pcc i386 -fPIC compiler error on src/complex/catanf.c

pcc i386 -fPIC compiler error on src/complex/catanf.c

[Isaac Dunham]

Hello,
With recent pcc (built from cvs since 1.1.0 was released), I've tried
building musl a few times, but the shared build consistently dies
on src/complex/catanf.c (command wrapped to avoid getting mangled
in the mail):
$ make
pcc -std=c99 -nostdinc -ffreestanding -fexcess-precision=standard    \
  -frounding-math -D_XOPEN_SOURCE=700 -I./arch/i386 -I./src/internal \
  -I./include  -Os -pipe -fomit-frame-pointer -fno-unwind-tables     \
  -fno-asynchronous-unwind-tables -Wa,--noexecstack -march=i486      \
  -Werror=implicit-function-declaration -Werror=implicit-int 	     \
  -Werror=pointer-sign -Werror=pointer-arith -fno-stack-protector    \
  -fPIC -DSHARED -c -o src/complex/catanf.lo src/complex/catanf.c
src/complex/catanf.c, line 80: compiler error: bad STCALL hidden reg
error: /usr/libexec/ccom terminated with status 1
Makefile:124: recipe for target 'src/complex/catanf.lo' failed
make: *** [src/complex/catanf.lo] Error 1

Line 80 is the closing bracket of _redupif(), which is fairly basic
floating point math.

If I try removing flags from the command until it compiles, -fPIC seems
to be what triggers the compiler error.

I'm compiling pcc with these flags:
$ ./configure --prefix=/usr --build=i486-alpine-linux-musl --target=i486-alpine-linux-musl --enable-tls --with-libdir=/usr/lib --with-incdir=/usr/include

With tls disabled (as per default), it still fails the same.

I can't really isolate the error in question, I'm afraid.

Thanks,
Isaac Dunham

Re: pcc i386 -fPIC compiler error on src/complex/catanf.c

[Iain Hibbert]

On Sat, 3 Jan 2015, Isaac Dunham wrote:

> Hello,
> With recent pcc (built from cvs since 1.1.0 was released), I've tried
> building musl a few times, but the shared build consistently dies
> on src/complex/catanf.c (command wrapped to avoid getting mangled
> in the mail):
> $ make
> pcc -std=c99 -nostdinc -ffreestanding -fexcess-precision=standard    \
>   -frounding-math -D_XOPEN_SOURCE=700 -I./arch/i386 -I./src/internal \
>   -I./include  -Os -pipe -fomit-frame-pointer -fno-unwind-tables     \
>   -fno-asynchronous-unwind-tables -Wa,--noexecstack -march=i486      \
>   -Werror=implicit-function-declaration -Werror=implicit-int 	     \
>   -Werror=pointer-sign -Werror=pointer-arith -fno-stack-protector    \
>   -fPIC -DSHARED -c -o src/complex/catanf.lo src/complex/catanf.c
> src/complex/catanf.c, line 80: compiler error: bad STCALL hidden reg
> error: /usr/libexec/ccom terminated with status 1
> Makefile:124: recipe for target 'src/complex/catanf.lo' failed
> make: *** [src/complex/catanf.lo] Error 1
>
> Line 80 is the closing bracket of _redupif(), which is fairly basic
> floating point math.

it will likely be something inside that function that does not show up
until the function is being emitted. STCALL is something to do with
structure arguments, which some kinds of floating point are big enough to
be considered.. so I guess some kind of double or long double arguments?

> I can't really isolate the error in question, I'm afraid.

maybe if you post the function, somebody else would be able to?

iain
_______________________________________________
Pcc mailing list
Pcc@lists.ludd.ltu.se
https://lists.ludd.ltu.se/cgi-bin/mailman/listinfo/pcc

Re: Re: [Pcc] pcc i386 -fPIC compiler error on src/complex/catanf.c

[Rich Felker]

On Sun, Jan 04, 2015 at 05:47:24PM +0000, Iain Hibbert wrote:
> On Sat, 3 Jan 2015, Isaac Dunham wrote:
> 
> > Hello,
> > With recent pcc (built from cvs since 1.1.0 was released), I've tried
> > building musl a few times, but the shared build consistently dies
> > on src/complex/catanf.c (command wrapped to avoid getting mangled
> > in the mail):
> > $ make
> > pcc -std=c99 -nostdinc -ffreestanding -fexcess-precision=standard    \
> >   -frounding-math -D_XOPEN_SOURCE=700 -I./arch/i386 -I./src/internal \
> >   -I./include  -Os -pipe -fomit-frame-pointer -fno-unwind-tables     \
> >   -fno-asynchronous-unwind-tables -Wa,--noexecstack -march=i486      \
> >   -Werror=implicit-function-declaration -Werror=implicit-int 	     \
> >   -Werror=pointer-sign -Werror=pointer-arith -fno-stack-protector    \
> >   -fPIC -DSHARED -c -o src/complex/catanf.lo src/complex/catanf.c
> > src/complex/catanf.c, line 80: compiler error: bad STCALL hidden reg
> > error: /usr/libexec/ccom terminated with status 1
> > Makefile:124: recipe for target 'src/complex/catanf.lo' failed
> > make: *** [src/complex/catanf.lo] Error 1
> >
> > Line 80 is the closing bracket of _redupif(), which is fairly basic
> > floating point math.
> 
> it will likely be something inside that function that does not show up
> until the function is being emitted. STCALL is something to do with
> structure arguments, which some kinds of floating point are big enough to
> be considered.. so I guess some kind of double or long double arguments?

