[Libreoffice-commits] core.git: Changes to 'refs/tags/libreoffice-4-2-milestone-7'
Tag 'libreoffice-4-2-milestone-7' created by Markus Mohrhard markus.mohrh...@googlemail.com at 2014-03-01 08:57 -0800 milestone 7 Changes since libreoffice-4-2-milestone-6: Wei Wei (1): GPU-Calc: remove Alloc_Host_Ptr for clmem of NAN vector --- sc/source/core/opencl/formulagroupcl.cxx |2 +- 1 file changed, 1 insertion(+), 1 deletion(-) --- ___ Libreoffice-commits mailing list libreoffice-comm...@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice-commits
[Libreoffice-commits] core.git: Changes to 'refs/tags/libreoffice-4-2-milestone-6'
Tag 'libreoffice-4-2-milestone-6' created by Jan Holesovsky ke...@collabora.com at 2014-02-28 19:15 -0800 milestone six Changes since libreoffice-4-2-milestone-5-3: --- 0 files changed --- ___ Libreoffice-commits mailing list libreoffice-comm...@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/libreoffice-commits
[Libreoffice-commits] core.git: 6 commits - sc/qa sc/source
sc/qa/unit/data/xls/opencl/logical/not.xls |binary
sc/qa/unit/data/xls/opencl/logical/xor.xls |binary
sc/qa/unit/opencl-test.cxx |5
sc/source/core/opencl/op_statistical.cxx | 612 -
4 files changed, 357 insertions(+), 260 deletions(-)
New commits:
commit 3d0152ba5f551e23b3de5a32e0c45e9d73f482ed
Author: Wei Wei wei...@multicorewareinc.com
Date: Wed Dec 18 16:34:56 2013 -0600
GPU Calc: remove the warning; shrink the test case
number for XOR and NOT
Change-Id: Ib184aa27a4bd4553886fb9422c937dac9c053a4e
Signed-off-by: I-Jui (Ray) Sung r...@multicorewareinc.com
diff --git a/sc/qa/unit/data/xls/opencl/logical/not.xls
b/sc/qa/unit/data/xls/opencl/logical/not.xls
index 651a8f9..7cd86a0 100644
Binary files a/sc/qa/unit/data/xls/opencl/logical/not.xls and
b/sc/qa/unit/data/xls/opencl/logical/not.xls differ
diff --git a/sc/qa/unit/data/xls/opencl/logical/xor.xls
b/sc/qa/unit/data/xls/opencl/logical/xor.xls
index fccba9b..7737727 100644
Binary files a/sc/qa/unit/data/xls/opencl/logical/xor.xls and
b/sc/qa/unit/data/xls/opencl/logical/xor.xls differ
diff --git a/sc/qa/unit/opencl-test.cxx b/sc/qa/unit/opencl-test.cxx
index 9d28359..09c8a48 100644
--- a/sc/qa/unit/opencl-test.cxx
+++ b/sc/qa/unit/opencl-test.cxx
@@ -4942,7 +4942,7 @@ void ScOpenclTest:: testLogicalFormulaNot()
ScDocShellRef xDocShRes = loadDoc(opencl/logical/not., XLS);
ScDocument *pDocRes = xDocShRes-GetDocument();
CPPUNIT_ASSERT(pDocRes);
-for (SCROW i = 0; i 3; ++i)
+for (SCROW i = 0; i 3000; ++i)
{
double fLibre = pDoc-GetValue(ScAddress(1, i, 0));
double fExcel = pDocRes-GetValue(ScAddress(1, i, 0));
@@ -4964,7 +4964,7 @@ void ScOpenclTest:: testLogicalFormulaXor()
ScDocShellRef xDocShRes = loadDoc(opencl/logical/xor., XLS);
ScDocument *pDocRes = xDocShRes-GetDocument();
CPPUNIT_ASSERT(pDocRes);
-for (SCROW i = 0; i 3; ++i)
+for (SCROW i = 0; i 3000; ++i)
{
double fLibre = pDoc-GetValue(ScAddress(1, i, 0));
double fExcel = pDocRes-GetValue(ScAddress(1, i, 0));
diff --git a/sc/source/core/opencl/op_statistical.cxx
b/sc/source/core/opencl/op_statistical.cxx
index 47c5ccf..c3b8a5d 100644
--- a/sc/source/core/opencl/op_statistical.cxx
+++ b/sc/source/core/opencl/op_statistical.cxx
@@ -3798,7 +3798,7 @@ void OpPearson::GenSlidingWindowFunction(
ss \n;
for (i = 0; i vSubArguments.size(); i++)
{
-FormulaToken *pCur = vSubArguments[i]-GetFormulaToken();
+pCur = vSubArguments[i]-GetFormulaToken();
assert(pCur);
if (pCur-GetType() == formula::svDoubleVectorRef)
{
@@ -4066,7 +4066,7 @@ vSubArguments)
ss double tmp = 0;\n;
for (unsigned i = 0; i vSubArguments.size(); i++)
{
-FormulaToken *pCur = vSubArguments[i]-GetFormulaToken();
+pCur = vSubArguments[i]-GetFormulaToken();
assert(pCur);
if (pCur-GetType() == formula::svDoubleVectorRef)
{
@@ -4334,13 +4334,13 @@ void OpRsq::GenSlidingWindowFunction(
ss \n;
for (i = 0; i vSubArguments.size(); i++)
