On Mon, Nov 26, 2018 at 9:52 PM Jakub Jelinek <ja...@redhat.com> wrote:
>
> Hi!
>
> The following patch makes def_builtin2 behave consistently wrt.
> OPTION_MASK_ISA_64BIT requirement to def_builtin.
> The latter has:
> tree decl = NULL_TREE;
>
> if (!(mask & OPTION_MASK_ISA_64BIT) || TARGET_64BIT)
> {
> ...
> }
>
> return decl;
> and so if some builtin requires TARGET_64BIT, it is never defined for -m32;
> that is an option affecting the whole TU.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2018-11-26 Jakub Jelinek <ja...@redhat.com>
>
> PR target/88195
> * config/i386/i386.c (def_builtin2): If tcode == VOID_FTYPE_UINT64
> and !TARGET_64BIT, return NULL_TREE.
>
> * gcc.target/i386/pr88195.c: New test.
OK.
Thanks,
Uros.
> --- gcc/config/i386/i386.c.jj 2018-11-22 22:35:05.741447372 +0100
> +++ gcc/config/i386/i386.c 2018-11-26 14:31:28.074501784 +0100
> @@ -30179,9 +30179,13 @@ def_builtin2 (HOST_WIDE_INT mask, const
> {
> tree decl = NULL_TREE;
>
> - ix86_builtins_isa[(int) code].isa2 = mask;
> if (tcode == VOID_FTYPE_UINT64)
> - ix86_builtins_isa[(int) code].isa = OPTION_MASK_ISA_64BIT;
> + {
> + if (!TARGET_64BIT)
> + return decl;
> + ix86_builtins_isa[(int) code].isa = OPTION_MASK_ISA_64BIT;
> + }
> + ix86_builtins_isa[(int) code].isa2 = mask;
>
> if (mask == 0
> || (mask & ix86_isa_flags2) != 0
> --- gcc/testsuite/gcc.target/i386/pr88195.c.jj 2018-11-26 14:41:29.098641863
> +0100
> +++ gcc/testsuite/gcc.target/i386/pr88195.c 2018-11-26 14:41:00.366113191
> +0100
> @@ -0,0 +1,8 @@
> +/* PR target/88195 */
> +/* { dg-options "-mptwrite" } */
> +
> +void
> +foo (void)
> +{
> + __builtin_ia32_ptwrite64 (1); /* { dg-warning "implicit declaration
> of function" "" { target ia32 } } */
> +}
>
> Jakub
