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

Add the GPU rigid body pipeline from https://github.com/erwincoumans/experiments as a Bullet 3.x preview for Bullet 2.80

Browse Source
This commit is contained in:
erwin.coumans
2012-03-05 00:54:32 +00:00
parent 73c4646b40
commit 571af41cf6
257 changed files with 55106 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

73
Extras/RigidBodyGpuPipeline/opencl/primitives/AdlPrimitives/Search/BoundSearch.h Normal file
View File

@@ -0,0 +1,73 @@
/*
Copyright (c) 2012 Advanced Micro Devices, Inc.
This software is provided 'as-is', without any express or implied warranty.
In no event will the authors be held liable for any damages arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it freely,
subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
//Originally written by Takahiro Harada
#pragma once
#include <Adl/Adl.h>
#include <AdlPrimitives/Math/Math.h>
#include <AdlPrimitives/Sort/SortData.h>
#include <AdlPrimitives/Fill/Fill.h>
namespace adl
{
class BoundSearchBase
{
public:
enum Option
{
BOUND_LOWER,
BOUND_UPPER,
COUNT,
};
};
template<DeviceType TYPE>
class BoundSearch : public BoundSearchBase
{
public:
typedef Launcher::BufferInfo BufferInfo;
struct Data
{
const Device* m_device;
Kernel* m_lowerSortDataKernel;
Kernel* m_upperSortDataKernel;
Kernel* m_subtractKernel;
Buffer<int4>* m_constBuffer;
Buffer<u32>* m_lower;
Buffer<u32>* m_upper;
typename Fill<TYPE>::Data* m_fillData;
};
static
Data* allocate(const Device* deviceData, int maxSize = 0);
static
void deallocate(Data* data);
// src has to be src[i].m_key <= src[i+1].m_key
static
void execute(Data* data, Buffer<SortData>& src, u32 nSrc, Buffer<u32>& dst, u32 nDst, Option option = BOUND_LOWER );
// static
// void execute(Data* data, Buffer<u32>& src, Buffer<u32>& dst, int n, Option option = );
};
#include <AdlPrimitives/Search/BoundSearchHost.inl>
#include <AdlPrimitives/Search/BoundSearch.inl>
};