{
-FormulaToken *pCur = vSubArguments[i]-GetFormulaToken();
+pCur = vSubArguments[i]-GetFormulaToken();
assert(pCur);
if (pCur-GetType() == formula::svDoubleVectorRef)
{
const formula::DoubleVectorRefToken* pDVR =
dynamic_castconst formula::DoubleVectorRefToken *(pCur);
-size_t nCurWindowSize = pDVR-GetRefRowSize();
+nCurWindowSize = pDVR-GetRefRowSize();
ss for (int i = ;
if (!pDVR-IsStartFixed() pDVR-IsEndFixed()) {
#ifdef ISNAN
@@ -5354,13 +5354,13 @@ void OpKurt::
GenSlidingWindowFunction(std::stringstream ss,
ss double tmp = 0;\n;
for (unsigned i = 0; i vSubArguments.size(); i++)
{
-FormulaToken *pCur = vSubArguments[i]-GetFormulaToken();
+pCur = vSubArguments[i]-GetFormulaToken();
assert(pCur);
if (pCur-GetType() == formula::svDoubleVectorRef)
{
const formula::DoubleVectorRefToken* pDVR =
dynamic_castconst formula::DoubleVectorRefToken *(pCur);
-size_t nCurWindowSize = pDVR-GetRefRowSize();
+nCurWindowSize = pDVR-GetRefRowSize();
ss for (int i = 0; i nCurWindowSize;
ss ; i++)\n;
ss {\n;
commit ef25611384dc4fe872a6b31806fea668f9ab645d
Author: hongyu zhong hon...@multicorewareinc.com
Date: Tue Dec 10 10:50:49 2013 +0800
GPU Calc: Optimized NORMDIST
AMLOEXT-284
Change-Id: I5028ebbee91d6c7a64e9993b97b5f942f039d92c
Signed-off-by: haochen haoc...@multicorewareinc.com
Signed-off-by: Wei Wei wei...@multicorewareinc.com
Signed-off-by: I-Jui (Ray) Sung r...@multicorewareinc.com
diff --git a/sc/source/core
[Libreoffice-commits] core.git: 2 commits - sc/qa sc/source
sc/qa/unit/data/ods/opencl/compiler/nested.ods |binary
sc/source/core/opencl/formulagroupcl.cxx| 57 ++--
sc/source/core/opencl/formulagroupcl_public.hxx |1
sc/source/core/opencl/opbase.cxx|6 ++
sc/source/core/opencl/opbase.hxx|1
5 files changed, 42 insertions(+), 23 deletions(-)
New commits:
commit 8b7853cc3a1727d6b0a9f7050b26680678e98de0
Author: Wei Wei wei...@multicorewareinc.com
Date: Fri Nov 22 17:16:49 2013 -0600
GPU Calc: Sum of product doesn't check out-of-bound accesses
AMLOEXT-244 FIX
Change-Id: I5f49f7acccaabd2a97d8ac4bfba4b973889278f1
diff --git a/sc/source/core/opencl/formulagroupcl.cxx
b/sc/source/core/opencl/formulagroupcl.cxx
index b0bb011..3405459 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -444,13 +444,26 @@ public:
}
virtual void GenSlidingWindowFunction(std::stringstream ) {}
-virtual std::string GenSlidingWindowDeclRef(bool=false) const
+virtual std::string GenSlidingWindowDeclRef(bool nested=false) const
{
+size_t nArrayLength = mpDVR-GetArrayLength();
std::stringstream ss;
if (!bIsStartFixed !bIsEndFixed)
+{
+if (nested)
+ss ((i+gid0) nArrayLength ?;
ss Base::GetName() [i + gid0];
+if (nested)
+ss :NAN);
+}
else
+{
+if (nested)
+ss (i nArrayLength ?;
ss Base::GetName() [i];
+if (nested)
+ss :NAN);
+}
return ss.str();
}
/// Controls how the elements in the DoubleVectorRef are traversed
@@ -533,9 +546,9 @@ return nCurWindowSize;
if(count==0){
temp1 if(i + gid0
mpDVR-GetArrayLength();
temp1 ){\n\t\t;
-temp1 tmp = ;
+temp1 tmp = legalize(;
temp1
mpCodeGen-Gen2(GenSlidingWindowDeclRef(), tmp);
-temp1 ;\n\t\t\t;
+temp1 , tmp);\n\t\t\t;
temp1 }\n\t;
}
ss temp1.str();
@@ -548,9 +561,9 @@ return nCurWindowSize;
if(count==nCurWindowSize/outLoopSize*outLoopSize){
temp2 if(i + gid0mpDVR-GetArrayLength();
temp2 ){\n\t\t;
-temp2 tmp = ;
+temp2 tmp = legalize(;
temp2 mpCodeGen-Gen2(GenSlidingWindowDeclRef(),
tmp);
-temp2 ;\n\t\t\t;
+temp2 , tmp);\n\t\t\t;
temp2 }\n\t;
}
ss temp2.str();
@@ -571,9 +584,9 @@ return nCurWindowSize;
for(int count=0; count outLoopSize; count++){
ss i = outLoop*outLoopSize+count;\n\t;
if(count==0){
-temp1 tmp = ;
+temp1 tmp = legalize(;
temp1
mpCodeGen-Gen2(GenSlidingWindowDeclRef(), tmp);
-temp1 ;\n\t\t\t;
+temp1 , tmp);\n\t\t\t;
}
ss temp1.str();
}
@@ -583,9 +596,9 @@ return nCurWindowSize;
for(unsigned int count=nCurWindowSize/outLoopSize*outLoopSize;
count nCurWindowSize; count++){
