tkonolige commented on code in PR #11066:
URL: https://github.com/apache/tvm/pull/11066#discussion_r858097347
##########
tests/python/unittest/test_runtime_profiling.py:
##########
@@ -257,6 +259,50 @@ def test_profile_function(target, dev):
assert report[metric].value > 0
[email protected]_targets("llvm")
+def test_estimate_peak_fma_flops(target, dev):
+ # This test uses vectorized instructions so we need a target that supports
them
+ if target == "llvm":
+ target = "llvm -mattr=+fma,+avx2"
Review Comment:
Turning off vectorization does have an effect. It drops flops down to
~1MFLOP/s.
Not sure how to check if the processor supports an instruction, I'll looked
through the codebase and couldn't find any examples. avx2 and fma are at least
10 years old, so I think we can assume x86 targets support them (totally could
be wrong here though).
--
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]
