On Mon, Oct 11, 2021 at 07:11:13PM -0500, Segher Boessenkool wrote:
> On Mon, Aug 23, 2021 at 02:03:10PM -0500, Paul A. Clarke wrote:
> > Some compatibility implementations of x86 intrinsics include
> > Power intrinsics which require POWER8. Guard them.
>
> > emmintrin.h:
> > - _mm_cmpord_pd: Remove code which was ostensibly for pre-POWER8,
> > but which indeed depended on POWER8 (vec_cmpgt(v2du)/vcmpgtud).
> > The "POWER8" version works fine on pre-POWER8.
>
> Huh. It just generates xvcmpeqdp I suppose?
Yes.
> > - _mm_mul_epu32: vec_mule(v4su) uses vmuleuw.
>
> Did this fail on p7? If not, add a test that *does*?
Do you mean fail if not for "dg-require-effective-target p8vector_hw"?
We have that, in gcc/testsuite/gcc.target/powerpc/sse2-pmuludq-1.c.
> > pmmintrin.h:
> > - _mm_movehdup_ps: vec_mergeo(v4su) uses vmrgow.
> > - _mm_moveldup_ps: vec_mergee(v4su) uses vmrgew.
>
> Similar.
gcc/testsuite/gcc.target/powerpc/sse3-movshdup.c
gcc/testsuite/gcc.target/powerpc/sse3-movsldup.c
> > smmintrin.h:
> > - _mm_cmpeq_epi64: vec_cmpeq(v2di) uses vcmpequd.
> > - _mm_mul_epi32: vec_mule(v4si) uses vmuluwm.
> > - _mm_cmpgt_epi64: vec_cmpgt(v2di) uses vcmpgtsd.
> > tmmintrin.h:
> > - _mm_sign_epi8: vec_neg(v4si) uses vsububm.
> > - _mm_sign_epi16: vec_neg(v4si) uses vsubuhm.
> > - _mm_sign_epi32: vec_neg(v4si) uses vsubuwm.
> > Note that the above three could actually be supported pre-POWER8,
> > but current GCC does not support them before POWER8.
> > - _mm_sign_pi8: depends on _mm_sign_epi8.
> > - _mm_sign_pi16: depends on _mm_sign_epi16.
> > - _mm_sign_pi32: depends on _mm_sign_epi32.
>
> And more.
gcc/testsuite/gcc.target/powerpc/sse4_1-pcmpeqq.c
gcc/testsuite/gcc.target/powerpc/sse4_1-pmuldq.c
gcc/testsuite/gcc.target/powerpc/sse4_2-pcmpgtq.c
- although this one will _actually_ fail on P7, as it only requires
"vsx_hw". I'll fix this.
gcc/testsuite/gcc.target/powerpc/ssse3-psignb.c
gcc/testsuite/gcc.target/powerpc/ssse3-psignw.c
gcc/testsuite/gcc.target/powerpc/ssse3-psignd.c
> > gcc
> > PR target/101893
>
> This is a different bug (the vgbdd one)?
PR 101893 is the same issue: things not being properly masked by
#ifdefs.
> All looks good, but we need such failing tests :-)
Thanks for the review! Let me know what you mean by "failing tests".
("Would fail if not for ..."?)
PC
