Update Bullet to the latest commit 126b676
This commit is contained in:
Oussama
Rémi Verschelde
parent
a6722cf362
commit
22b7c9dfa8
612 changed files with 114715 additions and 103413 deletions
|
|
@ -1,87 +1,86 @@
|
|||
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
|
||||
static const char* boundSearchKernelsCL= \
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\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"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" u32 m_key; \n"
|
||||
" u32 m_value;\n"
|
||||
"}SortData;\n"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" u32 m_nSrc;\n"
|
||||
" u32 m_nDst;\n"
|
||||
" u32 m_padding[2];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void SearchSortDataLowerKernel(__global SortData* src, __global u32 *dst, \n"
|
||||
" unsigned int nSrc, unsigned int nDst)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\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"
|
||||
" SortData iData = (gIdx==0)? first: src[gIdx-1];\n"
|
||||
" SortData jData = (gIdx==nSrc)? end: src[gIdx];\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"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void SearchSortDataUpperKernel(__global SortData* src, __global u32 *dst, \n"
|
||||
" unsigned int nSrc, unsigned int nDst)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX+1;\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"
|
||||
" SortData iData = src[gIdx-1];\n"
|
||||
" SortData jData = (gIdx==nSrc)? end: src[gIdx];\n"
|
||||
" if( iData.m_key != jData.m_key )\n"
|
||||
" {\n"
|
||||
" u32 k = iData.m_key;\n"
|
||||
" {\n"
|
||||
" dst[k] = gIdx;\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"
|
||||
" unsigned int nSrc, unsigned int nDst)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" \n"
|
||||
" if( gIdx < nDst )\n"
|
||||
" {\n"
|
||||
" C[gIdx] = A[gIdx] - B[gIdx];\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
;
|
||||
static const char* boundSearchKernelsCL =
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\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"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" u32 m_key; \n"
|
||||
" u32 m_value;\n"
|
||||
"}SortData;\n"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" u32 m_nSrc;\n"
|
||||
" u32 m_nDst;\n"
|
||||
" u32 m_padding[2];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void SearchSortDataLowerKernel(__global SortData* src, __global u32 *dst, \n"
|
||||
" unsigned int nSrc, unsigned int nDst)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\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"
|
||||
" SortData iData = (gIdx==0)? first: src[gIdx-1];\n"
|
||||
" SortData jData = (gIdx==nSrc)? end: src[gIdx];\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"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void SearchSortDataUpperKernel(__global SortData* src, __global u32 *dst, \n"
|
||||
" unsigned int nSrc, unsigned int nDst)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX+1;\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"
|
||||
" SortData iData = src[gIdx-1];\n"
|
||||
" SortData jData = (gIdx==nSrc)? end: src[gIdx];\n"
|
||||
" if( iData.m_key != jData.m_key )\n"
|
||||
" {\n"
|
||||
" u32 k = iData.m_key;\n"
|
||||
" {\n"
|
||||
" dst[k] = gIdx;\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"
|
||||
" unsigned int nSrc, unsigned int nDst)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" \n"
|
||||
" if( gIdx < nDst )\n"
|
||||
" {\n"
|
||||
" C[gIdx] = A[gIdx] - B[gIdx];\n"
|
||||
" }\n"
|
||||
"}\n";
|
||||
|
|
|
|||
|
|
@ -1,132 +1,131 @@
|
|||
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
