On Fri, 19 Jan 2024, Jakub Jelinek wrote:
> Hi!
>
> Like for GIMPLE_PHIs or calls, even for GIMPLE_ASMs we want
> a corresponding VAR_DECL assigned for lhs SSA_NAMEs of loads
> from memory, as even GIMPLE_ASM relies on those VAR_DECLs to exist.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
OK.
> 2024-01-19 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/113464
> * gimple-lower-bitint.cc (gimple_lower_bitint): Don't try to
> optimize loads into GIMPLE_ASM stmts.
>
> * gcc.dg/bitint-75.c: New test.
>
> --- gcc/gimple-lower-bitint.cc.jj 2024-01-18 08:44:08.337270271 +0100
> +++ gcc/gimple-lower-bitint.cc 2024-01-18 19:57:11.791976322 +0100
> @@ -6249,7 +6249,8 @@ gimple_lower_bitint (void)
> if (is_gimple_debug (use_stmt))
> continue;
> if (gimple_code (use_stmt) == GIMPLE_PHI
> - || is_gimple_call (use_stmt))
> + || is_gimple_call (use_stmt)
> + || gimple_code (use_stmt) == GIMPLE_ASM)
> {
> optimizable_load = false;
> break;
> --- gcc/testsuite/gcc.dg/bitint-75.c.jj 2024-01-18 20:08:21.710557536
> +0100
> +++ gcc/testsuite/gcc.dg/bitint-75.c 2024-01-18 20:07:18.017447734 +0100
> @@ -0,0 +1,11 @@
> +/* PR tree-optimization/113464 */
> +/* { dg-do compile { target bitint65535 } } */
> +/* { dg-options "-O2 -w -std=gnu23" } */
> +
> +_BitInt(65532) i;
> +
> +void
> +foo (void)
> +{
> + __asm__ ("" : "+r" (i)); /* { dg-error "impossible constraint" } */
> +}
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)
