m128 a_ = _mm_mul_ps(wkr_, xr_);
const __m128 b_ = _mm_mul_ps(wki_, xi_);
const __m128 c_ = _mm_mul_ps(wkr_, xi_);
const __m128 d_ = _mm_mul_ps(wki_, xr_);
const __m128 yr_ = _mm_sub_ps(a_, b_); // 2-126, 4-124, 6-122, 8-120,
const __m128 yi_ = _mm_add_ps(c_, d_); // 3-127, 5-125, 7-123, 9-121,
// Update 'a'.
// a[j2 + 0] -= yr;
// a[j2 + 1] -= yi;
// a[k2 + 0] += yr;
// a[k2 + 1] -= yi;
const __m128 a_j2_p0n = _mm_sub_ps(a_j2_p0, yr_); // 2, 4, 6, 8,
const __m128 a_j2_p1n = _mm_sub_ps(a_j2_p1, yi_); // 3, 5, 7, 9,
const __m128 a_k2_p0n = _mm_add_ps(a_k2_p0, yr_); // 126, 124, 122, 120,
const __m128 a_k2_p1n = _mm_sub_ps(a_k2_p1, yi_); // 127, 125, 123, 121,
// Shuffle in right order and store.
const __m128 a_j2_0n = _mm_unpacklo_ps(a_j2_p0n, a_j2_p1n);
// 2, 3, 4, 5,
const __m128 a_j2_4n = _mm_unpackhi_ps(a_j2_p0n, a_j2_p1n);
// 6, 7, 8, 9,
const __m128 a_k2_0nt = _mm_unpackhi_ps(a_k2_p0n, a_k2_p1n);
// 122, 123, 120, 121,
const __m128 a_k2_4nt = _mm_unpacklo_ps(a_k2_p0n, a_k2_p1n);
// 126, 127, 124, 125,
const __m128 a_k2_0n = _mm_shuffle_ps(a_k2_0nt, a_k2_0nt,
_MM_SHUFFLE(1, 0, 3 ,2)); // 120, 121, 122, 123,
const __m128 a_k2_4n = _mm_shuffle_ps(a_k2_4nt, a_k2_4nt,
_MM_SHUFFLE(1, 0, 3 ,2)); // 124, 125, 126, 127,
_mm_storeu_ps(&a[0 + j2], a_j2_0n);
_mm_storeu_ps(&a[4 + j2], a_j2_4n);
_mm_storeu_ps(&a[122 - j2], a_k2_0n);
_mm_storeu_ps(&a[126 - j2], a_k2_4n);
}
// Scalar code for the remaining items.
for (; j2 < 64; j1 += 1, j2 += 2) {
k2 = 128 - j2;
k1 = 32 - j1;
wkr = 0.5f - c[k1];
wki = c[j1];
xr = a[j2 + 0] - a[k2 + 0];
xi = a[j2 + 1] + a[k2 + 1];
yr = wkr * xr - wki * xi;
yi = wkr * xi + wki * xr;
a[j2 + 0] -= yr;
a[j2 + 1] -= yi;
a[k2 + 0] += yr;
a[k2 + 1] -= yi;
}
}
static void rftbsub_128_SSE2(float *a) {
const float *c = rdft_w + 32;
int j1, j2, k1, k2;
float wkr, wki, xr, xi, yr, yi;
static const ALIGN16_BEG float ALIGN16_END k_half[4] =
{0.5f, 0.5f, 0.5f, 0.5f};
const __m128 mm_half = _mm_load_ps(k_half);
a[1] = -a[1];
// Vectorized code (four at once).
// Note: commented number are indexes for the first iteration of the loop.
for (j1 = 1, j2 = 2; j2 + 7 < 64; j1 += 4, j2 += 8) {
// Load 'wk'.
const __m128 c_j1 = _mm_loadu_ps(&c[ j1]); // 1, 2, 3, 4,
const __m128 c_k1 = _mm_loadu_ps(&c[29 - j1]); // 28, 29, 30, 31,
const __m128 wkrt = _mm_sub_ps(mm_half, c_k1); // 28, 29, 30, 31,
const __m128 wkr_ =
_mm_shuffle_ps(wkrt, wkrt, _MM_SHUFFLE(0, 1, 2, 3)); // 31, 30, 29, 28,
const __m128 wki_ = c_j1; // 1, 2, 3, 4,
// Load and shuffle 'a'.
const __m128 a_j2_0 = _mm_loadu_ps(&a[0 + j2]); // 2, 3, 4, 5,
const __m128 a_j2_4 = _mm_loadu_ps(&a[4 + j2]); // 6, 7, 8, 9,
const __m128 a_k2_0 = _mm_loadu_ps(&a[122 - j2]); // 120, 121, 122, 123,
const __m128 a_k2_4 = _mm_loadu_ps(&a[126 - j2]); // 124, 125, 126, 127,
const __m128 a_j2_p0 = _mm_shuffle_ps(a_j2_0, a_j2_4,
_MM_SHUFFLE(2, 0, 2 ,0)); // 2, 4, 6, 8,
const __m128 a_j2_p1 = _mm_shuffle_ps(a_j2_0, a_j2_4,
_MM_SHUFFLE(3, 1, 3 ,1)); // 3, 5, 7, 9,
const __m128 a_k2_p0 = _mm_shuffle_ps(a_k2_4, a_k2_0,
_MM_SHUFFLE(0, 2, 0 ,2)); // 126, 124, 122, 120,
const __m128 a_k2_p1 = _mm_shuffle_ps(a_k2_4, a_k2_0,
_MM_SHUFFLE(1, 3, 1 ,3)); // 127, 125, 123, 121,
// Calculate 'x'.
const __m128 xr_ = _mm_sub_ps(a_j2_p0, a_k2_p0);
// 2-126, 4-124, 6-122, 8-120,
const __m128 xi_ = _mm_add_ps(a_j2_p1, a_k2_p1);
// 3-127, 5-125, 7-123, 9-121,
// Calculate product into 'y'.
// yr = wkr * xr + wki * xi;
// yi = wkr * xi - wki * xr;
const __m128 a_ = _mm_mul_ps(wkr_, xr_);
const __m128 b_ = _mm_mul_ps(wki_, xi_);
const __m128 c_ = _mm_mul_ps(wkr_, xi_);
const __m128 d_ = _mm_mul_ps(wki_, xr_);
const __m128 yr_ = _mm_add_ps(a_, b_); // 2-126, 4-124, 6-122, 8-120,
const __m128 yi_ = _mm_sub_ps(c_, d_); // 3-127, 5-125, 7-123, 9-121,
// Update 'a'.
// a[j2 + 0] = a[j2 + 0] - yr;
// a[j2 + 1] = yi - a[j2 + 1];
// a[k2 + 0] = yr + a[k2 + 0];
// a[k2 + 1] = yi - a[k2 + 1];
const __m128 a_j2_p0n = _mm_sub_ps(a_j2_p0, yr_); // 2, 4, 6, 8,
const __m128 a_j2_p1n = _mm_sub_ps(yi_, a_j2_p1); // 3, 5, 7, 9,
const __m128 a_k2_p0n = _mm_add_ps(a_k2_p0, yr_); // 126, 124, 122, 120,
const __m128 a_k2_p1n = _mm_sub_ps(yi_, a_k2_p1); // 127, 125, 123, 121,
// Shuffle in right order and store.
const __m128 a_j2_0n = _mm_unpacklo_ps(a_j2_p0n, a_j2_p1n);
// 2, 3, 4, 5,
const __m128 a_j2_4n = _mm_unpackhi_ps(a_j2_p0n, a_j2_p1n);
// 6, 7, 8, 9,
const __m128 a_k2_0nt = _mm_unpackhi_ps(a_k2_p0n, a_k2_p1n);
// 122, 123, 120, 121,
const __m128 a_k2_4nt = _mm_unpacklo_ps(a_k2_p0n, a_k2_p1n);
// 126, 127, 124, 125,
const __m128 a_k2_0n = _mm_shuffle_ps(a_k2_0nt, a_k2_0nt,
_MM_SHUFFLE(1, 0, 3 ,2)); // 120, 121, 122, 123,
const __m128 a_k2_4n = _mm_shuffle_ps(a_k2_4nt, a_k2_4nt,
_MM_SHUFFLE(1, 0, 3 ,2)); // 124, 125, 126, 127,
_mm_storeu_ps(&a[0 + j2], a_j2_0n);
_mm_storeu_ps(&a[4 + j2], a_j2_4n);
_mm_storeu_ps(&a[122 - j2], a_k2_0n);
_mm_storeu_ps(&a[126 - j2], a_k2_4n);
}
// Scalar code for the remaining items.
for (; j2 < 64; j1 += 1, j2 += 2) {
k2 = 128 - j2;
k1 = 32 - j1;
wkr = 0.5f - c[k1];
wki = c[j1];
xr = a[j2 + 0] - a[k2 + 0];
xi = a[j2 + 1] + a[k2 + 1];
yr = wkr * xr + wki * xi;
yi = wkr * xi - wki * xr;
a[j2 + 0] = a[j2 + 0] - yr;
a[j2 + 1] = yi - a[j2 + 1];
a[k2 + 0] = yr + a[k2 + 0];
a[k2 + 1] = yi - a[k2 + 1];
}
a[65] = -a[65];
}
void aec_rdft_init_sse2(void) {
cft1st_128 = cft1st_128_SSE2;
cftmdl_128 = cftmdl_128_SSE2;
rftfsub_128 = rftfsub_128_SSE2;
rftbsub_128 = rftbsub_128_SSE2;
}
#endif // WEBRTC_USE_SS2

901
webrtc/modules/audio_processing/aec/echo_cancellation.c Normal file
View File

