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

RCCL 2.4 update

Browse Source 
[ROCm/rccl commit: f11c8f60cd]
This commit is contained in:
Wenkai Du
2019-07-05 15:43:00 -07:00
parent a66ab68630
commit 6047487815
95 changed files with 7829 additions and 614 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/alloc.h
+9 -6
View File
@@ -12,14 +12,14 @@
#include <sys/mman.h>
static inline ncclResult_t ncclCudaHostAlloc(void** ptr, void** devPtr, size_t size) {
CUDACHECK(cudaHostAlloc(ptr, size, cudaHostAllocMapped));
CUDACHECK(hipHostMalloc(ptr, size, hipHostMallocMapped));
memset(*ptr, 0, size);
*devPtr = *ptr;
return ncclSuccess;
}
static inline ncclResult_t ncclCudaHostFree(void* ptr) {
CUDACHECK(cudaFreeHost(ptr));
CUDACHECK(hipHostFree(ptr));
return ncclSuccess;
}
@@ -36,15 +36,18 @@ static ncclResult_t ncclCalloc(T** ptr, size_t nelem) {
}
template <typename T>
static ncclResult_t ncclCudaCalloc(T** ptr, size_t nelem) {
CUDACHECK(cudaMalloc(ptr, nelem*sizeof(T)));
CUDACHECK(cudaMemset(*ptr, 0, nelem*sizeof(T)));
static ncclResult_t ncclCudaCalloc(T** ptr, size_t nelem, bool isFineGrain = false) {
if (isFineGrain)
CUDACHECK(hipExtMallocWithFlags((void**)ptr, nelem*sizeof(T), hipDeviceMallocFinegrained));
else
CUDACHECK(hipMalloc(ptr, nelem*sizeof(T)));
CUDACHECK(hipMemset(*ptr, 0, nelem*sizeof(T)));
return ncclSuccess;
}
template <typename T>
static ncclResult_t ncclCudaMemcpy(T* dst, T* src, size_t nelem) {
CUDACHECK(cudaMemcpy(dst, src, nelem*sizeof(T), cudaMemcpyDefault));
CUDACHECK(hipMemcpy(dst, src, nelem*sizeof(T), hipMemcpyDefault));
return ncclSuccess;
}