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

Merge branch 'develop' into 2.13.4

Browse Source
This commit is contained in:
Gilbert Lee
2022-09-09 23:07:04 +00:00
parent a79d9e3586 dd56135a9a
commit 009e79623f
12 changed files with 261 additions and 28 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
@@ -387,7 +387,7 @@ 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));
}
@@ -566,7 +566,7 @@ 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) {
@@ -704,7 +704,7 @@ 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
+4 -4
View File
@@ -156,7 +156,7 @@ 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;
@@ -425,7 +425,7 @@ static ncclResult_t p2pSendProxySetup(struct ncclProxyConnection* connection, st
NCCLCHECK(ncclCalloc(&proxyInfo, 1));
connection->transportResources = proxyInfo;
NCCLCHECK(ncclCudaCalloc(&proxyInfo->ceDevBuff, comm->buffSizes[NCCL_PROTO_SIMPLE], true));
NCCLCHECK(ncclCudaCalloc(&proxyInfo->ceDevBuff, comm->buffSizes[NCCL_PROTO_SIMPLE], comm->sideStream, true));
char shmPath[PATH_MAX];
shmPath[0] = '\0';
@@ -443,7 +443,7 @@ static ncclResult_t p2pSendProxySetup(struct ncclProxyConnection* connection, st
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) {
@@ -462,7 +462,7 @@ static ncclResult_t p2pRecvProxySetup(struct ncclProxyConnection* connection, st
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) {
src/transport/shm.cc
+2 -2
View File
@@ -210,7 +210,7 @@ static ncclResult_t shmSendProxyConnect(struct ncclProxyConnection* connection,
NCCLCHECK(ncclCalloc(&proxyInfo, 1));
if (reqSize != sizeof(struct shmProxyInfo)) return ncclInternalError;
memcpy(proxyInfo, reqBuff, reqSize);
NCCLCHECK(ncclCudaCalloc(&proxyInfo->devFifo, comm->buffSizes[NCCL_PROTO_SIMPLE]));
NCCLCHECK(ncclCudaCalloc(&proxyInfo->devFifo, comm->buffSizes[NCCL_PROTO_SIMPLE], comm->sideStream));
NCCLCHECK(ncclCudaHostCalloc(&proxyInfo->ceRecvMem, 1));
CUDACHECK(hipStreamCreateWithFlags(&proxyInfo->stream, hipStreamNonBlocking));
for (int i=0; i<NCCL_STEPS; i++) {
@@ -228,7 +228,7 @@ static ncclResult_t shmRecvProxyConnect(struct ncclProxyConnection* connection,
NCCLCHECK(ncclCalloc(&proxyInfo, 1));
if (reqSize != sizeof(struct shmProxyInfo)) return ncclInternalError;
memcpy(proxyInfo, reqBuff, reqSize);
NCCLCHECK(ncclCudaCalloc(&proxyInfo->devFifo, comm->buffSizes[NCCL_PROTO_SIMPLE]));
NCCLCHECK(ncclCudaCalloc(&proxyInfo->devFifo, comm->buffSizes[NCCL_PROTO_SIMPLE], comm->sideStream));
NCCLCHECK(ncclCudaHostCalloc(&proxyInfo->ceRecvMem, 1));
CUDACHECK(hipStreamCreateWithFlags(&proxyInfo->stream, hipStreamNonBlocking));
for (int i=0; i<NCCL_STEPS; i++) {