[Bug middle-end/89972] [8/9 Regression] ICE in expand_call, at calls.c:4229
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89972 Marek Polacek changed: What|Removed |Added Status|NEW |WAITING --- Comment #7 from Marek Polacek --- I suppose we need to find out the answer first -> WAITING.
[Bug middle-end/89972] [8/9 Regression] ICE in expand_call, at calls.c:4229
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89972 --- Comment #6 from Marek Polacek --- That'd be much appreciated, I was puzzled as to what we should do when I first took a look at this.
[Bug middle-end/89972] [8/9 Regression] ICE in expand_call, at calls.c:4229
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89972 --- Comment #5 from Jakub Jelinek --- (In reply to H.J. Lu from comment #4) > > So, do we want to ignore the TYPE_EMPTY_P arguments even for argument > > alignment computations (both at the caller and callee)? > > We should ask it in x86-64 psABI group. Can you please do that? Thanks.
[Bug middle-end/89972] [8/9 Regression] ICE in expand_call, at calls.c:4229
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89972
--- Comment #4 from H.J. Lu ---
(In reply to Jakub Jelinek from comment #3)
> Looking at
> struct S { long a[0] __attribute__ ((aligned (32))); };
> long double u;
> void baz (struct S *);
> void bar (long double x, struct S y, long double z)
> {
> u = x + z;
> baz ();
> }
> this doesn't ICE, but gcc emits loads from rsp+32 and rsp+64, while clang
> from rbp+16 and rbp+32.
> So, do we want to ignore the TYPE_EMPTY_P arguments even for argument
> alignment computations (both at the caller and callee)?
We should ask it in x86-64 psABI group.
> Do we want some -Wpsabi warning for this?
I think so.
[Bug middle-end/89972] [8/9 Regression] ICE in expand_call, at calls.c:4229
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89972
--- Comment #3 from Jakub Jelinek ---
Looking at
struct S { long a[0] __attribute__ ((aligned (32))); };
long double u;
void baz (struct S *);
void bar (long double x, struct S y, long double z)
{
u = x + z;
baz ();
}
this doesn't ICE, but gcc emits loads from rsp+32 and rsp+64, while clang from
rbp+16 and rbp+32.
So, do we want to ignore the TYPE_EMPTY_P arguments even for argument alignment
computations (both at the caller and callee)?
Do we want some -Wpsabi warning for this?
[Bug middle-end/89972] [8/9 Regression] ICE in expand_call, at calls.c:4229
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89972
Jakub Jelinek changed:
What|Removed |Added
Keywords||ABI
CC||andi-gcc at firstfloor dot org,
||hjl.tools at gmail dot com,
||hubicka at gcc dot gnu.org,
||jakub at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek ---
Guess this is primarily an ABI issue, whether such arguments shouldn't be
passed at all even if they have the extra alignments or not.
On:
struct S { long a[0] __attribute__ ((aligned (32))); };
void bar (long double, struct S, long double);
void foo (void)
{
struct S b;
bar (8.0L, b, 9.0L);
}
clang doesn't agree with icc, clang passes 8.0L at rsp and 9.0 at rsp+16, while
icc at rsp and rsp+32 (and gcc ICEs).
[Bug middle-end/89972] [8/9 Regression] ICE in expand_call, at calls.c:4229
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89972 Richard Biener changed: What|Removed |Added Priority|P3 |P2 Component|c |middle-end