@ -0,0 +1,901 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* Contains the API functions for the AEC.
*/
#include "echo_cancellation.h"
#include <math.h>
#ifdef AEC_DEBUG
#include <stdio.h>
#endif
#include <stdlib.h>
#include <string.h>
#include "aec_core.h"
#include "resampler.h"
#include "ring_buffer.h"
#define BUF_SIZE_FRAMES 50 // buffer size (frames)
// Maximum length of resampled signal. Must be an integer multiple of frames
// (ceil(1/(1 + MIN_SKEW)*2) + 1)*FRAME_LEN
// The factor of 2 handles wb, and the + 1 is as a safety margin
#define MAX_RESAMP_LEN (5 * FRAME_LEN)
static const int bufSizeSamp = BUF_SIZE_FRAMES * FRAME_LEN; // buffer size (samples)
static const int sampMsNb = 8; // samples per ms in nb
// Target suppression levels for nlp modes
// log{0.001, 0.00001, 0.00000001}
static const float targetSupp[3] = {-6.9f, -11.5f, -18.4f};
static const float minOverDrive[3] = {1.0f, 2.0f, 5.0f};
static const int initCheck = 42;
typedef struct {
int delayCtr;
int sampFreq;
int splitSampFreq;
int scSampFreq;
float sampFactor; // scSampRate / sampFreq
short nlpMode;
short autoOnOff;
short activity;
short skewMode;
short bufSizeStart;
//short bufResetCtr; // counts number of noncausal frames
int knownDelay;
// Stores the last frame added to the farend buffer
short farendOld[2][FRAME_LEN];
short initFlag; // indicates if AEC has been initialized
// Variables used for averaging far end buffer size
short counter;
short sum;
short firstVal;
short checkBufSizeCtr;
// Variables used for delay shifts
short msInSndCardBuf;
short filtDelay;
int timeForDelayChange;
int ECstartup;
int checkBuffSize;
int delayChange;
short lastDelayDiff;
#ifdef AEC_DEBUG
FILE *bufFile;
FILE *delayFile;
FILE *skewFile;
FILE *preCompFile;
FILE *postCompFile;
#endif // AEC_DEBUG
// Structures
void *farendBuf;
void *resampler;
int skewFrCtr;
int resample; // if the skew is small enough we don't resample
int highSkewCtr;
float skew;
int lastError;
aec_t *aec;
} aecpc_t;
// Estimates delay to set the position of the farend buffer read pointer
// (controlled by knownDelay)
static int EstBufDelay(aecpc_t *aecInst, short msInSndCardBuf);
// Stuffs the farend buffer if the estimated delay is too large
static int DelayComp(aecpc_t *aecInst);
WebRtc_Word32 WebRtcAec_Create(void **aecInst)
{
aecpc_t *aecpc;
if (aecInst == NULL) {
return -1;
}
aecpc = malloc(sizeof(aecpc_t));
*aecInst = aecpc;
if (aecpc == NULL) {
return -1;
}
if (WebRtcAec_CreateAec(&aecpc->aec) == -1) {
WebRtcAec_Free(aecpc);
aecpc = NULL;
return -1;
}
if (WebRtcApm_CreateBuffer(&aecpc->farendBuf, bufSizeSamp) == -1) {
WebRtcAec_Free(aecpc);
aecpc = NULL;
return -1;
}
if (WebRtcAec_CreateResampler(&aecpc->resampler) == -1) {
WebRtcAec_Free(aecpc);
aecpc = NULL;
return -1;
}
aecpc->initFlag = 0;
aecpc->lastError = 0;
#ifdef AEC_DEBUG
aecpc->aec->farFile = fopen("aecFar.pcm","wb");
aecpc->aec->nearFile = fopen("aecNear.pcm","wb");
aecpc->aec->outFile = fopen("aecOut.pcm","wb");
aecpc->aec->outLpFile = fopen("aecOutLp.pcm","wb");
aecpc->bufFile = fopen("aecBuf.dat", "wb");
aecpc->skewFile = fopen("aecSkew.dat", "wb");
aecpc->delayFile = fopen("aecDelay.dat", "wb");
aecpc->preCompFile = fopen("preComp.pcm", "wb");
aecpc->postCompFile = fopen("postComp.pcm", "wb");
#endif // AEC_DEBUG
return 0;
}
WebRtc_Word32 WebRtcAec_Free(void *aecInst)
{
aecpc_t *aecpc = aecInst;
if (aecpc == NULL) {
return -1;
}
#ifdef AEC_DEBUG
fclose(aecpc->aec->farFile);
fclose(aecpc->aec->nearFile);
fclose(aecpc->aec->outFile);
fclose(aecpc->aec->outLpFile);
fclose(aecpc->bufFile);
fclose(aecpc->skewFile);
fclose(aecpc->delayFile);
fclose(aecpc->preCompFile);
fclose(aecpc->postCompFile);
#endif // AEC_DEBUG
WebRtcAec_FreeAec(aecpc->aec);
WebRtcApm_FreeBuffer(aecpc->farendBuf);
WebRtcAec_FreeResampler(aecpc->resampler);
free(aecpc);
return 0;
}
WebRtc_Word32 WebRtcAec_Init(void *aecInst, WebRtc_Word32 sampFreq, WebRtc_Word32 scSampFreq)
{
aecpc_t *aecpc = aecInst;
AecConfig aecConfig;
if (aecpc == NULL) {
return -1;
}
if (sampFreq != 8000 && sampFreq != 16000 && sampFreq != 32000) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
aecpc->sampFreq = sampFreq;
if (scSampFreq < 1 || scSampFreq > 96000) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
aecpc->scSampFreq = scSampFreq;
// Initialize echo canceller core
if (WebRtcAec_InitAec(aecpc->aec, aecpc->sampFreq) == -1) {
aecpc->lastError = AEC_UNSPECIFIED_ERROR;
return -1;
}
// Initialize farend buffer
if (WebRtcApm_InitBuffer(aecpc->farendBuf) == -1) {
aecpc->lastError = AEC_UNSPECIFIED_ERROR;
return -1;
}
if (WebRtcAec_InitResampler(aecpc->resampler, aecpc->scSampFreq) == -1) {
aecpc->lastError = AEC_UNSPECIFIED_ERROR;
return -1;
}
aecpc->initFlag = initCheck; // indicates that initialization has been done
if (aecpc->sampFreq == 32000) {
aecpc->splitSampFreq = 16000;
}
else {
aecpc->splitSampFreq = sampFreq;
}
aecpc->skewFrCtr = 0;
aecpc->activity = 0;
aecpc->delayChange = 1;
aecpc->delayCtr = 0;
aecpc->sum = 0;
aecpc->counter = 0;
aecpc->checkBuffSize = 1;
aecpc->firstVal = 0;
aecpc->ECstartup = 1;
aecpc->bufSizeStart = 0;
aecpc->checkBufSizeCtr = 0;
aecpc->filtDelay = 0;
aecpc->timeForDelayChange =0;
aecpc->knownDelay = 0;
aecpc->lastDelayDiff = 0;
aecpc->skew = 0;
aecpc->resample = kAecFalse;
aecpc->highSkewCtr = 0;
aecpc->sampFactor = (aecpc->scSampFreq * 1.0f) / aecpc->splitSampFreq;
memset(&aecpc->farendOld[0][0], 0, 160);
// Default settings.
aecConfig.nlpMode = kAecNlpModerate;
aecConfig.skewMode = kAecFalse;
aecConfig.metricsMode = kAecFalse;
aecConfig.delay_logging = kAecFalse;
if (WebRtcAec_set_config(aecpc, aecConfig) == -1) {
aecpc->lastError = AEC_UNSPECIFIED_ERROR;
return -1;
}
return 0;
}
// only buffer L band for farend
WebRtc_Word32 WebRtcAec_BufferFarend(void *aecInst, const WebRtc_Word16 *farend,
WebRtc_Word16 nrOfSamples)
{
aecpc_t *aecpc = aecInst;
WebRtc_Word32 retVal = 0;
short newNrOfSamples;
short newFarend[MAX_RESAMP_LEN];
float skew;
if (aecpc == NULL) {
return -1;
}
if (farend == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
// number of samples == 160 for SWB input
if (nrOfSamples != 80 && nrOfSamples != 160) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
skew = aecpc->skew;
// TODO: Is this really a good idea?
if (!aecpc->ECstartup) {
DelayComp(aecpc);
}
if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
// Resample and get a new number of samples
newNrOfSamples = WebRtcAec_ResampleLinear(aecpc->resampler,
farend,
nrOfSamples,
skew,
newFarend);
WebRtcApm_WriteBuffer(aecpc->farendBuf, newFarend, newNrOfSamples);
#ifdef AEC_DEBUG
fwrite(farend, 2, nrOfSamples, aecpc->preCompFile);
fwrite(newFarend, 2, newNrOfSamples, aecpc->postCompFile);
#endif
}
else {
WebRtcApm_WriteBuffer(aecpc->farendBuf, farend, nrOfSamples);
}
return retVal;
}
WebRtc_Word32 WebRtcAec_Process(void *aecInst, const WebRtc_Word16 *nearend,
const WebRtc_Word16 *nearendH, WebRtc_Word16 *out, WebRtc_Word16 *outH,
WebRtc_Word16 nrOfSamples, WebRtc_Word16 msInSndCardBuf, WebRtc_Word32 skew)
{
aecpc_t *aecpc = aecInst;
WebRtc_Word32 retVal = 0;
short i;
short farend[FRAME_LEN];
short nmbrOfFilledBuffers;
short nBlocks10ms;
short nFrames;
#ifdef AEC_DEBUG
short msInAECBuf;
#endif
// Limit resampling to doubling/halving of signal
const float minSkewEst = -0.5f;
const float maxSkewEst = 1.0f;
if (aecpc == NULL) {
return -1;
}
if (nearend == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (out == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
// number of samples == 160 for SWB input
if (nrOfSamples != 80 && nrOfSamples != 160) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
// Check for valid pointers based on sampling rate
if (aecpc->sampFreq == 32000 && nearendH == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (msInSndCardBuf < 0) {
msInSndCardBuf = 0;
aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
retVal = -1;
}
else if (msInSndCardBuf > 500) {
msInSndCardBuf = 500;
aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
retVal = -1;
}
msInSndCardBuf += 10;
aecpc->msInSndCardBuf = msInSndCardBuf;
if (aecpc->skewMode == kAecTrue) {
if (aecpc->skewFrCtr < 25) {
aecpc->skewFrCtr++;
}
else {
retVal = WebRtcAec_GetSkew(aecpc->resampler, skew, &aecpc->skew);
if (retVal == -1) {
aecpc->skew = 0;
aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
}
aecpc->skew /= aecpc->sampFactor*nrOfSamples;
if (aecpc->skew < 1.0e-3 && aecpc->skew > -1.0e-3) {
aecpc->resample = kAecFalse;
}
else {
aecpc->resample = kAecTrue;
}
if (aecpc->skew < minSkewEst) {
aecpc->skew = minSkewEst;
}
else if (aecpc->skew > maxSkewEst) {
aecpc->skew = maxSkewEst;
}
#ifdef AEC_DEBUG
fwrite(&aecpc->skew, sizeof(aecpc->skew), 1, aecpc->skewFile);
#endif
}
}
nFrames = nrOfSamples / FRAME_LEN;
nBlocks10ms = nFrames / aecpc->aec->mult;
if (aecpc->ECstartup) {
if (nearend != out) {
// Only needed if they don't already point to the same place.
memcpy(out, nearend, sizeof(short) * nrOfSamples);
}
nmbrOfFilledBuffers = WebRtcApm_get_buffer_size(aecpc->farendBuf) / FRAME_LEN;
// The AEC is in the start up mode
// AEC is disabled until the soundcard buffer and farend buffers are OK
// Mechanism to ensure that the soundcard buffer is reasonably stable.
if (aecpc->checkBuffSize) {
aecpc->checkBufSizeCtr++;
// Before we fill up the far end buffer we require the amount of data on the
// sound card to be stable (+/-8 ms) compared to the first value. This
// comparison is made during the following 4 consecutive frames. If it seems
// to be stable then we start to fill up the far end buffer.
if (aecpc->counter == 0) {
aecpc->firstVal = aecpc->msInSndCardBuf;
aecpc->sum = 0;
}
if (abs(aecpc->firstVal - aecpc->msInSndCardBuf) <
WEBRTC_SPL_MAX(0.2 * aecpc->msInSndCardBuf, sampMsNb)) {
aecpc->sum += aecpc->msInSndCardBuf;
aecpc->counter++;
}
else {
aecpc->counter = 0;
}
if (aecpc->counter*nBlocks10ms >= 6) {
// The farend buffer size is determined in blocks of 80 samples
// Use 75% of the average value of the soundcard buffer
aecpc->bufSizeStart = WEBRTC_SPL_MIN((int) (0.75 * (aecpc->sum *
aecpc->aec->mult) / (aecpc->counter * 10)), BUF_SIZE_FRAMES);
// buffersize has now been determined
aecpc->checkBuffSize = 0;
}
if (aecpc->checkBufSizeCtr * nBlocks10ms > 50) {
// for really bad sound cards, don't disable echocanceller for more than 0.5 sec
aecpc->bufSizeStart = WEBRTC_SPL_MIN((int) (0.75 * (aecpc->msInSndCardBuf *
aecpc->aec->mult) / 10), BUF_SIZE_FRAMES);
aecpc->checkBuffSize = 0;
}
}
// if checkBuffSize changed in the if-statement above
if (!aecpc->checkBuffSize) {
// soundcard buffer is now reasonably stable
// When the far end buffer is filled with approximately the same amount of
// data as the amount on the sound card we end the start up phase and start
// to cancel echoes.
if (nmbrOfFilledBuffers == aecpc->bufSizeStart) {
aecpc->ECstartup = 0; // Enable the AEC
}
else if (nmbrOfFilledBuffers > aecpc->bufSizeStart) {
WebRtcApm_FlushBuffer(aecpc->farendBuf, WebRtcApm_get_buffer_size(aecpc->farendBuf) -
aecpc->bufSizeStart * FRAME_LEN);
aecpc->ECstartup = 0;
}
}
}
else {
// AEC is enabled
// Note only 1 block supported for nb and 2 blocks for wb
for (i = 0; i < nFrames; i++) {
nmbrOfFilledBuffers = WebRtcApm_get_buffer_size(aecpc->farendBuf) / FRAME_LEN;
// Check that there is data in the far end buffer
if (nmbrOfFilledBuffers > 0) {
// Get the next 80 samples from the farend buffer
WebRtcApm_ReadBuffer(aecpc->farendBuf, farend, FRAME_LEN);
// Always store the last frame for use when we run out of data
memcpy(&(aecpc->farendOld[i][0]), farend, FRAME_LEN * sizeof(short));
}
else {
// We have no data so we use the last played frame
memcpy(farend, &(aecpc->farendOld[i][0]), FRAME_LEN * sizeof(short));
}
// Call buffer delay estimator when all data is extracted,
// i.e. i = 0 for NB and i = 1 for WB or SWB
if ((i == 0 && aecpc->splitSampFreq == 8000) ||
(i == 1 && (aecpc->splitSampFreq == 16000))) {
EstBufDelay(aecpc, aecpc->msInSndCardBuf);
}
// Call the AEC
WebRtcAec_ProcessFrame(aecpc->aec, farend, &nearend[FRAME_LEN * i], &nearendH[FRAME_LEN * i],
&out[FRAME_LEN * i], &outH[FRAME_LEN * i], aecpc->knownDelay);
}
}
#ifdef AEC_DEBUG
msInAECBuf = WebRtcApm_get_buffer_size(aecpc->farendBuf) / (sampMsNb*aecpc->aec->mult);
fwrite(&msInAECBuf, 2, 1, aecpc->bufFile);
fwrite(&(aecpc->knownDelay), sizeof(aecpc->knownDelay), 1, aecpc->delayFile);
#endif
return retVal;
}
WebRtc_Word32 WebRtcAec_set_config(void *aecInst, AecConfig config)
{
aecpc_t *aecpc = aecInst;
if (aecpc == NULL) {
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
if (config.skewMode != kAecFalse && config.skewMode != kAecTrue) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
aecpc->skewMode = config.skewMode;
if (config.nlpMode != kAecNlpConservative && config.nlpMode !=
kAecNlpModerate && config.nlpMode != kAecNlpAggressive) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
aecpc->nlpMode = config.nlpMode;
aecpc->aec->targetSupp = targetSupp[aecpc->nlpMode];
aecpc->aec->minOverDrive = minOverDrive[aecpc->nlpMode];
if (config.metricsMode != kAecFalse && config.metricsMode != kAecTrue) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
aecpc->aec->metricsMode = config.metricsMode;
if (aecpc->aec->metricsMode == kAecTrue) {
WebRtcAec_InitMetrics(aecpc->aec);
}
if (config.delay_logging != kAecFalse && config.delay_logging != kAecTrue) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
aecpc->aec->delay_logging_enabled = config.delay_logging;
if (aecpc->aec->delay_logging_enabled == kAecTrue) {
memset(aecpc->aec->delay_histogram, 0, sizeof(aecpc->aec->delay_histogram));
}
return 0;
}
WebRtc_Word32 WebRtcAec_get_config(void *aecInst, AecConfig *config)
{
aecpc_t *aecpc = aecInst;
if (aecpc == NULL) {
return -1;
}
if (config == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
config->nlpMode = aecpc->nlpMode;
config->skewMode = aecpc->skewMode;
config->metricsMode = aecpc->aec->metricsMode;
config->delay_logging = aecpc->aec->delay_logging_enabled;
return 0;
}
WebRtc_Word32 WebRtcAec_get_echo_status(void *aecInst, WebRtc_Word16 *status)
{
aecpc_t *aecpc = aecInst;
if (aecpc == NULL) {
return -1;
}
if (status == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
*status = aecpc->aec->echoState;
return 0;
}
WebRtc_Word32 WebRtcAec_GetMetrics(void *aecInst, AecMetrics *metrics)
{
const float upweight = 0.7f;
float dtmp;
short stmp;
aecpc_t *aecpc = aecInst;
if (aecpc == NULL) {
return -1;
}
if (metrics == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
// ERL
metrics->erl.instant = (short) aecpc->aec->erl.instant;
if ((aecpc->aec->erl.himean > offsetLevel) && (aecpc->aec->erl.average > offsetLevel)) {
// Use a mix between regular average and upper part average
dtmp = upweight * aecpc->aec->erl.himean + (1 - upweight) * aecpc->aec->erl.average;
metrics->erl.average = (short) dtmp;
}
else {
metrics->erl.average = offsetLevel;
}
metrics->erl.max = (short) aecpc->aec->erl.max;
if (aecpc->aec->erl.min < (offsetLevel * (-1))) {
metrics->erl.min = (short) aecpc->aec->erl.min;
}
else {
metrics->erl.min = offsetLevel;
}
// ERLE
metrics->erle.instant = (short) aecpc->aec->erle.instant;
if ((aecpc->aec->erle.himean > offsetLevel) && (aecpc->aec->erle.average > offsetLevel)) {
// Use a mix between regular average and upper part average
dtmp = upweight * aecpc->aec->erle.himean + (1 - upweight) * aecpc->aec->erle.average;
metrics->erle.average = (short) dtmp;
}
else {
metrics->erle.average = offsetLevel;
}
metrics->erle.max = (short) aecpc->aec->erle.max;
if (aecpc->aec->erle.min < (offsetLevel * (-1))) {
metrics->erle.min = (short) aecpc->aec->erle.min;
} else {
metrics->erle.min = offsetLevel;
}
// RERL
if ((metrics->erl.average > offsetLevel) && (metrics->erle.average > offsetLevel)) {
stmp = metrics->erl.average + metrics->erle.average;
}
else {
stmp = offsetLevel;
}
metrics->rerl.average = stmp;
// No other statistics needed, but returned for completeness
metrics->rerl.instant = stmp;
metrics->rerl.max = stmp;
metrics->rerl.min = stmp;
// A_NLP
metrics->aNlp.instant = (short) aecpc->aec->aNlp.instant;
if ((aecpc->aec->aNlp.himean > offsetLevel) && (aecpc->aec->aNlp.average > offsetLevel)) {
// Use a mix between regular average and upper part average
dtmp = upweight * aecpc->aec->aNlp.himean + (1 - upweight) * aecpc->aec->aNlp.average;
metrics->aNlp.average = (short) dtmp;
}
else {
metrics->aNlp.average = offsetLevel;
}
metrics->aNlp.max = (short) aecpc->aec->aNlp.max;
if (aecpc->aec->aNlp.min < (offsetLevel * (-1))) {
metrics->aNlp.min = (short) aecpc->aec->aNlp.min;
}
else {
metrics->aNlp.min = offsetLevel;
}
return 0;
}
int WebRtcAec_GetDelayMetrics(void* handle, int* median, int* std) {
aecpc_t* self = handle;
int i = 0;
int delay_values = 0;
int num_delay_values = 0;
int my_median = 0;
const int kMsPerBlock = (PART_LEN * 1000) / self->splitSampFreq;
float l1_norm = 0;
if (self == NULL) {
return -1;
}
if (median == NULL) {
self->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (std == NULL) {
self->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (self->initFlag != initCheck) {
self->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
if (self->aec->delay_logging_enabled == 0) {
// Logging disabled
self->lastError = AEC_UNSUPPORTED_FUNCTION_ERROR;
return -1;
}
// Get number of delay values since last update
for (i = 0; i < kMaxDelay; i++) {
num_delay_values += self->aec->delay_histogram[i];
}
if (num_delay_values == 0) {
// We have no new delay value data
*median = -1;
*std = -1;
return 0;
}
delay_values = num_delay_values >> 1; // Start value for median count down
// Get median of delay values since last update
for (i = 0; i < kMaxDelay; i++) {
delay_values -= self->aec->delay_histogram[i];
if (delay_values < 0) {
my_median = i;
break;
}
}
*median = my_median * kMsPerBlock;
// Calculate the L1 norm, with median value as central moment
for (i = 0; i < kMaxDelay; i++) {
l1_norm += (float) (fabs(i - my_median) * self->aec->delay_histogram[i]);
}
*std = (int) (l1_norm / (float) num_delay_values + 0.5f) * kMsPerBlock;
// Reset histogram
memset(self->aec->delay_histogram, 0, sizeof(self->aec->delay_histogram));
return 0;
}
WebRtc_Word32 WebRtcAec_get_version(WebRtc_Word8 *versionStr, WebRtc_Word16 len)
{
const char version[] = "AEC 2.5.0";
const short versionLen = (short)strlen(version) + 1; // +1 for null-termination
if (versionStr == NULL) {
return -1;
}
if (versionLen > len) {
return -1;
}
strncpy(versionStr, version, versionLen);
return 0;
}
WebRtc_Word32 WebRtcAec_get_error_code(void *aecInst)
{
aecpc_t *aecpc = aecInst;
if (aecpc == NULL) {
return -1;
}
return aecpc->lastError;
}
static int EstBufDelay(aecpc_t *aecpc, short msInSndCardBuf)
{
short delayNew, nSampFar, nSampSndCard;
short diff;
nSampFar = WebRtcApm_get_buffer_size(aecpc->farendBuf);
nSampSndCard = msInSndCardBuf * sampMsNb * aecpc->aec->mult;
delayNew = nSampSndCard - nSampFar;
// Account for resampling frame delay
if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
delayNew -= kResamplingDelay;
}
if (delayNew < FRAME_LEN) {
WebRtcApm_FlushBuffer(aecpc->farendBuf, FRAME_LEN);
delayNew += FRAME_LEN;
}
aecpc->filtDelay = WEBRTC_SPL_MAX(0, (short)(0.8*aecpc->filtDelay + 0.2*delayNew));
diff = aecpc->filtDelay - aecpc->knownDelay;
if (diff > 224) {
if (aecpc->lastDelayDiff < 96) {
aecpc->timeForDelayChange = 0;
}
else {
aecpc->timeForDelayChange++;
}
}
else if (diff < 96 && aecpc->knownDelay > 0) {
if (aecpc->lastDelayDiff > 224) {
aecpc->timeForDelayChange = 0;
}
else {
aecpc->timeForDelayChange++;
}
}
else {
aecpc->timeForDelayChange = 0;
}
aecpc->lastDelayDiff = diff;
if (aecpc->timeForDelayChange > 25) {
aecpc->knownDelay = WEBRTC_SPL_MAX((int)aecpc->filtDelay - 160, 0);
}
return 0;
}
static int DelayComp(aecpc_t *aecpc)
{
int nSampFar, nSampSndCard, delayNew, nSampAdd;
const int maxStuffSamp = 10 * FRAME_LEN;
nSampFar = WebRtcApm_get_buffer_size(aecpc->farendBuf);
nSampSndCard = aecpc->msInSndCardBuf * sampMsNb * aecpc->aec->mult;
delayNew = nSampSndCard - nSampFar;
// Account for resampling frame delay
if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
delayNew -= kResamplingDelay;
}
if (delayNew > FAR_BUF_LEN - FRAME_LEN*aecpc->aec->mult) {
// The difference of the buffersizes is larger than the maximum
// allowed known delay. Compensate by stuffing the buffer.
nSampAdd = (int)(WEBRTC_SPL_MAX((int)(0.5 * nSampSndCard - nSampFar),
FRAME_LEN));
nSampAdd = WEBRTC_SPL_MIN(nSampAdd, maxStuffSamp);
WebRtcApm_StuffBuffer(aecpc->farendBuf, nSampAdd);
aecpc->delayChange = 1; // the delay needs to be updated
}
return 0;
}

