[PATCH] D106521: [Docs] Mention how to run lit tests in parallel

Thu Jul 22 00:52:11 PDT 2021

teemperor added inline comments.

================
Comment at: llvm/docs/TestingGuide.rst:186
+-DLLVM_LIT_ARGS="-sv jN" to CMake.
+
 For more information on using the :program:`lit` tool, see ``llvm-lit --help``
----------------
mehdi_amini wrote:
> Isn't lit running in parallel by default?
> `ninja check` spawn as many workers as core on my machine...
> 
> ```
> $ ./bin/llvm-lit ../mlir/test
> -- Testing: 1036 tests, 96 workers --
> ```
> 
I think this patch is trying to document how to limit the number of workers on a machine to something less than the number of hardware threads. At least the description says that this is based on some suggestion of mine and the last time I commented on anything related to `LIT_OPTS` was when I told someone on the Discord's 'beginner' channel how to make `lit` only run tests on one core (please correct me if I got that wrong @xgupta ).

The current way the patch is formulating this makes it actually sound like parallelism in lit is opt-in, so I think that could be rephrased.

I also wonder how useful this information actually is as I don't think many people care about limiting the number of test threads. We already point to the `lit` docs which document all of this, but maybe we should just briefly mention `LIT_OPTS` as a quick way to configure `lit`? We could have `export LIT_OPTS="-j1"` as an example then.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106521/new/

https://reviews.llvm.org/D106521