|
||||
static const char* copyKernelsCL= \
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"\n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\n"
|
||||
"\n"
|
||||
"#pragma OPENCL EXTENSION cl_amd_printf : enable\n"
|
||||
"#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable\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"
|
||||
"#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)\n"
|
||||
"#define AtomInc(x) atom_inc(&(x))\n"
|
||||
"#define AtomInc1(x, out) out = atom_inc(&(x))\n"
|
||||
"\n"
|
||||
"#define make_uint4 (uint4)\n"
|
||||
"#define make_uint2 (uint2)\n"
|
||||
"#define make_int2 (int2)\n"
|
||||
"\n"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" int m_n;\n"
|
||||
" int m_padding[3];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"\n"
|
||||
"\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void Copy1F4Kernel(__global float4* dst, __global float4* src, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( gIdx < cb.m_n )\n"
|
||||
" {\n"
|
||||
" float4 a0 = src[gIdx];\n"
|
||||
"\n"
|
||||
" dst[ gIdx ] = a0;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void Copy2F4Kernel(__global float4* dst, __global float4* src, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( 2*gIdx <= cb.m_n )\n"
|
||||
" {\n"
|
||||
" float4 a0 = src[gIdx*2+0];\n"
|
||||
" float4 a1 = src[gIdx*2+1];\n"
|
||||
"\n"
|
||||
" dst[ gIdx*2+0 ] = a0;\n"
|
||||
" dst[ gIdx*2+1 ] = a1;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void Copy4F4Kernel(__global float4* dst, __global float4* src, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( 4*gIdx <= cb.m_n )\n"
|
||||
" {\n"
|
||||
" int idx0 = gIdx*4+0;\n"
|
||||
" int idx1 = gIdx*4+1;\n"
|
||||
" int idx2 = gIdx*4+2;\n"
|
||||
" int idx3 = gIdx*4+3;\n"
|
||||
"\n"
|
||||
" float4 a0 = src[idx0];\n"
|
||||
" float4 a1 = src[idx1];\n"
|
||||
" float4 a2 = src[idx2];\n"
|
||||
" float4 a3 = src[idx3];\n"
|
||||
"\n"
|
||||
" dst[ idx0 ] = a0;\n"
|
||||
" dst[ idx1 ] = a1;\n"
|
||||
" dst[ idx2 ] = a2;\n"
|
||||
" dst[ idx3 ] = a3;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void CopyF1Kernel(__global float* dstF1, __global float* srcF1, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( gIdx < cb.m_n )\n"
|
||||
" {\n"
|
||||
" float a0 = srcF1[gIdx];\n"
|
||||
"\n"
|
||||
" dstF1[ gIdx ] = a0;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void CopyF2Kernel(__global float2* dstF2, __global float2* srcF2, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( gIdx < cb.m_n )\n"
|
||||
" {\n"
|
||||
" float2 a0 = srcF2[gIdx];\n"
|
||||
"\n"
|
||||
" dstF2[ gIdx ] = a0;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"\n"
|
||||
;
|
||||
static const char* copyKernelsCL =
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"\n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\n"
|
||||
"\n"
|
||||
"#pragma OPENCL EXTENSION cl_amd_printf : enable\n"
|
||||
"#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable\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"
|
||||
"#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)\n"
|
||||
"#define AtomInc(x) atom_inc(&(x))\n"
|
||||
"#define AtomInc1(x, out) out = atom_inc(&(x))\n"
|
||||
"\n"
|
||||
"#define make_uint4 (uint4)\n"
|
||||
"#define make_uint2 (uint2)\n"
|
||||
"#define make_int2 (int2)\n"
|
||||
"\n"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" int m_n;\n"
|
||||
" int m_padding[3];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"\n"
|
||||
"\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void Copy1F4Kernel(__global float4* dst, __global float4* src, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( gIdx < cb.m_n )\n"