278
webrtc/modules/audio_processing/aec/interface/echo_cancellation.h Normal file
View File

@ -0,0 +1,278 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_INTERFACE_ECHO_CANCELLATION_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_INTERFACE_ECHO_CANCELLATION_H_
#include "typedefs.h"
// Errors
#define AEC_UNSPECIFIED_ERROR 12000
#define AEC_UNSUPPORTED_FUNCTION_ERROR 12001
#define AEC_UNINITIALIZED_ERROR 12002
#define AEC_NULL_POINTER_ERROR 12003
#define AEC_BAD_PARAMETER_ERROR 12004
// Warnings
#define AEC_BAD_PARAMETER_WARNING 12050
enum {
kAecNlpConservative = 0,
kAecNlpModerate,
kAecNlpAggressive
};
enum {
kAecFalse = 0,
kAecTrue
};
typedef struct {
WebRtc_Word16 nlpMode; // default kAecNlpModerate
WebRtc_Word16 skewMode; // default kAecFalse
WebRtc_Word16 metricsMode; // default kAecFalse
int delay_logging; // default kAecFalse
//float realSkew;
} AecConfig;
typedef struct {
WebRtc_Word16 instant;
WebRtc_Word16 average;
WebRtc_Word16 max;
WebRtc_Word16 min;
} AecLevel;
typedef struct {
AecLevel rerl;
AecLevel erl;
AecLevel erle;
AecLevel aNlp;
} AecMetrics;
#ifdef __cplusplus
extern "C" {
#endif
/*
* Allocates the memory needed by the AEC. The memory needs to be initialized
* separately using the WebRtcAec_Init() function.
*
* Inputs Description
* -------------------------------------------------------------------
* void **aecInst Pointer to the AEC instance to be created
* and initialized
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_Create(void **aecInst);
/*
* This function releases the memory allocated by WebRtcAec_Create().
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_Free(void *aecInst);
/*
* Initializes an AEC instance.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
* WebRtc_Word32 sampFreq Sampling frequency of data
* WebRtc_Word32 scSampFreq Soundcard sampling frequency
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_Init(void *aecInst,
WebRtc_Word32 sampFreq,
WebRtc_Word32 scSampFreq);
/*
* Inserts an 80 or 160 sample block of data into the farend buffer.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
* WebRtc_Word16 *farend In buffer containing one frame of
* farend signal for L band
* WebRtc_Word16 nrOfSamples Number of samples in farend buffer
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_BufferFarend(void *aecInst,
const WebRtc_Word16 *farend,
WebRtc_Word16 nrOfSamples);
/*
* Runs the echo canceller on an 80 or 160 sample blocks of data.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
* WebRtc_Word16 *nearend In buffer containing one frame of
* nearend+echo signal for L band
* WebRtc_Word16 *nearendH In buffer containing one frame of
* nearend+echo signal for H band
* WebRtc_Word16 nrOfSamples Number of samples in nearend buffer
* WebRtc_Word16 msInSndCardBuf Delay estimate for sound card and
* system buffers
* WebRtc_Word16 skew Difference between number of samples played
* and recorded at the soundcard (for clock skew
* compensation)
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word16 *out Out buffer, one frame of processed nearend
* for L band
* WebRtc_Word16 *outH Out buffer, one frame of processed nearend
* for H band
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_Process(void *aecInst,
const WebRtc_Word16 *nearend,
const WebRtc_Word16 *nearendH,
WebRtc_Word16 *out,
WebRtc_Word16 *outH,
WebRtc_Word16 nrOfSamples,
WebRtc_Word16 msInSndCardBuf,
WebRtc_Word32 skew);
/*
* This function enables the user to set certain parameters on-the-fly.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
* AecConfig config Config instance that contains all
* properties to be set
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_set_config(void *aecInst, AecConfig config);
/*
* Gets the on-the-fly paramters.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
*
* Outputs Description
* -------------------------------------------------------------------
* AecConfig *config Pointer to the config instance that
* all properties will be written to
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_get_config(void *aecInst, AecConfig *config);
/*
* Gets the current echo status of the nearend signal.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word16 *status 0: Almost certainly nearend single-talk
* 1: Might not be neared single-talk
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_get_echo_status(void *aecInst, WebRtc_Word16 *status);
/*
* Gets the current echo metrics for the session.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
*
* Outputs Description
* -------------------------------------------------------------------
* AecMetrics *metrics Struct which will be filled out with the
* current echo metrics.
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_GetMetrics(void *aecInst, AecMetrics *metrics);
/*
* Gets the current delay metrics for the session.
*
* Inputs Description
* -------------------------------------------------------------------
* void* handle Pointer to the AEC instance
*
* Outputs Description
* -------------------------------------------------------------------
* int* median Delay median value.
* int* std Delay standard deviation.
*
* int return 0: OK
* -1: error
*/
int WebRtcAec_GetDelayMetrics(void* handle, int* median, int* std);
/*
* Gets the last error code.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecInst Pointer to the AEC instance
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 11000-11100: error code
*/
WebRtc_Word32 WebRtcAec_get_error_code(void *aecInst);
/*
* Gets a version string.
*
* Inputs Description
* -------------------------------------------------------------------
* char *versionStr Pointer to a string array
* WebRtc_Word16 len The maximum length of the string
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word8 *versionStr Pointer to a string array
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAec_get_version(WebRtc_Word8 *versionStr, WebRtc_Word16 len);
#ifdef __cplusplus
}
#endif
#endif /* WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_INTERFACE_ECHO_CANCELLATION_H_ */

