[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)
[ROCm/rccl commit: d009ab144e]
Tento commit je obsažen v:
@@ -146,6 +146,9 @@ __device__ __forceinline__ void mscclRunInterpreter(
|
||||
}
|
||||
if (bytes) copyToShmem8(tid%WARP_SIZE, dst, src, bytes);
|
||||
}
|
||||
#ifdef ENABLE_WARP_SPEED
|
||||
ncclShmem.warpComm = 0;
|
||||
#endif
|
||||
__syncthreads(); // publish shmem
|
||||
|
||||
#if defined(ENABLE_NPKIT)
|
||||
|
||||
Odkázat v novém úkolu
Zablokovat Uživatele