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

Merge remote-tracking branch 'nccl/master' into develop

Browse Source 
[ROCm/rccl commit: 36343be84f]
This commit is contained in:
BertanDogancay
2025-01-23 11:48:18 -06:00
committed by Corey Derochie
parent e3a29f5eab 5ca1b6c160
commit 1b000665df
122 changed files with 9009 additions and 4702 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/gdrwrap.h
+6 -5
View File
@@ -9,6 +9,7 @@
#define NCCL_GDRWRAP_H_
#include "nccl.h"
#include "alloc.h"
#include <stdint.h> // for standard [u]intX_t types
#include <stdio.h>
#include <stdlib.h>
@@ -171,7 +172,7 @@ static ncclResult_t ncclGdrCudaCalloc(T** ptr, T** devPtr, size_t nelem, void**
char *devMem;
void *gdrMap;
mapSize = sizeof(T)*nelem;
mapSize = ncclSizeOfT<T>()*nelem;
// GDRCOPY Pinned buffer has to be a minimum of a GPU_PAGE_SIZE
ALIGN_SIZE(mapSize, GPU_PAGE_SIZE);
@@ -252,7 +253,7 @@ static ncclResult_t ncclGdrCudaCalloc(T** ptr, T** devPtr, size_t nelem, void**
char *devMem;
void *gdrMap;
mapSize = sizeof(T)*nelem;
mapSize = ncclSizeOfT<T>()*nelem;
// GDRCOPY Pinned buffer has to be a minimum of a GPU_PAGE_SIZE
ALIGN_SIZE(mapSize, GPU_PAGE_SIZE);
@@ -261,7 +262,7 @@ static ncclResult_t ncclGdrCudaCalloc(T** ptr, T** devPtr, size_t nelem, void**
uint64_t alignedAddr = (((uint64_t) devMem) + GPU_PAGE_OFFSET) & GPU_PAGE_MASK;
size_t align = alignedAddr - (uint64_t)devMem;
//TRACE(NCCL_INIT, "GDRCOPY: Pin buffer 0x%lx (%p) align %zi size %zi", alignedAddr, devMem, align, mapSize);
//TRACE(NCCL_INIT, "GDRCOPY: Pin buffer 0x%lx (%p) align %zu size %zu", alignedAddr, devMem, align, mapSize);
NCCLCHECK(wrap_gdr_pin_buffer(ncclGdrCopy, alignedAddr, mapSize, 0, 0, &mh));
NCCLCHECK(wrap_gdr_map(ncclGdrCopy, mh, &gdrMap, mapSize));
@@ -284,7 +285,7 @@ static ncclResult_t ncclGdrCudaCalloc(T** ptr, T** devPtr, size_t nelem, void**
*ptr = (T *)((char *)gdrMap+off);
if (devPtr) *devPtr = (T *)(devMem+off+align);
TRACE(NCCL_INIT, "GDRCOPY : allocated devMem %p gdrMap %p offset %lx mh %lx mapSize %zi at %p",
TRACE(NCCL_INIT, "GDRCOPY : allocated devMem %p gdrMap %p offset %lx mh %lx mapSize %zu at %p",
md->gdrDevMem, md->gdrMap, md->gdrOffset, md->gdrMh.h, md->gdrMapSize, *ptr);
return ncclSuccess;
@@ -293,7 +294,7 @@ static ncclResult_t ncclGdrCudaCalloc(T** ptr, T** devPtr, size_t nelem, void**
template <typename T>
static ncclResult_t ncclGdrCudaCopy(void *gdrHandle, T* dst, T* src, size_t nelem) {
gdr_mem_desc_t *md = (gdr_mem_desc_t*)gdrHandle;
NCCLCHECK(wrap_gdr_copy_to_mapping(md->gdrMh, dst, src, nelem*sizeof(T)));
NCCLCHECK(wrap_gdr_copy_to_mapping(md->gdrMh, dst, src, nelem*ncclSizeOfT<T>()));
return ncclSuccess;
}