|
||||
" {\n"
|
||||
" float4 a0 = src[gIdx];\n"
|
||||
"\n"
|
||||
" dst[ gIdx ] = a0;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void Copy2F4Kernel(__global float4* dst, __global float4* src, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( 2*gIdx <= cb.m_n )\n"
|
||||
" {\n"
|
||||
" float4 a0 = src[gIdx*2+0];\n"
|
||||
" float4 a1 = src[gIdx*2+1];\n"
|
||||
"\n"
|
||||
" dst[ gIdx*2+0 ] = a0;\n"
|
||||
" dst[ gIdx*2+1 ] = a1;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void Copy4F4Kernel(__global float4* dst, __global float4* src, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( 4*gIdx <= cb.m_n )\n"
|
||||
" {\n"
|
||||
" int idx0 = gIdx*4+0;\n"
|
||||
" int idx1 = gIdx*4+1;\n"
|
||||
" int idx2 = gIdx*4+2;\n"
|
||||
" int idx3 = gIdx*4+3;\n"
|
||||
"\n"
|
||||
" float4 a0 = src[idx0];\n"
|
||||
" float4 a1 = src[idx1];\n"
|
||||
" float4 a2 = src[idx2];\n"
|
||||
" float4 a3 = src[idx3];\n"
|
||||
"\n"
|
||||
" dst[ idx0 ] = a0;\n"
|
||||
" dst[ idx1 ] = a1;\n"
|
||||
" dst[ idx2 ] = a2;\n"
|
||||
" dst[ idx3 ] = a3;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void CopyF1Kernel(__global float* dstF1, __global float* srcF1, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( gIdx < cb.m_n )\n"
|
||||
" {\n"
|
||||
" float a0 = srcF1[gIdx];\n"
|
||||
"\n"
|
||||
" dstF1[ gIdx ] = a0;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void CopyF2Kernel(__global float2* dstF2, __global float2* srcF2, \n"
|
||||
" ConstBuffer cb)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
"\n"
|
||||
" if( gIdx < cb.m_n )\n"
|
||||
" {\n"
|
||||
" float2 a0 = srcF2[gIdx];\n"
|
||||
"\n"
|
||||
" dstF2[ gIdx ] = a0;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"\n"
|
||||
"\n";
|
||||
|
|
|
|||
|
|
@ -1,91 +1,90 @@
|
|||
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
|
||||
static const char* fillKernelsCL= \
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\n"
|
||||
"#pragma OPENCL EXTENSION cl_amd_printf : enable\n"
|
||||
"#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable\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"
|
||||
"#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)\n"
|
||||
"#define AtomInc(x) atom_inc(&(x))\n"
|
||||
"#define AtomInc1(x, out) out = atom_inc(&(x))\n"
|
||||
"#define make_uint4 (uint4)\n"
|
||||
"#define make_uint2 (uint2)\n"
|
||||
"#define make_int2 (int2)\n"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" union\n"
|
||||
" {\n"
|
||||
" int4 m_data;\n"
|
||||
" uint4 m_unsignedData;\n"
|
||||
" float m_floatData;\n"
|
||||
" };\n"
|
||||
" int m_offset;\n"
|
||||
" int m_n;\n"
|
||||
" int m_padding[2];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillIntKernel(__global int* dstInt, int num_elements, int value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num_elements )\n"
|
||||
" {\n"
|
||||
" dstInt[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillFloatKernel(__global float* dstFloat, int num_elements, float value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num_elements )\n"
|
||||
" {\n"
|
||||
" dstFloat[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillUnsignedIntKernel(__global unsigned int* dstInt, const int num, const unsigned int value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num )\n"
|
||||
" {\n"
|
||||
" dstInt[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillInt2Kernel(__global int2* dstInt2, const int num, const int2 value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num )\n"
|
||||
" {\n"
|
||||
