[PATCH] rs6000/test: Add emulated gather test case

Hongtao Liu crazylht@gmail.com
Thu Nov 25 05:17:35 GMT 2021


On Thu, Nov 25, 2021 at 11:21 AM Kewen.Lin via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> Hi,
>
> This patch is to add a test case similar to the one in i386
> to add testing coverage for 510.parest_r hotspots.
>
> As evaluated, the emulated gather capability of vectorizer
> (r12-2733) can help to speed up SPEC2017 510.parest_r on
> Power8/9/10 by 5% to 9% with option sets Ofast unroll and
> Ofast lto.  But since rs6000 missed unpacking support for
> unsigned int before, it can only vectorize the hotspots
> until r12-3134.
>
> By checking why r12-2733 doesn't immediately show its impact
> for SPEC2017 510.parest_r while the associated test case
> already can get vectorized on rs6000 at that time, I realized
> the associated test case use int as INDEXTYPE while the
> hotspots actually use unsigned int.  So different from the one
> in i386, this patch uses unsigned int as INDEXTYPE since the
> unpack support for unsigned int (r12-3134) also matters for
> the hotspots vectorization.  Not sure if it's worth to updating
> the one in i386 as well?
It looks like the same testcase added in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531
>
> Tested on powerpc64le-linux-gnu P9 and powerpc64-linux-gnu P8.
>
> Is it ok for trunk?
>
> BR,
> Kewen
> -----
> gcc/testsuite/ChangeLog:
>
>         * gcc.target/powerpc/vect-gather-1.c: New test.
>
> diff --git a/gcc/testsuite/gcc.target/powerpc/vect-gather-1.c b/gcc/testsuite/gcc.target/powerpc/vect-gather-1.c
> new file mode 100644
> index 00000000000..bf98045ab03
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/powerpc/vect-gather-1.c
> @@ -0,0 +1,20 @@
> +/* { dg-do compile } */
> +/* Profitable from Power8 since it supports efficient unaligned load.  */
> +/* { dg-options "-Ofast -mdejagnu-cpu=power8 -fdump-tree-vect-details -fdump-tree-forwprop4" } */
> +
> +#ifndef INDEXTYPE
> +#define INDEXTYPE unsigned int
> +#endif
> +double vmul(INDEXTYPE *rowstart, INDEXTYPE *rowend,
> +           double *luval, double *dst)
> +{
> +  double res = 0;
> +  for (const INDEXTYPE * col = rowstart; col != rowend; ++col, ++luval)
> +        res += *luval * dst[*col];
> +  return res;
> +}
> +
> +/* With gather emulation this should be profitable to vectorize from Power8.  */
> +/* { dg-final { scan-tree-dump "loop vectorized" "vect" } } */
> +/* The index vector loads and promotions should be scalar after forwprop.  */
> +/* { dg-final { scan-tree-dump-not "vec_unpack" "forwprop4" } } */
> --
> 2.25.1
>


-- 
BR,
Hongtao


More information about the Gcc-patches mailing list