Replace __HIP_PLATFORM_HCC__ with __HIP_PLATFORM_AMD__ (#1154)
[ROCm/rccl commit: 9e0c9b4ed8]
Αυτή η υποβολή περιλαμβάνεται σε:
@@ -48,7 +48,7 @@ extern __shared__ struct mscclShmemData mscclShmem;
|
||||
#endif
|
||||
|
||||
inline __device__ static void barrier(int nthreads) {
|
||||
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
|
||||
#if defined(__HIP_PLATFORM_AMD__) || defined(__HCC__) || defined(__HIPCC__)
|
||||
assert(nthreads == NCCL_MAX_NTHREADS);
|
||||
__asm__ __volatile__("s_waitcnt vmcnt(0) lgkmcnt(0)\ns_barrier");
|
||||
#else
|
||||
|
||||
Αναφορά σε νέο ζήτημα
Block a user