[clfft] 10/21: code cleanup
Ghislain Vaillant
ghisvail-guest at moszumanska.debian.org
Wed Mar 16 13:14:04 UTC 2016
This is an automated email from the git hooks/post-receive script.
ghisvail-guest pushed a commit to branch master
in repository clfft.
commit ffa4245c9afc338a047860df64f772392c57f20d
Author: Timmy <timmy.liu at amd.com>
Date: Mon Feb 29 13:19:35 2016 -0600
code cleanup
---
src/library/action.transpose.cpp | 6 ------
src/library/generator.transpose.cpp | 2 +-
src/library/plan.cpp | 26 ++++++++++++--------------
src/tests/accuracy_test_pow2.cpp | 26 +++++++++++++-------------
4 files changed, 26 insertions(+), 34 deletions(-)
diff --git a/src/library/action.transpose.cpp b/src/library/action.transpose.cpp
index baf4022..5806812 100644
--- a/src/library/action.transpose.cpp
+++ b/src/library/action.transpose.cpp
@@ -220,7 +220,6 @@ clfftStatus FFTGeneratedTransposeNonSquareAction::generateKernel(FFTRepo& fftRep
}
}
OPENCL_V(clfft_transpose_generator::genTransposeKernelLeadingDimensionBatched(this->signature, programCode, lwSize, reShapeFactor), _T("genTransposeKernel() failed!"));
- //std::cout << programCode << std::endl;//TIMMY
}
else if (this->signature.nonSquareKernelType == NON_SQUARE_TRANS_TRANSPOSE_BATCHED)
{
@@ -244,13 +243,9 @@ clfftStatus FFTGeneratedTransposeNonSquareAction::generateKernel(FFTRepo& fftRep
}
}
OPENCL_V(clfft_transpose_generator::genTransposeKernelBatched(this->signature, programCode, lwSize, reShapeFactor), _T("genTransposeKernel() failed!"));
- //std::cout << programCode << std::endl;//TIMMY
}
else
{
- //No pre-callback possible in swap kernel
- //assert(!this->signature.fft_hasPreCallback);
-
//pre-callback is possible in swap kernel now
if (this->signature.fft_hasPreCallback && this->signature.fft_preCallback.localMemSize > 0)
{
@@ -271,7 +266,6 @@ clfftStatus FFTGeneratedTransposeNonSquareAction::generateKernel(FFTRepo& fftRep
}
}
OPENCL_V(clfft_transpose_generator::genSwapKernel(this->signature, programCode, lwSize, reShapeFactor), _T("genSwapKernel() failed!"));
- //std::cout << programCode << std::endl;//TIMMY
}
cl_int status = CL_SUCCESS;
diff --git a/src/library/generator.transpose.cpp b/src/library/generator.transpose.cpp
index b07907d..a8f417d 100644
--- a/src/library/generator.transpose.cpp
+++ b/src/library/generator.transpose.cpp
@@ -38,7 +38,7 @@ void OffsetCalc(std::stringstream& transKernel, const FFTKernelGenKeyParams& par
for (size_t i = params.fft_DataDim - 2; i > 0; i--)
{
- clKernWrite(transKernel, 3) << offset << " += (g_index/numGroupsY_" << i << ")*" << stride[i + 1] << ";" << std::endl;//TIMMY
+ clKernWrite(transKernel, 3) << offset << " += (g_index/numGroupsY_" << i << ")*" << stride[i + 1] << ";" << std::endl;
//clKernWrite(transKernel, 3) << offset << " += (g_index/numGroupsY_" << i << ")*" << 1048576 << ";" << std::endl;
clKernWrite(transKernel, 3) << "g_index = g_index % numGroupsY_" << i << ";" << std::endl;
}
diff --git a/src/library/plan.cpp b/src/library/plan.cpp
index 895f1d8..bd65bab 100644
--- a/src/library/plan.cpp
+++ b/src/library/plan.cpp
@@ -775,7 +775,7 @@ clfftStatus clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
trans2Plan->oDist = clLengths[1] * trans2Plan->outStride[1];
trans2Plan->gen = transGen;
- //if(transGen != Transpose_NONSQUARE)//Timmy was commented
+ //if(transGen != Transpose_NONSQUARE)
trans2Plan->large1D = fftPlan->length[0];
trans2Plan->transflag = true;
@@ -830,7 +830,7 @@ clfftStatus clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
row2Plan->iDist *= fftPlan->length[index];
row2Plan->oDist *= fftPlan->length[index];
}
- //Timmy was group commented
+
//if (transGen == Transpose_NONSQUARE)
//{
// row2Plan->large1D = fftPlan->length[0];
@@ -1971,17 +1971,17 @@ clfftStatus clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
*/
enum NON_SQUARE_KERNEL_ORDER
{
- SWAP_AND_TRANSPOSE,
- TRANSPOSE_AND_SWAP,
- TRANSPOSE_LEADING_AND_SWAP,
+ SWAP_AND_TRANSPOSE, // A.