I would guess it's treating complex floating point objects like
structures. However...

> > I can't really isolate the error in question, I'm afraid.
> 
> maybe if you post the function, somebody else would be able to?

...the function where the error is reported is not using any complex
objects:

http://git.musl-libc.org/cgit/musl/tree/src/complex/catanf.c?id=v1.1.5

So something else must be going on.

Rich

Re: Re: [Pcc] pcc i386 -fPIC compiler error on src/complex/catanf.c

[Jens Gustedt]

[-- Attachment #1: Type: text/plain, Size: 1515 bytes --]

Am Sonntag, den 04.01.2015, 13:01 -0500 schrieb Rich Felker:
> On Sun, Jan 04, 2015 at 05:47:24PM +0000, Iain Hibbert wrote:
> > it will likely be something inside that function that does not show up
> > until the function is being emitted. STCALL is something to do with
> > structure arguments, which some kinds of floating point are big enough to
> > be considered.. so I guess some kind of double or long double arguments?
> 
> I would guess it's treating complex floating point objects like
> structures. However...
> 
> > > I can't really isolate the error in question, I'm afraid.
> > 
> > maybe if you post the function, somebody else would be able to?
> 
> ...the function where the error is reported is not using any complex
> objects:
> 
> http://git.musl-libc.org/cgit/musl/tree/src/complex/catanf.c?id=v1.1.5

You are refering to _redupif?

This is a `static` function that is used exactly once. Perhaps the
compiler "augments" it to `float complex` return type, because the
context where it is used is of that type. It would be interesting to
see if the error disappears if the return value is first stored in a
`volatile float`, and only then assigned to `w`.

Jens

-- 
:: INRIA Nancy Grand Est ::: AlGorille ::: ICube/ICPS :::
:: ::::::::::::::: office Strasbourg : +33 368854536   ::
:: :::::::::::::::::::::: gsm France : +33 651400183   ::
:: ::::::::::::::: gsm international : +49 15737185122 ::
:: http://icube-icps.unistra.fr/index.php/Jens_Gustedt ::



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: pcc i386 -fPIC compiler error on src/complex/catanf.c

[Anders Magnusson]

Iain Hibbert skrev den 2015-01-04 18:47:
> On Sat, 3 Jan 2015, Isaac Dunham wrote:
>
>> Hello,
>> With recent pcc (built from cvs since 1.1.0 was released), I've tried
>> building musl a few times, but the shared build consistently dies
>> on src/complex/catanf.c (command wrapped to avoid getting mangled
>> in the mail):
>> $ make
>> pcc -std=c99 -nostdinc -ffreestanding -fexcess-precision=standard    \
>>    -frounding-math -D_XOPEN_SOURCE=700 -I./arch/i386 -I./src/internal \
>>    -I./include  -Os -pipe -fomit-frame-pointer -fno-unwind-tables     \
>>    -fno-asynchronous-unwind-tables -Wa,--noexecstack -march=i486      \
>>    -Werror=implicit-function-declaration -Werror=implicit-int 	     \
>>    -Werror=pointer-sign -Werror=pointer-arith -fno-stack-protector    \
>>    -fPIC -DSHARED -c -o src/complex/catanf.lo src/complex/catanf.c
>> src/complex/catanf.c, line 80: compiler error: bad STCALL hidden reg
>> error: /usr/libexec/ccom terminated with status 1
>> Makefile:124: recipe for target 'src/complex/catanf.lo' failed
>> make: *** [src/complex/catanf.lo] Error 1
>>
>> Line 80 is the closing bracket of _redupif(), which is fairly basic
>> floating point math.
> it will likely be something inside that function that does not show up
> until the function is being emitted. STCALL is something to do with
> structure arguments, which some kinds of floating point are big enough to
> be considered.. so I guess some kind of double or long double arguments?
>
This assertion occurs because pass2 code expects a hidden argument to be
present when calling a struct return function, but the hidden arg cannot be
found in the argument list, and in this special case the hidden arg shall be
in a register.