233
webrtc/modules/audio_processing/aec/resampler.c Normal file
View File

@ -0,0 +1,233 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/* Resamples a signal to an arbitrary rate. Used by the AEC to compensate for clock
* skew by resampling the farend signal.
*/
#include <assert.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include "resampler.h"
#include "aec_core.h"
enum { kFrameBufferSize = FRAME_LEN * 4 };
enum { kEstimateLengthFrames = 400 };
typedef struct {
short buffer[kFrameBufferSize];
float position;
int deviceSampleRateHz;
int skewData[kEstimateLengthFrames];
int skewDataIndex;
float skewEstimate;
} resampler_t;
static int EstimateSkew(const int* rawSkew,
int size,
int absLimit,
float *skewEst);
int WebRtcAec_CreateResampler(void **resampInst)
{
resampler_t *obj = malloc(sizeof(resampler_t));
*resampInst = obj;
if (obj == NULL) {
return -1;
}
return 0;
}
int WebRtcAec_InitResampler(void *resampInst, int deviceSampleRateHz)
{
resampler_t *obj = (resampler_t*) resampInst;
memset(obj->buffer, 0, sizeof(obj->buffer));
obj->position = 0.0;
obj->deviceSampleRateHz = deviceSampleRateHz;
memset(obj->skewData, 0, sizeof(obj->skewData));
obj->skewDataIndex = 0;
obj->skewEstimate = 0.0;
return 0;
}
int WebRtcAec_FreeResampler(void *resampInst)
{
resampler_t *obj = (resampler_t*) resampInst;
free(obj);
return 0;
}
int WebRtcAec_ResampleLinear(void *resampInst,
const short *inspeech,
int size,
float skew,
short *outspeech)
{
resampler_t *obj = (resampler_t*) resampInst;
short *y;
float be, tnew, interp;
int tn, outsize, mm;
if (size < 0 || size > 2 * FRAME_LEN) {
return -1;
}
// Add new frame data in lookahead
memcpy(&obj->buffer[FRAME_LEN + kResamplingDelay],
inspeech,
size * sizeof(short));
// Sample rate ratio
be = 1 + skew;
// Loop over input frame
mm = 0;
y = &obj->buffer[FRAME_LEN]; // Point at current frame
tnew = be * mm + obj->position;
tn = (int) tnew;
while (tn < size) {
// Interpolation
interp = y[tn] + (tnew - tn) * (y[tn+1] - y[tn]);
if (interp > 32767) {
interp = 32767;
}
else if (interp < -32768) {
interp = -32768;
}
outspeech[mm] = (short) interp;
mm++;
tnew = be * mm + obj->position;
tn = (int) tnew;
}
outsize = mm;
obj->position += outsize * be - size;
// Shift buffer
memmove(obj->buffer,
&obj->buffer[size],
(kFrameBufferSize - size) * sizeof(short));
return outsize;
}
int WebRtcAec_GetSkew(void *resampInst, int rawSkew, float *skewEst)
{
resampler_t *obj = (resampler_t*)resampInst;
int err = 0;
if (obj->skewDataIndex < kEstimateLengthFrames) {
obj->skewData[obj->skewDataIndex] = rawSkew;
obj->skewDataIndex++;
}
else if (obj->skewDataIndex == kEstimateLengthFrames) {
err = EstimateSkew(obj->skewData,
kEstimateLengthFrames,
obj->deviceSampleRateHz,
skewEst);
obj->skewEstimate = *skewEst;
obj->skewDataIndex++;
}
else {
*skewEst = obj->skewEstimate;
}
return err;
}
int EstimateSkew(const int* rawSkew,
int size,
int deviceSampleRateHz,
float *skewEst)
{
const int absLimitOuter = (int)(0.04f * deviceSampleRateHz);
const int absLimitInner = (int)(0.0025f * deviceSampleRateHz);
int i = 0;
int n = 0;
float rawAvg = 0;
float err = 0;
float rawAbsDev = 0;
int upperLimit = 0;
int lowerLimit = 0;
float cumSum = 0;
float x = 0;
float x2 = 0;
float y = 0;
float xy = 0;
float xAvg = 0;
float denom = 0;
float skew = 0;
*skewEst = 0; // Set in case of error below.
for (i = 0; i < size; i++) {
if ((rawSkew[i] < absLimitOuter && rawSkew[i] > -absLimitOuter)) {
n++;
rawAvg += rawSkew[i];
}
}
if (n == 0) {
return -1;
}
assert(n > 0);
rawAvg /= n;
for (i = 0; i < size; i++) {
if ((rawSkew[i] < absLimitOuter && rawSkew[i] > -absLimitOuter)) {
err = rawSkew[i] - rawAvg;
rawAbsDev += err >= 0 ? err : -err;
}
}
assert(n > 0);
rawAbsDev /= n;
upperLimit = (int)(rawAvg + 5 * rawAbsDev + 1); // +1 for ceiling.
lowerLimit = (int)(rawAvg - 5 * rawAbsDev - 1); // -1 for floor.
n = 0;
for (i = 0; i < size; i++) {
if ((rawSkew[i] < absLimitInner && rawSkew[i] > -absLimitInner) ||
(rawSkew[i] < upperLimit && rawSkew[i] > lowerLimit)) {
n++;
cumSum += rawSkew[i];
x += n;
x2 += n*n;
y += cumSum;
xy += n * cumSum;
}
}
if (n == 0) {
return -1;
}
assert(n > 0);
xAvg = x / n;
denom = x2 - xAvg*x;
if (denom != 0) {
skew = (xy - xAvg*y) / denom;
}
*skewEst = skew;
return 0;
}

32
webrtc/modules/audio_processing/aec/resampler.h Normal file
View File

@ -0,0 +1,32 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_RESAMPLER_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_RESAMPLER_H_
enum { kResamplingDelay = 1 };
// Unless otherwise specified, functions return 0 on success and -1 on error
int WebRtcAec_CreateResampler(void **resampInst);
int WebRtcAec_InitResampler(void *resampInst, int deviceSampleRateHz);
int WebRtcAec_FreeResampler(void *resampInst);
// Estimates skew from raw measurement.
int WebRtcAec_GetSkew(void *resampInst, int rawSkew, float *skewEst);
// Resamples input using linear interpolation.
// Returns size of resampled array.
int WebRtcAec_ResampleLinear(void *resampInst,
const short *inspeech,
int size,
float skew,
short *outspeech);
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_RESAMPLER_H_

9
webrtc/modules/audio_processing/aecm/Makefile.am Normal file
View File

@ -0,0 +1,9 @@
noinst_LTLIBRARIES = libaecm.la
libaecm_la_SOURCES = interface/echo_control_mobile.h \
echo_control_mobile.c \
aecm_core.c \
aecm_core.h
libaecm_la_CFLAGS = $(AM_CFLAGS) $(COMMON_CFLAGS) \
-I$(top_srcdir)/src/common_audio/signal_processing_library/main/interface \
-I$(top_srcdir)/src/modules/audio_processing/utility

34
webrtc/modules/audio_processing/aecm/aecm.gypi Normal file
View File

@ -0,0 +1,34 @@
# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
{
'targets': [
{
'target_name': 'aecm',
'type': '<(library)',
'dependencies': [
'<(webrtc_root)/common_audio/common_audio.gyp:spl',
'apm_util'
],
'include_dirs': [
'interface',
],
'direct_dependent_settings': {
'include_dirs': [
'interface',
],
},
'sources': [
'interface/echo_control_mobile.h',
'echo_control_mobile.c',
'aecm_core.c',
'aecm_core.h',
],
},
],
}