128
Extras/RigidBodyGpuPipeline/opencl/primitives/AdlPrimitives/Search/BoundSearch.inl Normal file
View File

@@ -0,0 +1,128 @@
/*
Copyright (c) 2012 Advanced Micro Devices, Inc.
This software is provided 'as-is', without any express or implied warranty.
In no event will the authors be held liable for any damages arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it freely,
subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
//Originally written by Takahiro Harada
#define PATH "..\\..\\opencl\\primitives\\AdlPrimitives\\Search\\BoundSearchKernels"
#define KERNEL0 "SearchSortDataLowerKernel"
#define KERNEL1 "SearchSortDataUpperKernel"
#define KERNEL2 "SubtractKernel"
#include <AdlPrimitives/Search/BoundSearchKernelsCL.h>
#include <AdlPrimitives/Search/BoundSearchKernelsDX11.h>
template<DeviceType TYPE>
typename BoundSearch<TYPE>::Data* BoundSearch<TYPE>::allocate(const Device* device, int maxSize)
{
ADLASSERT( TYPE == device->m_type );
const char* src[] =
#if defined(ADL_LOAD_KERNEL_FROM_STRING)
{boundSearchKernelsCL, boundSearchKernelsDX11};
#else
{0,0};
#endif
Data* data = new Data;
data->m_device = device;
data->m_lowerSortDataKernel = device->getKernel( PATH, KERNEL0, 0, src[TYPE] );
data->m_upperSortDataKernel = device->getKernel( PATH, KERNEL1, 0, src[TYPE] );
data->m_constBuffer = new Buffer<int4>( device, 1, BufferBase::BUFFER_CONST );
if( maxSize )
{
data->m_subtractKernel = device->getKernel( PATH, KERNEL2, 0, src[TYPE] );
}
data->m_lower = (maxSize == 0)? 0: new Buffer<u32>( device, maxSize );
data->m_upper = (maxSize == 0)? 0: new Buffer<u32>( device, maxSize );
data->m_fillData = (maxSize == 0)? 0: Fill<TYPE>::allocate( device );
return data;
}
template<DeviceType TYPE>
void BoundSearch<TYPE>::deallocate(Data* data)
{
delete data->m_constBuffer;
if( data->m_lower ) delete data->m_lower;
if( data->m_upper ) delete data->m_upper;
if( data->m_fillData ) Fill<TYPE>::deallocate( data->m_fillData );
delete data;
}
template<DeviceType TYPE>
void BoundSearch<TYPE>::execute(Data* data, Buffer<SortData>& src, u32 nSrc, Buffer<u32>& dst, u32 nDst, Option option )
{
int4 constBuffer;
constBuffer.x = nSrc;
constBuffer.y = nDst;
Buffer<SortData>* srcNative = BufferUtils::map<TYPE, true>( data->m_device, &src );
Buffer<u32>* dstNative = BufferUtils::map<TYPE, false>( data->m_device, &dst );
if( option == BOUND_LOWER )
{
BufferInfo bInfo[] = { BufferInfo( srcNative, true ), BufferInfo( dstNative ) };
Launcher launcher( data->m_device, data->m_lowerSortDataKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(Launcher::BufferInfo) );
launcher.setConst( *data->m_constBuffer, constBuffer );
launcher.launch1D( nSrc, 64 );
}
else if( option == BOUND_UPPER )
{
BufferInfo bInfo[] = { BufferInfo( srcNative, true ), BufferInfo( dstNative ) };
Launcher launcher( data->m_device, data->m_upperSortDataKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(Launcher::BufferInfo) );
launcher.setConst( *data->m_constBuffer, constBuffer );
launcher.launch1D( nSrc+1, 64 );
}
else if( option == COUNT )
{
ADLASSERT( data->m_lower );
ADLASSERT( data->m_upper );
ADLASSERT( data->m_lower->getSize() <= (int)nDst );
ADLASSERT( data->m_upper->getSize() <= (int)nDst );
int zero = 0;
Fill<TYPE>::execute( data->m_fillData, (Buffer<int>&)*data->m_lower, zero, nDst );
Fill<TYPE>::execute( data->m_fillData, (Buffer<int>&)*data->m_upper, zero, nDst );
execute( data, src, nSrc, *data->m_lower, nDst, BOUND_LOWER );
execute( data, src, nSrc, *data->m_upper, nDst, BOUND_UPPER );
{
BufferInfo bInfo[] = { BufferInfo( data->m_upper, true ), BufferInfo( data->m_lower, true ), BufferInfo( dstNative ) };
Launcher launcher( data->m_device, data->m_subtractKernel );
launcher.setBuffers( bInfo, sizeof(bInfo)/sizeof(Launcher::BufferInfo) );
launcher.setConst( *data->m_constBuffer, constBuffer );
launcher.launch1D( nDst, 64 );
}
}
else
{
ADLASSERT( 0 );
}
BufferUtils::unmap<false>( srcNative, &src );
BufferUtils::unmap<true>( dstNative, &dst );
}
#undef PATH
#undef KERNEL0
#undef KERNEL1
#undef KERNEL2

