AltiVec too pedantic about type qualifiers

AltiVec too pedantic about type qualifiers

[Daniel Egger]

Hija,

I just discovered that AltiVec is quite pedantic and anal about type
qualifiers.

#include <altivec.h>

void test (const char *test)
{
  vector signed char vec;

  vec = vec_ld (0, test);
}

delivers:

test.c: In function `test':
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: warning: passing arg 2 of `__builtin_altivec_lvx' discards
qualifiers 
from pointer target type
test.c:7: incompatible types in assignment

IMHO this should at a maximum deliver on of the warning and no
error at all, given the fact that I just tried to hint the compiler
that the pointer is not used to write to memory.

In real world code this also leads to 

none.c:283: incompatible types in assignment
none.c:296: Internal compiler error in gen_reg_rtx, at emit-rtl.c:654
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions.

but unfortunately I cannot release this code and I haven't been able to
synthesize a testcase which bails out with an compiler error yet.

-- 
Servus,
       Daniel

Re: AltiVec too pedantic about type qualifiers

[Aldy Hernandez]

On Monday, February 25, 2002, at 10:40  AM, Daniel Egger wrote:

> Hija,
>
> I just discovered that AltiVec is quite pedantic and anal about type
> qualifiers.
>
> #include <altivec.h>
>
> void test (const char *test)
> {
>   vector signed char vec;
>
>   vec = vec_ld (0, test);
> }
>

const problem again.

but if you fix the const, you'll get an incompatible assignment.
this is because you're probably building on darwin which has
char defaulting to unsigned char, so vec_ld will return a
"vector unsigned char".

this is an inconsistency in darwin having chars being unsigned.

so you'll have to do:

void test (signed char *test)
            ^^^^^^
{
   vector signed char vec;
   vec = vec_ld (0, test);
}

i think there's probably a flag to have gcc's chars default to
signed (anyone??)

> none.c:283: incompatible types in assignment
> none.c:296: Internal compiler error in gen_reg_rtx, at emit-rtl.c:654
> Please submit a full bug report,
> with preprocessed source if appropriate.
> See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions.
>
> but unfortunately I cannot release this code and I haven't been able to
> synthesize a testcase which bails out with an compiler error yet.

ok, *this* is a valid error.  you're going to have to simplify it
enough for me to reproduce.  if you want, send it off list and
i'll fix it.

>
> --
> Servus,
>        Daniel
>
>
--
Aldy Hernandez                                E-mail: aldyh@redhat.com
Professional Gypsy Lost in Australia
Red Hat, Inc.

Re: AltiVec too pedantic about type qualifiers

[Daniel Egger]

Am Mon, 2002-02-25 um 03.10 schrieb Aldy Hernandez:

> but if you fix the const, you'll get an incompatible assignment.
> this is because you're probably building on darwin which has
> char defaulting to unsigned char, so vec_ld will return a
> "vector unsigned char".

Negative. This is linux-ppc.

> this is an inconsistency in darwin having chars being unsigned.

This is no problem here. :)
 
> i think there's probably a flag to have gcc's chars default to
> signed (anyone??)

-fsigned-char
 
-- 
Servus,
       Daniel

Re: AltiVec too pedantic about type qualifiers

[Andrew Pinski]

On PPC, all char's are unsigned by default.

Thanks,
Andrew Pinski
On Monday, February 25, 2002, at 03:40 , Daniel Egger wrote:

> Am Mon, 2002-02-25 um 03.10 schrieb Aldy Hernandez:
>
>> but if you fix the const, you'll get an incompatible assignment.
>> this is because you're probably building on darwin which has
>> char defaulting to unsigned char, so vec_ld will return a
>> "vector unsigned char".
>
> Negative. This is linux-ppc.
>
>> this is an inconsistency in darwin having chars being unsigned.
>
> This is no problem here. :)
>
>> i think there's probably a flag to have gcc's chars default to
>> signed (anyone??)
>
> -fsigned-char
>
> --
> Servus,
>        Daniel
>
>
>

Re: AltiVec too pedantic about type qualifiers

[Daniel Egger]

Am Mon, 2002-02-25 um 22.09 schrieb Andrew Pinski:

> On PPC, all char's are unsigned by default.

I never intended to claim something different.
 
-- 
Servus,
       Daniel

Re: AltiVec too pedantic about type qualifiers

[Aldy Hernandez]

On Tuesday, February 26, 2002, at 08:09  AM, Andrew Pinski wrote:

> On PPC, all char's are unsigned by default.

that explains it.