7f96005985
Commented out optimized path in sampler as it doesn't work now that we are using Morton ordering for the data in chunks. However, we can probably reinstate such a fast path if we give some thought as to how it should be done.
2015-03-29 09:58:28 +02:00
d3b2dab1ac
Switched to using lookup tables for Morton encoding.
2015-03-29 00:17:27 +01:00
d99ed5e624
Implemented morton encoding using standard bit-twidling approach.
2015-03-28 09:46:05 +01:00
1c17a7147b
Made test do 10 times more iterations.
2015-03-28 08:42:06 +01:00
9256f3deb5
Fixed compile warning.
2015-03-27 21:23:31 +01:00
322bedc009
Improved test for PagedVolume::Chunk performance.
2015-03-26 23:45:01 +01:00
1d24b189ca
Refactoed test code.
2015-03-25 17:12:11 +01:00
931c6cd3ec
Added tests for chunk performance.
2015-03-25 16:40:05 +01:00
6516c00e62
Merge branch 'feature/optimize-paged-volume' into develop
2015-03-21 16:30:07 +01:00
3facd4df41
Removed commented out code.
2015-03-21 16:27:43 +01:00
d6c708f869
Merge branch 'develop' into feature/optimize-paged-volume
2015-03-21 16:24:46 +01:00
b027cf1a0c
Moved common code into function.
2015-03-21 16:22:23 +01:00
778238d11d
Moved the test for whether we are accessing the same voxel as last time.
2015-03-21 14:57:48 +01:00
d477bec540
Revert "Replaced Vector3D with integer as key to map."
...
This reverts commit e82d6beca1
2015-03-21 14:41:15 +01:00
672c375a7a
Revert "Work on using a bitfield to set up chunk key."
...
This reverts commit 8bd8f8ba7a
2015-03-21 14:41:10 +01:00
5fc0317260
Revert "Added typedef for chunk key type."
...
This reverts commit 6419c5827b
2015-03-21 14:41:04 +01:00
92eaaae765
Revert "Decided to always use a 64-bit chunk key, rather than trying to make it configurable."
...
This reverts commit 69f6f4ac37
2015-03-21 14:40:57 +01:00
ceeb8f70ce
Revert "Going back to building key by shifting instead of using bitfield."
...
This reverts commit 0d638f9837
2015-03-21 14:40:49 +01:00
0c619ebec7
Revert "Added typedef for ChunkKey."
...
This reverts commit 905ec27f47
2015-03-21 14:40:30 +01:00
f574563672
Revert "New, safer method of packing which makes careful use of casting to avoid problems with e.g. signed integer sign extension."
...
This reverts commit fd451be2dd
2015-03-21 14:40:11 +01:00
fd451be2dd
New, safer method of packing which makes careful use of casting to avoid problems with e.g. signed integer sign extension.
2015-03-21 08:48:45 +01:00
905ec27f47
Added typedef for ChunkKey.
2015-03-21 08:12:02 +01:00
0d638f9837
Going back to building key by shifting instead of using bitfield.
2015-03-21 08:05:58 +01:00
0e995b5140
Fixed some compiler warnings.
2015-03-21 07:40:32 +01:00
69f6f4ac37
Decided to always use a 64-bit chunk key, rather than trying to make it configurable.
2015-03-20 23:09:38 +01:00
6419c5827b
Added typedef for chunk key type.
2015-03-20 16:59:25 +01:00
8bd8f8ba7a
Work on using a bitfield to set up chunk key.
2015-03-20 15:48:46 +01:00
cd48c6f611
Made method const.
2015-03-19 23:08:35 +01:00
e82d6beca1
Replaced Vector3D with integer as key to map.
...
Chunks of voxel data are stored in a map, and it is quite common to need to search the map for a particular chunk. The key type used to be a Vector3D (i.e. the position of the chunk in 3D space) which makes conceptual sense but is relatively slow. Using a Vector3D as a key seems to have overhead, probably in terms of copying and performing comparisons. It seems to be significantly faster to use an integer as a key, so we now take the 3D position and pack it into a single integer by bitshifting.
Naturally this reduces the range of positions we can store - a 32-bit int can only encode 3 x 10-bit values, which means our volume can only be 1024 chunks in each direction (with a chunk often being 32x32x32 voxels). This should still be large enough for most uses, but an upcoming change will allow 64-bit keys to be used (at least on 64-bit builds) which then allows 21 bits of precision per component. This is so large that it's almost infinite for any practical purposes.
2015-03-15 09:32:42 +01:00
d305038c27
Replaced loop with fill.
2015-03-11 23:47:32 +01:00
741234e4a5
Small speed improvement by storing variables separately (rather than in Vector3D) to void construction/comparison overhead.
2015-03-09 23:52:56 +01:00
99d0a226c8
Tidying up.
2015-03-08 23:48:55 +01:00
72abcd8e9c
Chunks are now stored with unique_ptr rather than shared_ptr.
2015-03-08 23:30:12 +01:00
639eb6620f
Add 'this' pointer disambiguation.
...
I guess this didn't fail on MSVC due to the different name lookup rules
the use.
Fixes issue 56.
2015-03-08 12:51:00 +00:00
c98ad8c948
Merge branch 'feature/remove-wrap-modes' into develop
2015-03-07 21:19:21 +01:00
43c203a2c9
Removed declaration of function for which we already removed the definition.
2015-03-07 17:42:31 +01:00
aaa6b1dc15
A PagedVolume must now always be provided with a Pager when it is constructed.
2015-03-07 17:01:07 +01:00
5a5b2b3875
Spotted an optimization for computing the voxel position.
2015-03-06 23:03:38 +01:00
2c0d9cb9e7
Moved contents of initialize() into constructor.
2015-03-05 23:51:16 +01:00
c804190d84
Removed the ability to set the target memory usage after construction, and this is now just done through the constructor. I don't think the functionality was useful, and this simplifies more logic.
2015-03-05 23:42:50 +01:00
89550fcd44
Fixed warnings.
2015-03-05 00:04:22 +01:00
61bffc9783
Removed some not useful functions.
2015-03-04 23:31:24 +01:00
1213a4047a
Fully stripped out references to wrap modes.
2015-03-04 23:13:37 +01:00
bd6efe8c3c
Stuff related to valid regions is being moved from BaseVolum to RawVolume, as PagedVolume is now infinite.
2015-03-04 22:42:14 +01:00
d3618ca688
Removed tracking of whether the PagedVolume::Sampler is currently valid.
2015-03-01 23:32:22 +01:00
396d1cfc59
Ambient occlusion test now uses RawVolume, as it need a fixed size volume to create a temporary array.
2015-03-01 09:51:45 +01:00
e7f41b7e0f
More fixing/tidying of test, related to removing volume size limit.
2015-03-01 08:14:30 +01:00
bfc7dfdc1b
Removed the ability to specify a region in the PagedVolume's constructor, and updated the tests and examples where required.
2015-02-28 23:31:23 +01:00
7e03c3e05b
Compile fixes and tweaks for tests.
2015-02-28 13:43:17 +01:00
09c15173f9
Simplify -std=c++0x flag logic
...
This could be improved in CMake 3.1 by using CXX_STANDARD and
CXX_STANDARD_REQUIRED (see http://stackoverflow.com/a/20165220/96704 )
2015-02-28 10:55:01 +00:00