111
Extras/RigidBodyGpuPipeline/opencl/primitives/AdlPrimitives/Search/BoundSearchHost.inl Normal file
View File

@@ -0,0 +1,111 @@
/*
Copyright (c) 2012 Advanced Micro Devices, Inc.
This software is provided 'as-is', without any express or implied warranty.
In no event will the authors be held liable for any damages arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it freely,
subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
//Originally written by Takahiro Harada
template<>
class BoundSearch<TYPE_HOST> : public BoundSearchBase
{
public:
typedef Launcher::BufferInfo BufferInfo;
struct Data
{
const Device* m_device;
};
static
Data* allocate(const Device* deviceData, int maxSize = 0)
{
ADLASSERT( deviceData->m_type == TYPE_HOST );
Data* data = new Data;
data->m_device = deviceData;
return data;
}
static
void deallocate(Data* data)
{
delete data;
}
static
void execute(Data* data, Buffer<SortData>& rawSrc, u32 nSrc, Buffer<u32>& rawDst, u32 nDst, Option option = BOUND_LOWER)
{
ADLASSERT( rawSrc.getType() == TYPE_HOST );
ADLASSERT( rawDst.getType() == TYPE_HOST );
HostBuffer<SortData>& src = *(HostBuffer<SortData>*)&rawSrc;
HostBuffer<u32>& dst = *(HostBuffer<u32>*)&rawDst;
for(int i=0; i<nSrc-1; i++)
ADLASSERT( src[i].m_key <= src[i+1].m_key );
if( option == BOUND_LOWER )
{
for(u32 i=0; i<nSrc; i++)
{
SortData& iData = (i==0)? SortData(-1,-1): src[i-1];
SortData& jData = (i==nSrc)? SortData(nDst, nDst): src[i];
if( iData.m_key != jData.m_key )
{
// for(u32 k=iData.m_key+1; k<=min(jData.m_key,nDst-1); k++)
u32 k = jData.m_key;
{
dst[k] = i;
}
}
}
}
else if( option == BOUND_UPPER )
{
for(u32 i=0; i<nSrc+1; i++)
{
SortData& iData = (i==0)? SortData(0,0): src[i-1];
SortData& jData = (i==nSrc)? SortData(nDst, nDst): src[i];
if( iData.m_key != jData.m_key )
{
// for(u32 k=iData.m_key; k<min(jData.m_key,nDst); k++)
u32 k = iData.m_key;
{
dst[k] = i;
}
}
}
}
else if( option == COUNT )
{
HostBuffer<u32> lower( data->m_device, nDst );
HostBuffer<u32> upper( data->m_device, nDst );
for(u32 i=0; i<nDst; i++) { lower[i] = upper[i] = 0; }
execute( data, rawSrc, nSrc, lower, nDst, BOUND_LOWER );
execute( data, rawSrc, nSrc, upper, nDst, BOUND_UPPER );
for(u32 i=0; i<nDst; i++) { dst[i] = upper[i] - lower[i]; }
}
else
{
ADLASSERT( 0 );
}
}
// static
// void execute(Data* data, Buffer<u32>& src, Buffer<u32>& dst, int n, Option option = );
};

112
Extras/RigidBodyGpuPipeline/opencl/primitives/AdlPrimitives/Search/BoundSearchKernels.cl Normal file
View File

@@ -0,0 +1,112 @@
/*
Copyright (c) 2012 Advanced Micro Devices, Inc.
This software is provided 'as-is', without any express or implied warranty.
In no event will the authors be held liable for any damages arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it freely,
subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
//Originally written by Takahiro Harada
typedef unsigned int u32;
#define GET_GROUP_IDX get_group_id(0)
#define GET_LOCAL_IDX get_local_id(0)
#define GET_GLOBAL_IDX get_global_id(0)
#define GET_GROUP_SIZE get_local_size(0)
#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)
typedef struct
{
u32 m_key;
u32 m_value;
}SortData;
typedef struct
{
u32 m_nSrc;
u32 m_nDst;
u32 m_padding[2];
} ConstBuffer;
__attribute__((reqd_work_group_size(64,1,1)))
__kernel
void SearchSortDataLowerKernel(__global SortData* src, __global u32 *dst,
ConstBuffer cb)
{
int gIdx = GET_GLOBAL_IDX;
u32 nSrc = cb.m_nSrc;
u32 nDst = cb.m_nDst;
if( gIdx < nSrc )
{
SortData first; first.m_key = (u32)(-1); first.m_value = (u32)(-1);
SortData end; end.m_key = nDst; end.m_value = nDst;
SortData iData = (gIdx==0)? first: src[gIdx-1];
SortData jData = (gIdx==nSrc)? end: src[gIdx];
if( iData.m_key != jData.m_key )
{
// for(u32 k=iData.m_key+1; k<=min(jData.m_key, nDst-1); k++)
u32 k = jData.m_key;
{
dst[k] = gIdx;
}
}
}
}
__attribute__((reqd_work_group_size(64,1,1)))
__kernel
void SearchSortDataUpperKernel(__global SortData* src, __global u32 *dst,
ConstBuffer cb)
{
int gIdx = GET_GLOBAL_IDX;
u32 nSrc = cb.m_nSrc;
u32 nDst = cb.m_nDst;
if( gIdx < nSrc+1 )
{
SortData first; first.m_key = 0; first.m_value = 0;
SortData end; end.m_key = nDst; end.m_value = nDst;
SortData iData = (gIdx==0)? first: src[gIdx-1];
SortData jData = (gIdx==nSrc)? end: src[gIdx];
if( iData.m_key != jData.m_key )
{
// for(u32 k=iData.m_key; k<min(jData.m_key, nDst); k++)
u32 k = iData.m_key;
{
dst[k] = gIdx;
}
}
}
}
__attribute__((reqd_work_group_size(64,1,1)))
__kernel
void SubtractKernel(__global u32* A, __global u32 *B, __global u32 *C,
ConstBuffer cb)
{
int gIdx = GET_GLOBAL_IDX;
u32 nSrc = cb.m_nSrc;
u32 nDst = cb.m_nDst;
if( gIdx < nDst )
{
C[gIdx] = A[gIdx] - B[gIdx];
}
}

