jcf94 commented on a change in pull request #6184:
URL: https://github.com/apache/incubator-tvm/pull/6184#discussion_r467778942
##########
File path: src/auto_scheduler/compute_dag.cc
##########
@@ -342,11 +343,16 @@ AccessAnalyzer::AccessAnalyzer(const Array<te::Tensor>&
tensors) {
has_expensive_op |= HasExpensiveOp(expr);
}
if (has_expensive_op || has_branch[op]) {
- is_strict_inlineable = false;
+ is_strictly_inlineable = false;
+ }
+
+ // constant tensor is strict-inlineable
+ if (node->read_from[op].empty()) {
+ is_strictly_inlineable = true;
}
Review comment:
@merrymercy The transform_matrices A, B of the winograd conv2d is a
constant tensor but with branch, so the order of `has_branch[op]` and
`read_from[op]` check will result on different `is_strictly_inlineable`.
Should this intend to be strictly inlined?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