1932
webrtc/modules/audio_processing/aecm/aecm_core.c Normal file
View File

File diff suppressed because it is too large Load Diff

358
webrtc/modules/audio_processing/aecm/aecm_core.h Normal file
View File

@ -0,0 +1,358 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
// Performs echo control (suppression) with fft routines in fixed-point
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_SOURCE_AECM_CORE_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_SOURCE_AECM_CORE_H_
#define AECM_DYNAMIC_Q // turn on/off dynamic Q-domain
//#define AECM_WITH_ABS_APPROX
//#define AECM_SHORT // for 32 sample partition length (otherwise 64)
#include "typedefs.h"
#include "signal_processing_library.h"
// Algorithm parameters
#define FRAME_LEN 80 // Total frame length, 10 ms
#ifdef AECM_SHORT
#define PART_LEN 32 // Length of partition
#define PART_LEN_SHIFT 6 // Length of (PART_LEN * 2) in base 2
#else
#define PART_LEN 64 // Length of partition
#define PART_LEN_SHIFT 7 // Length of (PART_LEN * 2) in base 2
#endif
#define PART_LEN1 (PART_LEN + 1) // Unique fft coefficients
#define PART_LEN2 (PART_LEN << 1) // Length of partition * 2
#define PART_LEN4 (PART_LEN << 2) // Length of partition * 4
#define FAR_BUF_LEN PART_LEN4 // Length of buffers
#define MAX_DELAY 100
// Counter parameters
#ifdef AECM_SHORT
#define CONV_LEN 1024 // Convergence length used at startup
#else
#define CONV_LEN 512 // Convergence length used at startup
#endif
#define CONV_LEN2 (CONV_LEN << 1) // Convergence length * 2 used at startup
// Energy parameters
#define MAX_BUF_LEN 64 // History length of energy signals
#define FAR_ENERGY_MIN 1025 // Lowest Far energy level: At least 2 in energy
#define FAR_ENERGY_DIFF 929 // Allowed difference between max and min
#define ENERGY_DEV_OFFSET 0 // The energy error offset in Q8
#define ENERGY_DEV_TOL 400 // The energy estimation tolerance in Q8
#define FAR_ENERGY_VAD_REGION 230 // Far VAD tolerance region
// Stepsize parameters
#define MU_MIN 10 // Min stepsize 2^-MU_MIN (far end energy dependent)
#define MU_MAX 1 // Max stepsize 2^-MU_MAX (far end energy dependent)
#define MU_DIFF 9 // MU_MIN - MU_MAX
// Channel parameters
#define MIN_MSE_COUNT 20 // Min number of consecutive blocks with enough far end
// energy to compare channel estimates
#define MIN_MSE_DIFF 29 // The ratio between adapted and stored channel to
// accept a new storage (0.8 in Q-MSE_RESOLUTION)
#define MSE_RESOLUTION 5 // MSE parameter resolution
#define RESOLUTION_CHANNEL16 12 // W16 Channel in Q-RESOLUTION_CHANNEL16
#define RESOLUTION_CHANNEL32 28 // W32 Channel in Q-RESOLUTION_CHANNEL
#define CHANNEL_VAD 16 // Minimum energy in frequency band to update channel
// Suppression gain parameters: SUPGAIN_ parameters in Q-(RESOLUTION_SUPGAIN)
#define RESOLUTION_SUPGAIN 8 // Channel in Q-(RESOLUTION_SUPGAIN)
#define SUPGAIN_DEFAULT (1 << RESOLUTION_SUPGAIN) // Default suppression gain
#define SUPGAIN_ERROR_PARAM_A 3072 // Estimation error parameter (Maximum gain) (8 in Q8)
#define SUPGAIN_ERROR_PARAM_B 1536 // Estimation error parameter (Gain before going down)
#define SUPGAIN_ERROR_PARAM_D SUPGAIN_DEFAULT // Estimation error parameter
// (Should be the same as Default) (1 in Q8)
#define SUPGAIN_EPC_DT 200 // = SUPGAIN_ERROR_PARAM_C * ENERGY_DEV_TOL
// Defines for "check delay estimation"
#define CORR_WIDTH 31 // Number of samples to correlate over.
#define CORR_MAX 16 // Maximum correlation offset
#define CORR_MAX_BUF 63
#define CORR_DEV 4
#define CORR_MAX_LEVEL 20
#define CORR_MAX_LOW 4
#define CORR_BUF_LEN (CORR_MAX << 1) + 1
// Note that CORR_WIDTH + 2*CORR_MAX <= MAX_BUF_LEN
#define ONE_Q14 (1 << 14)
// NLP defines
#define NLP_COMP_LOW 3277 // 0.2 in Q14
#define NLP_COMP_HIGH ONE_Q14 // 1 in Q14
extern const WebRtc_Word16 WebRtcAecm_kSqrtHanning[];
typedef struct {
WebRtc_Word16 real;
WebRtc_Word16 imag;
} complex16_t;
typedef struct
{
int farBufWritePos;
int farBufReadPos;
int knownDelay;
int lastKnownDelay;
int firstVAD; // Parameter to control poorly initialized channels
void *farFrameBuf;
void *nearNoisyFrameBuf;
void *nearCleanFrameBuf;
void *outFrameBuf;
WebRtc_Word16 farBuf[FAR_BUF_LEN];
WebRtc_Word16 mult;
WebRtc_UWord32 seed;
// Delay estimation variables
void* delay_estimator;
WebRtc_UWord16 currentDelay;
WebRtc_Word16 nlpFlag;
WebRtc_Word16 fixedDelay;
WebRtc_UWord32 totCount;
WebRtc_Word16 dfaCleanQDomain;
WebRtc_Word16 dfaCleanQDomainOld;
WebRtc_Word16 dfaNoisyQDomain;
WebRtc_Word16 dfaNoisyQDomainOld;
WebRtc_Word16 nearLogEnergy[MAX_BUF_LEN];
WebRtc_Word16 farLogEnergy;
WebRtc_Word16 echoAdaptLogEnergy[MAX_BUF_LEN];
WebRtc_Word16 echoStoredLogEnergy[MAX_BUF_LEN];
// The extra 16 or 32 bytes in the following buffers are for alignment based Neon code.
// It's designed this way since the current GCC compiler can't align a buffer in 16 or 32
// byte boundaries properly.
WebRtc_Word16 channelStored_buf[PART_LEN1 + 8];
WebRtc_Word16 channelAdapt16_buf[PART_LEN1 + 8];
WebRtc_Word32 channelAdapt32_buf[PART_LEN1 + 8];
WebRtc_Word16 xBuf_buf[PART_LEN2 + 16]; // farend
WebRtc_Word16 dBufClean_buf[PART_LEN2 + 16]; // nearend
WebRtc_Word16 dBufNoisy_buf[PART_LEN2 + 16]; // nearend
WebRtc_Word16 outBuf_buf[PART_LEN + 8];
// Pointers to the above buffers
WebRtc_Word16 *channelStored;
WebRtc_Word16 *channelAdapt16;
WebRtc_Word32 *channelAdapt32;
WebRtc_Word16 *xBuf;
WebRtc_Word16 *dBufClean;
WebRtc_Word16 *dBufNoisy;
WebRtc_Word16 *outBuf;
WebRtc_Word32 echoFilt[PART_LEN1];
WebRtc_Word16 nearFilt[PART_LEN1];
WebRtc_Word32 noiseEst[PART_LEN1];
int noiseEstTooLowCtr[PART_LEN1];
int noiseEstTooHighCtr[PART_LEN1];
WebRtc_Word16 noiseEstCtr;
WebRtc_Word16 cngMode;
WebRtc_Word32 mseAdaptOld;
WebRtc_Word32 mseStoredOld;
WebRtc_Word32 mseThreshold;
WebRtc_Word16 farEnergyMin;
WebRtc_Word16 farEnergyMax;
WebRtc_Word16 farEnergyMaxMin;
WebRtc_Word16 farEnergyVAD;
WebRtc_Word16 farEnergyMSE;
int currentVADValue;
WebRtc_Word16 vadUpdateCount;
WebRtc_Word16 startupState;
WebRtc_Word16 mseChannelCount;
WebRtc_Word16 supGain;
WebRtc_Word16 supGainOld;
WebRtc_Word16 supGainErrParamA;
WebRtc_Word16 supGainErrParamD;
WebRtc_Word16 supGainErrParamDiffAB;
WebRtc_Word16 supGainErrParamDiffBD;
#ifdef AEC_DEBUG
FILE *farFile;
FILE *nearFile;
FILE *outFile;
#endif
} AecmCore_t;
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_CreateCore(...)
//
// Allocates the memory needed by the AECM. The memory needs to be
// initialized separately using the WebRtcAecm_InitCore() function.
//
// Input:
// - aecm : Instance that should be created
//
// Output:
// - aecm : Created instance
//
// Return value : 0 - Ok
// -1 - Error
//
int WebRtcAecm_CreateCore(AecmCore_t **aecm);
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_InitCore(...)
//
// This function initializes the AECM instant created with WebRtcAecm_CreateCore(...)
// Input:
// - aecm : Pointer to the AECM instance
// - samplingFreq : Sampling Frequency
//
// Output:
// - aecm : Initialized instance
//
// Return value : 0 - Ok
// -1 - Error
//
int WebRtcAecm_InitCore(AecmCore_t * const aecm, int samplingFreq);
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_FreeCore(...)
//
// This function releases the memory allocated by WebRtcAecm_CreateCore()
// Input:
// - aecm : Pointer to the AECM instance
//
// Return value : 0 - Ok
// -1 - Error
// 11001-11016: Error
//
int WebRtcAecm_FreeCore(AecmCore_t *aecm);
int WebRtcAecm_Control(AecmCore_t *aecm, int delay, int nlpFlag);
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_InitEchoPathCore(...)
//
// This function resets the echo channel adaptation with the specified channel.
// Input:
// - aecm : Pointer to the AECM instance
// - echo_path : Pointer to the data that should initialize the echo path
//
// Output:
// - aecm : Initialized instance
//
void WebRtcAecm_InitEchoPathCore(AecmCore_t* aecm, const WebRtc_Word16* echo_path);
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_ProcessFrame(...)
//
// This function processes frames and sends blocks to WebRtcAecm_ProcessBlock(...)
//
// Inputs:
// - aecm : Pointer to the AECM instance
// - farend : In buffer containing one frame of echo signal
// - nearendNoisy : In buffer containing one frame of nearend+echo signal without NS
// - nearendClean : In buffer containing one frame of nearend+echo signal with NS
//
// Output:
// - out : Out buffer, one frame of nearend signal :
//
//
int WebRtcAecm_ProcessFrame(AecmCore_t * aecm, const WebRtc_Word16 * farend,
const WebRtc_Word16 * nearendNoisy,
const WebRtc_Word16 * nearendClean,
WebRtc_Word16 * out);
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_ProcessBlock(...)
//
// This function is called for every block within one frame
// This function is called by WebRtcAecm_ProcessFrame(...)
//
// Inputs:
// - aecm : Pointer to the AECM instance
// - farend : In buffer containing one block of echo signal
// - nearendNoisy : In buffer containing one frame of nearend+echo signal without NS
// - nearendClean : In buffer containing one frame of nearend+echo signal with NS
//
// Output:
// - out : Out buffer, one block of nearend signal :
//
//
int WebRtcAecm_ProcessBlock(AecmCore_t * aecm, const WebRtc_Word16 * farend,
const WebRtc_Word16 * nearendNoisy,
const WebRtc_Word16 * noisyClean,
WebRtc_Word16 * out);
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_BufferFarFrame()
//
// Inserts a frame of data into farend buffer.
//
// Inputs:
// - aecm : Pointer to the AECM instance
// - farend : In buffer containing one frame of farend signal
// - farLen : Length of frame
//
void WebRtcAecm_BufferFarFrame(AecmCore_t * const aecm, const WebRtc_Word16 * const farend,
const int farLen);
///////////////////////////////////////////////////////////////////////////////////////////////
// WebRtcAecm_FetchFarFrame()
//
// Read the farend buffer to account for known delay
//
// Inputs:
// - aecm : Pointer to the AECM instance
// - farend : In buffer containing one frame of farend signal
// - farLen : Length of frame
// - knownDelay : known delay
//
void WebRtcAecm_FetchFarFrame(AecmCore_t * const aecm, WebRtc_Word16 * const farend,
const int farLen, const int knownDelay);
///////////////////////////////////////////////////////////////////////////////////////////////
// Some internal functions shared by ARM NEON and generic C code:
//
void WebRtcAecm_CalcLinearEnergies(AecmCore_t* aecm,
const WebRtc_UWord16* far_spectrum,
WebRtc_Word32* echoEst,
WebRtc_UWord32* far_energy,
WebRtc_UWord32* echo_energy_adapt,
WebRtc_UWord32* echo_energy_stored);
void WebRtcAecm_StoreAdaptiveChannel(AecmCore_t* aecm,
const WebRtc_UWord16* far_spectrum,
WebRtc_Word32* echo_est);
void WebRtcAecm_ResetAdaptiveChannel(AecmCore_t *aecm);
void WebRtcAecm_WindowAndFFT(WebRtc_Word16* fft,
const WebRtc_Word16* time_signal,
complex16_t* freq_signal,
int time_signal_scaling);
void WebRtcAecm_InverseFFTAndWindow(AecmCore_t* aecm,
WebRtc_Word16* fft,
complex16_t* efw,
WebRtc_Word16* output,
const WebRtc_Word16* nearendClean);
#endif