" dstInt2[ gIdx + offset] = make_int2( value.x, value.y );\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillInt4Kernel(__global int4* dstInt4, const int num, const int4 value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num )\n"
|
||||
" {\n"
|
||||
" dstInt4[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
;
|
||||
static const char* fillKernelsCL =
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\n"
|
||||
"#pragma OPENCL EXTENSION cl_amd_printf : enable\n"
|
||||
"#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable\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"
|
||||
"#define GROUP_MEM_FENCE mem_fence(CLK_LOCAL_MEM_FENCE)\n"
|
||||
"#define AtomInc(x) atom_inc(&(x))\n"
|
||||
"#define AtomInc1(x, out) out = atom_inc(&(x))\n"
|
||||
"#define make_uint4 (uint4)\n"
|
||||
"#define make_uint2 (uint2)\n"
|
||||
"#define make_int2 (int2)\n"
|
||||
"typedef struct\n"
|
||||
"{\n"
|
||||
" union\n"
|
||||
" {\n"
|
||||
" int4 m_data;\n"
|
||||
" uint4 m_unsignedData;\n"
|
||||
" float m_floatData;\n"
|
||||
" };\n"
|
||||
" int m_offset;\n"
|
||||
" int m_n;\n"
|
||||
" int m_padding[2];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillIntKernel(__global int* dstInt, int num_elements, int value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num_elements )\n"
|
||||
" {\n"
|
||||
" dstInt[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillFloatKernel(__global float* dstFloat, int num_elements, float value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num_elements )\n"
|
||||
" {\n"
|
||||
" dstFloat[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillUnsignedIntKernel(__global unsigned int* dstInt, const int num, const unsigned int value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num )\n"
|
||||
" {\n"
|
||||
" dstInt[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillInt2Kernel(__global int2* dstInt2, const int num, const int2 value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num )\n"
|
||||
" {\n"
|
||||
" dstInt2[ gIdx + offset] = make_int2( value.x, value.y );\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__kernel\n"
|
||||
"__attribute__((reqd_work_group_size(64,1,1)))\n"
|
||||
"void FillInt4Kernel(__global int4* dstInt4, const int num, const int4 value, const int offset)\n"
|
||||
"{\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" if( gIdx < num )\n"
|
||||
" {\n"
|
||||
" dstInt4[ offset+gIdx ] = value;\n"
|
||||
" }\n"
|
||||
"}\n";
|
||||
|
|
|
|||
|
|
@ -1,129 +1,128 @@
|
|||
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
|
||||
static const char* prefixScanKernelsCL= \
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\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"
|
||||
"// takahiro end\n"
|
||||
"#define WG_SIZE 128 \n"
|
||||
"#define m_numElems x\n"
|
||||
"#define m_numBlocks y\n"
|
||||
"#define m_numScanBlocks z\n"
|
||||
"/*typedef struct\n"
|
||||
"{\n"
|
||||
" uint m_numElems;\n"
|
||||
" uint m_numBlocks;\n"
|
||||
" uint m_numScanBlocks;\n"
|
||||
" uint m_padding[1];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"*/\n"
|
||||
"u32 ScanExclusive(__local u32* data, u32 n, int lIdx, int lSize)\n"
|
||||
"{\n"
|
||||
" u32 blocksum;\n"
|
||||
" int offset = 1;\n"
|
||||
" for(int nActive=n>>1; nActive>0; nActive>>=1, offset<<=1)\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for(int iIdx=lIdx; iIdx<nActive; iIdx+=lSize)\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" data[bi] += data[ai];\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" if( lIdx == 0 )\n"
|
||||
" {\n"
|
||||
" blocksum = data[ n-1 ];\n"
|
||||
" data[ n-1 ] = 0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" offset >>= 1;\n"
|
||||
