> Am 18.06.2025 um 21:48 schrieb Jakub Jelinek <ja...@redhat.com>:
>
> Hi!
>
> The following testcase shows a regression caused by the r10-577 change
> made for cris. Before that change, the MEM holding (in this case 3 byte)
> struct parameter was BITS_PER_WORD aligned, now it is just BITS_PER_UNIT
> aligned and that causes significantly worse generated code.
> So, the MAX (DECL_ALIGN (parm), BITS_PER_WORD) extra alignment clearly
> doesn't help just STRICT_ALIGNMENT targets, but other targets as well.
> Of course, it isn't worth doing stack realignment in the rare case of
> MAX_SUPPORTED_STACK_ALIGNMENT < BITS_PER_WORD targets like cris, so the
> patch only bumps the alignment if it won't go the
>> MAX_SUPPORTED_STACK_ALIGNMENT path because of that optimization.
>
> The change on the testcase is:
> bar:
> - movl %edi, %eax
> - movzbl %dil, %r8d
> - movl %esi, %ecx
> - movzbl %sil, %r10d
> - movl %edx, %r9d
> - movzbl %dl, %r11d
> - shrl $16, %edi
> - andl $65280, %ecx
> - shrl $16, %esi
> - shrl $16, %edx
> - andl $65280, %r9d
> - orq %r10, %rcx
> - movzbl %dl, %edx
> - movzbl %sil, %esi
> - andl $65280, %eax
> - movzbl %dil, %edi
> - salq $16, %rdx
> - orq %r11, %r9
> - salq $16, %rsi
> - orq %r8, %rax
> - salq $16, %rdi
> - orq %r9, %rdx
> - orq %rcx, %rsi
> - orq %rax, %rdi
> jmp foo
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
How does this interact with -mincoming-stack-boundary? Is this, and thus when
we need stack realignment, visible here? Do we know whether we need to realign
the stack anyway for other reasons?
Richard
> 2025-06-18 Jakub Jelinek <ja...@redhat.com>
>
> PR target/120689
> * function.cc (assign_parm_setup_block): Align parm to at least
> word alignment even on !STRICT_ALIGNMENT targets, as long as
> BITS_PER_WORD is not larger than MAX_SUPPORTED_STACK_ALIGNMENT.
>
> * gcc.target/i386/pr120689.c: New test.
>
> --- gcc/function.cc.jj 2025-05-20 08:14:06.105410349 +0200
> +++ gcc/function.cc 2025-06-18 10:25:53.841280559 +0200
> @@ -2937,7 +2937,7 @@ assign_parm_setup_block (struct assign_p
> if (stack_parm == 0)
> {
> HOST_WIDE_INT parm_align
> - = (STRICT_ALIGNMENT
> + = ((STRICT_ALIGNMENT || BITS_PER_WORD <= MAX_SUPPORTED_STACK_ALIGNMENT)
> ? MAX (DECL_ALIGN (parm), BITS_PER_WORD) : DECL_ALIGN (parm));
>
> SET_DECL_ALIGN (parm, parm_align);
> --- gcc/testsuite/gcc.target/i386/pr120689.c.jj 2025-06-18
> 10:29:39.744346367 +0200
> +++ gcc/testsuite/gcc.target/i386/pr120689.c 2025-06-18 10:33:53.504111156
> +0200
> @@ -0,0 +1,17 @@
> +/* PR target/120689 */
> +/* { dg-do compile { target lp64 } } */
> +/* { dg-options "-O2 -mtune=generic -fno-stack-protector -masm=att" } */
> +/* { dg-final { scan-assembler-not "\t\(movzbl\|shrl\|salq\|orq\)\t" } } */
> +
> +struct S { char a, b, c; };
> +
> +[[gnu::noipa]]
> +void foo (struct S x, struct S y, struct S z)
> +{
> +}
> +
> +void
> +bar (struct S x, struct S y, struct S z)
> +{
> + [[gnu::musttail]] return foo (x, y, z);
> +}
>
> Jakub
>