+ TRANSPOSE_AND_SWAP, // B.
+ TRANSPOSE_LEADING_AND_SWAP, // C.
};
NON_SQUARE_KERNEL_ORDER currKernelOrder;
- //controling the transpose and swap kernel order
+ // controling the transpose and swap kernel order
+ // if leading dim is larger than the other dim it makes sense to swap and transpose
if (clLengths[0] > clLengths[1])
{
currKernelOrder = SWAP_AND_TRANSPOSE;
- std::cout << "SWAP_AND_TRANSPOSE" << std::endl;
}
else
{
@@ -1990,20 +1990,18 @@ clfftStatus clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
//currently tranpose twiddling is only supported in below case
//TODO support tranpose twiddling for all cases.
currKernelOrder = TRANSPOSE_LEADING_AND_SWAP;
- std::cout << "TRANSPOSE_LEADING_AND_SWAP" << std::endl;
}
else
{
currKernelOrder = TRANSPOSE_AND_SWAP;
- std::cout << "TRANSPOSE_AND_SWAP" << std::endl;
}
}
//if the original input data is more than 1d only TRANSPOSE_LEADING_AND_SWAP order is supported
- //TODO need to fix this here. related to sub batch size.
+ //TODO need to fix this here. related to multi dim batch size.
if (fftPlan->length.size() > 2)
currKernelOrder = TRANSPOSE_LEADING_AND_SWAP;
- else
- std::cout << "new order" << std::endl;
+ //ends tranpose kernel order
+
//Transpose stage 1
OPENCL_V(clfftCreateDefaultPlanInternal(&fftPlan->planTX, fftPlan->context, CLFFT_2D, clLengths),
_T("CreateDefaultPlan transpose_nsq_stage1 plan failed"));
@@ -2027,7 +2025,7 @@ clfftStatus clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
trans1Plan->oDist = fftPlan->oDist;
trans1Plan->gen = Transpose_NONSQUARE;
if(currKernelOrder == SWAP_AND_TRANSPOSE)
- trans1Plan->nonSquareKernelType = NON_SQUARE_TRANS_SWAP;// was NON_SQUARE_TRANS_TRANSPOSE_BATCHED_LEADING;
+ trans1Plan->nonSquareKernelType = NON_SQUARE_TRANS_SWAP;
else if (currKernelOrder == TRANSPOSE_AND_SWAP)
trans1Plan->nonSquareKernelType = NON_SQUARE_TRANS_TRANSPOSE_BATCHED;
else
@@ -2100,7 +2098,7 @@ clfftStatus clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
trans2Plan->oDist = fftPlan->oDist;
trans2Plan->gen = Transpose_NONSQUARE;
if (currKernelOrder == SWAP_AND_TRANSPOSE)
- trans2Plan->nonSquareKernelType = NON_SQUARE_TRANS_TRANSPOSE_BATCHED; //was NON_SQUARE_TRANS_SWAP;
+ trans2Plan->nonSquareKernelType = NON_SQUARE_TRANS_TRANSPOSE_BATCHED;
else if(currKernelOrder == TRANSPOSE_AND_SWAP)