" for(int nActive=1; nActive<n; nActive<<=1, offset>>=1 )\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for( int iIdx = lIdx; iIdx<nActive; iIdx += lSize )\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" u32 temp = data[ai];\n"
|
||||
" data[ai] = data[bi];\n"
|
||||
" data[bi] += temp;\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" return blocksum;\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void LocalScanKernel(__global u32* dst, __global u32 *src, __global u32 *sumBuffer,\n"
|
||||
" uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local u32 ldsData[WG_SIZE*2];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" ldsData[2*lIdx] = ( 2*gIdx < cb.m_numElems )? src[2*gIdx]: 0;\n"
|
||||
" ldsData[2*lIdx + 1] = ( 2*gIdx+1 < cb.m_numElems )? src[2*gIdx + 1]: 0;\n"
|
||||
" u32 sum = ScanExclusive(ldsData, WG_SIZE*2, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" if( lIdx == 0 ) sumBuffer[GET_GROUP_IDX] = sum;\n"
|
||||
" if( (2*gIdx) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx] = ldsData[2*lIdx];\n"
|
||||
" }\n"
|
||||
" if( (2*gIdx + 1) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx + 1] = ldsData[2*lIdx + 1];\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void AddOffsetKernel(__global u32 *dst, __global u32 *blockSum, uint4 cb)\n"
|
||||
"{\n"
|
||||
" const u32 blockSize = WG_SIZE*2;\n"
|
||||
" int myIdx = GET_GROUP_IDX+1;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" u32 iBlockSum = blockSum[myIdx];\n"
|
||||
" int endValue = min((myIdx+1)*(blockSize), cb.m_numElems);\n"
|
||||
" for(int i=myIdx*blockSize+lIdx; i<endValue; i+=GET_GROUP_SIZE)\n"
|
||||
" {\n"
|
||||
" dst[i] += iBlockSum;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void TopLevelScanKernel(__global u32* dst, uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local u32 ldsData[2048];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" int lSize = GET_GROUP_SIZE;\n"
|
||||
" for(int i=lIdx; i<cb.m_numScanBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" ldsData[i] = (i<cb.m_numBlocks)? dst[i]:0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" u32 sum = ScanExclusive(ldsData, cb.m_numScanBlocks, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" for(int i=lIdx; i<cb.m_numBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" dst[i] = ldsData[i];\n"
|
||||
" }\n"
|
||||
" if( gIdx == 0 )\n"
|
||||
" {\n"
|
||||
" dst[cb.m_numBlocks] = sum;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
;
|
||||
static const char* prefixScanKernelsCL =
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\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"
|
||||
"// takahiro end\n"
|
||||
"#define WG_SIZE 128 \n"
|
||||
"#define m_numElems x\n"
|
||||
"#define m_numBlocks y\n"
|
||||
"#define m_numScanBlocks z\n"
|
||||
"/*typedef struct\n"
|
||||
"{\n"
|
||||
" uint m_numElems;\n"
|
||||
" uint m_numBlocks;\n"
|
||||
" uint m_numScanBlocks;\n"
|
||||
" uint m_padding[1];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"*/\n"
|
||||
"u32 ScanExclusive(__local u32* data, u32 n, int lIdx, int lSize)\n"
|
||||
"{\n"
|
||||
" u32 blocksum;\n"
|
||||
" int offset = 1;\n"
|
||||
" for(int nActive=n>>1; nActive>0; nActive>>=1, offset<<=1)\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for(int iIdx=lIdx; iIdx<nActive; iIdx+=lSize)\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" data[bi] += data[ai];\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" if( lIdx == 0 )\n"
|
||||
" {\n"
|
||||
" blocksum = data[ n-1 ];\n"
|
||||
" data[ n-1 ] = 0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" offset >>= 1;\n"
|
||||
