AI/rocm-systems
2
0 Fork 0
程式碼 問題 合併請求 Actions 套件 專案 版本發布 Wiki 動態

Implement RCCL Replayer (#817)

瀏覽代碼 
* Implement RCCL Replayer


[ROCm/rccl commit: 8bab4f04b7]
此提交包含在：
Bertan Dogancay authored and GitHub committed 2023-07-24 16:26:22 -06:00
1 父節點 dfda1d6fab
當前提交 0508fa569a
5 file changed
+666 -3
下載 Patch 檔 下載差異檔 展開所有檔案 折疊所有檔案

No files matched your search

projects/rccl/src/enqueue.cc
+5 -3
查看文件
@@ -1616,9 +1616,11 @@ ncclResult_t ncclEnqueueCheck(struct ncclInfo* info) {
}
NCCLCHECKGOTO(ArgsCheck(info), ret, fail);
INFO(NCCL_COLL,"%s: opCount %lx sendbuff %p recvbuff %p count %zi datatype %d op %d root %d comm %p [nranks=%d] stream %p",
info->opName, info->comm->opCount, info->sendbuff, info->recvbuff, info->count,
info->datatype, info->op, info->root, info->comm, info->comm->nRanks, info->stream);
INFO(NCCL_COLL,"%s: opCount %lx sendbuff %p recvbuff %p count %zi datatype %d op %d root %d comm %p [nranks=%d] stream %p task %d globalrank %d",
info->opName, info->comm->opCount, info->sendbuff, info->recvbuff, info->count,
info->datatype, info->op, info->root, info->comm, info->comm->nRanks, info->stream,
info->comm->tasks.nTasksP2p + info->comm->tasks.nTasksColl,
info->comm->localRankToRank[info->comm->localRank]);
TRACE_CALL("nccl%s(%" PRIx64 ",%" PRIx64 ",%zi,%d,%d,%d,%p,%p)", info->opName, reinterpret_cast<int64_t>(info->sendbuff), reinterpret_cast<int64_t>(info->recvbuff), info->count, info->datatype, info->op, info->root, info->comm, info->stream);
NCCLCHECKGOTO(taskAppend(info->comm, info), ret, fail);