On Fri, Aug 21, 2020 at 8:41 AM Hongtao Liu <crazy...@gmail.com> wrote:
>
> On Fri, Aug 21, 2020 at 9:15 PM Uros Bizjak <ubiz...@gmail.com> wrote:
> >
> > > > > gcc/
> > > > > PR target/88808
> > > > > * config/i386/i386.c (ix86_preferred_reload_class): Allow
> > > > > QImode data go into mask registers.
> > > > > * config/i386/i386.md: (*movhi_internal): Adjust constraints
> > > > > for mask registers.
> > > > > (*movqi_internal): Ditto.
> > > > > (*anddi_1): Support mask register operations
> > > > > (*and<mode>_1): Ditto.
> > > > > (*andqi_1): Ditto.
> > > > > (*andn<mode>_1): Ditto.
> > > > > (*<code><mode>_1): Ditto.
> > > > > (*<code>qi_1): Ditto.
> > > > > (*one_cmpl<mode>2_1): Ditto.
> > > > > (*one_cmplsi2_1_zext): Ditto.
> > > > > (*one_cmplqi2_1): Ditto.
> > > > > (define_peephole2): Move constant 0/-1 directly into mask
> > > > > registers.
> > > > > * config/i386/predicates.md (mask_reg_operand): New predicate.
> > > > > * config/i386/sse.md (define_split): Add post-reload splitters
> > > > > that would convert "generic" patterns to mask patterns.
> > > > > (*knotsi_1_zext): New define_insn.
> > > > >
> > > > > gcc/testsuite/
> > > > > * gcc.target/i386/bitwise_mask_op-1.c: New test.
> > > > > * gcc.target/i386/bitwise_mask_op-2.c: New test.
> > > > > * gcc.target/i386/bitwise_mask_op-3.c: New test.
> > > > > * gcc.target/i386/avx512bw-pr88465.c: New testcase.
> > > > > * gcc.target/i386/avx512bw-kunpckwd-1.c: Adjust testcase.
> > > > > * gcc.target/i386/avx512bw-kunpckwd-3.c: Ditto.
> > > > > * gcc.target/i386/avx512dq-kmovb-5.c: Ditto.
> > > > > * gcc.target/i386/avx512f-kmovw-5.c: Ditto.
> > > >
> > > > A little nit, please put new splitters after the instruction pattern.
> > > >
> > > > OK for the whole patch set with the above change,
> > > >
> > >
> > > Yes, thanks for the review.
> >
> > Please note that your patch introduces several testsuite fails with -m32:
> >
> > gcc -O2 -mavx512bitalg -mavx512bw -m32 -g avx512bitalg-vpopcntb-1.c
> >
>
> I can't reproduce this failure.
>
> > Program received signal SIGILL, Illegal instruction.
> > 0x080490ac in __get_cpuid_count (__edx=<synthetic pointer>,
> > __ecx=<synthetic pointer>, __ebx=<synthetic pointer>, __eax=<synthetic
> > pointer>,
> > __subleaf=0, __leaf=7) at
> > /hdd/uros/gcc-build-fast/gcc/include/cpuid.h:316
> > 316 __cpuid_count (__leaf, __subleaf, *__eax, *__ebx, *__ecx, *__edx);
> >
> > 0x080490a3 <+51>: cpuid
> > 0x080490a5 <+53>: mov $0x1,%eax
> > 0x080490aa <+58>: mov %ecx,%esi
> > => 0x080490ac <+60>: kmovd %ebx,%k0
> > 0x080490b0 <+64>: mov %edi,%ecx
> > 0x080490b2 <+66>: mov %edi,%ebx
> >
> > kmov insn is generated for __cpuid_count function, where the binary
> > determines, if the new instructions are supported. The binary will
> > crash in the detection code if the processor lacks AVX512
> > instructions.
> >
>
> IMHO, the testcase shouldn't be run on processors without AVX512BW.
> Because in avx512bitalg-vpopcntb-1.c, there's /* {
> dg-require-effective-target avx512bw } */.
>
dg-require-effective-target avx512bw checks assembler support.
__cpuid_count can't use any mask instructions. Please run this test
on CPU without AVX512.
--
H.J.