I would like a preprocessed output from the above to diagnose further (which
means to just add a -E to the line above).

-- Ragge
_______________________________________________
Pcc mailing list
Pcc@lists.ludd.ltu.se
https://lists.ludd.ltu.se/cgi-bin/mailman/listinfo/pcc

Re: pcc i386 -fPIC compiler error on src/complex/catanf.c

[Isaac Dunham]

[-- Attachment #1: Type: text/plain, Size: 1992 bytes --]

On Sun, Jan 04, 2015 at 08:24:54PM +0100, Anders Magnusson wrote:
> Iain Hibbert skrev den 2015-01-04 18:47:
> >On Sat, 3 Jan 2015, Isaac Dunham wrote:
> >
> >>Hello,
> >>With recent pcc (built from cvs since 1.1.0 was released), I've tried
> >>building musl a few times, but the shared build consistently dies
> >>on src/complex/catanf.c (command wrapped to avoid getting mangled
> >>in the mail):
> >>$ make
> >>pcc -std=c99 -nostdinc -ffreestanding -fexcess-precision=standard    \
> >>   -frounding-math -D_XOPEN_SOURCE=700 -I./arch/i386 -I./src/internal \
> >>   -I./include  -Os -pipe -fomit-frame-pointer -fno-unwind-tables     \
> >>   -fno-asynchronous-unwind-tables -Wa,--noexecstack -march=i486      \
> >>   -Werror=implicit-function-declaration -Werror=implicit-int 	     \
> >>   -Werror=pointer-sign -Werror=pointer-arith -fno-stack-protector    \
> >>   -fPIC -DSHARED -c -o src/complex/catanf.lo src/complex/catanf.c
> >>src/complex/catanf.c, line 80: compiler error: bad STCALL hidden reg
> >>error: /usr/libexec/ccom terminated with status 1
> >>Makefile:124: recipe for target 'src/complex/catanf.lo' failed
> >>make: *** [src/complex/catanf.lo] Error 1
> >>
> >>Line 80 is the closing bracket of _redupif(), which is fairly basic
> >>floating point math.
> >it will likely be something inside that function that does not show up
> >until the function is being emitted. STCALL is something to do with
> >structure arguments, which some kinds of floating point are big enough to
> >be considered.. so I guess some kind of double or long double arguments?
> >
> This assertion occurs because pass2 code expects a hidden argument to be
> present when calling a struct return function, but the hidden arg cannot be
> found in the argument list, and in this special case the hidden arg shall be
> in a register.
> 
> I would like a preprocessed output from the above to diagnose further (which
> means to just add a -E to the line above).

Ok, will send.

Thanks!
Isaac Dunham

[-- Attachment #2: catanf.i.gz --]
[-- Type: application/octet-stream, Size: 2638 bytes --]

[-- Attachment #3: Type: text/plain, Size: 141 bytes --]

_______________________________________________
Pcc mailing list
Pcc@lists.ludd.ltu.se
https://lists.ludd.ltu.se/cgi-bin/mailman/listinfo/pcc

Re: pcc i386 -fPIC compiler error on src/complex/catanf.c

[Isaac Dunham]

[-- Attachment #1: Type: text/plain, Size: 705 bytes --]

Sorry for the noise, but the catanf.i.gz above was from a slightly modified
catanf.c (which had the same behavior as the original code).
I had changed this from upstream musl:
===
--- a/src/complex/catanf.c
+++ b/src/complex/catanf.c
@@ -61,12 +61,12 @@ static const double DP1 = 3.140625;
 static const double DP2 = 9.67502593994140625E-4;
 static const double DP3 = 1.509957990978376432E-7;
 
-static float _redupif(float xx)
+static float _redupif(float x)
 {
-	float x, t;
+	float t;
 	long i;
 
-	x = xx;
+	//x = xx;
 	t = x/(float)M_PI;
 	if (t >= 0.0f)
 		t += 0.5f;
===

The preprocessed code has a corresponding change; I'm attaching the gzip'd
result of the original code.