314
webrtc/modules/audio_processing/aecm/aecm_core_neon.c Normal file
View File

@ -0,0 +1,314 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#if defined(WEBRTC_ANDROID) && defined(WEBRTC_ARCH_ARM_NEON)
#include "aecm_core.h"
#include <arm_neon.h>
#include <assert.h>
// Square root of Hanning window in Q14.
static const WebRtc_Word16 kSqrtHanningReversed[] __attribute__ ((aligned (8))) = {
16384, 16373, 16354, 16325,
16286, 16237, 16179, 16111,
16034, 15947, 15851, 15746,
15631, 15506, 15373, 15231,
15079, 14918, 14749, 14571,
14384, 14189, 13985, 13773,
13553, 13325, 13089, 12845,
12594, 12335, 12068, 11795,
11514, 11227, 10933, 10633,
10326, 10013, 9695, 9370,
9040, 8705, 8364, 8019,
7668, 7313, 6954, 6591,
6224, 5853, 5478, 5101,
4720, 4337, 3951, 3562,
3172, 2780, 2386, 1990,
1594, 1196, 798, 399
};
void WebRtcAecm_WindowAndFFT(WebRtc_Word16* fft,
const WebRtc_Word16* time_signal,
complex16_t* freq_signal,
int time_signal_scaling)
{
int i, j;
int16x4_t tmp16x4_scaling = vdup_n_s16(time_signal_scaling);
__asm__("vmov.i16 d21, #0" ::: "d21");
for(i = 0, j = 0; i < PART_LEN; i += 4, j += 8)
{
int16x4_t tmp16x4_0;
int16x4_t tmp16x4_1;
int32x4_t tmp32x4_0;
/* Window near end */
// fft[j] = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT((time_signal[i]
// << time_signal_scaling), WebRtcAecm_kSqrtHanning[i], 14);
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&time_signal[i]));
tmp16x4_0 = vshl_s16(tmp16x4_0, tmp16x4_scaling);
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&WebRtcAecm_kSqrtHanning[i]));
tmp32x4_0 = vmull_s16(tmp16x4_0, tmp16x4_1);
__asm__("vshrn.i32 d20, %q0, #14" : : "w"(tmp32x4_0) : "d20");
__asm__("vst2.16 {d20, d21}, [%0, :128]" : : "r"(&fft[j]) : "q10");
// fft[PART_LEN2 + j] = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(
// (time_signal[PART_LEN + i] << time_signal_scaling),
// WebRtcAecm_kSqrtHanning[PART_LEN - i], 14);
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&time_signal[i + PART_LEN]));
tmp16x4_0 = vshl_s16(tmp16x4_0, tmp16x4_scaling);
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&kSqrtHanningReversed[i]));
tmp32x4_0 = vmull_s16(tmp16x4_0, tmp16x4_1);
__asm__("vshrn.i32 d20, %q0, #14" : : "w"(tmp32x4_0) : "d20");
__asm__("vst2.16 {d20, d21}, [%0, :128]" : : "r"(&fft[PART_LEN2 + j]) : "q10");
}
WebRtcSpl_ComplexBitReverse(fft, PART_LEN_SHIFT);
WebRtcSpl_ComplexFFT(fft, PART_LEN_SHIFT, 1);
// Take only the first PART_LEN2 samples, and switch the sign of the imaginary part.
for(i = 0, j = 0; j < PART_LEN2; i += 8, j += 16)
{
__asm__("vld2.16 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&fft[j]) : "q10", "q11");
__asm__("vneg.s16 d22, d22" : : : "q10");
__asm__("vneg.s16 d23, d23" : : : "q11");
__asm__("vst2.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&freq_signal[i].real): "q10", "q11");
}
}
void WebRtcAecm_InverseFFTAndWindow(AecmCore_t* aecm,
WebRtc_Word16* fft,
complex16_t* efw,
WebRtc_Word16* output,
const WebRtc_Word16* nearendClean)
{
int i, j, outCFFT;
WebRtc_Word32 tmp32no1;
// Synthesis
for(i = 0, j = 0; i < PART_LEN; i += 4, j += 8)
{
// We overwrite two more elements in fft[], but it's ok.
__asm__("vld2.16 {d20, d21}, [%0, :128]" : : "r"(&(efw[i].real)) : "q10");
__asm__("vmov q11, q10" : : : "q10", "q11");
__asm__("vneg.s16 d23, d23" : : : "q11");
__asm__("vst2.16 {d22, d23}, [%0, :128]" : : "r"(&fft[j]): "q11");
__asm__("vrev64.16 q10, q10" : : : "q10");
__asm__("vst2.16 {d20, d21}, [%0]" : : "r"(&fft[PART_LEN4 - j - 6]): "q10");
}
fft[PART_LEN2] = efw[PART_LEN].real;
fft[PART_LEN2 + 1] = -efw[PART_LEN].imag;
// Inverse FFT, result should be scaled with outCFFT.
WebRtcSpl_ComplexBitReverse(fft, PART_LEN_SHIFT);
outCFFT = WebRtcSpl_ComplexIFFT(fft, PART_LEN_SHIFT, 1);
// Take only the real values and scale with outCFFT.
for (i = 0, j = 0; i < PART_LEN2; i += 8, j+= 16)
{
__asm__("vld2.16 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&fft[j]) : "q10", "q11");
__asm__("vst1.16 {d20, d21}, [%0, :128]" : : "r"(&fft[i]): "q10");
}
int32x4_t tmp32x4_2;
__asm__("vdup.32 %q0, %1" : "=w"(tmp32x4_2) : "r"((WebRtc_Word32)
(outCFFT - aecm->dfaCleanQDomain)));
for (i = 0; i < PART_LEN; i += 4)
{
int16x4_t tmp16x4_0;
int16x4_t tmp16x4_1;
int32x4_t tmp32x4_0;
int32x4_t tmp32x4_1;
// fft[i] = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT_WITH_ROUND(
// fft[i], WebRtcAecm_kSqrtHanning[i], 14);
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&fft[i]));
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&WebRtcAecm_kSqrtHanning[i]));
__asm__("vmull.s16 %q0, %P1, %P2" : "=w"(tmp32x4_0) : "w"(tmp16x4_0), "w"(tmp16x4_1));
__asm__("vrshr.s32 %q0, %q1, #14" : "=w"(tmp32x4_0) : "0"(tmp32x4_0));
// tmp32no1 = WEBRTC_SPL_SHIFT_W32((WebRtc_Word32)fft[i],
// outCFFT - aecm->dfaCleanQDomain);
__asm__("vshl.s32 %q0, %q1, %q2" : "=w"(tmp32x4_0) : "0"(tmp32x4_0), "w"(tmp32x4_2));
// fft[i] = (WebRtc_Word16)WEBRTC_SPL_SAT(WEBRTC_SPL_WORD16_MAX,
// tmp32no1 + outBuf[i], WEBRTC_SPL_WORD16_MIN);
// output[i] = fft[i];
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&aecm->outBuf[i]));
__asm__("vmovl.s16 %q0, %P1" : "=w"(tmp32x4_1) : "w"(tmp16x4_0));
__asm__("vadd.i32 %q0, %q1" : : "w"(tmp32x4_0), "w"(tmp32x4_1));
__asm__("vqshrn.s32 %P0, %q1, #0" : "=w"(tmp16x4_0) : "w"(tmp32x4_0));
__asm__("vst1.16 %P0, [%1, :64]" : : "w"(tmp16x4_0), "r"(&fft[i]));
__asm__("vst1.16 %P0, [%1, :64]" : : "w"(tmp16x4_0), "r"(&output[i]));
// tmp32no1 = WEBRTC_SPL_MUL_16_16_RSFT(
// fft[PART_LEN + i], WebRtcAecm_kSqrtHanning[PART_LEN - i], 14);
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&fft[PART_LEN + i]));
__asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&kSqrtHanningReversed[i]));
__asm__("vmull.s16 %q0, %P1, %P2" : "=w"(tmp32x4_0) : "w"(tmp16x4_0), "w"(tmp16x4_1));
__asm__("vshr.s32 %q0, %q1, #14" : "=w"(tmp32x4_0) : "0"(tmp32x4_0));
// tmp32no1 = WEBRTC_SPL_SHIFT_W32(tmp32no1, outCFFT - aecm->dfaCleanQDomain);
__asm__("vshl.s32 %q0, %q1, %q2" : "=w"(tmp32x4_0) : "0"(tmp32x4_0), "w"(tmp32x4_2));
// outBuf[i] = (WebRtc_Word16)WEBRTC_SPL_SAT(
// WEBRTC_SPL_WORD16_MAX, tmp32no1, WEBRTC_SPL_WORD16_MIN);
__asm__("vqshrn.s32 %P0, %q1, #0" : "=w"(tmp16x4_0) : "w"(tmp32x4_0));
__asm__("vst1.16 %P0, [%1, :64]" : : "w"(tmp16x4_0), "r"(&aecm->outBuf[i]));
}
// Copy the current block to the old position (outBuf is shifted elsewhere).
for (i = 0; i < PART_LEN; i += 16)
{
__asm__("vld1.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&aecm->xBuf[i + PART_LEN]) : "q10");
__asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&aecm->xBuf[i]): "q10");
}
for (i = 0; i < PART_LEN; i += 16)
{
__asm__("vld1.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&aecm->dBufNoisy[i + PART_LEN]) : "q10");
__asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&aecm->dBufNoisy[i]): "q10");
}
if (nearendClean != NULL) {
for (i = 0; i < PART_LEN; i += 16)
{
__asm__("vld1.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&aecm->dBufClean[i + PART_LEN]) : "q10");
__asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&aecm->dBufClean[i]): "q10");
}
}
}
void WebRtcAecm_CalcLinearEnergies(AecmCore_t* aecm,
const WebRtc_UWord16* far_spectrum,
WebRtc_Word32* echo_est,
WebRtc_UWord32* far_energy,
WebRtc_UWord32* echo_energy_adapt,
WebRtc_UWord32* echo_energy_stored)
{
int i;
register WebRtc_UWord32 far_energy_r;
register WebRtc_UWord32 echo_energy_stored_r;
register WebRtc_UWord32 echo_energy_adapt_r;
uint32x4_t tmp32x4_0;
__asm__("vmov.i32 q14, #0" : : : "q14"); // far_energy
__asm__("vmov.i32 q8, #0" : : : "q8"); // echo_energy_stored
__asm__("vmov.i32 q9, #0" : : : "q9"); // echo_energy_adapt
for(i = 0; i < PART_LEN -7; i += 8)
{
// far_energy += (WebRtc_UWord32)(far_spectrum[i]);
__asm__("vld1.16 {d26, d27}, [%0]" : : "r"(&far_spectrum[i]) : "q13");
__asm__("vaddw.u16 q14, q14, d26" : : : "q14", "q13");
__asm__("vaddw.u16 q14, q14, d27" : : : "q14", "q13");
// Get estimated echo energies for adaptive channel and stored channel.
// echoEst[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
__asm__("vld1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelStored[i]) : "q12");
__asm__("vmull.u16 q10, d26, d24" : : : "q12", "q13", "q10");
__asm__("vmull.u16 q11, d27, d25" : : : "q12", "q13", "q11");
__asm__("vst1.32 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&echo_est[i]):
"q10", "q11");
// echo_energy_stored += (WebRtc_UWord32)echoEst[i];
__asm__("vadd.u32 q8, q10" : : : "q10", "q8");
__asm__("vadd.u32 q8, q11" : : : "q11", "q8");
// echo_energy_adapt += WEBRTC_SPL_UMUL_16_16(
// aecm->channelAdapt16[i], far_spectrum[i]);
__asm__("vld1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelAdapt16[i]) : "q12");
__asm__("vmull.u16 q10, d26, d24" : : : "q12", "q13", "q10");
__asm__("vmull.u16 q11, d27, d25" : : : "q12", "q13", "q11");
__asm__("vadd.u32 q9, q10" : : : "q9", "q15");
__asm__("vadd.u32 q9, q11" : : : "q9", "q11");
}
__asm__("vadd.u32 d28, d29" : : : "q14");
__asm__("vpadd.u32 d28, d28" : : : "q14");
__asm__("vmov.32 %0, d28[0]" : "=r"(far_energy_r): : "q14");
__asm__("vadd.u32 d18, d19" : : : "q9");
__asm__("vpadd.u32 d18, d18" : : : "q9");
__asm__("vmov.32 %0, d18[0]" : "=r"(echo_energy_adapt_r): : "q9");
__asm__("vadd.u32 d16, d17" : : : "q8");
__asm__("vpadd.u32 d16, d16" : : : "q8");
__asm__("vmov.32 %0, d16[0]" : "=r"(echo_energy_stored_r): : "q8");
// Get estimated echo energies for adaptive channel and stored channel.
echo_est[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
*echo_energy_stored = echo_energy_stored_r + (WebRtc_UWord32)echo_est[i];
*far_energy = far_energy_r + (WebRtc_UWord32)(far_spectrum[i]);
*echo_energy_adapt = echo_energy_adapt_r + WEBRTC_SPL_UMUL_16_16(
aecm->channelAdapt16[i], far_spectrum[i]);
}
void WebRtcAecm_StoreAdaptiveChannel(AecmCore_t* aecm,
const WebRtc_UWord16* far_spectrum,
WebRtc_Word32* echo_est)
{
int i;
// During startup we store the channel every block.
// Recalculate echo estimate.
for(i = 0; i < PART_LEN -7; i += 8)
{
// aecm->channelStored[i] = acem->channelAdapt16[i];
// echo_est[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
__asm__("vld1.16 {d26, d27}, [%0]" : : "r"(&far_spectrum[i]) : "q13");
__asm__("vld1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelAdapt16[i]) : "q12");
__asm__("vst1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelStored[i]) : "q12");
__asm__("vmull.u16 q10, d26, d24" : : : "q12", "q13", "q10");
__asm__("vmull.u16 q11, d27, d25" : : : "q12", "q13", "q11");
__asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&echo_est[i]) : "q10", "q11");
}
aecm->channelStored[i] = aecm->channelAdapt16[i];
echo_est[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
}
void WebRtcAecm_ResetAdaptiveChannel(AecmCore_t* aecm)
{
int i;
for(i = 0; i < PART_LEN -7; i += 8)
{
// aecm->channelAdapt16[i] = aecm->channelStored[i];
// aecm->channelAdapt32[i] = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)
// aecm->channelStored[i], 16);
__asm__("vld1.16 {d24, d25}, [%0, :128]" : :
"r"(&aecm->channelStored[i]) : "q12");
__asm__("vst1.16 {d24, d25}, [%0, :128]" : :
"r"(&aecm->channelAdapt16[i]) : "q12");
__asm__("vshll.s16 q10, d24, #16" : : : "q12", "q13", "q10");
__asm__("vshll.s16 q11, d25, #16" : : : "q12", "q13", "q11");
__asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : :
"r"(&aecm->channelAdapt32[i]): "q10", "q11");
}
aecm->channelAdapt16[i] = aecm->channelStored[i];
aecm->channelAdapt32[i] = WEBRTC_SPL_LSHIFT_W32(
(WebRtc_Word32)aecm->channelStored[i], 16);
}
#endif // #if defined(WEBRTC_ANDROID) && defined(WEBRTC_ARCH_ARM_NEON)

800
webrtc/modules/audio_processing/aecm/echo_control_mobile.c Normal file
View File

