https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118076
--- Comment #5 from Jonathan Gruber <jonathan.gruber.jg at gmail dot com> ---
For completeness, I also checked for when the call to the struct-accepting
function is not a tail call. This is just the same test case as before, but the
return value of extern_func is bitwise negated to make it a non-tail call:
struct S {
void *x, *y, *z, *w;
};
extern int extern_func(struct S s);
int non_tail_fwd(void *x, void *y, void *z, void *w) {
struct S s = { x, y, z, w };
return ~extern_func(s);
}
Unsurprisingly, we see the same bug. Below is the generated assembly for the
target architectures I tested, all with -O3 optimization (we see the same bug
on other optimization levels, too).
gcc (on x86_64):
.cfi_startproc
subq $72, %rsp
.cfi_def_cfa_offset 80
movq %rdi, 32(%rsp)
movq %rsi, 40(%rsp)
movdqa 32(%rsp), %xmm0
movq %rdx, 48(%rsp)
movq %rcx, 56(%rsp)
movups %xmm0, (%rsp)
movdqa 48(%rsp), %xmm0
movups %xmm0, 16(%rsp)
call extern_func@PLT
addq $72, %rsp
.cfi_def_cfa_offset 8
notl %eax
ret
.cfi_endproc
aarch64-linux-gnu-gcc (on x86_64):
.cfi_startproc
stp x29, x30, [sp, -80]!
.cfi_def_cfa_offset 80
.cfi_offset 29, -80
.cfi_offset 30, -72
mov x29, sp
stp x0, x1, [sp, 48]
add x0, sp, 16
stp x2, x3, [sp, 64]
ldp q30, q31, [sp, 48]
str q30, [sp, 16]
str q31, [x0, 16]
bl extern_func
mvn w0, w0
ldp x29, x30, [sp], 80
.cfi_restore 30
.cfi_restore 29
.cfi_def_cfa_offset 0
ret
.cfi_endproc
riscv64-linux-gnu-gcc (on x86_64):
.cfi_startproc
addi sp,sp,-80
.cfi_def_cfa_offset 80
mv a5,a0
mv a0,sp
sd ra,72(sp)
.cfi_offset 1, -8
sd a5,0(sp)
sd a1,8(sp)
sd a2,16(sp)
sd a3,24(sp)
call extern_func@plt
ld ra,72(sp)
.cfi_restore 1
not a0,a0
sext.w a0,a0
addi sp,sp,80
.cfi_def_cfa_offset 0
jr ra
.cfi_endproc
