From 22b7c9dfa80d0f7abca40f061865c2ab3c136a74 Mon Sep 17 00:00:00 2001
From: Oussama <o.boukhelf@gmail.com>
Date: Thu, 3 Jan 2019 14:26:51 +0100
Subject: Update Bullet to the latest commit 126b676

---
 .../kernels/PrefixScanKernelsCL.h                  | 255 ++++++++++-----------
 1 file changed, 127 insertions(+), 128 deletions(-)

(limited to 'thirdparty/bullet/Bullet3OpenCL/ParallelPrimitives/kernels/PrefixScanKernelsCL.h')

diff --git a/thirdparty/bullet/Bullet3OpenCL/ParallelPrimitives/kernels/PrefixScanKernelsCL.h b/thirdparty/bullet/Bullet3OpenCL/ParallelPrimitives/kernels/PrefixScanKernelsCL.h
index 27baab8331..fc5e7b865c 100644
--- a/thirdparty/bullet/Bullet3OpenCL/ParallelPrimitives/kernels/PrefixScanKernelsCL.h
+++ b/thirdparty/bullet/Bullet3OpenCL/ParallelPrimitives/kernels/PrefixScanKernelsCL.h
@@ -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";
-- 
cgit v1.2.3