zheng-da commented on a change in pull request #10451: [MXNET-432] Add Foreach
URL: https://github.com/apache/incubator-mxnet/pull/10451#discussion_r189731937
##########
File path: src/imperative/imperative_utils.h
##########
@@ -456,17 +458,24 @@ inline void PushOperator(const OpStatePtr& state,
if (fcompute_ex != nullptr && dispatch_mode == DispatchMode::kFComputeEx) {
const auto& run = [=](RunContext rctx,
engine::CallbackOnComplete on_complete) {
- OpContext opctx{is_train, rctx, on_complete, requested};
+ bool need_grad = Imperative::Get()->is_recording();
+ OpContext opctx{need_grad, is_train, rctx, on_complete, requested};
#if MXNET_USE_MKLDNN == 1
InvalidateOutputs(outputs, req);
#endif
fcompute_ex(state, opctx, inputs, req, outputs);
- if (ctx.dev_mask() == gpu::kDevMask && exec_type == ExecType::kSync) {
+ if (ctx.dev_mask() == gpu::kDevMask && exec_type == ExecType::kSync
+ && rctx.get_stream<gpu>()) {
rctx.get_stream<gpu>()->Wait();
}
};
- if (exec_type == ExecType::kSync) {
+ // For operators with subgraphs, we need to invoke them in the main thread
+ // instead of the threaded engine.
+ if (!attrs.subgraphs.empty()) {
Review comment:
I think it can happen. For example, if we hybridize a block with control
flow operators, the execution of these operators will happen here.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