@ -0,0 +1,800 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <stdlib.h>
//#include <string.h>
#include "echo_control_mobile.h"
#include "aecm_core.h"
#include "ring_buffer.h"
#ifdef AEC_DEBUG
#include <stdio.h>
#endif
#ifdef MAC_IPHONE_PRINT
#include <time.h>
#include <stdio.h>
#elif defined ARM_WINM_LOG
#include "windows.h"
extern HANDLE logFile;
#endif
#define BUF_SIZE_FRAMES 50 // buffer size (frames)
// Maximum length of resampled signal. Must be an integer multiple of frames
// (ceil(1/(1 + MIN_SKEW)*2) + 1)*FRAME_LEN
// The factor of 2 handles wb, and the + 1 is as a safety margin
#define MAX_RESAMP_LEN (5 * FRAME_LEN)
static const int kBufSizeSamp = BUF_SIZE_FRAMES * FRAME_LEN; // buffer size (samples)
static const int kSampMsNb = 8; // samples per ms in nb
// Target suppression levels for nlp modes
// log{0.001, 0.00001, 0.00000001}
static const int kInitCheck = 42;
typedef struct
{
int sampFreq;
int scSampFreq;
short bufSizeStart;
int knownDelay;
// Stores the last frame added to the farend buffer
short farendOld[2][FRAME_LEN];
short initFlag; // indicates if AEC has been initialized
// Variables used for averaging far end buffer size
short counter;
short sum;
short firstVal;
short checkBufSizeCtr;
// Variables used for delay shifts
short msInSndCardBuf;
short filtDelay;
int timeForDelayChange;
int ECstartup;
int checkBuffSize;
int delayChange;
short lastDelayDiff;
WebRtc_Word16 echoMode;
#ifdef AEC_DEBUG
FILE *bufFile;
FILE *delayFile;
FILE *preCompFile;
FILE *postCompFile;
#endif // AEC_DEBUG
// Structures
void *farendBuf;
int lastError;
AecmCore_t *aecmCore;
} aecmob_t;
// Estimates delay to set the position of the farend buffer read pointer
// (controlled by knownDelay)
static int WebRtcAecm_EstBufDelay(aecmob_t *aecmInst, short msInSndCardBuf);
// Stuffs the farend buffer if the estimated delay is too large
static int WebRtcAecm_DelayComp(aecmob_t *aecmInst);
WebRtc_Word32 WebRtcAecm_Create(void **aecmInst)
{
aecmob_t *aecm;
if (aecmInst == NULL)
{
return -1;
}
aecm = malloc(sizeof(aecmob_t));
*aecmInst = aecm;
if (aecm == NULL)
{
return -1;
}
if (WebRtcAecm_CreateCore(&aecm->aecmCore) == -1)
{
WebRtcAecm_Free(aecm);
aecm = NULL;
return -1;
}
if (WebRtcApm_CreateBuffer(&aecm->farendBuf, kBufSizeSamp) == -1)
{
WebRtcAecm_Free(aecm);
aecm = NULL;
return -1;
}
aecm->initFlag = 0;
aecm->lastError = 0;
#ifdef AEC_DEBUG
aecm->aecmCore->farFile = fopen("aecFar.pcm","wb");
aecm->aecmCore->nearFile = fopen("aecNear.pcm","wb");
aecm->aecmCore->outFile = fopen("aecOut.pcm","wb");
//aecm->aecmCore->outLpFile = fopen("aecOutLp.pcm","wb");
aecm->bufFile = fopen("aecBuf.dat", "wb");
aecm->delayFile = fopen("aecDelay.dat", "wb");
aecm->preCompFile = fopen("preComp.pcm", "wb");
aecm->postCompFile = fopen("postComp.pcm", "wb");
#endif // AEC_DEBUG
return 0;
}
WebRtc_Word32 WebRtcAecm_Free(void *aecmInst)
{
aecmob_t *aecm = aecmInst;
if (aecm == NULL)
{
return -1;
}
#ifdef AEC_DEBUG
fclose(aecm->aecmCore->farFile);
fclose(aecm->aecmCore->nearFile);
fclose(aecm->aecmCore->outFile);
//fclose(aecm->aecmCore->outLpFile);
fclose(aecm->bufFile);
fclose(aecm->delayFile);
fclose(aecm->preCompFile);
fclose(aecm->postCompFile);
#endif // AEC_DEBUG
WebRtcAecm_FreeCore(aecm->aecmCore);
WebRtcApm_FreeBuffer(aecm->farendBuf);
free(aecm);
return 0;
}
WebRtc_Word32 WebRtcAecm_Init(void *aecmInst, WebRtc_Word32 sampFreq)
{
aecmob_t *aecm = aecmInst;
AecmConfig aecConfig;
if (aecm == NULL)
{
return -1;
}
if (sampFreq != 8000 && sampFreq != 16000)
{
aecm->lastError = AECM_BAD_PARAMETER_ERROR;
return -1;
}
aecm->sampFreq = sampFreq;
// Initialize AECM core
if (WebRtcAecm_InitCore(aecm->aecmCore, aecm->sampFreq) == -1)
{
aecm->lastError = AECM_UNSPECIFIED_ERROR;
return -1;
}
// Initialize farend buffer
if (WebRtcApm_InitBuffer(aecm->farendBuf) == -1)
{
aecm->lastError = AECM_UNSPECIFIED_ERROR;
return -1;
}
aecm->initFlag = kInitCheck; // indicates that initialization has been done
aecm->delayChange = 1;
aecm->sum = 0;
aecm->counter = 0;
aecm->checkBuffSize = 1;
aecm->firstVal = 0;
aecm->ECstartup = 1;
aecm->bufSizeStart = 0;
aecm->checkBufSizeCtr = 0;
aecm->filtDelay = 0;
aecm->timeForDelayChange = 0;
aecm->knownDelay = 0;
aecm->lastDelayDiff = 0;
memset(&aecm->farendOld[0][0], 0, 160);
// Default settings.
aecConfig.cngMode = AecmTrue;
aecConfig.echoMode = 3;
if (WebRtcAecm_set_config(aecm, aecConfig) == -1)
{
aecm->lastError = AECM_UNSPECIFIED_ERROR;
return -1;
}
return 0;
}
WebRtc_Word32 WebRtcAecm_BufferFarend(void *aecmInst, const WebRtc_Word16 *farend,
WebRtc_Word16 nrOfSamples)
{
aecmob_t *aecm = aecmInst;
WebRtc_Word32 retVal = 0;
if (aecm == NULL)
{
return -1;
}
if (farend == NULL)
{
aecm->lastError = AECM_NULL_POINTER_ERROR;
return -1;
}
if (aecm->initFlag != kInitCheck)
{
aecm->lastError = AECM_UNINITIALIZED_ERROR;
return -1;
}
if (nrOfSamples != 80 && nrOfSamples != 160)
{
aecm->lastError = AECM_BAD_PARAMETER_ERROR;
return -1;
}
// TODO: Is this really a good idea?
if (!aecm->ECstartup)
{
WebRtcAecm_DelayComp(aecm);
}
WebRtcApm_WriteBuffer(aecm->farendBuf, farend, nrOfSamples);
return retVal;
}
WebRtc_Word32 WebRtcAecm_Process(void *aecmInst, const WebRtc_Word16 *nearendNoisy,
const WebRtc_Word16 *nearendClean, WebRtc_Word16 *out,
WebRtc_Word16 nrOfSamples, WebRtc_Word16 msInSndCardBuf)
{
aecmob_t *aecm = aecmInst;
WebRtc_Word32 retVal = 0;
short i;
short farend[FRAME_LEN];
short nmbrOfFilledBuffers;
short nBlocks10ms;
short nFrames;
#ifdef AEC_DEBUG
short msInAECBuf;
#endif
#ifdef ARM_WINM_LOG
__int64 freq, start, end, diff;
unsigned int milliseconds;
DWORD temp;
#elif defined MAC_IPHONE_PRINT
// double endtime = 0, starttime = 0;
struct timeval starttime;
struct timeval endtime;
static long int timeused = 0;
static int timecount = 0;
#endif
if (aecm == NULL)
{
return -1;
}
if (nearendNoisy == NULL)
{
aecm->lastError = AECM_NULL_POINTER_ERROR;
return -1;
}
if (out == NULL)
{
aecm->lastError = AECM_NULL_POINTER_ERROR;
return -1;
}
if (aecm->initFlag != kInitCheck)
{
aecm->lastError = AECM_UNINITIALIZED_ERROR;
return -1;
}
if (nrOfSamples != 80 && nrOfSamples != 160)
{
aecm->lastError = AECM_BAD_PARAMETER_ERROR;
return -1;
}
if (msInSndCardBuf < 0)
{
msInSndCardBuf = 0;
aecm->lastError = AECM_BAD_PARAMETER_WARNING;
retVal = -1;
} else if (msInSndCardBuf > 500)
{
msInSndCardBuf = 500;
aecm->lastError = AECM_BAD_PARAMETER_WARNING;
retVal = -1;
}
msInSndCardBuf += 10;
aecm->msInSndCardBuf = msInSndCardBuf;
nFrames = nrOfSamples / FRAME_LEN;
nBlocks10ms = nFrames / aecm->aecmCore->mult;
if (aecm->ECstartup)
{
if (nearendClean == NULL)
{
memcpy(out, nearendNoisy, sizeof(short) * nrOfSamples);
} else
{
memcpy(out, nearendClean, sizeof(short) * nrOfSamples);
}
nmbrOfFilledBuffers = WebRtcApm_get_buffer_size(aecm->farendBuf) / FRAME_LEN;
// The AECM is in the start up mode
// AECM is disabled until the soundcard buffer and farend buffers are OK
// Mechanism to ensure that the soundcard buffer is reasonably stable.
if (aecm->checkBuffSize)
{
aecm->checkBufSizeCtr++;
// Before we fill up the far end buffer we require the amount of data on the
// sound card to be stable (+/-8 ms) compared to the first value. This
// comparison is made during the following 4 consecutive frames. If it seems
// to be stable then we start to fill up the far end buffer.
if (aecm->counter == 0)
{
aecm->firstVal = aecm->msInSndCardBuf;
aecm->sum = 0;
}
if (abs(aecm->firstVal - aecm->msInSndCardBuf)
< WEBRTC_SPL_MAX(0.2 * aecm->msInSndCardBuf, kSampMsNb))
{
aecm->sum += aecm->msInSndCardBuf;
aecm->counter++;
} else
{
aecm->counter = 0;
}
if (aecm->counter * nBlocks10ms >= 6)
{
// The farend buffer size is determined in blocks of 80 samples
// Use 75% of the average value of the soundcard buffer
aecm->bufSizeStart
= WEBRTC_SPL_MIN((3 * aecm->sum
* aecm->aecmCore->mult) / (aecm->counter * 40), BUF_SIZE_FRAMES);
// buffersize has now been determined
aecm->checkBuffSize = 0;
}
if (aecm->checkBufSizeCtr * nBlocks10ms > 50)
{
// for really bad sound cards, don't disable echocanceller for more than 0.5 sec
aecm->bufSizeStart = WEBRTC_SPL_MIN((3 * aecm->msInSndCardBuf
* aecm->aecmCore->mult) / 40, BUF_SIZE_FRAMES);
aecm->checkBuffSize = 0;
}
}
// if checkBuffSize changed in the if-statement above
if (!aecm->checkBuffSize)
{
// soundcard buffer is now reasonably stable
// When the far end buffer is filled with approximately the same amount of
// data as the amount on the sound card we end the start up phase and start
// to cancel echoes.
if (nmbrOfFilledBuffers == aecm->bufSizeStart)
{
aecm->ECstartup = 0; // Enable the AECM
} else if (nmbrOfFilledBuffers > aecm->bufSizeStart)
{
WebRtcApm_FlushBuffer(
aecm->farendBuf,
WebRtcApm_get_buffer_size(aecm->farendBuf)
- aecm->bufSizeStart * FRAME_LEN);
aecm->ECstartup = 0;
}
}
} else
{
// AECM is enabled
// Note only 1 block supported for nb and 2 blocks for wb
for (i = 0; i < nFrames; i++)
{
nmbrOfFilledBuffers = WebRtcApm_get_buffer_size(aecm->farendBuf) / FRAME_LEN;
// Check that there is data in the far end buffer
if (nmbrOfFilledBuffers > 0)
{
// Get the next 80 samples from the farend buffer
WebRtcApm_ReadBuffer(aecm->farendBuf, farend, FRAME_LEN);
// Always store the last frame for use when we run out of data
memcpy(&(aecm->farendOld[i][0]), farend, FRAME_LEN * sizeof(short));
} else
{
// We have no data so we use the last played frame
memcpy(farend, &(aecm->farendOld[i][0]), FRAME_LEN * sizeof(short));
}
// Call buffer delay estimator when all data is extracted,
// i,e. i = 0 for NB and i = 1 for WB
if ((i == 0 && aecm->sampFreq == 8000) || (i == 1 && aecm->sampFreq == 16000))
{
WebRtcAecm_EstBufDelay(aecm, aecm->msInSndCardBuf);
}
#ifdef ARM_WINM_LOG
// measure tick start
QueryPerformanceFrequency((LARGE_INTEGER*)&freq);
QueryPerformanceCounter((LARGE_INTEGER*)&start);
#elif defined MAC_IPHONE_PRINT
// starttime = clock()/(double)CLOCKS_PER_SEC;
gettimeofday(&starttime, NULL);
#endif
// Call the AECM
/*WebRtcAecm_ProcessFrame(aecm->aecmCore, farend, &nearend[FRAME_LEN * i],
&out[FRAME_LEN * i], aecm->knownDelay);*/
if (nearendClean == NULL)
{
if (WebRtcAecm_ProcessFrame(aecm->aecmCore,
farend,
&nearendNoisy[FRAME_LEN * i],
NULL,
&out[FRAME_LEN * i]) == -1)
{
return -1;
}
} else
{
if (WebRtcAecm_ProcessFrame(aecm->aecmCore,
farend,
&nearendNoisy[FRAME_LEN * i],
&nearendClean[FRAME_LEN * i],
&out[FRAME_LEN * i]) == -1)
{
return -1;
}
}
#ifdef ARM_WINM_LOG
// measure tick end
QueryPerformanceCounter((LARGE_INTEGER*)&end);
if(end > start)
{
diff = ((end - start) * 1000) / (freq/1000);
milliseconds = (unsigned int)(diff & 0xffffffff);
WriteFile (logFile, &milliseconds, sizeof(unsigned int), &temp, NULL);
}
#elif defined MAC_IPHONE_PRINT
// endtime = clock()/(double)CLOCKS_PER_SEC;
// printf("%f\n", endtime - starttime);
gettimeofday(&endtime, NULL);
if( endtime.tv_usec > starttime.tv_usec)
{
timeused += endtime.tv_usec - starttime.tv_usec;
} else
{
timeused += endtime.tv_usec + 1000000 - starttime.tv_usec;
}
if(++timecount == 1000)
{
timecount = 0;
printf("AEC: %ld\n", timeused);
timeused = 0;
}
#endif
}
}
#ifdef AEC_DEBUG
msInAECBuf = WebRtcApm_get_buffer_size(aecm->farendBuf) / (kSampMsNb*aecm->aecmCore->mult);
fwrite(&msInAECBuf, 2, 1, aecm->bufFile);
fwrite(&(aecm->knownDelay), sizeof(aecm->knownDelay), 1, aecm->delayFile);
#endif
return retVal;
}
WebRtc_Word32 WebRtcAecm_set_config(void *aecmInst, AecmConfig config)
{
aecmob_t *aecm = aecmInst;
if (aecm == NULL)
{
return -1;
}
if (aecm->initFlag != kInitCheck)
{
aecm->lastError = AECM_UNINITIALIZED_ERROR;
return -1;
}
if (config.cngMode != AecmFalse && config.cngMode != AecmTrue)
{
aecm->lastError = AECM_BAD_PARAMETER_ERROR;
return -1;
}
aecm->aecmCore->cngMode = config.cngMode;
if (config.echoMode < 0 || config.echoMode > 4)
{
aecm->lastError = AECM_BAD_PARAMETER_ERROR;
return -1;
}
aecm->echoMode = config.echoMode;
if (aecm->echoMode == 0)
{
aecm->aecmCore->supGain = SUPGAIN_DEFAULT >> 3;
aecm->aecmCore->supGainOld = SUPGAIN_DEFAULT >> 3;
aecm->aecmCore->supGainErrParamA = SUPGAIN_ERROR_PARAM_A >> 3;
aecm->aecmCore->supGainErrParamD = SUPGAIN_ERROR_PARAM_D >> 3;
aecm->aecmCore->supGainErrParamDiffAB = (SUPGAIN_ERROR_PARAM_A >> 3)
- (SUPGAIN_ERROR_PARAM_B >> 3);
aecm->aecmCore->supGainErrParamDiffBD = (SUPGAIN_ERROR_PARAM_B >> 3)
- (SUPGAIN_ERROR_PARAM_D >> 3);
} else if (aecm->echoMode == 1)
{
aecm->aecmCore->supGain = SUPGAIN_DEFAULT >> 2;
aecm->aecmCore->supGainOld = SUPGAIN_DEFAULT >> 2;
aecm->aecmCore->supGainErrParamA = SUPGAIN_ERROR_PARAM_A >> 2;
aecm->aecmCore->supGainErrParamD = SUPGAIN_ERROR_PARAM_D >> 2;
aecm->aecmCore->supGainErrParamDiffAB = (SUPGAIN_ERROR_PARAM_A >> 2)
- (SUPGAIN_ERROR_PARAM_B >> 2);
aecm->aecmCore->supGainErrParamDiffBD = (SUPGAIN_ERROR_PARAM_B >> 2)
- (SUPGAIN_ERROR_PARAM_D >> 2);
} else if (aecm->echoMode == 2)
{
aecm->aecmCore->supGain = SUPGAIN_DEFAULT >> 1;
aecm->aecmCore->supGainOld = SUPGAIN_DEFAULT >> 1;
aecm->aecmCore->supGainErrParamA = SUPGAIN_ERROR_PARAM_A >> 1;
aecm->aecmCore->supGainErrParamD = SUPGAIN_ERROR_PARAM_D >> 1;
aecm->aecmCore->supGainErrParamDiffAB = (SUPGAIN_ERROR_PARAM_A >> 1)
- (SUPGAIN_ERROR_PARAM_B >> 1);
aecm->aecmCore->supGainErrParamDiffBD = (SUPGAIN_ERROR_PARAM_B >> 1)
- (SUPGAIN_ERROR_PARAM_D >> 1);
} else if (aecm->echoMode == 3)
{
aecm->aecmCore->supGain = SUPGAIN_DEFAULT;
aecm->aecmCore->supGainOld = SUPGAIN_DEFAULT;
aecm->aecmCore->supGainErrParamA = SUPGAIN_ERROR_PARAM_A;
aecm->aecmCore->supGainErrParamD = SUPGAIN_ERROR_PARAM_D;
aecm->aecmCore->supGainErrParamDiffAB = SUPGAIN_ERROR_PARAM_A - SUPGAIN_ERROR_PARAM_B;
aecm->aecmCore->supGainErrParamDiffBD = SUPGAIN_ERROR_PARAM_B - SUPGAIN_ERROR_PARAM_D;
} else if (aecm->echoMode == 4)
{
aecm->aecmCore->supGain = SUPGAIN_DEFAULT << 1;
aecm->aecmCore->supGainOld = SUPGAIN_DEFAULT << 1;
aecm->aecmCore->supGainErrParamA = SUPGAIN_ERROR_PARAM_A << 1;
aecm->aecmCore->supGainErrParamD = SUPGAIN_ERROR_PARAM_D << 1;
aecm->aecmCore->supGainErrParamDiffAB = (SUPGAIN_ERROR_PARAM_A << 1)
- (SUPGAIN_ERROR_PARAM_B << 1);
aecm->aecmCore->supGainErrParamDiffBD = (SUPGAIN_ERROR_PARAM_B << 1)
- (SUPGAIN_ERROR_PARAM_D << 1);
}
return 0;
}
WebRtc_Word32 WebRtcAecm_get_config(void *aecmInst, AecmConfig *config)
{
aecmob_t *aecm = aecmInst;
if (aecm == NULL)
{
return -1;
}
if (config == NULL)
{
aecm->lastError = AECM_NULL_POINTER_ERROR;
return -1;
}
if (aecm->initFlag != kInitCheck)
{
aecm->lastError = AECM_UNINITIALIZED_ERROR;
return -1;
}
config->cngMode = aecm->aecmCore->cngMode;
config->echoMode = aecm->echoMode;
return 0;
}
WebRtc_Word32 WebRtcAecm_InitEchoPath(void* aecmInst,
const void* echo_path,
size_t size_bytes)
{
aecmob_t *aecm = aecmInst;
const WebRtc_Word16* echo_path_ptr = echo_path;
if ((aecm == NULL) || (echo_path == NULL))
{
aecm->lastError = AECM_NULL_POINTER_ERROR;
return -1;
}
if (size_bytes != WebRtcAecm_echo_path_size_bytes())
{
// Input channel size does not match the size of AECM
aecm->lastError = AECM_BAD_PARAMETER_ERROR;
return -1;
}
if (aecm->initFlag != kInitCheck)
{
aecm->lastError = AECM_UNINITIALIZED_ERROR;
return -1;
}
WebRtcAecm_InitEchoPathCore(aecm->aecmCore, echo_path_ptr);
return 0;
}
WebRtc_Word32 WebRtcAecm_GetEchoPath(void* aecmInst,
void* echo_path,
size_t size_bytes)
{
aecmob_t *aecm = aecmInst;
WebRtc_Word16* echo_path_ptr = echo_path;
if ((aecm == NULL) || (echo_path == NULL))
{
aecm->lastError = AECM_NULL_POINTER_ERROR;
return -1;
}
if (size_bytes != WebRtcAecm_echo_path_size_bytes())
{
// Input channel size does not match the size of AECM
aecm->lastError = AECM_BAD_PARAMETER_ERROR;
return -1;
}
if (aecm->initFlag != kInitCheck)
{
aecm->lastError = AECM_UNINITIALIZED_ERROR;
return -1;
}
memcpy(echo_path_ptr, aecm->aecmCore->channelStored, size_bytes);
return 0;
}
size_t WebRtcAecm_echo_path_size_bytes()
{
return (PART_LEN1 * sizeof(WebRtc_Word16));
}
WebRtc_Word32 WebRtcAecm_get_version(WebRtc_Word8 *versionStr, WebRtc_Word16 len)
{
const char version[] = "AECM 1.2.0";
const short versionLen = (short)strlen(version) + 1; // +1 for null-termination
if (versionStr == NULL)
{
return -1;
}
if (versionLen > len)
{
return -1;
}
strncpy(versionStr, version, versionLen);
return 0;
}
WebRtc_Word32 WebRtcAecm_get_error_code(void *aecmInst)
{
aecmob_t *aecm = aecmInst;
if (aecm == NULL)
{
return -1;
}
return aecm->lastError;
}
static int WebRtcAecm_EstBufDelay(aecmob_t *aecm, short msInSndCardBuf)
{
short delayNew, nSampFar, nSampSndCard;
short diff;
nSampFar = WebRtcApm_get_buffer_size(aecm->farendBuf);
nSampSndCard = msInSndCardBuf * kSampMsNb * aecm->aecmCore->mult;
delayNew = nSampSndCard - nSampFar;
if (delayNew < FRAME_LEN)
{
WebRtcApm_FlushBuffer(aecm->farendBuf, FRAME_LEN);
delayNew += FRAME_LEN;
}
aecm->filtDelay = WEBRTC_SPL_MAX(0, (8 * aecm->filtDelay + 2 * delayNew) / 10);
diff = aecm->filtDelay - aecm->knownDelay;
if (diff > 224)
{
if (aecm->lastDelayDiff < 96)
{
aecm->timeForDelayChange = 0;
} else
{
aecm->timeForDelayChange++;
}
} else if (diff < 96 && aecm->knownDelay > 0)
{
if (aecm->lastDelayDiff > 224)
{
aecm->timeForDelayChange = 0;
} else
{
aecm->timeForDelayChange++;
}
} else
{
aecm->timeForDelayChange = 0;
}
aecm->lastDelayDiff = diff;
if (aecm->timeForDelayChange > 25)
{
aecm->knownDelay = WEBRTC_SPL_MAX((int)aecm->filtDelay - 160, 0);
}
return 0;
}
static int WebRtcAecm_DelayComp(aecmob_t *aecm)
{
int nSampFar, nSampSndCard, delayNew, nSampAdd;
const int maxStuffSamp = 10 * FRAME_LEN;
nSampFar = WebRtcApm_get_buffer_size(aecm->farendBuf);
nSampSndCard = aecm->msInSndCardBuf * kSampMsNb * aecm->aecmCore->mult;
delayNew = nSampSndCard - nSampFar;
if (delayNew > FAR_BUF_LEN - FRAME_LEN * aecm->aecmCore->mult)
{
// The difference of the buffer sizes is larger than the maximum
// allowed known delay. Compensate by stuffing the buffer.
nSampAdd = (int)(WEBRTC_SPL_MAX(((nSampSndCard >> 1) - nSampFar),
FRAME_LEN));
nSampAdd = WEBRTC_SPL_MIN(nSampAdd, maxStuffSamp);
WebRtcApm_StuffBuffer(aecm->farendBuf, nSampAdd);
aecm->delayChange = 1; // the delay needs to be updated
}
return 0;
}