104
Extras/RigidBodyGpuPipeline/opencl/primitives/AdlPrimitives/Search/BoundSearchKernels.hlsl Normal file
View File

@@ -0,0 +1,104 @@
/*
Copyright (c) 2012 Advanced Micro Devices, Inc.
This software is provided 'as-is', without any express or implied warranty.
In no event will the authors be held liable for any damages arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it freely,
subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
//Originally written by Takahiro Harada
typedef uint u32;
#define GET_GROUP_IDX groupIdx.x
#define GET_LOCAL_IDX localIdx.x
#define GET_GLOBAL_IDX globalIdx.x
#define GROUP_LDS_BARRIER GroupMemoryBarrierWithGroupSync()
#define DEFAULT_ARGS uint3 globalIdx : SV_DispatchThreadID, uint3 localIdx : SV_GroupThreadID, uint3 groupIdx : SV_GroupID
#define AtomInc(x) InterlockedAdd(x, 1)
#define AtomInc1(x, out) InterlockedAdd(x, 1, out)
typedef struct
{
u32 m_key;
u32 m_value;
}SortData;
cbuffer SortCB : register( b0 )
{
u32 m_nSrc;
u32 m_nDst;
u32 m_padding[2];
};
StructuredBuffer<SortData> src : register( t0 );
RWStructuredBuffer<u32> dst : register( u0 );
[numthreads(64, 1, 1)]
void SearchSortDataLowerKernel( DEFAULT_ARGS )
{
int gIdx = GET_GLOBAL_IDX;
u32 nSrc = m_nSrc;
u32 nDst = m_nDst;
if( gIdx < nSrc )
{
SortData iData;
SortData jData;
if( gIdx==0 ) iData.m_key = iData.m_value = (u32)-1;
else iData = src[gIdx-1];
if( gIdx==nSrc ) jData.m_key = jData.m_value = nDst;
else jData = src[gIdx];
if( iData.m_key != jData.m_key )
{
// for(u32 k=iData.m_key+1; k<=min(jData.m_key, nDst-1); k++)
u32 k = jData.m_key;
{
dst[k] = gIdx;
}
}
}
}
[numthreads(64, 1, 1)]
void SearchSortDataUpperKernel( DEFAULT_ARGS )
{
int gIdx = GET_GLOBAL_IDX;
u32 nSrc = m_nSrc;
u32 nDst = m_nDst;
if( gIdx < nSrc+1 )
{
SortData iData;
SortData jData;
if( gIdx==0 ) iData.m_key = iData.m_value = 0;
else iData = src[gIdx-1];
if( gIdx==nSrc ) jData.m_key = jData.m_value = nDst;
else jData = src[gIdx];
if( iData.m_key != jData.m_key )
{
// for(u32 k=iData.m_key; k<min(jData.m_key, nDst); k++)
u32 k = iData.m_key;
{
dst[k] = gIdx;
}
}
}
}

