https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #11 from Andrew Pinski ---
The bpf verifier is just plain broken when it comes to subreg usage. So after
every 32bit usage you need to output a zero extend. To fix that.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #10 from Andrew Pinski ---
The only way to fix this is to add extra zero extends all the time.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
Jose E. Marchesi changed:
What|Removed |Added
Ever confirmed|0 |1
Status|RESOLVED
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #8 from Andrew Pinski ---
(In reply to Andrew Pinski from comment #7)
> The problem is the verifier:
> Mär 26 23:57:12 H systemd[1]: 17: (ac) w0 ^= w3 ;
> R0_w=scalar()
> R3_w=scalar(smin=smin32=0,smax=umax=smax32
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #7 from Andrew Pinski ---
The problem is the verifier:
Mär 26 23:57:12 H systemd[1]: 17: (ac) w0 ^= w3 ;
R0_w=scalar() R3_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=1,var_off=(0x0;
0x1))
Mär 26 23:57:12 H syst
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
Andrew Pinski changed:
What|Removed |Added
Resolution|--- |INVALID
Status|UNCONFIRMED
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #5 from Andrew Pinski ---
here is one which makes similar code for both gcc and clang:
```
int foo2 ()
{
int quux = bar ();
unsigned long t3 = (unsigned int)quux;
t3+=-1;
asm("":"+r"(t3));
t3 = t3>>63;
unsigned char t =
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #4 from Andrew Pinski ---
```
int foo2 ()
{
int quux = bar ();
unsigned long t3 = (unsigned int)quux;
t3+=-1;
t3 = t3>>63;
unsigned char t = t3;
unsigned long t4 = (unsigned int)baz;
t4 = -t4;
t4 = t4>>63;
unsigned
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #3 from Andrew Pinski ---
```
int foo1 ()
{
int quux = bar ();
_Bool t = quux == 0;
_Bool t1 = baz != 0;
int t2 = t;
t2 ^= t1;
return t2;
}
```
Produces the same resulting asm.
The way clang produces comparisons is:
```
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #2 from Andrew Pinski ---
> - Disable the optimization for BPF until the verifier fixes the limitation if
> possible.
NO.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #1 from Andrew Pinski ---
There is nothing magical about phiopt (in this case) is doing that can't be
produced by an user.
What does the final code from the backend looks like?
11 matches
Mail list logo