trans2Plan->nonSquareKernelType = NON_SQUARE_TRANS_SWAP;
else
diff --git a/src/tests/accuracy_test_pow2.cpp b/src/tests/accuracy_test_pow2.cpp
index e00a5fc..4c8a0c6 100644
--- a/src/tests/accuracy_test_pow2.cpp
+++ b/src/tests/accuracy_test_pow2.cpp
@@ -1298,7 +1298,7 @@ TEST_F(accuracy_test_pow2_double, large_1D_forward_in_place_complex_planar_to_co
// *****************************************************
// *****************************************************
-#define CLFFT_TEST_HUGE
+//#define CLFFT_TEST_HUGE
#ifdef CLFFT_TEST_HUGE
#define HUGE_TEST_MAKE(test_name, len, bat) \
@@ -1344,20 +1344,20 @@ void test_name() \
SP_HUGE_TEST( huge_sp_test_1, 1048576, 11 )
SP_HUGE_TEST( huge_sp_test_2, 1048576*2, 7 )
-//SP_HUGE_TEST( huge_sp_test_3, 1048576*4, 3 )
-//SP_HUGE_TEST( huge_sp_test_4, 1048576*8, 5 )
-//SP_HUGE_TEST( huge_sp_test_5, 1048576*16, 3 )
-//SP_HUGE_TEST( huge_sp_test_6, 1048576*32, 2 )
-//SP_HUGE_TEST( huge_sp_test_7, 1048576*64, 1 )
+SP_HUGE_TEST( huge_sp_test_3, 1048576*4, 3 )
+SP_HUGE_TEST( huge_sp_test_4, 1048576*8, 5 )
+SP_HUGE_TEST( huge_sp_test_5, 1048576*16, 3 )
+SP_HUGE_TEST( huge_sp_test_6, 1048576*32, 2 )
+SP_HUGE_TEST( huge_sp_test_7, 1048576*64, 1 )
DP_HUGE_TEST( huge_dp_test_1, 524288, 11 )
DP_HUGE_TEST( huge_dp_test_2, 524288*2, 7 )
-//DP_HUGE_TEST( huge_dp_test_3, 524288*4, 3 )
-//DP_HUGE_TEST( huge_dp_test_4, 524288*8, 5 )
-//DP_HUGE_TEST( huge_dp_test_5, 524288*16, 3 )
-//DP_HUGE_TEST( huge_dp_test_6, 524288*32, 2 )
-//DP_HUGE_TEST( huge_dp_test_7, 524288*64, 1 )
-/*
+DP_HUGE_TEST( huge_dp_test_3, 524288*4, 3 )
+DP_HUGE_TEST( huge_dp_test_4, 524288*8, 5 )
+DP_HUGE_TEST( huge_dp_test_5, 524288*16, 3 )
+DP_HUGE_TEST( huge_dp_test_6, 524288*32, 2 )
+DP_HUGE_TEST( huge_dp_test_7, 524288*64, 1 )
+
SP_HUGE_TEST( large_sp_test_1, 8192, 11 )
SP_HUGE_TEST( large_sp_test_2, 8192*2, 7 )
SP_HUGE_TEST( large_sp_test_3, 8192*4, 3 )
@@ -1373,7 +1373,7 @@ DP_HUGE_TEST( large_dp_test_4, 4096*8, 5 )
DP_HUGE_TEST( large_dp_test_5, 4096*16, 3 )
DP_HUGE_TEST( large_dp_test_6, 4096*32, 21 )
DP_HUGE_TEST( large_dp_test_7, 4096*64, 17 )
-*/
+
#endif
// *****************************************************
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/clfft.git
More information about the debian-science-commits
mailing list