" for(int nActive=1; nActive<n; nActive<<=1, offset>>=1 )\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for( int iIdx = lIdx; iIdx<nActive; iIdx += lSize )\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" u32 temp = data[ai];\n"
|
||||
" data[ai] = data[bi];\n"
|
||||
" data[bi] += temp;\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" return blocksum;\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void LocalScanKernel(__global u32* dst, __global u32 *src, __global u32 *sumBuffer,\n"
|
||||
" uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local u32 ldsData[WG_SIZE*2];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" ldsData[2*lIdx] = ( 2*gIdx < cb.m_numElems )? src[2*gIdx]: 0;\n"
|
||||
" ldsData[2*lIdx + 1] = ( 2*gIdx+1 < cb.m_numElems )? src[2*gIdx + 1]: 0;\n"
|
||||
" u32 sum = ScanExclusive(ldsData, WG_SIZE*2, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" if( lIdx == 0 ) sumBuffer[GET_GROUP_IDX] = sum;\n"
|
||||
" if( (2*gIdx) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx] = ldsData[2*lIdx];\n"
|
||||
" }\n"
|
||||
" if( (2*gIdx + 1) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx + 1] = ldsData[2*lIdx + 1];\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void AddOffsetKernel(__global u32 *dst, __global u32 *blockSum, uint4 cb)\n"
|
||||
"{\n"
|
||||
" const u32 blockSize = WG_SIZE*2;\n"
|
||||
" int myIdx = GET_GROUP_IDX+1;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" u32 iBlockSum = blockSum[myIdx];\n"
|
||||
" int endValue = min((myIdx+1)*(blockSize), cb.m_numElems);\n"
|
||||
" for(int i=myIdx*blockSize+lIdx; i<endValue; i+=GET_GROUP_SIZE)\n"
|
||||
" {\n"
|
||||
" dst[i] += iBlockSum;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void TopLevelScanKernel(__global u32* dst, uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local u32 ldsData[2048];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" int lSize = GET_GROUP_SIZE;\n"
|
||||
" for(int i=lIdx; i<cb.m_numScanBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" ldsData[i] = (i<cb.m_numBlocks)? dst[i]:0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" u32 sum = ScanExclusive(ldsData, cb.m_numScanBlocks, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" for(int i=lIdx; i<cb.m_numBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" dst[i] = ldsData[i];\n"
|
||||
" }\n"
|
||||
" if( gIdx == 0 )\n"
|
||||
" {\n"
|
||||
" dst[cb.m_numBlocks] = sum;\n"
|
||||
" }\n"
|
||||
"}\n";
|
||||
|
|
|
|||
|
|
@ -1,129 +1,128 @@
|
|||
//this file is autogenerated using stringify.bat (premake --stringify) in the build folder of this project
|
||||
static const char* prefixScanKernelsFloat4CL= \
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\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"
|
||||
"// takahiro end\n"
|
||||
"#define WG_SIZE 128 \n"
|
||||
"#define m_numElems x\n"
|
||||
"#define m_numBlocks y\n"
|
||||
"#define m_numScanBlocks z\n"
|
||||
"/*typedef struct\n"
|
||||
"{\n"
|
||||
" uint m_numElems;\n"
|
||||
" uint m_numBlocks;\n"
|
||||
" uint m_numScanBlocks;\n"
|
||||
" uint m_padding[1];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"*/\n"
|
||||
"float4 ScanExclusiveFloat4(__local float4* data, u32 n, int lIdx, int lSize)\n"
|
||||
"{\n"
|
||||
" float4 blocksum;\n"
|
||||
" int offset = 1;\n"
|
||||
" for(int nActive=n>>1; nActive>0; nActive>>=1, offset<<=1)\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for(int iIdx=lIdx; iIdx<nActive; iIdx+=lSize)\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" data[bi] += data[ai];\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" if( lIdx == 0 )\n"
|
||||
" {\n"
|
||||
" blocksum = data[ n-1 ];\n"
