[GitHub] [tvm] giuseros commented on a change in pull request #6981: [AutoScheduler] Accelerate feature extraction for winograd

[GitBox <gi...@apache.org>]

giuseros commented on a change in pull request #6981:
URL: https://github.com/apache/tvm/pull/6981#discussion_r531523344



##########
File path: src/auto_scheduler/feature.cc
##########
@@ -669,7 +669,7 @@ class PerStoreFeatureExtractor : public StmtExprVisitor {
     math_op_counter(node->value);
     std::vector<float> mem_bytes_list;
     std::vector<float> compute_ops_list;
-    int cur_compute_ops;
+    double cur_compute_ops;

Review comment:
       Why moving this to double?

##########
File path: include/tvm/te/schedule.h
##########
@@ -375,9 +375,11 @@ class Schedule : public ObjectRef {
    *  Insert necessary RebaseNode to make sure all leaf_iter_vars
    *  are in form [0, extent)
    *
+   * \param feature_extraction_mode Whether to enable feature extraction mode for
+   *   fast compilation and feature extraction in autotvm/auto-scheduler.
    * \return A normalized schedule, can be same as current one.
    */
-  Schedule normalize();
+  Schedule normalize(bool feature_extraction_mode);

Review comment:
       I agree with others that a different API would be better. But if we stick with this interface,  could we default `feature_extraction_mode` to `False` (in the same way you did for the Python API)?




----------------------------------------------------------------
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:
users@infra.apache.org