remove useless sync (#809)

Binyang2014 · web-flow · commit 9e177b388c1a · 2026-05-20T16:49:49.000-07:00
diff --git a/src/ext/collectives/allreduce/allreduce_allpair_packet.cu b/src/ext/collectives/allreduce/allreduce_allpair_packet.cu
@@ -63,7 +63,6 @@ __global__ void allreduceAllPairs(T* buff, T* scratch, T* resultBuff, DeviceHand
     }
     dst[idx] = mscclpp::downcastVector<T, AccumT, uint32_t>(acc);
   }
-  __syncthreads();
   if (threadIdx.x == 0) {
     ((uint32_t*)flags)[blockIdx.x] = flag + 1;
   }

Original file line number	Diff line number	Diff line change
`@@ -63,7 +63,6 @@ __global__ void allreduceAllPairs(T* buff, T* scratch, T* resultBuff, DeviceHand`
`63`	`63`	`}`
`64`	`64`	`dst[idx] = mscclpp::downcastVector<T, AccumT, uint32_t>(acc);`
`65`	`65`	`}`
`66`		`- __syncthreads();`
`67`	`66`	`if (threadIdx.x == 0) {`
`68`	`67`	`((uint32_t*)flags)[blockIdx.x] = flag + 1;`
`69`	`68`	`}`