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

Updating stream caching (#614)

Browse Source 
- Adding non-captured hipStream for use in setup
This commit is contained in:
gilbertlee-amd
2022-09-09 16:30:15 -06:00
committed by GitHub
parent 65d78e9a1d
commit dd56135a9a
10 changed files with 30 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/transport/coll_net.cc
+1 -1
View File
@@ -351,7 +351,7 @@ static ncclResult_t sharedBuffersInit(struct ncclComm* comm, int cuda, char** gp
*size = state->size;
if (cuda && state->cudaBuff == NULL) {
NCCLCHECK(ncclCudaCalloc(&state->cudaBuff, *size, cuda));
NCCLCHECK(ncclCudaCalloc(&state->cudaBuff, *size, comm->sideStream, cuda));
}
if (!cuda && state->hostBuff == NULL) {
NCCLCHECK(ncclCudaHostCalloc(&state->hostBuff, *size));
src/transport/net.cc
+3 -3
View File
@@ -396,7 +396,7 @@ static ncclResult_t sharedBuffersInit(struct ncclComm* comm, int cuda, int local
if (size) *size = state->size;
if (cuda && state->cudaBuff == NULL) {
NCCLCHECK(ncclCudaCalloc(&state->cudaBuff, state->size, cuda));
NCCLCHECK(ncclCudaCalloc(&state->cudaBuff, state->size, comm->sideStream, cuda));
if (sameProcess == 0) {
CUDACHECK(hipIpcGetMemHandle(&state->ipc, state->cudaBuff));
}
@@ -575,7 +575,7 @@ static ncclResult_t sendProxyConnect(struct ncclProxyConnection* connection, str
if (!map->sameProcess) {
ALIGN_SIZE(map->mems[NCCL_NET_MAP_DEVMEM].size, CUDA_IPC_MIN);
}
NCCLCHECK(ncclCudaCalloc(&map->mems[NCCL_NET_MAP_DEVMEM].gpuPtr, map->mems[NCCL_NET_MAP_DEVMEM].size, resources->useGdr));
NCCLCHECK(ncclCudaCalloc(&map->mems[NCCL_NET_MAP_DEVMEM].gpuPtr, map->mems[NCCL_NET_MAP_DEVMEM].size, comm->sideStream, resources->useGdr));
map->mems[NCCL_NET_MAP_DEVMEM].cpuPtr = map->mems[NCCL_NET_MAP_DEVMEM].gpuPtr;
}
if (!map->sameProcess) {
@@ -689,7 +689,7 @@ static ncclResult_t recvProxyConnect(struct ncclProxyConnection* connection, str
if (map->mems[NCCL_NET_MAP_DEVMEM].size) {
if (resources->shared == 0) {
NCCLCHECK(ncclCudaCalloc(&map->mems[NCCL_NET_MAP_DEVMEM].gpuPtr, map->mems[NCCL_NET_MAP_DEVMEM].size, resources->useGdr));
NCCLCHECK(ncclCudaCalloc(&map->mems[NCCL_NET_MAP_DEVMEM].gpuPtr, map->mems[NCCL_NET_MAP_DEVMEM].size, comm->sideStream, resources->useGdr));
map->mems[NCCL_NET_MAP_DEVMEM].cpuPtr = map->mems[NCCL_NET_MAP_DEVMEM].gpuPtr;
}
}
src/transport/p2p.cc
+2 -2
View File
@@ -115,7 +115,7 @@ ncclResult_t p2pCanConnect(int* ret, struct ncclTopoSystem* topo, struct ncclTop
// Check that legacy IPC support is available (WSL WAR)
char *dummy;
hipIpcMemHandle_t ipc;
NCCLCHECK(ncclCudaCalloc(&dummy, CUDA_IPC_MIN));
CUDACHECK(hipMalloc(&dummy, CUDA_IPC_MIN));
if (hipIpcGetMemHandle(&ipc, dummy) != hipSuccess) {
INFO(NCCL_INIT|NCCL_P2P,"Legacy IPC not supported");
*ret = 0;
@@ -347,7 +347,7 @@ static ncclResult_t p2pProxySetup(struct ncclProxyConnection* connection, struct
int size = *((int*)reqBuff);
if (respSize != sizeof(struct ncclP2pBuff)) return ncclInternalError;
struct ncclP2pBuff* p2pBuff = (struct ncclP2pBuff*)respBuff;
NCCLCHECK(ncclCudaCalloc((char**)&p2pBuff->directPtr, size, true));
NCCLCHECK(ncclCudaCalloc((char**)&p2pBuff->directPtr, size, comm->sideStream, true));
connection->transportResources = p2pBuff->directPtr;
hipError_t res = hipIpcGetMemHandle(&p2pBuff->devIpc, p2pBuff->directPtr);
if (res != hipSuccess) {