[PATCH] D121727: [NVPTX] Integrate ptxas to LIT tests

[Artem Belevich via Phabricator via llvm-commits]

tra added a comment.

In D121727#3383523 <https://reviews.llvm.org/D121727#3383523>, @asavonic wrote:

> We should probably have both: substitution to `true` works for *all* existing tests where can use ptxas as a sanity check, and a feature allows us to write tests for machine code.

Wouldn't it be redundant? If `REQUIRES: ptxas` is satisfied, there's no need for using `true` as a substitute. If it's not satisfied, then we would not run such test and therefore `true` would not be useful either.

Using `ptxas` in a test w/o `REQUIRES: ptxas` would be a user error, IMO as the test would be relying on something that's not expected to be available by default and would not do anything useful.
I would actually make this assertion even stronger -- allowing a test to run and return a success when we in fact didn't test anything would be a wrong thing to do. We should correctly report such test as not executed.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D121727/new/

https://reviews.llvm.org/D121727