ss i = count;\n\t;
if(count==nCurWindowSize/outLoopSize*outLoopSize){
-temp2 tmp = ;
+temp2 tmp = legalize(;
temp2 mpCodeGen-Gen2(GenSlidingWindowDeclRef(),
tmp);
-temp2 ;\n\t\t\t;
+temp2 , tmp);\n\t\t\t;
}
ss temp2.str();
}
@@ -673,13 +686,13 @@ public:
ss tmp = mpCodeGen-GetBottom() ;\n;
ss int loopOffset = l*512;\n;
ss if((loopOffset + lidx + offset + 256) end) {\n;
-ss tmp = mpCodeGen-Gen2(
-A[loopOffset + lidx + offset], tmp) ;\n;
-ss tmp = mpCodeGen-Gen2(
-A[loopOffset + lidx + offset + 256], tmp) ;\n;
+ss tmp = legalize( mpCodeGen-Gen2(
+A[loopOffset + lidx + offset], tmp) , tmp);\n;
+ss tmp = legalize( mpCodeGen-Gen2(
+A[loopOffset + lidx + offset + 256], tmp) , tmp);\n;
ss } else if ((loopOffset + lidx + offset) end)\n;
-ss tmp = mpCodeGen-Gen2(
-A[loopOffset + lidx + offset], tmp) ;\n;
+ss tmp
[Libreoffice-commits] core.git: Branch 'libreoffice-4-2' - 2 commits - sc/qa sc/source
sc/qa/unit/data/ods/opencl/compiler/nested.ods |binary
sc/source/core/opencl/formulagroupcl.cxx| 57 ++--
sc/source/core/opencl/formulagroupcl_public.hxx |1
sc/source/core/opencl/opbase.cxx|6 ++
sc/source/core/opencl/opbase.hxx|1
5 files changed, 42 insertions(+), 23 deletions(-)
New commits:
commit c9fcb7933c9a4233c0ccb0951068840db1cb9ad6
Author: Wei Wei wei...@multicorewareinc.com
Date: Fri Nov 22 17:18:21 2013 -0600
GPU Calc: Testcase for Sums of Product that have out-of-bound window
AMLOEXT-244 BUG
Change-Id: Iefce62484b2189b37b747bfd4b739115dafbc40d
diff --git a/sc/qa/unit/data/ods/opencl/compiler/nested.ods
b/sc/qa/unit/data/ods/opencl/compiler/nested.ods
index 5957844..83310fa5 100644
Binary files a/sc/qa/unit/data/ods/opencl/compiler/nested.ods and
b/sc/qa/unit/data/ods/opencl/compiler/nested.ods differ
commit 2947549f2361a5462346f4f88fc207213c180bd8
Author: Wei Wei wei...@multicorewareinc.com
Date: Fri Nov 22 17:16:49 2013 -0600
GPU Calc: Sum of product doesn't check out-of-bound accesses
AMLOEXT-244 FIX
Change-Id: I5f49f7acccaabd2a97d8ac4bfba4b973889278f1
diff --git a/sc/source/core/opencl/formulagroupcl.cxx
b/sc/source/core/opencl/formulagroupcl.cxx
index b0bb011..3405459 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -444,13 +444,26 @@ public:
}
virtual void GenSlidingWindowFunction(std::stringstream ) {}
-virtual std::string GenSlidingWindowDeclRef(bool=false) const
+virtual std::string GenSlidingWindowDeclRef(bool nested=false) const
{
+size_t nArrayLength = mpDVR-GetArrayLength();
std::stringstream ss;
if (!bIsStartFixed !bIsEndFixed)
+{
+if (nested)
+ss ((i+gid0) nArrayLength ?;
ss Base::GetName() [i + gid0];
+if (nested)
+ss :NAN);
+}
else
+{
+if (nested)
+ss (i nArrayLength ?;
ss Base::GetName() [i];
+if (nested)
+ss :NAN);
+}
return ss.str();
}
/// Controls how the elements in the DoubleVectorRef are traversed
@@ -533,9 +546,9 @@ return nCurWindowSize;
if(count==0){
temp1 if(i + gid0
mpDVR-GetArrayLength();
temp1 ){\n\t\t;
-temp1 tmp = ;
+temp1 tmp = legalize(;
temp1
mpCodeGen-Gen2(GenSlidingWindowDeclRef(), tmp);
-temp1 ;\n\t\t\t;
+temp1 , tmp);\n\t\t\t;
temp1 }\n\t;
}
ss temp1.str();
@@ -548,9 +561,9 @@ return nCurWindowSize;
if(count==nCurWindowSize/outLoopSize*outLoopSize){
temp2 if(i + gid0mpDVR-GetArrayLength();
temp2 ){\n\t\t;
-temp2 tmp = ;
+temp2 tmp = legalize(;
temp2 mpCodeGen-Gen2(GenSlidingWindowDeclRef(),
tmp);
-temp2 ;\n\t\t\t;
+temp2 , tmp);\n\t\t\t;
temp2 }\n\t;
}
ss temp2.str();
@@ -571,9 +584,9 @@ return nCurWindowSize;
for(int count=0; count outLoopSize; count++){
ss i = outLoop*outLoopSize+count;\n\t;
if(count==0){
-temp1 tmp = ;
+temp1 tmp = legalize(;
temp1
mpCodeGen-Gen2(GenSlidingWindowDeclRef(), tmp);
-temp1 ;\n\t\t\t;
+temp1 , tmp);\n\t\t\t;
}
ss temp1.str();
}
@@ -583,9 +596,9 @@ return nCurWindowSize;
for(unsigned int count=nCurWindowSize/outLoopSize*outLoopSize;
count nCurWindowSize; count++){
ss i = count;\n\t;
if(count==nCurWindowSize/outLoopSize*outLoopSize){