|
||||
" data[ n-1 ] = 0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" offset >>= 1;\n"
|
||||
" for(int nActive=1; nActive<n; nActive<<=1, offset>>=1 )\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for( int iIdx = lIdx; iIdx<nActive; iIdx += lSize )\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" float4 temp = data[ai];\n"
|
||||
" data[ai] = data[bi];\n"
|
||||
" data[bi] += temp;\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" return blocksum;\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void LocalScanKernel(__global float4* dst, __global float4* src, __global float4* sumBuffer, uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local float4 ldsData[WG_SIZE*2];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" ldsData[2*lIdx] = ( 2*gIdx < cb.m_numElems )? src[2*gIdx]: 0;\n"
|
||||
" ldsData[2*lIdx + 1] = ( 2*gIdx+1 < cb.m_numElems )? src[2*gIdx + 1]: 0;\n"
|
||||
" float4 sum = ScanExclusiveFloat4(ldsData, WG_SIZE*2, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" if( lIdx == 0 ) \n"
|
||||
" sumBuffer[GET_GROUP_IDX] = sum;\n"
|
||||
" if( (2*gIdx) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx] = ldsData[2*lIdx];\n"
|
||||
" }\n"
|
||||
" if( (2*gIdx + 1) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx + 1] = ldsData[2*lIdx + 1];\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void AddOffsetKernel(__global float4* dst, __global float4* blockSum, uint4 cb)\n"
|
||||
"{\n"
|
||||
" const u32 blockSize = WG_SIZE*2;\n"
|
||||
" int myIdx = GET_GROUP_IDX+1;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" float4 iBlockSum = blockSum[myIdx];\n"
|
||||
" int endValue = min((myIdx+1)*(blockSize), cb.m_numElems);\n"
|
||||
" for(int i=myIdx*blockSize+lIdx; i<endValue; i+=GET_GROUP_SIZE)\n"
|
||||
" {\n"
|
||||
" dst[i] += iBlockSum;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void TopLevelScanKernel(__global float4* dst, uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local float4 ldsData[2048];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" int lSize = GET_GROUP_SIZE;\n"
|
||||
" for(int i=lIdx; i<cb.m_numScanBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" ldsData[i] = (i<cb.m_numBlocks)? dst[i]:0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" float4 sum = ScanExclusiveFloat4(ldsData, cb.m_numScanBlocks, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" for(int i=lIdx; i<cb.m_numBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" dst[i] = ldsData[i];\n"
|
||||
" }\n"
|
||||
" if( gIdx == 0 )\n"
|
||||
" {\n"
|
||||
" dst[cb.m_numBlocks] = sum;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
;
|
||||
static const char* prefixScanKernelsFloat4CL =
|
||||
"/*\n"
|
||||
"Copyright (c) 2012 Advanced Micro Devices, Inc. \n"
|
||||
"This software is provided 'as-is', without any express or implied warranty.\n"
|
||||
"In no event will the authors be held liable for any damages arising from the use of this software.\n"
|
||||
"Permission is granted to anyone to use this software for any purpose, \n"
|
||||
"including commercial applications, and to alter it and redistribute it freely, \n"
|
||||
"subject to the following restrictions:\n"
|
||||
"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.\n"
|
||||
"2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.\n"
|
||||
"3. This notice may not be removed or altered from any source distribution.\n"
|
||||
"*/\n"
|
||||
"//Originally written by Takahiro Harada\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"
|
||||
"// takahiro end\n"
|
||||
"#define WG_SIZE 128 \n"
|
||||
"#define m_numElems x\n"
|
||||
"#define m_numBlocks y\n"
|
