gcc 3.3.2 ICE on altivec - bug category question

gcc 3.3.2 ICE on altivec - bug category question

[Jakub Bogusz]

I'm getting an ICE using gcc 3.3.2 (doesn't matter here if PR
target/11793 fix is applied or not, I tried both versions) on
ppc-linux system, on the following code (already preprocessed and
simplified from original one):


typedef __attribute__((vector_size(16))) signed int vector_s_t;

static vector_s_t my_vec_and (vector_s_t const A, vector_s_t const B)
{
return __builtin_altivec_vand(A, B);
}


$ gcc -c alt.i -maltivec
alt.i: In function `my_vec_and':
alt.i:6: error: unrecognizable insn:
(insn 5 4 30 0 (nil) (set (reg:V4SI 119)
        (mem/u/f:V4SI (plus:SI (reg/f:SI 67 ap)
                (const_int 8 [0x8])) [0 B+0 S16 A128])) -1 (nil)
    (nil))
alt.i:6: internal compiler error: in extract_insn, at recog.c:2175
Please submit a full bug report,
with preprocessed source if appropriate.

But:
$ gcc -c alt.i -maltivec -mabi=altivec
$

Is suppose that altivec ABI can be needed to return vector... but
I haven't found such information - is it true?
And thus, does it fall under "ice-on-valid-code" or "ice-on-invalid-code"?
Can it be expected to be fixed in 3.3.x line, or 3.4/3.5?


-- 
Jakub Bogusz    http://cyber.cs.net.pl/~qboosh/
PLD Team        http://www.pld-linux.org/

Re: gcc 3.3.2 ICE on altivec - bug category question

[Segher Boessenkool]

On 19-jan-04, at 21:01, Jakub Bogusz wrote:

> I'm getting an ICE using gcc 3.3.2 (doesn't matter here if PR
> target/11793 fix is applied or not, I tried both versions) on
> ppc-linux system, on the following code (already preprocessed and
> simplified from original one):

<snip>

This is the same bug that causes the altivec-5.c testcase to fail.
I really should fix it ;-)

> Is suppose that altivec ABI can be needed to return vector... but
> I haven't found such information - is it true?
> And thus, does it fall under "ice-on-valid-code" or 
> "ice-on-invalid-code"?

Interesting question, I'm not sure really.  It might be that with the
plain-old ABI a pointer to a temporary is used, just like with structs.

But the ICE is unrelated, so it is ice-on-valid.

> Can it be expected to be fixed in 3.3.x line, or 3.4/3.5?

This on will probably not go on 3.3, sorry.


Segher

Re: gcc 3.3.2 ICE on altivec - bug category question

[Dara Hazeghi]

I'd call this ICE-on-invalid myself... Suggest you
file a bug report regardless though. I just checked
and its still present on 3.4 branch/mainline, so
clearly it still is an issue. Cheers,

Dara

__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus