Bug in cgtools 7.3.5 compiling libjpeg.

Eric,

The compiler development team is working on a fix for this bug and were wondering if you could provide a linkable test case that we can run to verify if the whole problem is fixed. Are you able to share the full project or a cutdown version of a builable project that reproduces the issue?

AartiG,

I've cut down our project to just libjpeg6b. To compile, run the commands

./configure CC=cl6x_wrapper CFLAGS="--abi=coffabi -mv6600 --mem_model:data=far -mo -pden -pds 179 -pds 552 --gcc -o3 --symdebug:none

and then

make

on a Linux system (cygwin may also work, but I haven't tested it). cl6x_wrapper is a simple Perl script that translates gcc command line options into the corresponding cl6x options (at least those required for configure to work). It assumes that cl6x can be found in the current PATH and C6X_C_DIR is set correctly.

I cannot run the configure script on my linux machine, Here is what I got::

[wzhao@dflvdt0024]testcase%./configure CC=cl6x_wrapper CFLAGS="--abi=coffabi -mv6600 --mem_model:data=far -mo -pden -pds 179 -pds 552 --gcc -o3 --symdebug:none

>  

It hangs there.

Can you just give me a ready to run Makefile to try? I mean after the above ./configure

Thanks,

Wei

Looks like it needs a closing double-quote.

I add the closing double quote.

Configure is done. And I type gmake, I got the following, any clue where it got wrong?

cl6x_wrapper --abi=coffabi -mv6600 --mem_model:data=far -mo -pden -pds 179 -pds 552 --gcc -o3 --symdebug:none  -I.   -c -o wrjpgcom.o wrjpgcom.c
--> cl6x --abi=coffabi -mv6600 --mem_model:data=far -mo -pden -pds 179 -pds 552 --gcc -o3 --symdebug:none -i. -c wrjpgcom.c -ft /tmp/cg6xtmp_dR6ySStR -eo=.o -fr /tmp/cg6xtmp_dR6ySStR
--> mv "/tmp/cg6xtmp_dR6ySStR/wrjpgcom.o" "wrjpgcom.o"
cl6x_wrapper  -o wrjpgcom wrjpgcom.o
WARNING: "wrjpgcom" is in the list of programs that are not to be built.
--> touch "wrjpgcom"
[wzhao@dflvdt0024]testcase%ls -lt |more
total 4000
-rw-rw----  1 wzhao lang  15402 Oct 24 16:26 wrjpgcom.o
-rw-rw----  1 wzhao lang  15871 Oct 24 16:25 rdjpgcom.o
-rw-rw----  1 wzhao lang  23687 Oct 24 16:25 transupp.o
-rw-rw----  1 wzhao lang      0 Oct 24 16:25 wrjpgcom
-rw-rw----  1 wzhao lang      0 Oct 24 16:25 rdjpgcom
-rw-rw----  1 wzhao lang      0 Oct 24 16:25 jpegtran
-rw-rw----  1 wzhao lang  24113 Oct 24 16:25 jpegtran.o

This looks fine to me. What you should have (if everything went right) is a libjpeg.a - the example programs aren't linked because that would require a linker command file, and we don't need them in any case. The original bug was that compiling jfdctint.c for COFF with an optimization level >1 produced an internal compiler error - if this no longer occurs, then I would assume this bug has been fixed.

Eric,

Get it to compiler is just one goal of this fix. I also would like to verifiy that the fix can get your code running correctly.

So are you using 7.3.5 now? How about I give you a engineering release of the cg6x, you can place it in your bin/ directory and try to build and run your whole project. I want to make sure that this fix is validate.

Thanks,

Wei

Wei,

I'm currently using 7.3.6, which can build a runable version of our code using the COFF object format. Building in ELF format fails due to a linker issue (this is SDSCM00044738 in TI's bug tracker). As our code is proprietary, I unfortunately cannot make it available to you for testing. If you have an engineering release you would like me to test, I can do that.

Eric,

Try it and it should get your project build and then see if everything works.

Any problem, please let me know.

Wei

Eric,

It seems I can not get the filed uploaded to the forum.

Is there a way I can send the file to you via FTP or some other ways.

Wei

Wei,

the last time I got an engineering release from TI was via the ftp server ftppickup.ti.com. Perhaps this is an option ?

Wei,

I've attached my email address as a .png image (forgive my paranoia, but I don't want to make it too easy for spammers to harvest).