-temp2 tmp = ;
+temp2 tmp = legalize(;
temp2 mpCodeGen-Gen2(GenSlidingWindowDeclRef(),
tmp);
-temp2 ;\n\t\t\t;
+temp2 , tmp);\n\t\t\t;
}
ss temp2.str();
}
@@ -673,13 +686,13 @@ public:
ss tmp = mpCodeGen-GetBottom() ;\n;
ss int loopOffset = l*512;\n;
ss if((loopOffset + lidx + offset + 256) end) {\n;
-ss tmp = mpCodeGen-Gen2(
-A[loopOffset + lidx + offset], tmp
[Libreoffice-commits] core.git: Branch 'feature/calc-group-interpreter-4' - 2 commits - sc/qa sc/source
sc/qa/unit/data/ods/opencl/compiler/nested.ods |binary
sc/source/core/opencl/formulagroupcl.cxx| 57 ++--
sc/source/core/opencl/formulagroupcl_public.hxx |1
sc/source/core/opencl/opbase.cxx|6 ++
sc/source/core/opencl/opbase.hxx|1
5 files changed, 42 insertions(+), 23 deletions(-)
New commits:
commit 6769a8a4c0ffb3dfaa236988cf9915805b4abeac
Author: Wei Wei wei...@multicorewareinc.com
Date: Fri Nov 22 17:18:21 2013 -0600
GPU Calc: Testcase for Sums of Product that have out-of-bound window
AMLOEXT-244 BUG
Change-Id: Iefce62484b2189b37b747bfd4b739115dafbc40d
diff --git a/sc/qa/unit/data/ods/opencl/compiler/nested.ods
b/sc/qa/unit/data/ods/opencl/compiler/nested.ods
index 5957844..83310fa5 100644
Binary files a/sc/qa/unit/data/ods/opencl/compiler/nested.ods and
b/sc/qa/unit/data/ods/opencl/compiler/nested.ods differ
commit 521acd05aba7fb35a244a6302584a950d4904fbe
Author: Wei Wei wei...@multicorewareinc.com
Date: Fri Nov 22 17:16:49 2013 -0600
GPU Calc: Sum of product doesn't check out-of-bound accesses
AMLOEXT-244 FIX
Change-Id: I5f49f7acccaabd2a97d8ac4bfba4b973889278f1
diff --git a/sc/source/core/opencl/formulagroupcl.cxx
b/sc/source/core/opencl/formulagroupcl.cxx
index 2c795cc..ad6ccc1 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -444,13 +444,26 @@ public:
}
virtual void GenSlidingWindowFunction(std::stringstream ) {}
-virtual std::string GenSlidingWindowDeclRef(bool=false) const
+virtual std::string GenSlidingWindowDeclRef(bool nested=false) const
{
+size_t nArrayLength = mpDVR-GetArrayLength();
std::stringstream ss;
if (!bIsStartFixed !bIsEndFixed)
+{
+if (nested)
+ss ((i+gid0) nArrayLength ?;
ss Base::GetName() [i + gid0];
+if (nested)
+ss :NAN);
+}
else
+{
+if (nested)
+ss (i nArrayLength ?;
ss Base::GetName() [i];
+if (nested)
+ss :NAN);
+}
return ss.str();
}
/// Controls how the elements in the DoubleVectorRef are traversed
@@ -533,9 +546,9 @@ return nCurWindowSize;
if(count==0){
temp1 if(i + gid0
mpDVR-GetArrayLength();
temp1 ){\n\t\t;
-temp1 tmp = ;
+temp1 tmp = legalize(;
temp1
mpCodeGen-Gen2(GenSlidingWindowDeclRef(), tmp);
-temp1 ;\n\t\t\t;
+temp1 , tmp);\n\t\t\t;
temp1 }\n\t;
}
ss temp1.str();
@@ -548,9 +561,9 @@ return nCurWindowSize;
if(count==nCurWindowSize/outLoopSize*outLoopSize){
temp2 if(i + gid0mpDVR-GetArrayLength();
temp2 ){\n\t\t;
-temp2 tmp = ;
+temp2 tmp = legalize(;
temp2 mpCodeGen-Gen2(GenSlidingWindowDeclRef(),
tmp);
-temp2 ;\n\t\t\t;
+temp2 , tmp);\n\t\t\t;
temp2 }\n\t;
}
ss temp2.str();
@@ -571,9 +584,9 @@ return nCurWindowSize;
for(int count=0; count outLoopSize; count++){
ss i = outLoop*outLoopSize+count;\n\t;
if(count==0){
-temp1 tmp = ;
+temp1 tmp = legalize(;
temp1
mpCodeGen-Gen2(GenSlidingWindowDeclRef(), tmp);
-temp1 ;\n\t\t\t;
+temp1 , tmp);\n\t\t\t;
}
ss temp1.str();
}
@@ -583,9 +596,9 @@ return nCurWindowSize;
for(unsigned int count=nCurWindowSize/outLoopSize*outLoopSize;
count nCurWindowSize; count++){
ss i = count;\n\t;
if(count==nCurWindowSize/outLoopSize*outLoopSize){
-temp2 tmp = ;
+temp2 tmp = legalize(;
temp2 mpCodeGen-Gen2(GenSlidingWindowDeclRef(),
tmp);
-temp2 ;\n\t\t\t;
+temp2 , tmp);\n\t\t\t;
}
ss temp2.str();
}
@@ -673,13 +686,13 @@ public:
ss tmp = mpCodeGen-GetBottom() ;\n;
ss int loopOffset = l*512;\n;
ss if((loopOffset + lidx + offset + 256) end) {\n;
-ss tmp = mpCodeGen-Gen2(
-A[loopOffset + lidx + offset], tmp
[Libreoffice-commits] core.git: sc/source
