[clfft] 97/107: fixing some more bugs

[Ghislain Vaillant]

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

ghisvail-guest pushed a commit to branch master
in repository clfft.

commit e7d49b0ca8d201de43f89f7e6d7ddde00f20e20e
Author: bragadeesh <bragadeesh.natarajan at amd>
Date:   Wed Jun 24 19:55:54 2015 -0500

    fixing some more bugs
---
 src/library/generator.stockham.cpp | 20 ++++++++++++--------
 src/library/plan.cpp               | 10 ++++++++--
 2 files changed, 20 insertions(+), 10 deletions(-)

diff --git a/src/library/generator.stockham.cpp b/src/library/generator.stockham.cpp
index 4016acc..93d073d 100644
--- a/src/library/generator.stockham.cpp
+++ b/src/library/generator.stockham.cpp
@@ -957,7 +957,7 @@ namespace StockhamGenerator
 
 							passStr += rType; passStr += " TR, TI;\n\t\t";
 
-							if(realSpecial)
+							if(realSpecial && (flag == SR_TWMUL_3STEP))
 							{
 								if(fwd)
 								{
@@ -969,21 +969,21 @@ namespace StockhamGenerator
 									passStr += regImagIndex; passStr += ");\n\t\t";
 
 									passStr += "}\n\t\telse\n\t\t{\n\t\t";
-
-									passStr += "TR =  (W.x * "; passStr += regRealIndex; passStr += ") + (W.y * ";
+									
+									passStr += "TR = (W.x * "; passStr += regRealIndex; passStr += ") + (W.y * ";
 									passStr += regImagIndex; passStr += ");\n\t\t";
-									passStr += "TI = -(W.y * "; passStr += regRealIndex; passStr += ") + (W.x * ";
+									passStr += "TI = (W.y * "; passStr += regRealIndex; passStr += ") - (W.x * ";
 									passStr += regImagIndex; passStr += ");\n\t\t";
-
+									
 									passStr += "}\n\t\t";
 								}
 								else
 								{
 									passStr += "if(t==0)\n\t\t{\n\t\t";
 
-									passStr += "TR =  (W.x * "; passStr += regRealIndex; passStr += ") + (W.y * ";
+									passStr += "TR = (W.x * "; passStr += regRealIndex; passStr += ") + (W.y * ";
 									passStr += regImagIndex; passStr += ");\n\t\t";
-									passStr += "TI = -(W.y * "; passStr += regRealIndex; passStr += ") + (W.x * ";
+									passStr += "TI = (W.y * "; passStr += regRealIndex; passStr += ") - (W.x * ";
 									passStr += regImagIndex; passStr += ");\n\t\t";
 
 									passStr += "}\n\t\telse\n\t\t{\n\t\t";
@@ -2502,7 +2502,11 @@ namespace StockhamGenerator
 			inReal  = (params.fft_inputLayout == CLFFT_REAL) ? true : false;
 			outReal = (params.fft_outputLayout == CLFFT_REAL) ? true : false;
 
-			size_t large1D = params.fft_N[0] * params.fft_N[1];
+			size_t large1D = 0;
+			if(params.fft_realSpecial)
+				large1D = params.fft_N[0] * params.fft_realSpecial_Nr;
+			else
+				large1D = params.fft_N[0] * params.fft_N[1];
 
 			// Pragma
 			str += ClPragma<PR>();
diff --git a/src/library/plan.cpp b/src/library/plan.cpp
index 62004e4..fcf0c3a 100644
--- a/src/library/plan.cpp
+++ b/src/library/plan.cpp
@@ -811,10 +811,16 @@ clfftStatus	clfftBakePlan( clfftPlanHandle plHandle, cl_uint numQueues, cl_comma
 					if( (smallerDim % 64 == 0) || (biggerDim % 64 == 0) )
 						padding = 64;
 
+
 					if (fftPlan->tmpBufSize==0 )
 					{
-						fftPlan->tmpBufSize = (smallerDim + padding) * biggerDim *
-							fftPlan->batchsize * fftPlan->ElementSize() / 2;
+						size_t Nf = (1 + smallerDim/2) * biggerDim;
+						fftPlan->tmpBufSize = (smallerDim + padding) * biggerDim / 2;
+
+						if(fftPlan->tmpBufSize < Nf) 
+							fftPlan->tmpBufSize = Nf;
+
+						fftPlan->tmpBufSize *= ( fftPlan->batchsize * fftPlan->ElementSize() );
 
 						for (size_t index=1; index < fftPlan->length.size(); index++)
 						{

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