AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update API trace information for hipLaunchKernel

Browse Source 
[ROCm/hip commit: 883a3fe831]
This commit is contained in:
pensun
2016-04-20 09:57:55 -05:00
parent b0c6b89dd3
commit 3a25eeca87
1 changed files with 3 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/hip/include/hcc_detail/hip_runtime.h
+3 -10
View File
@@ -401,13 +401,6 @@ __device__ int __all( int input);
__device__ int __any( int input);
__device__ unsigned long long int __ballot( int input);
// __ldg function
template <typename T>
__device__ __forceinline__ T __ldg( const T * addr)
{
return *addr;
}
// warp shuffle functions
#ifdef __cplusplus
@@ -431,7 +424,7 @@ __device__ float __shfl_xor(float input, int lane_mask, int width);
#endif
__host__ __device__ int min(int arg1, int arg2);
__host__ __device__ int max(int arg1, int arg2);
__host__ __device__ int max(int arg1, int arg2);
//TODO - add a couple fast math operations here, the set here will grow :
__device__ float __cosf(float x);
@@ -522,8 +515,8 @@ do {\
lp.cf = &cf; \
hipStream_t trueStream = (ihipPreLaunchKernel(_stream, &lp.av)); \
if (HIP_TRACE_API) {\
fprintf(stderr, KGRN "<<hip-api: hipLaunchKernel '%s' gridDim:[%d.%d.%d] groupDim:[%d.%d.%d] groupMem:+%d stream=%p\n" KNRM, \
#_kernelName, lp.gridDim.z, lp.gridDim.y, lp.gridDim.x, lp.groupDim.z, lp.groupDim.y, lp.groupDim.x, lp.groupMemBytes, (void*)(_stream));\
fprintf(stderr, KGRN "<<hip-api: hipLaunchKernel '%s' gridDim:(%d,%d,%d) groupDim:(%d,%d,%d) groupMem:+%d stream=%p\n" KNRM, \
#_kernelName, lp.gridDim.x, lp.gridDim.y, lp.gridDim.z, lp.groupDim.x, lp.groupDim.y, lp.groupDim.z, lp.groupMemBytes, (void*)(_stream));\
}\
_kernelName (lp, __VA_ARGS__);\
ihipPostLaunchKernel(trueStream, cf);\