sc/source/core/opencl/formulagroupcl.cxx | 37 +++
sc/source/core/opencl/op_math.cxx|4 ++-
2 files changed, 21 insertions(+), 20 deletions(-)
New commits:
commit ff983712ecaf91b3dc1ef432fc3fc64de2e29f13
Author: Wei Wei wei...@multicorewareinc.com
Date: Sun Nov 17 20:43:18 2013 -0600
GPU calc: refactor code for sum reduction and
sumifs naming rule
Change-Id: I685d263337bebe236befa5e5f45356336936c998
Signed-off-by: I-Jui (Ray) Sung r...@multicorewareinc.com
diff --git a/sc/source/core/opencl/formulagroupcl.cxx
b/sc/source/core/opencl/formulagroupcl.cxx
index 43aa30d..a25f729 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -411,7 +411,7 @@ class DynamicKernelSlidingArgument: public Base
public:
DynamicKernelSlidingArgument(const std::string s,
FormulaTreeNodeRef ft, boost::shared_ptrSlidingFunctionBase
CodeGen):
-Base(s, ft), mpCodeGen(CodeGen), needReductionKernel(true),
mpClmem2(NULL)
+Base(s, ft), mpCodeGen(CodeGen), mpClmem2(NULL)
{
FormulaToken *t = ft-GetFormulaToken();
if (t-GetType() != formula::svDoubleVectorRef)
@@ -428,7 +428,7 @@ public:
(!GetStartFixed() !GetEndFixed()) ) ;
}
virtual void GenSlidingWindowFunction(std::stringstream ss) {
-if (dynamic_castOpSum*(mpCodeGen.get()))
+if (dynamic_castOpSum*(mpCodeGen.get()) NeedParallelReduction())
{
std::string name = Base::GetName();
ss __kernel void name;
@@ -436,20 +436,23 @@ public:
__global double *result,int arrayLength,int windowSize){\n;
ss double tmp, current_result = 0.0;\n;
ss int writePos = get_group_id(1);\n;
-ss int offset = get_group_id(1);\n;
ss int lidx = get_local_id(0);\n;
ss __local double shm_buf[256];\n;
-ss if (arrayLength == windowSize)\n;
-ss offset = 0;\n;
+if (mpDVR-IsStartFixed() mpDVR-IsEndFixed())
+ss int offset = 0;\n;
+else if (!mpDVR-IsStartFixed() !mpDVR-IsEndFixed())
+ss int offset = get_group_id(1);\n;
+else
+throw Unhandled();
ss barrier(CLK_LOCAL_MEM_FENCE);\n;
ss int loop = arrayLength/512 + 1;\n;
ss for (int l=0; lloop; l++){\n;
ss tmp = 0.0;\n;
ss int loopOffset = l*512;\n;
-ss if((loopOffset + lidx + offset + 256) ( offset +
windowSize))\n;
+ss if((loopOffset + lidx + offset + 256) min( offset +
windowSize, arrayLength))\n;
ss tmp = A[loopOffset + lidx + offset] +
A[loopOffset + lidx + offset + 256];\n;
-ss else if ((loopOffset + lidx + offset) ( offset +
windowSize))\n;
+ss else if ((loopOffset + lidx + offset) min(offset +
windowSize, arrayLength))\n;
ss tmp = A[loopOffset + lidx + offset];\n;
ss shm_buf[lidx] = tmp;\n;
ss barrier(CLK_LOCAL_MEM_FENCE);\n;
@@ -487,38 +490,35 @@ public:
{
if (!bIsStartFixed !bIsEndFixed)
{
-// set 100 as a threshold for invoking reduction kernel
-// Ray: temporarily turn off parallel sum reduction
-if (false /*nCurWindowSize 100*/)
+// set 100 as a temporary threshold for invoking reduction
+// kernel in NeedParalleLReduction function
+if (NeedParallelReduction())
{
std::string temp = Base::GetName() + [gid0];
ss tmp = ;
ss mpCodeGen-Gen2(temp, tmp);
ss ;\n\t;
needBody = false;
-needReductionKernel = false;
return nCurWindowSize;
}
}
if (bIsStartFixed bIsEndFixed)
{
-// set 100 as a threshold for invoking reduction kernel
-// Ray: temporarily turn off parallel sum reduction
-if (false /* nCurWindowSize 100 */)
+// set 100 as a temporary threshold for invoking reduction
+// kernel in NeedParalleLReduction function
+if (NeedParallelReduction())
{
std::string temp = Base::GetName() + [0];
ss tmp = ;
ss mpCodeGen-Gen2(temp, tmp);
ss ;\n\t;
needBody = false;
-needReductionKernel = false;
return nCurWindowSize;
}
}
}
needBody = true;
-needReductionKernel
[Libreoffice-commits] core.git: Branch 'feature/calc-group-interpreter-4' - sc/source
sc/source/core/opencl/formulagroupcl.cxx | 37 +++
sc/source/core/opencl/op_math.cxx|4 ++-
2 files changed, 21 insertions(+), 20 deletions(-)
New commits:
commit 12172c73b45e453c72e3afdc4459a6642711d23e
