Internal Compiler error at line 1090 under any optimization level. Segment Faults. $ powerpc-wrs-vxworks-gcc --save-temps -Wall -include "altivec.h" -B/dna/ppc-gcc/i686/vxworks/lib/gcc-lib -D_AltiVec -D__VXWORKS__ -fno-builtin -mstrict-align -mabi=altivec -maltivec -mcpu=7400 -mvxworks -O0 -mregnames -DCPU=PPC604 -DRW_MULTI_THREAD -Id:/Torn ado/target/h -It:/opt/gedae/gedae_sys_4.0/include -It:/opt/gedae/gedae_sys_4.0/include/embeddable -I ../../../../VSIPL-1.5.0/src/ include -DM_PI=3.14159265358979323846 -DM_PI_2=1.57079632679489661923 -DM_PI_4=0.78539816339744830962 -DM_PIl=3.1415926535897932384 626433832795029L -DM_PI_2l=1.5707963267948966192313216916397514L -DM_PI_4l=0.7853981633974483096156608458198757L -c -I../../. ./../VSIPL-1.5.0/src/include/auto -I../../../../VSIPL-1.5.0/src/include -I../../../src/include -o small2.nas_gemp_f.o small2.nas_ge mp_f.c small2.nas_gemp_f.c: In function `nas_gempnn_f': small2.nas_gemp_f.c:38: warning: comparison between signed and unsigned small2.nas_gemp_f.c:40: warning: comparison between signed and unsigned small2.nas_gemp_f.c:42: warning: comparison between signed and unsigned small2.nas_gemp_f.c:1090: internal compiler error: Segmentation fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://www.gnu.org/software/gcc/bugs.html> for instructions. Release: gcc version 3.3 20021118 (experimental) Environment: Configured with: ../../src/gcc-core/configure --disable-shared --disable-multilib --enable-languages=c --with-gnu-as --with-gnu-ld --target=powerpc-wrs-vxworks --with-newlib --prefix=/dna/ppc-gcc/i Cygwin 1.3.13 hosted on Windows NT 4.0 Update 6, vxWorks 5.4 target platformm , Hardware Dell Optiplex GX110 Pentium 3 256 MB RAM, 600MHZ, How-To-Repeat: Compile File
Fix: Unknown
Responsible-Changed-From-To: unassigned->aldyh Responsible-Changed-Why: Altivec
From: Aldy Hernandez <aldyh@redhat.com> To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, vtaylor@dna-cs.com Cc: Subject: Re: c/8703: Compiler Segmentation Fault cross-compiling PPC Altivec accelerated code Date: Thu, 5 Dec 2002 18:44:43 -0800 It looks like you have a complicated expression that's expanding in such a way that it's running out of memory. The file *is* 441megs after all. Could you simplify the testcase to its bare minimum, and perhaps send an *un*preprocessed source (with header files of course). Aldy
State-Changed-From-To: open->feedback State-Changed-Why: Aldy already asked for feedback, so let's reflect this in the state of this report: ------------------------ It looks like you have a complicated expression that's expanding in such a way that it's running out of memory. The file *is* 441megs after all. Could you simplify the testcase to its bare minimum, and perhaps send an *un*preprocessed source (with header files of course).
Hello, would it be possible for you to check whether this problem still occurs under gcc mainline? Aldy's done a lot of work fixing Altivec bugs, so it'd be good to know whether this was one of them. Thanks, Dara
Actually I think this bug is still there because gcc is expanding many macros, a workaround is to use tempary variables, this will also speed up the compile. Also send the *un*preprocessed source file.
Subject: RE: Compiler Segmentation Fault cross-compiling PPC Altivec accelerated code I will double check with the new line. I am not sure what temporary variables you mean? Could you enlighten me? > -----Original Message----- > From: pinskia@physics.uc.edu [mailto:gcc-bugzilla@gcc.gnu.org] > Sent: Sunday, June 01, 2003 11:57 PM > To: vtaylor@dna-cs.com > Subject: [Bug c/8703] Compiler Segmentation Fault cross-compiling PPC > Altivec accelerated code > > > PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org. > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=8703 > > > > ------- Additional Comments From pinskia@physics.uc.edu > 2003-06-02 04:57 ------- > Actually I think this bug is still there because gcc is expanding > many macros, a > workaround is to use tempary variables, this will also speed up > the compile. > > Also send the *un*preprocessed source file. > > > > ------- You are receiving this mail because: ------- > You reported the bug, or are watching the reporter. >
instead of vec_stx(vec_add(vec_ld(...),vec_ld(...)),..) do vector int temp = vec_ld(...); vector int temp1 = vec_ld(...); vector int temp2 = vec_add(temp, temp1); vec_stx(temp,..);
We need the non preprocessed source file. It will be called something.c.
No feed back in 3 months (T-8 days) plus this really has been in waiting for the unpreprocessed source file for over 6 months.