Re: [PATCH v2, rs6000] Add multiply-add expand pattern [PR103109]

["Kewen.Lin" <linkw@linux.ibm.com>]

on 2022/8/10 05:34, Segher Boessenkool wrote:
> On Tue, Aug 09, 2022 at 11:14:16AM +0800, Kewen.Lin wrote:
>> on 2022/8/8 14:04, HAO CHEN GUI wrote:
>>> +/* { dg-do run { target { has_arch_ppc64 } } } */
>>> +/* { dg-options "-O2 -mdejagnu-cpu=power9 -save-temps" } */
>>> +/* { dg-require-effective-target int128 } */
>>> +/* { dg-require-effective-target p9modulo_hw } */
>>> +/* { dg-final { scan-assembler-times {\mmaddld\M} 2 } } */
>>> +/* { dg-final { scan-assembler-times {\mmaddhd\M} 1 } } */
>>> +/* { dg-final { scan-assembler-times {\mmaddhdu\M} 1 } } */
>>> +
>>
>> Maybe it's good to split this case into two, one for compiling and the other for running.
>> Since the generated asm is a test point here, with one separated case for compiling, we
>> can still have that part of test coverage on hosts which are unable to run this case.
>> You can move functions multiply_add and multiply_addu into one common header file, then
>> include it in both source files.
> 
> Yeah, good point.  You cannot make dg-do do different things on
> different targets.  Fortunatelt just duplicating this test and then
> removing the things not relevant to run resp. compile testing makes
> things even more clear :-)
> 
>> Nit: better to add one explicit "return 0;" to avoid possible warning.
> 
> This is in main(), the C standard requires this to work without return
> (and it is common).  But, before C99 the implicit return value from
> main() was undefined, so yes, it could warn then.  Does it?
> 

Yes, exactly, with explicit -std=c89 -Wreturn-type it will have a warning:

warning: control reaches end of non-void function...

BR,
Kewen