Author: Wei Wei wei...@multicorewareinc.com
Date: Sun Nov 17 20:43:18 2013 -0600
GPU calc: refactor code for sum reduction and
sumifs naming rule
Change-Id: I685d263337bebe236befa5e5f45356336936c998
Signed-off-by: I-Jui (Ray) Sung r...@multicorewareinc.com
diff --git a/sc/source/core/opencl/formulagroupcl.cxx
b/sc/source/core/opencl/formulagroupcl.cxx
index dfb2037..eb9a28f 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -411,7 +411,7 @@ class DynamicKernelSlidingArgument: public Base
public:
DynamicKernelSlidingArgument(const std::string s,
FormulaTreeNodeRef ft, boost::shared_ptrSlidingFunctionBase
CodeGen):
-Base(s, ft), mpCodeGen(CodeGen), needReductionKernel(true),
mpClmem2(NULL)
+Base(s, ft), mpCodeGen(CodeGen), mpClmem2(NULL)
{
FormulaToken *t = ft-GetFormulaToken();
if (t-GetType() != formula::svDoubleVectorRef)
@@ -428,7 +428,7 @@ public:
(!GetStartFixed() !GetEndFixed()) ) ;
}
virtual void GenSlidingWindowFunction(std::stringstream ss) {
-if (dynamic_castOpSum*(mpCodeGen.get()))
+if (dynamic_castOpSum*(mpCodeGen.get()) NeedParallelReduction())
{
std::string name = Base::GetName();
ss __kernel void name;
@@ -436,20 +436,23 @@ public:
__global double *result,int arrayLength,int windowSize){\n;
ss double tmp, current_result = 0.0;\n;
ss int writePos = get_group_id(1);\n;
-ss int offset = get_group_id(1);\n;
ss int lidx = get_local_id(0);\n;
ss __local double shm_buf[256];\n;
-ss if (arrayLength == windowSize)\n;
-ss offset = 0;\n;
+if (mpDVR-IsStartFixed() mpDVR-IsEndFixed())
+ss int offset = 0;\n;
+else if (!mpDVR-IsStartFixed() !mpDVR-IsEndFixed())
+ss int offset = get_group_id(1);\n;
+else
+throw Unhandled();
ss barrier(CLK_LOCAL_MEM_FENCE);\n;
ss int loop = arrayLength/512 + 1;\n;
ss for (int l=0; lloop; l++){\n;
ss tmp = 0.0;\n;
ss int loopOffset = l*512;\n;
-ss if((loopOffset + lidx + offset + 256) ( offset +
windowSize))\n;
+ss if((loopOffset + lidx + offset + 256) min( offset +
windowSize, arrayLength))\n;
ss tmp = A[loopOffset + lidx + offset] +
A[loopOffset + lidx + offset + 256];\n;
-ss else if ((loopOffset + lidx + offset) ( offset +
windowSize))\n;
+ss else if ((loopOffset + lidx + offset) min(offset +
windowSize, arrayLength))\n;
ss tmp = A[loopOffset + lidx + offset];\n;
ss shm_buf[lidx] = tmp;\n;
ss barrier(CLK_LOCAL_MEM_FENCE);\n;
@@ -487,38 +490,35 @@ public:
{
if (!bIsStartFixed !bIsEndFixed)
{
-// set 100 as a threshold for invoking reduction kernel
-// Ray: temporarily turn off parallel sum reduction
-if (false /*nCurWindowSize 100*/)
+// set 100 as a temporary threshold for invoking reduction
+// kernel in NeedParalleLReduction function
+if (NeedParallelReduction())
{
std::string temp = Base::GetName() + [gid0];
ss tmp = ;
ss mpCodeGen-Gen2(temp, tmp);
ss ;\n\t;
needBody = false;
-needReductionKernel = false;
return nCurWindowSize;
}
}
if (bIsStartFixed bIsEndFixed)
{
-// set 100 as a threshold for invoking reduction kernel
-// Ray: temporarily turn off parallel sum reduction
-if (false /* nCurWindowSize 100 */)
+// set 100 as a temporary threshold for invoking reduction
+// kernel in NeedParalleLReduction function
+if (NeedParallelReduction())
{
std::string temp = Base::GetName() + [0];
ss tmp = ;
ss mpCodeGen-Gen2(temp, tmp);
ss ;\n\t;
needBody = false;
-needReductionKernel = false;
return nCurWindowSize;
}
}
}
needBody = true;
-needReductionKernel
[Libreoffice-commits] core.git: Branch 'feature/calc-group-interpreter-4' - 2 commits - sc/source
sc/source/core/opencl/formulagroupcl.cxx | 429 +--
sc/source/core/opencl/op_math.cxx| 225 +---
sc/source/core/opencl/op_math.hxx|4
sc/source/core/opencl/opbase.cxx | 43 +++
sc/source/core/opencl/opbase.hxx |8
5 files changed, 590 insertions(+), 119 deletions(-)
New commits:
commit 483da7cdb5082821541b1897ad81b8ddf55ff1a7
Author: Wei Wei wei...@multicorewareinc.com
Date: Fri Nov 15 17:33:19 2013 -0600