102
Extras/RigidBodyGpuPipeline/opencl/primitives/AdlPrimitives/Search/BoundSearchKernelsCL.h Normal file
View File

@@ -0,0 +1,102 @@
static const char* boundSearchKernelsCL= \
"/*\n"
" 2011 Takahiro Harada\n"
"*/\n"
"\n"
"typedef unsigned int u32;\n"
"#define GET_GROUP_IDX get_group_id(0)\n"
"#define GET_LOCAL_IDX get_local_id(0)\n"
"#define GET_GLOBAL_IDX get_global_id(0)\n"
"#define GET_GROUP_SIZE get_local_size(0)\n"
"#define GROUP_LDS_BARRIER barrier(CLK_LOCAL_MEM_FENCE)\n"
"\n"
"typedef struct\n"
"{\n"
" u32 m_key; \n"
" u32 m_value;\n"
"}SortData;\n"
"\n"
"\n"
"\n"
"typedef struct\n"
"{\n"
" u32 m_nSrc;\n"
" u32 m_nDst;\n"
" u32 m_padding[2];\n"
"} ConstBuffer;\n"
"\n"
"\n"
"\n"
"__attribute__((reqd_work_group_size(64,1,1)))\n"
"__kernel\n"
"void SearchSortDataLowerKernel(__global SortData* src, __global u32 *dst, \n"
" ConstBuffer cb)\n"
"{\n"
" int gIdx = GET_GLOBAL_IDX;\n"
" u32 nSrc = cb.m_nSrc;\n"
" u32 nDst = cb.m_nDst;\n"
"\n"
" if( gIdx < nSrc )\n"
" {\n"
" SortData first; first.m_key = (u32)(-1); first.m_value = (u32)(-1);\n"
" SortData end; end.m_key = nDst; end.m_value = nDst;\n"
"\n"
" SortData iData = (gIdx==0)? first: src[gIdx-1];\n"
" SortData jData = (gIdx==nSrc)? end: src[gIdx];\n"
"\n"
" if( iData.m_key != jData.m_key )\n"
" {\n"
"// for(u32 k=iData.m_key+1; k<=min(jData.m_key, nDst-1); k++)\n"
" u32 k = jData.m_key;\n"
" {\n"
" dst[k] = gIdx;\n"
" }\n"
" }\n"
" }\n"
"}\n"
"\n"
"\n"
"__attribute__((reqd_work_group_size(64,1,1)))\n"
"__kernel\n"
"void SearchSortDataUpperKernel(__global SortData* src, __global u32 *dst, \n"
" ConstBuffer cb)\n"
"{\n"
" int gIdx = GET_GLOBAL_IDX;\n"
" u32 nSrc = cb.m_nSrc;\n"
" u32 nDst = cb.m_nDst;\n"
"\n"
" if( gIdx < nSrc+1 )\n"
" {\n"
" SortData first; first.m_key = 0; first.m_value = 0;\n"
" SortData end; end.m_key = nDst; end.m_value = nDst;\n"
"\n"
" SortData iData = (gIdx==0)? first: src[gIdx-1];\n"
" SortData jData = (gIdx==nSrc)? end: src[gIdx];\n"
"\n"
" if( iData.m_key != jData.m_key )\n"
" {\n"
"// for(u32 k=iData.m_key; k<min(jData.m_key, nDst); k++)\n"
" u32 k = iData.m_key;\n"
" {\n"
" dst[k] = gIdx;\n"
" }\n"
" }\n"
" }\n"
"}\n"
"\n"
"__attribute__((reqd_work_group_size(64,1,1)))\n"
"__kernel\n"
"void SubtractKernel(__global u32* A, __global u32 *B, __global u32 *C, \n"
" ConstBuffer cb)\n"
"{\n"
" int gIdx = GET_GLOBAL_IDX;\n"
" u32 nSrc = cb.m_nSrc;\n"
" u32 nDst = cb.m_nDst;\n"
"\n"
" if( gIdx < nDst )\n"
" {\n"
" C[gIdx] = A[gIdx] - B[gIdx];\n"
" }\n"
"}\n"
"\n"
;

