expand: Align PARM_DECLs again to at least BITS_PER_WORD if possible [PR120689]

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 patch keeps the gcc 15 behavior for avr, pru, m68k and cris (at
least some options for those) and restores the behavior before r10-577 on
other targets.

The change on the testcase on x86_64 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

2025-06-19  Jakub Jelinek  <jakub@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.

diff --git a/gcc/function.cc b/gcc/function.cc
index a5b245a98e915416da21e27c74e59c41cb12e071..a3a74b44b9165eb2b682b001676f8414d087e578 100644 (file)
--- a/gcc/function.cc
+++ b/gcc/function.cc
@@ -2937,7 +2937,7 @@ assign_parm_setup_block (struct assign_parm_data_all *all,
   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);

diff --git a/gcc/testsuite/gcc.target/i386/pr120689.c b/gcc/testsuite/gcc.target/i386/pr120689.c
new file mode 100644 (file)
index 0000000..cd10cdb
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr120689.c
@@ -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);
+}