GPU Calc: implemented parallel reduction for SUMIFS
For now only works for fixed and sliding fixed-sized windows.
Change-Id: I25e3f893a86d0e1723ae1e1633ff93926b8d
Signed-off-by: I-Jui (Ray) Sung r...@multicorewareinc.com
diff --git a/sc/source/core/opencl/formulagroupcl.cxx
b/sc/source/core/opencl/formulagroupcl.cxx
index 3b19886..427dd9e 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -419,42 +419,51 @@ public:
bIsStartFixed = mpDVR-IsStartFixed();
bIsEndFixed = mpDVR-IsEndFixed();
}
+virtual bool NeedParallelReduction(void) const
+{
+return GetWindowSize() 100
+( (GetStartFixed() GetEndFixed()) ||
+ (!GetStartFixed() !GetEndFixed()) ) ;
+}
virtual void GenSlidingWindowFunction(std::stringstream ss) {
-std::string name = Base::GetName();
-ss __kernel void name;
-ss _reduction(__global double* A,
-__global double *result,int arrayLength,int windowSize){\n;
-ss double tmp, current_result = 0.0;\n;
-ss int writePos = get_group_id(1);\n;
-ss int offset = get_group_id(1);\n;
-ss int lidx = get_local_id(0);\n;
-ss __local double shm_buf[256];\n;
-ss if (arrayLength == windowSize)\n;
-ss offset = 0;\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss int loop = arrayLength/512 + 1;\n;
-ss for (int l=0; lloop; l++){\n;
-ss tmp = 0.0;\n;
-ss int loopOffset = l*512;\n;
-ss if((loopOffset + lidx + offset + 256) ( offset +
windowSize))\n;
-ss tmp = A[loopOffset + lidx + offset] +
-A[loopOffset + lidx + offset + 256];\n;
-ss else if ((loopOffset + lidx + offset) ( offset +
windowSize))\n;
-ss tmp = A[loopOffset + lidx + offset];\n;
-ss shm_buf[lidx] = tmp;\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss for (int i = 128; i 0; i/=2) {\n;
-ss if (lidx i)\n;
-ss shm_buf[lidx] += shm_buf[lidx + i];\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss }\n;
-ss if (lidx == 0)\n;
-ss current_result += shm_buf[0];\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss }\n;
-ss if (lidx == 0)\n;
-ss result[writePos] = current_result;\n;
-ss }\n;
+if (dynamic_castOpSum*(mpCodeGen.get()))
+{
+std::string name = Base::GetName();
+ss __kernel void name;
+ss _reduction(__global double* A,
+__global double *result,int arrayLength,int windowSize){\n;
+ss double tmp, current_result = 0.0;\n;
+ss int writePos = get_group_id(1);\n;
+ss int offset = get_group_id(1);\n;
+ss int lidx = get_local_id(0);\n;
+ss __local double shm_buf[256];\n;
+ss if (arrayLength == windowSize)\n;
+ss offset = 0;\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss int loop = arrayLength/512 + 1;\n;
+ss for (int l=0; lloop; l++){\n;
+ss tmp = 0.0;\n;
+ss int loopOffset = l*512;\n;
+ss if((loopOffset + lidx + offset + 256) ( offset +
windowSize))\n;
+ss tmp = A[loopOffset + lidx + offset] +
+A[loopOffset + lidx + offset + 256];\n;
+ss else if ((loopOffset + lidx + offset) ( offset +
windowSize))\n;
+ss tmp = A[loopOffset + lidx + offset];\n;
+ss shm_buf[lidx] = tmp;\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss for (int i = 128; i 0; i/=2) {\n;
+ss if (lidx i)\n;
+ss shm_buf[lidx] += shm_buf[lidx + i];\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss }\n;
+ss if (lidx == 0)\n;
+ss current_result += shm_buf[0];\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss }\n;
+ss if (lidx == 0)\n;
+ss result[writePos] = current_result;\n;
+ss }\n
[Libreoffice-commits] core.git: 2 commits - sc/source
sc/source/core/opencl/formulagroupcl.cxx | 424 +--
sc/source/core/opencl/op_math.cxx| 225 +---
sc/source/core/opencl/op_math.hxx|4
sc/source/core/opencl/opbase.cxx | 43 +++
sc/source/core/opencl/opbase.hxx |8
5 files changed, 588 insertions(+), 116 deletions(-)
New commits:
commit 1e3bc2925c0ec1b03d6ae7cf3f281b0df3ec88d3
Author: Wei Wei wei...@multicorewareinc.com
Date: Fri Nov 15 17:33:19 2013 -0600
GPU Calc: implemented parallel reduction for SUMIFS
For now only works for fixed and sliding fixed-sized windows.
