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

fixed helpers and makefile

Browse Source
This commit is contained in:
Evghenii
2014-01-05 17:25:31 +01:00
parent 6e2d59e279
commit 730e0f7098
2 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
examples_ptx/common_gpu.mk
View File

@@ -33,8 +33,8 @@ CXX_OBJS=$(CXX_SRC:%.cpp=objs_gpu/%_gcc.o)
CU_OBJS=$(CU_SRC:%.cu=objs_gpu/%_cu.o)
#NVCC_OBJS=$(NVCC_SRC:%.cu=objs_gpu/%_nvcc.o)
CXX_SRC+=../ispc_malloc.cpp
CXX_OJS+=objs/ispc_malloc_gcc.o
CXX_SRC+=ispc_malloc.cpp
CXX_OBJS+=objs_gpu/ispc_malloc_gcc.o
# PTXGEN = $(HOME)/ptxgen
# PTXGEN += -opt=3

4
examples_ptx/cuda_helpers.cuh
View File

@@ -6,6 +6,8 @@
#define taskCount0 (gridDim.x*4)
#define taskIndex1 (blockIdx.y)
#define taskCount1 (gridDim.y)
#define taskIndex (taskIndex0 + taskCount0*taskIndex1)
#define taskIndex2 (blockIdx.z)
#define taskCount2 (gridDim.z)
#define taskIndex (taskIndex0 + taskCount0*(taskIndex1 + taskCount1*taskIndex2))
#define warpIdx (threadIdx.x >> 5)
#define launch(ntx,nty,ntz,func) if (programIndex==0) func<<<dim3(((ntx)+4-1)/4,nty,ntz),128>>>