On Mon, Dec 4, 2023 at 8:35 AM Jakub Jelinek <ja...@redhat.com> wrote:
>
> Hi!
>
> The following testcase ICEs, because the signbit<mode>2 expander uses an
> explicit SUBREG in the pattern around match_operand with register_operand
> predicate. If we are unlucky enough that expansion tries to expand it
> with some SUBREG as operands[1], we have two nested SUBREGs in the IL,
> which is not valid and causes ICE later.
>
> Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok
> for trunk?
>
> 2023-12-04 Jakub Jelinek <ja...@redhat.com>
>
> PR target/112816
> * config/i386/sse.md (signbit<mode>2): Force operands[1] into a REG.
>
> * gcc.target/i386/sse2-pr112816.c: New test.
OK.
Thanks,
Uros.
>
> --- gcc/config/i386/sse.md.jj 2023-12-01 08:10:42.311330174 +0100
> +++ gcc/config/i386/sse.md 2023-12-02 09:53:45.489970487 +0100
> @@ -5116,7 +5116,10 @@ (define_expand "signbit<mode>2"
> (match_operand:VF1_AVX2 1 "register_operand") 0)
> (match_dup 2)))]
> "TARGET_SSE2"
> - "operands[2] = GEN_INT (GET_MODE_UNIT_BITSIZE (<MODE>mode)-1);")
> +{
> + operands[1] = force_reg (<MODE>mode, operands[1]);
> + operands[2] = GEN_INT (GET_MODE_UNIT_BITSIZE (<MODE>mode)-1);
> +})
>
> ;; Also define scalar versions. These are used for abs, neg, and
> ;; conditional move. Using subregs into vector modes causes register
> --- gcc/testsuite/gcc.target/i386/sse2-pr112816.c.jj 2023-12-02
> 10:00:23.623394880 +0100
> +++ gcc/testsuite/gcc.target/i386/sse2-pr112816.c 2023-12-02
> 09:59:47.024909235 +0100
> @@ -0,0 +1,16 @@
> +/* PR target/112816 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -msse2" } */
> +
> +#define N 4
> +struct S { float x[N]; };
> +struct T { int x[N]; };
> +
> +struct T
> +foo (struct S x)
> +{
> + struct T res;
> + for (int i = 0; i < N; ++i)
> + res.x[i] = __builtin_signbit (x.x[i]) ? -1 : 0;
> + return res;
> +}
>
> Jakub
>