Change-Id: I25e3f893a86d0e1723ae1e1633ff93926b8d
Signed-off-by: I-Jui (Ray) Sung r...@multicorewareinc.com
diff --git a/sc/source/core/opencl/formulagroupcl.cxx
b/sc/source/core/opencl/formulagroupcl.cxx
index e4b6bfb..79c33ae 100644
--- a/sc/source/core/opencl/formulagroupcl.cxx
+++ b/sc/source/core/opencl/formulagroupcl.cxx
@@ -419,42 +419,51 @@ public:
bIsStartFixed = mpDVR-IsStartFixed();
bIsEndFixed = mpDVR-IsEndFixed();
}
+virtual bool NeedParallelReduction(void) const
+{
+return GetWindowSize() 100
+( (GetStartFixed() GetEndFixed()) ||
+ (!GetStartFixed() !GetEndFixed()) ) ;
+}
virtual void GenSlidingWindowFunction(std::stringstream ss) {
-std::string name = Base::GetName();
-ss __kernel void name;
-ss _reduction(__global double* A,
-__global double *result,int arrayLength,int windowSize){\n;
-ss double tmp, current_result = 0.0;\n;
-ss int writePos = get_group_id(1);\n;
-ss int offset = get_group_id(1);\n;
-ss int lidx = get_local_id(0);\n;
-ss __local double shm_buf[256];\n;
-ss if (arrayLength == windowSize)\n;
-ss offset = 0;\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss int loop = arrayLength/512 + 1;\n;
-ss for (int l=0; lloop; l++){\n;
-ss tmp = 0.0;\n;
-ss int loopOffset = l*512;\n;
-ss if((loopOffset + lidx + offset + 256) ( offset +
windowSize))\n;
-ss tmp = A[loopOffset + lidx + offset] +
-A[loopOffset + lidx + offset + 256];\n;
-ss else if ((loopOffset + lidx + offset) ( offset +
windowSize))\n;
-ss tmp = A[loopOffset + lidx + offset];\n;
-ss shm_buf[lidx] = tmp;\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss for (int i = 128; i 0; i/=2) {\n;
-ss if (lidx i)\n;
-ss shm_buf[lidx] += shm_buf[lidx + i];\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss }\n;
-ss if (lidx == 0)\n;
-ss current_result += shm_buf[0];\n;
-ss barrier(CLK_LOCAL_MEM_FENCE);\n;
-ss }\n;
-ss if (lidx == 0)\n;
-ss result[writePos] = current_result;\n;
-ss }\n;
+if (dynamic_castOpSum*(mpCodeGen.get()))
+{
+std::string name = Base::GetName();
+ss __kernel void name;
+ss _reduction(__global double* A,
+__global double *result,int arrayLength,int windowSize){\n;
+ss double tmp, current_result = 0.0;\n;
+ss int writePos = get_group_id(1);\n;
+ss int offset = get_group_id(1);\n;
+ss int lidx = get_local_id(0);\n;
+ss __local double shm_buf[256];\n;
+ss if (arrayLength == windowSize)\n;
+ss offset = 0;\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss int loop = arrayLength/512 + 1;\n;
+ss for (int l=0; lloop; l++){\n;
+ss tmp = 0.0;\n;
+ss int loopOffset = l*512;\n;
+ss if((loopOffset + lidx + offset + 256) ( offset +
windowSize))\n;
+ss tmp = A[loopOffset + lidx + offset] +
+A[loopOffset + lidx + offset + 256];\n;
+ss else if ((loopOffset + lidx + offset) ( offset +
windowSize))\n;
+ss tmp = A[loopOffset + lidx + offset];\n;
+ss shm_buf[lidx] = tmp;\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss for (int i = 128; i 0; i/=2) {\n;
+ss if (lidx i)\n;
+ss shm_buf[lidx] += shm_buf[lidx + i];\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss }\n;
+ss if (lidx == 0)\n;
+ss current_result += shm_buf[0];\n;
+ss barrier(CLK_LOCAL_MEM_FENCE);\n;
+ss }\n;
+ss if (lidx == 0)\n;
+ss result[writePos] = current_result;\n;
+ss }\n
