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

msccl: allocate scratch as ext-scope fine-grained (#968)

Browse Source 
[ROCm/rccl commit: 569d3f7d59]
This commit is contained in:
Wenkai Du
2023-11-16 07:57:25 -08:00
committed by GitHub
parent b5642f39ed
commit 4c2fa05a23
2 changed files with 2 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/collectives/device/msccl_kernel_impl.h
+1 -10
View File
@@ -315,11 +315,7 @@ __device__ __forceinline__ void mscclRunInterpreter(
int16_t dependentStep = mscclShmem.mscclTB.dependentStep[dependentPointer+tid];
uint64_t goalFlag = COMPUTE_FLAG(workIndex, iter, dependentStep);
while (true){
#if defined(__gfx942__)
uint64_t curFlag = __atomic_load_n(&(mscclFlags + dependentBid)->flag, __ATOMIC_ACQUIRE);
#else
uint64_t curFlag = __atomic_load_n(&(mscclFlags + dependentBid)->flag, __ATOMIC_RELAXED);
#endif
if (curFlag >= goalFlag && GET_WORKINDEX_FROM_FLAG(curFlag) == workIndex) break;
}
}
@@ -463,13 +459,8 @@ __device__ __forceinline__ void mscclRunInterpreter(
else
return;
}
if (t->hasDependence && tid == nthreads-1){
#if defined(__gfx942__)
__atomic_store_n(&mscclFlags[bid].flag, (uint64_t) COMPUTE_FLAG(workIndex, iter, step), __ATOMIC_RELEASE);
#else
if (t->hasDependence && tid == nthreads-1)
__atomic_store_n(&mscclFlags[bid].flag, (uint64_t) COMPUTE_FLAG(workIndex, iter, step), __ATOMIC_RELAXED);
#endif
}
step++;
}
}