antinucleon commented on a change in pull request #7313:
URL: https://github.com/apache/tvm/pull/7313#discussion_r560760923
##########
File path: python/tvm/auto_scheduler/measure.py
##########
@@ -1132,3 +1249,44 @@ def rpc_runner_run(
print("")
return results
+
+
+# The map stores special registered buffer for measurement
+# This can be used for sparse workloads when we cannot use random tensors for
measurment.
+global special_buffer_table
+special_buffer_table = {}
Review comment:
I am against to use global variable. In my personal [test
branch](https://github.com/antinucleon/tvm/tree/metal), I use pickle to dump
all buffers to `tmpdir/workload_key.pkl` to avoid global variable. In setting
with ARM64 Python + Multiprocessing, global variable behavior is not same to
what we expect on Linux.
----------------------------------------------------------------
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]