||||
"#define m_numScanBlocks z\n"
|
||||
"/*typedef struct\n"
|
||||
"{\n"
|
||||
" uint m_numElems;\n"
|
||||
" uint m_numBlocks;\n"
|
||||
" uint m_numScanBlocks;\n"
|
||||
" uint m_padding[1];\n"
|
||||
"} ConstBuffer;\n"
|
||||
"*/\n"
|
||||
"float4 ScanExclusiveFloat4(__local float4* data, u32 n, int lIdx, int lSize)\n"
|
||||
"{\n"
|
||||
" float4 blocksum;\n"
|
||||
" int offset = 1;\n"
|
||||
" for(int nActive=n>>1; nActive>0; nActive>>=1, offset<<=1)\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for(int iIdx=lIdx; iIdx<nActive; iIdx+=lSize)\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" data[bi] += data[ai];\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" if( lIdx == 0 )\n"
|
||||
" {\n"
|
||||
" blocksum = data[ n-1 ];\n"
|
||||
" data[ n-1 ] = 0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" offset >>= 1;\n"
|
||||
" for(int nActive=1; nActive<n; nActive<<=1, offset>>=1 )\n"
|
||||
" {\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" for( int iIdx = lIdx; iIdx<nActive; iIdx += lSize )\n"
|
||||
" {\n"
|
||||
" int ai = offset*(2*iIdx+1)-1;\n"
|
||||
" int bi = offset*(2*iIdx+2)-1;\n"
|
||||
" float4 temp = data[ai];\n"
|
||||
" data[ai] = data[bi];\n"
|
||||
" data[bi] += temp;\n"
|
||||
" }\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" return blocksum;\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void LocalScanKernel(__global float4* dst, __global float4* src, __global float4* sumBuffer, uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local float4 ldsData[WG_SIZE*2];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" ldsData[2*lIdx] = ( 2*gIdx < cb.m_numElems )? src[2*gIdx]: 0;\n"
|
||||
" ldsData[2*lIdx + 1] = ( 2*gIdx+1 < cb.m_numElems )? src[2*gIdx + 1]: 0;\n"
|
||||
" float4 sum = ScanExclusiveFloat4(ldsData, WG_SIZE*2, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" if( lIdx == 0 ) \n"
|
||||
" sumBuffer[GET_GROUP_IDX] = sum;\n"
|
||||
" if( (2*gIdx) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx] = ldsData[2*lIdx];\n"
|
||||
" }\n"
|
||||
" if( (2*gIdx + 1) < cb.m_numElems )\n"
|
||||
" {\n"
|
||||
" dst[2*gIdx + 1] = ldsData[2*lIdx + 1];\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void AddOffsetKernel(__global float4* dst, __global float4* blockSum, uint4 cb)\n"
|
||||
"{\n"
|
||||
" const u32 blockSize = WG_SIZE*2;\n"
|
||||
" int myIdx = GET_GROUP_IDX+1;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" float4 iBlockSum = blockSum[myIdx];\n"
|
||||
" int endValue = min((myIdx+1)*(blockSize), cb.m_numElems);\n"
|
||||
" for(int i=myIdx*blockSize+lIdx; i<endValue; i+=GET_GROUP_SIZE)\n"
|
||||
" {\n"
|
||||
" dst[i] += iBlockSum;\n"
|
||||
" }\n"
|
||||
"}\n"
|
||||
"__attribute__((reqd_work_group_size(WG_SIZE,1,1)))\n"
|
||||
"__kernel\n"
|
||||
"void TopLevelScanKernel(__global float4* dst, uint4 cb)\n"
|
||||
"{\n"
|
||||
" __local float4 ldsData[2048];\n"
|
||||
" int gIdx = GET_GLOBAL_IDX;\n"
|
||||
" int lIdx = GET_LOCAL_IDX;\n"
|
||||
" int lSize = GET_GROUP_SIZE;\n"
|
||||
" for(int i=lIdx; i<cb.m_numScanBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" ldsData[i] = (i<cb.m_numBlocks)? dst[i]:0;\n"
|
||||
" }\n"
|
||||
" GROUP_LDS_BARRIER;\n"
|
||||
" float4 sum = ScanExclusiveFloat4(ldsData, cb.m_numScanBlocks, GET_LOCAL_IDX, GET_GROUP_SIZE);\n"
|
||||
" for(int i=lIdx; i<cb.m_numBlocks; i+=lSize )\n"
|
||||
" {\n"
|
||||
" dst[i] = ldsData[i];\n"
|
||||
" }\n"
|
||||
" if( gIdx == 0 )\n"
|
||||
" {\n"
|
||||
" dst[cb.m_numBlocks] = sum;\n"
|
||||
" }\n"
|
||||
"}\n";
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load diff
Loading…
Add table
Add a link
Reference in a new issue