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

2.16.5-1

Browse Source 
Add support for 400Gbit NDR network adapters (CX7)
Handle EINTR in socket poll() function
Add NCCL_PROGRESS_APPENDOP_FREQ to control op append overhead
Resource cleanup fixes
Fix double free in case of init failure
Fix crash in ncclCommAbort
Revert AMD speed commit


[ROCm/rccl commit: f3d5166783]
This commit is contained in:
Sylvain Jeaugey
2023-02-02 12:52:47 -08:00
parent 2840ac0139
commit e6e8f2555c
12 changed files with 71 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/misc/shmutils.cc
+1 -1
View File
@@ -54,7 +54,7 @@ ncclResult_t ncclShmOpen(char* shmPath, size_t shmSize, void** shmPtr, void** de
const size_t realShmSize = shmSize + refSize;
*handle = *shmPtr = NULL; /* assume shmPtr and handle always set correctly by users. */
EQCHECKGOTO(tmphandle = (struct shmHandleInternal*)malloc(sizeof(struct shmHandleInternal)), NULL, ret, fail);
EQCHECKGOTO(tmphandle = (struct shmHandleInternal*)calloc(1, sizeof(struct shmHandleInternal)), NULL, ret, fail);
if (create) {
/* refcount > 0 means the caller tries to allocate a shared memory. This shared memory segment will have
* refcount references; when the peer attaches, it should pass -1 to reduce one reference count. When it