[clfft] 38/74: bug fix for twiddle factor computation in non square kernels.

[Ghislain Vaillant]

This is an automated email from the git hooks/post-receive script.

ghisvail-guest pushed a commit to branch debian/sid
in repository clfft.

commit dbd9745e8004ae69473ef4215ebd96c21e92b7d9
Author: santanu-thangaraj <t.santanu at gmail.com>
Date:   Wed Dec 9 22:09:10 2015 +0530

    bug fix for twiddle factor computation in non square kernels.
---
 src/library/generator.transpose.nonsquare.cpp |  4 ++--
 src/library/plan.cpp                          | 14 +++++++-------
 2 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/src/library/generator.transpose.nonsquare.cpp b/src/library/generator.transpose.nonsquare.cpp
index bfd926d..52aa64f 100644
--- a/src/library/generator.transpose.nonsquare.cpp
+++ b/src/library/generator.transpose.nonsquare.cpp
@@ -148,12 +148,12 @@ static clfftStatus genTwiddleMath(const FFTKernelGenKeyParams& params, std::stri
     if (params.fft_N[0] > params.fft_N[1])
     {
         clKernWrite(transKernel, 9) << dtComplex << " Wm = TW3step( ("<< params.fft_N[1] <<" * square_matrix_index + t_gx_p*32 + lidx) * (t_gy_p*32 + lidy + loop*8) );" << std::endl;
-        clKernWrite(transKernel, 9) << dtComplex << " Wt = TW3step( (t_gy_p*32 + lidx) * ("<< params.fft_N[1] <<" * square_matrix_index + t_gx_p*32 + lidy + loop*8) );" << std::endl;
+        clKernWrite(transKernel, 9) << dtComplex << " Wt = TW3step( ("<< params.fft_N[1] <<" * square_matrix_index + t_gy_p*32 + lidx) * (t_gx_p*32 + lidy + loop*8) );" << std::endl;
     }
     else
     {
         clKernWrite(transKernel, 9) << dtComplex << " Wm = TW3step( (t_gx_p*32 + lidx) * (" << params.fft_N[0] << " * square_matrix_index + t_gy_p*32 + lidy + loop*8) );" << std::endl;
-        clKernWrite(transKernel, 9) << dtComplex << " Wt = TW3step( (" << params.fft_N[0] << " * square_matrix_index + t_gy_p*32 + lidx) * (t_gx_p*32 + lidy + loop*8) );" << std::endl;
+        clKernWrite(transKernel, 9) << dtComplex << " Wt = TW3step( (t_gy_p*32 + lidx) * (" << params.fft_N[0] << " * square_matrix_index + t_gx_p*32 + lidy + loop*8) );" << std::endl;
     }
     clKernWrite(transKernel, 9) << dtComplex << " Tm, Tt;" << std::endl;
 
diff --git a/src/library/plan.cpp b/src/library/plan.cpp
index c018e78..9b5f6c4 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)
+				//	if(transGen != Transpose_NONSQUARE)
 						trans2Plan->large1D		  = fftPlan->length[0];
 
 					trans2Plan->transflag     = true;
@@ -831,11 +831,11 @@ clfftStatus	clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
 						row2Plan->oDist *= fftPlan->length[index];
 					}
 
-					if (transGen == Transpose_NONSQUARE)
-					{
-						row2Plan->large1D = fftPlan->length[0];
-						row2Plan->twiddleFront = true;
-					}
+//					if (transGen == Transpose_NONSQUARE)
+//					{
+//						row2Plan->large1D = fftPlan->length[0];
+//						row2Plan->twiddleFront = true;
+//					}
 
 					OPENCL_V(clfftBakePlan(fftPlan->planY, numQueues, commQueueFFT, NULL, NULL ),
 						_T( "BakePlan large1d second row plan failed" ) );
@@ -1935,7 +1935,7 @@ clfftStatus	clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
 						trans1Plan->gen = Transpose_NONSQUARE;
 						trans1Plan->nonSquareKernelType = NON_SQUARE_TRANS_TRANSPOSE;
 						trans1Plan->transflag = true;
-
+                        trans1Plan->large1D = fftPlan->large1D;
 						for (size_t index = 2; index < fftPlan->length.size(); index++)
 						{
 							trans1Plan->length.push_back(fftPlan->length[index]);

-- 
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/clfft.git