94
Extras/RigidBodyGpuPipeline/opencl/primitives/AdlPrimitives/Search/BoundSearchKernelsDX11.h Normal file
View File

@@ -0,0 +1,94 @@
static const char* boundSearchKernelsDX11= \
"/*\n"
" 2011 Takahiro Harada\n"
"*/\n"
"\n"
"typedef uint u32;\n"
"\n"
"#define GET_GROUP_IDX groupIdx.x\n"
"#define GET_LOCAL_IDX localIdx.x\n"
"#define GET_GLOBAL_IDX globalIdx.x\n"
"#define GROUP_LDS_BARRIER GroupMemoryBarrierWithGroupSync()\n"
"#define DEFAULT_ARGS uint3 globalIdx : SV_DispatchThreadID, uint3 localIdx : SV_GroupThreadID, uint3 groupIdx : SV_GroupID\n"
"#define AtomInc(x) InterlockedAdd(x, 1)\n"
"#define AtomInc1(x, out) InterlockedAdd(x, 1, out)\n"
"\n"
"\n"
"\n"
"typedef struct\n"
"{\n"
" u32 m_key; \n"
" u32 m_value;\n"
"}SortData;\n"
"\n"
"\n"
"\n"
"cbuffer SortCB : register( b0 )\n"
"{\n"
" u32 m_nSrc;\n"
" u32 m_nDst;\n"
" u32 m_padding[2];\n"
"};\n"
"\n"
"\n"
"StructuredBuffer<SortData> src : register( t0 );\n"
"RWStructuredBuffer<u32> dst : register( u0 );\n"
"\n"
"\n"
"[numthreads(64, 1, 1)]\n"
"void SearchSortDataLowerKernel( DEFAULT_ARGS )\n"
"{\n"
" int gIdx = GET_GLOBAL_IDX;\n"
" u32 nSrc = m_nSrc;\n"
" u32 nDst = m_nDst;\n"
"\n"
" if( gIdx < nSrc )\n"
" {\n"
" SortData iData;\n"
" SortData jData;\n"
" if( gIdx==0 ) iData.m_key = iData.m_value = (u32)-1;\n"
" else iData = src[gIdx-1];\n"
"\n"
" if( gIdx==nSrc ) jData.m_key = jData.m_value = nDst;\n"
" else jData = src[gIdx];\n"
"\n"
" if( iData.m_key != jData.m_key )\n"
" {\n"
"// for(u32 k=iData.m_key+1; k<=min(jData.m_key, nDst-1); k++)\n"
" u32 k = jData.m_key;\n"
" {\n"
" dst[k] = gIdx;\n"
" }\n"
" }\n"
" }\n"
"}\n"
"\n"
"[numthreads(64, 1, 1)]\n"
"void SearchSortDataUpperKernel( DEFAULT_ARGS )\n"
"{\n"
" int gIdx = GET_GLOBAL_IDX;\n"
" u32 nSrc = m_nSrc;\n"
" u32 nDst = m_nDst;\n"
"\n"
" if( gIdx < nSrc+1 )\n"
" {\n"
" SortData iData;\n"
" SortData jData;\n"
" if( gIdx==0 ) iData.m_key = iData.m_value = 0;\n"
" else iData = src[gIdx-1];\n"
"\n"
" if( gIdx==nSrc ) jData.m_key = jData.m_value = nDst;\n"
" else jData = src[gIdx];\n"
"\n"
" if( iData.m_key != jData.m_key )\n"
" {\n"
"// for(u32 k=iData.m_key; k<min(jData.m_key, nDst); k++)\n"
" u32 k = iData.m_key;\n"
" {\n"
" dst[k] = gIdx;\n"
" }\n"
" }\n"
" }\n"
"}\n"
"\n"
;