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: a79d9e3586]
This commit is contained in:
Wenkai Du
2022-09-09 01:20:52 +00:00
parent fe99249cde c9a50a9ec5
commit 7874a99c75
90 changed files with 5517 additions and 3115 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/proxy.h
+18 -10
View File
@@ -26,18 +26,26 @@ struct ncclProxyOp {
int channelId;
int nsteps;
ssize_t nbytes;
int root;
struct {
int root:30;
uint32_t connIndex:2;
};
int next;
uint64_t opCount;
int sliceSteps;
int chunkSteps;
int chunkSize;
ncclDataType_t dtype;
ncclRedOp_t redOp;
ncclPattern_t pattern; // uint8_t
uint8_t /*ncclDataType_t*/ dtype;
uint8_t /*ncclDevRedOp_t*/ redOp;
uint8_t /*ncclPattern_t*/ pattern;
uint8_t protocol;
uint16_t connIndex;
union {
uint64_t unused;
// For use by enqueue.cc
struct ncclProxyOp *enqNext;
};
};
static_assert(sizeof(struct ncclProxyOp) == 64, "Keep ProxyOp aligned with cache lines for effective prefetch");
@@ -73,9 +81,9 @@ struct ncclProxyArgs {
int sliceSteps;
int chunkSteps;
int chunkSize;
ncclDataType_t dtype;
ncclRedOp_t redOp;
ncclPattern_t pattern;
uint8_t /*ncclDataType_t*/ dtype;
uint8_t /*ncclDevRedOp_t*/ redOp;
uint8_t /*ncclPattern_t*/ pattern;
uint8_t protocol;
int state;
char* sharedBuff[NCCL_STEPS];
@@ -164,6 +172,7 @@ struct ncclProxyState {
pthread_t thread;
struct ncclSocket* listenSock;
int stop;
hipCtx_t cudaCtx;
// Used by main thread
union ncclSocketAddress* peerAddresses;
@@ -193,9 +202,8 @@ enum proxyMode {
proxyTo = 2
};
ncclResult_t ncclProxySaveColl(struct ncclComm* comm, struct ncclProxyOp* proxyOp, int nranks);
ncclResult_t ncclProxySaveOp(struct ncclComm* comm, struct ncclProxyOp* proxyOp, bool *justInquire);
ncclResult_t ncclProxyComputeP2p(struct ncclInfo* info, struct ncclProxyOp* proxyOp);
ncclResult_t ncclProxySaveP2p(struct ncclComm* comm, struct ncclProxyOp* proxyOp);
ncclResult_t ncclProxyStart(struct ncclComm* comm);
ncclResult_t ncclProxyInit(struct ncclComm* comm, struct ncclSocket* sock, union ncclSocketAddress* peerAddresses);
ncclResult_t ncclProxyCreate(struct ncclComm* comm);