Mousius commented on a change in pull request #9338:
URL: https://github.com/apache/tvm/pull/9338#discussion_r734418005
##########
File path: tests/python/relay/aot/aot_test_utils.py
##########
@@ -140,6 +140,11 @@ class AOTTestRunner(NamedTuple):
""",
includes=["uart.h"],
parameters={"NPU_VARIANT": "256"},
+ pass_config={
Review comment:
The flags here decide on the buffer sizes in TVM, in the precompile for
CMSIS-NN it passes `-mcpu=cortex-m55` which results in that configuration for
that device for the libraries linked in the tests. This is why I've covered
this in unit tests, our reference system tests cover `-mcpu=cortex-m55` end to
end but we test the buffer size is calculated properly for various
architectures. The best we could potentially do here is do `+nodsp` and
`+nomve` with `cortex-m55` but not the other CPUs, and that'd require
precompiling those variants of CMSIS-NN in the container.
--
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]
