* Changes to device code

* Changes to src/misc

* Changes to graph

* src/include changes

* src/transport changes

* changes in init, enqueue, proxy

* Changes to CMakeLists.txt

* Additional changes to device code

* Additional changes to net.cc

* adding 'compiler warning' tag to ease upstream merge'

* typo correction

* Addessing comments

* Additional changes for new commits

[ROCm/rccl commit: 3f8cac388e]
Этот коммит содержится в:
Avinash authored and GitHub committed 2025-08-05 17:36:23 -05:00
1 родитель df3b7e477f
Коммит f34d760613
28 files changed
+104 -106

No files matched your search

+2 -2
Просмотреть файл
@@ -1238,7 +1238,7 @@ static void waitWorkFifoAvailable(struct ncclComm* comm, uint32_t desiredProduce
warned = 1;
WARN("Waiting for work FIFO to become available. "
"Work fifo exhaustion can happen in large scale/high iteration count of alltoall. "
"In order to increase work FIFO size, set NCCL_WORK_FIFO_BYTES to higher number (current: %ld).\n\n"
"In order to increase work FIFO size, set NCCL_WORK_FIFO_BYTES to higher number (current: %d).\n\n"
"RCCL continues to retry...", comm->workFifoBytes);
}
@@ -1776,7 +1776,7 @@ ncclResult_t ncclLaunchFinish(struct ncclComm* comm) {
ncclIntruQueueConstruct(&planner->planQueue);
bool capturing = ncclCudaGraphValid(planner->capturingGraph);
cudaStream_t launchStream = planner->streams->stream; // First user stream gets launch
//cudaStream_t launchStream = planner->streams->stream; // First user stream gets launch // unused variable - compiler warning
cudaStream_t deviceStream, launchOrder;
if (capturing || planner->numStreams != 1) {