Steven M. Schultz wrote: > Hi - > > >>From: Wolfgang Fritz <[EMAIL PROTECTED]> >> >>>gcc -O6 -I. -c -o jdapimin.o jdapimin.c >>>/tmp/cc7IfzBn.s: Assembler messages: >>>/tmp/cc7IfzBn.s:619: Error: symbol `NOT_SUPPORTED' is already defined >>>gmake: *** [jdapimin.o] Error 1 >>> >> >>The reason for this it that gcc inlines the assembler function >>mmxsupport() at 2 places, so the symbol NOT_SUPPORTED is defined twice. > > > Hmmm, the 2 places are under different #ifdef conditions. > > The first is '#ifdef HAVE_MMX_INTEL_MNEMONICS' while the second > one is '#if defined(HAVE_MMX_ATT_MNEMONICS)' > > Why does gcc 3.3 include both sections of code? Effectively > "gcc-3.3 -O6" causes the compiler to apparently ignore the '#ifdef' > statements and include both blocks of code which causes NOT_SUPPORTED > to be defined twice. > > In order for NOT_SUPPORTED to be defined twice both the > #ifdef HAVE_MMX_INTEL_MNEMONICS and the #ifdef HAVE_MMX_ATT_MNEMONIC > sections would have to be compiled. > >
No. That's what I thought too, but it is not the case. If you change the symbol name in one of the #ifdefs, the error remains. If you do a gcc -O6 -I. -c -S -o jdapimin.s jdapimin.c and look at the assembler code in jdapimin.s, you see that the code of mmxsupport() really appears twice: First the function itself, then inlined in jpeg_CreateDecompress(). Seems that gcc-3.3 does a more aggressive inlining with -O6. Wolfgang >>A fix without changing the O option is to change the global symbol to a >>local one. See the attached patch. > > > That will work but I wonder if it is simply masking the problem? > > The current jpeg_mmx code seems to have a weird blend of support > for ATT mnemonics in some modules, ATT and INTEL in other modules > and ATT + INTEL in others. > > Thus the 'cpu_accel.c' module will return a meaningful value if > 'HAVE_MMX_ATT_MNEMONICS' is defined because there is no > HAVE_MMX_INTEL_MNEMONICS support in cpu_accel.c - thus if jpeg_mmx > is built on a system without ATT_MNEMONICS defined cpu_accel will > always return 0. > > Would a better approach be to only support 1 set of assembler > mnemonics? > > Cheers, > Steven Schultz > > > ------------------------------------------------------- > This SF.net email is sponsored by: eBay > Get office equipment for less on eBay! > http://adfarm.mediaplex.com/ad/ck/711-11697-6916-5 > _______________________________________________ > Mjpeg-users mailing list > [EMAIL PROTECTED] > https://lists.sourceforge.net/lists/listinfo/mjpeg-users > ------------------------------------------------------- This SF.net email is sponsored by: eBay Get office equipment for less on eBay! http://adfarm.mediaplex.com/ad/ck/711-11697-6916-5 _______________________________________________ Mjpeg-users mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/mjpeg-users
