On Fri, Jul 17, 2026 at 05:46:05AM -0500, Avinash Jayakar wrote:
> 
> The 2 patches in this sequence refactor the existing mma tests for
> rs6000. This is done to test them with the future and power10 cpu
> option separately. The future test cases depend on the following patch
> sequence:
> https://gcc.gnu.org/pipermail/gcc-patches/2026-July/724014.html
> 
> The patch has been split into two for better understanding:
> 1. Split the original mma builtin tests.
> 2. Split the tests that were reported by bugs.
> 
> Is this OK for trunk and gcc-16 after the above sequence of patches go
> in?
> 
> Avinash Jayakar (2):
>   testsuite: Split rs6000 mma tests into power10 and future
>   testsuite: Split rs6000 mma related test cases to p10 and future

As I said in another patch review post, please do not use the name
'future' in the filename of the new tests.

Please use dmf (for dense math facility) instead of future.  That way
if the dense math facility is added to a real processor, we won't have
to change the filenames of the tests.  Obviously, we will have to
change the options used, but we won't have to change the name of the
test.

I would suggest using nodmf for the existing ISA 3.1 (i.e. power10 and
power11) tests, rather than p10.

In addition, you will need to modify all of the p10/nodmf tests to
explicitly add '-mno-dense-math' to the options.  The reason is
-mdejagnu-cpu=power10 will only set the cpu to power10 if the default
options do not include the power10 options.  If the potential future
machine is released as the 'foobar' PowerPC, and the GCC compiler is
configured so that the default cpu is foobar, when the tests are run,
the dense-math option will be set.

Please rename the files and add the -mno-dense-math option and resubmit
the patches.

-- 
Michael Meissner, IBM
PO Box 98, Ayer, Massachusetts, USA, 01432
email: [email protected]

Reply via email to