Fix crash when PXN is enabled on some platforms (#1369)

[ROCm/rccl commit: b55b6be0cb]
This commit is contained in:
Wenkai Du
2024-10-11 09:02:59 -07:00
committed by GitHub
parent 5545392913
commit 1b988c1b31
6 changed files with 28 additions and 35 deletions
+1 -3
View File
@@ -61,8 +61,6 @@ struct ncclRecvMem {
uint64_t tail;
char pad1[CACHE_LINE_SIZE-sizeof(uint64_t)];
struct ncclConnFifo connFifo[NCCL_STEPS];
int sizesFifo[NCCL_STEPS];
int offsFifo[NCCL_STEPS];
int flush; // For GDRCopy-based flush
};
char pad4[MEM_ALIGN];
@@ -257,7 +255,7 @@ struct ncclComm {
int rank; // my rank in the communicator
int nRanks; // number of GPUs in communicator
int cudaDev; // my cuda device index
//int nvmlDev; // my nvml device index
int nvmlDev; // my nvml device index
int compCap; // compute capability of the GPU
int minCompCap, maxCompCap; // min/max compute capability in the communicator
int64_t busId; // my PCI bus ID in int format
+1
View File
@@ -36,6 +36,7 @@ struct ncclComm;
struct ncclPeerInfo {
int rank;
int cudaDev;
int nvmlDev;
int gdrSupport;
bool hasFineGrain;
uint64_t hostHash;