mbs-octoml commented on a change in pull request #9312:
URL: https://github.com/apache/tvm/pull/9312#discussion_r746899670
##########
File path: src/relay/op/memory/device_copy.cc
##########
@@ -103,6 +107,19 @@ DeviceCopyProps GetDeviceCopyProps(const CallNode*
call_node) {
} else {
return {call_node->args[0], src_dev_type, dst_dev_type};
}
+ } else if (call_node->op == CallLoweredOp()) {
+ /* Get device props for a TIR function */
+ CallLoweredProps call_lowered_props = GetCallLoweredProps(call_node);
+
+ if (call_lowered_props.attrs.metadata.count("source_device") == 1 &&
Review comment:
confirming looks like we can avoid this here
(eventually GetPrimitiveDeviceCopyProps from device_domains.cc will need to
get moved alongside this one (and renamed) but no need here)
##########
File path: src/relay/backend/te_compiler.cc
##########
@@ -628,15 +639,13 @@ class LowerTensorExprMutator : public
DeviceAwareExprMutator {
// TODO(mbs): Replace device_type with target so this lookup is
unnecessary.
target = GetTargetFromInteger(device_type, targets_);
}
-
+ Array<Expr> visited_args;
Review comment:
not a correctness issue, but we are revisiting the args (bad rebase).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
