aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

+1

Browse Source
This commit is contained in:
Evghenii
2014-01-29 13:38:50 +01:00
parent 36ee8911b4
commit 784eb2d15b
2 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples_ptx/radixSort/Makefile_gpu
View File

@@ -2,7 +2,7 @@ PROG=radixSort
ISPC_SRC=radixSort.ispc ISPC_SRC=radixSort.ispc
CU_SRC=radixSort.cu CU_SRC=radixSort.cu
NVCC_FLAGS=-Xptxas=-O1 # NVCC_FLAGS=-Xptxas=-O1
CXX_SRC=radixSort.cpp radixSort.cpp CXX_SRC=radixSort.cpp radixSort.cpp
PTXCC_REGMAX=64 PTXCC_REGMAX=64

1
examples_ptx/radixSort/radixSort.cu
View File

@@ -109,6 +109,7 @@ void sortPass(
const int key = mask & ((unsigned int)keys[i] >> bit); const int key = mask & ((unsigned int)keys[i] >> bit);
int scatter; int scatter;
/* not a vector friendly loop */ /* not a vector friendly loop */
#pragma unroll 1 /* needed, otherwise compiler unroll and optimizes the result :S */
for (int iv = 0; iv < programCount; iv++) for (int iv = 0; iv < programCount; iv++)
if (programIndex == iv) if (programIndex == iv)
scatter = digitOffsets[key]++; scatter = digitOffsets[key]++;