Make jpeg-mmx ready for -fPIC

Logged In: YES
user_id=155293

Hallo

With your patch I have problems linking the stuff together:
gcc -o djpeg djpeg.o wrppm.o wrgif.o wrtarga.o wrrle.o
wrbmp.o rdcolmap.o cdjpeg.o libjpeg-mmx.a
libjpeg-mmx.a(jidctint.o)(.text+0x46): In function
`domidct8x8llmW':
: undefined reference to `fix_054n184_054'
....
libjpeg-mmx.a(jidctint.o)(.text+0xce8): more undefined
references to `const_0x0808' follow
....
libjpeg-mmx.a(jdcolor.o)(.text+0x185): In function
`ycc_rgb_convert_mmx':
: undefined reference to `te0'
.
I tried to compile it on a Suse 9.2 Athlon 32Bit System.
gcc (GCC) 3.3.4 (pre 3.3.5 20040809)
NASM version 0.98.38 compiled on Oct 2 2004

If you need fuerther information please tell me.

Logged In: YES
user_id=313104

> Date: 2005-02-16 22:02
> Sender: waldviertler
> With your patch I have problems linking the stuff together:

Are you sure it's this patch that causes the problem?
I had basically the same, unrelated, issue with undefined
symbols, and I made a patch for gcc 3.4 that is posted under
item 1107275:
https://sourceforge.net/tracker/?func=detail&aid=1107275&group_id=5776&atid=305776

Logged In: YES
user_id=155293

Appling your other patch first, and thes errors go away,
but, now I have problems appling this patch:
cat jpeg-mmx-0.1.5-alt-pic-register-use.patch | patch -p0
patching file jpeg-mmx-0.1.5/jidctfst.c
patching file jpeg-mmx-0.1.5/jdcolor.c
Hunk #2 FAILED at 348.
1 out of 2 hunks FAILED -- saving rejects to file
jpeg-mmx-0.1.5/jdcolor.c.rej

patching file jpeg-mmx-0.1.5/jidctint.c
Hunk #2 FAILED at 2837.
1 out of 2 hunks FAILED -- saving rejects to file
jpeg-mmx-0.1.5/jidctint.c.rej
patching file jpeg-mmx-0.1.5/jdmerge.c
patching file jpeg-mmx-0.1.5/cpu_accel.c
patching file jpeg-mmx-0.1.5/jdapimin.c
Hunk #1 succeeded at 413 (offset 1 line).
Hunk #2 succeeded at 462 (offset 1 line).
Hunk #3 succeeded at 481 (offset 1 line).
Hunk #4 succeeded at 490 (offset 1 line).
patching file jpeg-mmx-0.1.5/jdsample.c

When I try to apply you patch maunaly the compiling fails.
When I ignore them compiling works well. (except the other
integer constant is too large for "long" type) messages from
your other patch.

PS: You can write me directly, or use the mailinglist. That
is usually a bit faster.

Logged In: YES
user_id=8236

The patch compiled correctly against the current CVS on my
machine. I let Bernhard decide if he accepts the patch into
the source, but it is looking good.

Logged In: YES
user_id=155293

Just added it to the CVS

Logged In: YES
user_id=155293

From what we know the jpeg-mmx is only used with the mjpegtools.
If there are other programms using it it would be nice if
you could tell us which use the jpeg-mmx.

The patches you sent made the jpeg-mmx uncompilabel on gcc
2.95.x (2.95.3 is know by now):
gcc -O6 -I. -c -o jidctint.o jidctint.c
jidctint.c: In function `domidct8x8llmW':
jidctint.c:2868: more than 10 operands in `asm'
gmake: *** [jidctint.o] Error 1
Is it possible that you check it with the old compiler ?

Have the change to run any tests on a 64Bit Machine ?
One person had tried to compile it on a Opter 250 with Suse
9.2 and gcc 3.3.4. And it didn't compile.

Logged In: NO

There is a GStreamer plugin that uses jpeg-mmx, and
GStreamer is all dynamically loaded.
Libquicktime and transcode also seem to use it.

I admit I didn't check my patch on any antiquated gcc
version. I'm afraid you simply can't have it both ways.

I don't have a 64bit machine to try my patch against.

Logged In: YES
user_id=313104

Strangely, on two files the patch was applied halfway, as
can be seen in 0.1.6. As a result, the routines got "pushl
%ebx\n" with no matching "popl %ebx\n" at the end (and ebx
still in the clobber clause, which allowed me to catch that
quickly)!
I guess this broken code gets little to no use. I attach the
fix here.

PIC register use (and code unbreaking) patch for jpeg-mmx 0.1.6