Fix broken profiling build (#263)
这个提交包含在:
@@ -99,7 +99,7 @@ __device__ void ncclAllReduceRingKernel(struct CollectiveArgs* args) {
|
||||
ACCUMULATE_COUNTER(directRecv);
|
||||
}
|
||||
#ifdef ENABLE_PROFILING
|
||||
if (tid == 0 && args->opCount > 0) __atomic_fetch_add(&(devProf->total_cycle), __rtc64() - clk, __ATOMIC_SEQ_CST);
|
||||
if (tid == 0) __atomic_fetch_add(&(devProf->total_cycle), __rtc64() - clk, __ATOMIC_SEQ_CST);
|
||||
#endif
|
||||
}
|
||||
|
||||
|
||||
在新工单中引用
屏蔽一个用户