Bart/bullet3
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

fixes in the DX11 cloth (for devices with physical wavefronts smaller than 32)

Browse Source 
another fix for out-of-bounds check (there was no assert for this previously)
This commit is contained in:
erwin.coumans
2011-08-03 20:21:23 +00:00
parent cf0d948d97
commit e83b5dac75
2 changed files with 21 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
src/BulletMultiThreaded/GpuSoftBodySolvers/DX11/HLSL/SolvePositionsSIMDBatched.hlsl
View File

@@ -1,5 +1,7 @@
MSTRINGIFY( MSTRINGIFY(
cbuffer SolvePositionsFromLinksKernelCB : register( b0 ) cbuffer SolvePositionsFromLinksKernelCB : register( b0 )
{ {
int startWaveInBatch; int startWaveInBatch;
@@ -41,16 +43,20 @@ SolvePositionsFromLinksKernel( uint3 Gid : SV_GroupID, uint3 DTid : SV_DispatchT
const int firstWavefrontInBlock = startWaveInBatch + Gid.x * WAVEFRONT_BLOCK_MULTIPLIER; const int firstWavefrontInBlock = startWaveInBatch + Gid.x * WAVEFRONT_BLOCK_MULTIPLIER;
const int localWavefront = wavefront - firstWavefrontInBlock; const int localWavefront = wavefront - firstWavefrontInBlock;
int batchesWithinWavefront = 0;
int verticesUsedByWave = 0;
int cond = wavefront < (startWaveInBatch + numWaves);
// Mask out in case there's a stray "wavefront" at the end that's been forced in through the multiplier // Mask out in case there's a stray "wavefront" at the end that's been forced in through the multiplier
if( wavefront < (startWaveInBatch + numWaves) ) if( cond)
{ {
// Load the batch counts for the wavefronts // Load the batch counts for the wavefronts
int2 batchesAndVerticesWithinWavefront = g_wavefrontBatchCountsVertexCounts[wavefront]; int2 batchesAndVerticesWithinWavefront = g_wavefrontBatchCountsVertexCounts[wavefront];
int batchesWithinWavefront = batchesAndVerticesWithinWavefront.x; batchesWithinWavefront = batchesAndVerticesWithinWavefront.x;
int verticesUsedByWave = batchesAndVerticesWithinWavefront.y; verticesUsedByWave = batchesAndVerticesWithinWavefront.y;
// Load the vertices for the wavefronts // Load the vertices for the wavefronts
for( int vertex = laneInWavefront; vertex < verticesUsedByWave; vertex+=WAVEFRONT_SIZE ) for( int vertex = laneInWavefront; vertex < verticesUsedByWave; vertex+=WAVEFRONT_SIZE )
@@ -62,10 +68,13 @@ SolvePositionsFromLinksKernel( uint3 Gid : SV_GroupID, uint3 DTid : SV_DispatchT
vertexInverseMassSharedData[localWavefront*MAX_NUM_VERTICES_PER_WAVE + vertex] = g_verticesInverseMass[vertexAddress]; vertexInverseMassSharedData[localWavefront*MAX_NUM_VERTICES_PER_WAVE + vertex] = g_verticesInverseMass[vertexAddress];
} }
}
// Ensure compiler does not re-order memory operations // Ensure compiler does not re-order memory operations
AllMemoryBarrier(); //AllMemoryBarrier();
AllMemoryBarrierWithGroupSync ();
if( cond)
{
// Loop through the batches performing the solve on each in LDS // Loop through the batches performing the solve on each in LDS
int baseDataLocationForWave = WAVEFRONT_SIZE * wavefront * MAX_BATCHES_PER_WAVE; int baseDataLocationForWave = WAVEFRONT_SIZE * wavefront * MAX_BATCHES_PER_WAVE;
@@ -128,6 +137,11 @@ SolvePositionsFromLinksKernel( uint3 Gid : SV_GroupID, uint3 DTid : SV_DispatchT
} }
} }
} }
);
);

2
src/BulletMultiThreaded/GpuSoftBodySolvers/DX11/btSoftBodySolver_DX11SIMDAware.cpp
View File

@@ -620,7 +620,7 @@ static void generateBatchesOfWavefronts( btAlignedObjectArray < btAlignedObjectA
mapOfVerticesInBatches.resize( batch + 1 ); mapOfVerticesInBatches.resize( batch + 1 );
// Resize maps with total number of vertices // Resize maps with total number of vertices
mapOfVerticesInBatches[batch].resize( numVertices, false ); mapOfVerticesInBatches[batch].resize( numVertices+1, false );
// Insert vertices into this batch too // Insert vertices into this batch too
for( int link = 0; link < wavefront.size(); ++link ) for( int link = 0; link < wavefront.size(); ++link )