Change __CUDACC_VER_*__ preprocessor directives to CUDA_VERSION because clang doesn't define the former.
This commit is contained in:
committed by
Sylvain Jeaugey
parent
469b69a5d0
commit
d08e9b5279
+1
-1
@@ -58,7 +58,7 @@ static void* const ncclKerns[ncclCollCount*ncclNumOps*ncclNumTypes*2] = {
|
||||
};
|
||||
|
||||
ncclResult_t ncclLaunchCooperativeKernelMultiDevice(struct cudaLaunchParams *paramsList, int* cudaDevs, int numDevices, int cgMode) {
|
||||
#if __CUDACC_VER_MAJOR__ >= 9
|
||||
#if CUDA_VERSION >= 9000
|
||||
if (cgMode & 0x01) {
|
||||
CUDACHECK(cudaLaunchCooperativeKernelMultiDevice(paramsList, numDevices,
|
||||
// These flags are to reduce the latency of using this API
|
||||
|
||||
Reference in New Issue
Block a user