Thanks,
Isaac Dunham

[-- Attachment #2: catanf.i.gz --]
[-- Type: application/octet-stream, Size: 2637 bytes --]

[-- Attachment #3: Type: text/plain, Size: 141 bytes --]

_______________________________________________
Pcc mailing list
Pcc@lists.ludd.ltu.se
https://lists.ludd.ltu.se/cgi-bin/mailman/listinfo/pcc

Re: pcc i386 -fPIC compiler error on src/complex/catanf.c

[Anders Magnusson]

Isaac Dunham skrev den 2015-01-04 22:04:
> On Sun, Jan 04, 2015 at 08:24:54PM +0100, Anders Magnusson wrote:
>> Iain Hibbert skrev den 2015-01-04 18:47:
>>> On Sat, 3 Jan 2015, Isaac Dunham wrote:
>>>
>>>> Hello,
>>>> With recent pcc (built from cvs since 1.1.0 was released), I've tried
>>>> building musl a few times, but the shared build consistently dies
>>>> on src/complex/catanf.c (command wrapped to avoid getting mangled
>>>> in the mail):
>>>> $ make
>>>> pcc -std=c99 -nostdinc -ffreestanding -fexcess-precision=standard    \
>>>>    -frounding-math -D_XOPEN_SOURCE=700 -I./arch/i386 -I./src/internal \
>>>>    -I./include  -Os -pipe -fomit-frame-pointer -fno-unwind-tables     \
>>>>    -fno-asynchronous-unwind-tables -Wa,--noexecstack -march=i486      \
>>>>    -Werror=implicit-function-declaration -Werror=implicit-int 	     \
>>>>    -Werror=pointer-sign -Werror=pointer-arith -fno-stack-protector    \
>>>>    -fPIC -DSHARED -c -o src/complex/catanf.lo src/complex/catanf.c
>>>> src/complex/catanf.c, line 80: compiler error: bad STCALL hidden reg
>>>> error: /usr/libexec/ccom terminated with status 1
>>>> Makefile:124: recipe for target 'src/complex/catanf.lo' failed
>>>> make: *** [src/complex/catanf.lo] Error 1
>>>>
>>>> Line 80 is the closing bracket of _redupif(), which is fairly basic
>>>> floating point math.
>>> it will likely be something inside that function that does not show up
>>> until the function is being emitted. STCALL is something to do with
>>> structure arguments, which some kinds of floating point are big enough to
>>> be considered.. so I guess some kind of double or long double arguments?
>>>
>> This assertion occurs because pass2 code expects a hidden argument to be
>> present when calling a struct return function, but the hidden arg cannot be
>> found in the argument list, and in this special case the hidden arg shall be
>> in a register.
>>
>> I would like a preprocessed output from the above to diagnose further (which
>> means to just add a -E to the line above).
> Ok, will send.
>
Thanks, fixed now.
Float complex struct return were not dealt with as it should in the PIC 
case.

-- Ragge
_______________________________________________
Pcc mailing list
Pcc@lists.ludd.ltu.se
https://lists.ludd.ltu.se/cgi-bin/mailman/listinfo/pcc

Re: [Pcc] pcc i386 -fPIC compiler error on src/complex/catanf.c

[Isaac Dunham]

On Mon, Jan 05, 2015 at 11:31:11AM +0100, Anders Magnusson wrote:
> Thanks, fixed now.
> Float complex struct return were not dealt with as it should in the PIC
> case.
> 

Thank you, but I'm afraid that now I'm getting another error for most code
when I compile with -fPIC:
$ pcc -fPIC argvname.c -o argvname
/tmp/ctm.HhEooN: Assembler messages:
/tmp/ctm.HhEooN:1: Error: no such instruction: `pucext'
/tmp/ctm.HhEooN:2: Error: junk at end of line, first unrecognized character is `0'
/tmp/ctm.HhEooN:3: Error: junk at end of line, first unrecognized character is `0'
/tmp/ctm.HhEooN:4: Error: junk at end of line, first unrecognized character is `0'
/tmp/ctm.HhEooN:5: Error: junk at end of line, first unrecognized character is `0'
/tmp/ctm.HhEooN:6: Error: junk at end of line, first unrecognized character is `0'
/tmp/ctm.HhEooN:7: Error: junk at end of line, first unrecognized character is `0'
/tmp/ctm.HhEooN:8: Error: junk at end of line, first unrecognized character is `0'
error: as terminated with status 1


argvname.c is a trivial example, as follows:
#include <stdio.h>

int main(int argc, char *argv[])
{
	puts(argv[0]);
}
EOF

I'm getting the impression that this was debugging code dumping some of
the compiler state (at line 174, arch/i386/local.c).
Is that correct?

Thanks,
Isaac Dunham

Re: [Pcc] pcc i386 -fPIC compiler error on src/complex/catanf.c

[Anders Magnusson]

Isaac Dunham skrev den 2015-01-06 00:25:
> On Mon, Jan 05, 2015 at 11:31:11AM +0100, Anders Magnusson wrote:
>> Thanks, fixed now.
>> Float complex struct return were not dealt with as it should in the PIC
>> case.
>>
> Thank you, but I'm afraid that now I'm getting another error for most code
> when I compile with -fPIC:
> $ pcc -fPIC argvname.c -o argvname
> /tmp/ctm.HhEooN: Assembler messages:
>
I did check in a debug printout by mistake :-/  Fixed now, thanks.

-- R