250
webrtc/modules/audio_processing/aecm/interface/echo_control_mobile.h Normal file
View File

@ -0,0 +1,250 @@
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_INTERFACE_ECHO_CONTROL_MOBILE_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_INTERFACE_ECHO_CONTROL_MOBILE_H_
#include "typedefs.h"
enum {
AecmFalse = 0,
AecmTrue
};
// Errors
#define AECM_UNSPECIFIED_ERROR 12000
#define AECM_UNSUPPORTED_FUNCTION_ERROR 12001
#define AECM_UNINITIALIZED_ERROR 12002
#define AECM_NULL_POINTER_ERROR 12003
#define AECM_BAD_PARAMETER_ERROR 12004
// Warnings
#define AECM_BAD_PARAMETER_WARNING 12100
typedef struct {
WebRtc_Word16 cngMode; // AECM_FALSE, AECM_TRUE (default)
WebRtc_Word16 echoMode; // 0, 1, 2, 3 (default), 4
} AecmConfig;
#ifdef __cplusplus
extern "C" {
#endif
/*
* Allocates the memory needed by the AECM. The memory needs to be
* initialized separately using the WebRtcAecm_Init() function.
*
* Inputs Description
* -------------------------------------------------------------------
* void **aecmInst Pointer to the AECM instance to be
* created and initialized
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_Create(void **aecmInst);
/*
* This function releases the memory allocated by WebRtcAecm_Create()
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecmInst Pointer to the AECM instance
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_Free(void *aecmInst);
/*
* Initializes an AECM instance.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecmInst Pointer to the AECM instance
* WebRtc_Word32 sampFreq Sampling frequency of data
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_Init(void* aecmInst,
WebRtc_Word32 sampFreq);
/*
* Inserts an 80 or 160 sample block of data into the farend buffer.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecmInst Pointer to the AECM instance
* WebRtc_Word16 *farend In buffer containing one frame of
* farend signal
* WebRtc_Word16 nrOfSamples Number of samples in farend buffer
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_BufferFarend(void* aecmInst,
const WebRtc_Word16* farend,
WebRtc_Word16 nrOfSamples);
/*
* Runs the AECM on an 80 or 160 sample blocks of data.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecmInst Pointer to the AECM instance
* WebRtc_Word16 *nearendNoisy In buffer containing one frame of
* reference nearend+echo signal. If
* noise reduction is active, provide
* the noisy signal here.
* WebRtc_Word16 *nearendClean In buffer containing one frame of
* nearend+echo signal. If noise
* reduction is active, provide the
* clean signal here. Otherwise pass a
* NULL pointer.
* WebRtc_Word16 nrOfSamples Number of samples in nearend buffer
* WebRtc_Word16 msInSndCardBuf Delay estimate for sound card and
* system buffers
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word16 *out Out buffer, one frame of processed nearend
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_Process(void* aecmInst,
const WebRtc_Word16* nearendNoisy,
const WebRtc_Word16* nearendClean,
WebRtc_Word16* out,
WebRtc_Word16 nrOfSamples,
WebRtc_Word16 msInSndCardBuf);
/*
* This function enables the user to set certain parameters on-the-fly
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecmInst Pointer to the AECM instance
* AecmConfig config Config instance that contains all
* properties to be set
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_set_config(void* aecmInst,
AecmConfig config);
/*
* This function enables the user to set certain parameters on-the-fly
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecmInst Pointer to the AECM instance
*
* Outputs Description
* -------------------------------------------------------------------
* AecmConfig *config Pointer to the config instance that
* all properties will be written to
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_get_config(void *aecmInst,
AecmConfig *config);
/*
* This function enables the user to set the echo path on-the-fly.
*
* Inputs Description
* -------------------------------------------------------------------
* void* aecmInst Pointer to the AECM instance
* void* echo_path Pointer to the echo path to be set
* size_t size_bytes Size in bytes of the echo path
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_InitEchoPath(void* aecmInst,
const void* echo_path,
size_t size_bytes);
/*
* This function enables the user to get the currently used echo path
* on-the-fly
*
* Inputs Description
* -------------------------------------------------------------------
* void* aecmInst Pointer to the AECM instance
* void* echo_path Pointer to echo path
* size_t size_bytes Size in bytes of the echo path
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_GetEchoPath(void* aecmInst,
void* echo_path,
size_t size_bytes);
/*
* This function enables the user to get the echo path size in bytes
*
* Outputs Description
* -------------------------------------------------------------------
* size_t return : size in bytes
*/
size_t WebRtcAecm_echo_path_size_bytes();
/*
* Gets the last error code.
*
* Inputs Description
* -------------------------------------------------------------------
* void *aecmInst Pointer to the AECM instance
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word32 return 11000-11100: error code
*/
WebRtc_Word32 WebRtcAecm_get_error_code(void *aecmInst);
/*
* Gets a version string
*
* Inputs Description
* -------------------------------------------------------------------
* char *versionStr Pointer to a string array
* WebRtc_Word16 len The maximum length of the string
*
* Outputs Description
* -------------------------------------------------------------------
* WebRtc_Word8 *versionStr Pointer to a string array
* WebRtc_Word32 return 0: OK
* -1: error
*/
WebRtc_Word32 WebRtcAecm_get_version(WebRtc_Word8 *versionStr,
WebRtc_Word16 len);
#ifdef __cplusplus
}
#endif
#endif /* WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_INTERFACE_ECHO_CONTROL_MOBILE_H_ */

10
webrtc/modules/audio_processing/agc/Makefile.am Normal file
View File

@ -0,0 +1,10 @@
noinst_LTLIBRARIES = libagc.la
libagc_la_SOURCES = interface/gain_control.h \
analog_agc.c \
analog_agc.h \
digital_agc.c \
digital_agc.h
libagc_la_CFLAGS = $(AM_CFLAGS) $(COMMON_CFLAGS) \
-I$(top_srcdir)/src/common_audio/signal_processing_library/main/interface \
-I$(top_srcdir)/src/modules/audio_processing/utility

34
webrtc/modules/audio_processing/agc/agc.gypi Normal file
View File

@ -0,0 +1,34 @@
# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
{
'targets': [
{
'target_name': 'agc',
'type': '<(library)',
'dependencies': [
'<(webrtc_root)/common_audio/common_audio.gyp:spl',
],
'include_dirs': [
'interface',
],
'direct_dependent_settings': {
'include_dirs': [
'interface',
],
},
'sources': [
'interface/gain_control.h',
'analog_agc.c',
'analog_agc.h',
'digital_agc.c',
'digital_agc.h',
],
},
],
}

Some files were not shown because too many files have changed in this diff Show More