ekrell@ulysses.att.com (02/04/89)
On a sun4os4 configuration, if you change CFLAGS in the Makefile to -O instead of -g, cc1 coredumps in the last stage (when you run make CC=stage2/gcc CFLAGS="-O -Bstage2/") on the file toplev.c. Without the -O it bootstraps ok, so I guess this is a bug in the optimizer.
rms@WHEATIES.AI.MIT.EDU (Richard Stallman) (02/04/89)
On a sun4os4 configuration, if you change CFLAGS in the Makefile to -O
instead of -g, cc1 coredumps in the last stage (when you run
make CC=stage2/gcc CFLAGS="-O -Bstage2/") on the file toplev.c.
Either I don't understand you, or you have described a change which
has no effect. Specifying -O twice is the same as specifying it once.kcwellsch@watmath.waterloo.edu (Ken Wellsch) (02/04/89)
> On a sun4os4 configuration, if you change CFLAGS in the Makefile to -O > instead of -g, cc1 coredumps in the last stage (when you run > make CC=stage2/gcc CFLAGS="-O -Bstage2/") on the file toplev.c. I expect you are changing the "-g" to a "-O" on the first "make" compilation step in compiling gcc which uses the native "cc". Don't. Another software person here discovered that it wasn't very hard to find examples of C code that when compiled using the stock "cc" on sun4-os4 WITH optimization, it produce WRONG code. We were really impressed with the code generated with gcc-1.32. BTW: it was not wrong on the samples he found failed with "cc -O".
ekrell@ulysses.att.com (02/04/89)
Let me explain myself. The Makefile distributed with gcc 1.33 uses CFLAGS=-g. I changed this line to CFLAGS=-O. Then I run: make make stage1 make CC=stage1/gcc CFLAGS="-O -Bstage1/" rm -fr stage1 make stage2 make CC=stage2/gcc CFLAGS="-O -Bstage2/" at this point, I get a core dump from stage2/cc1 when compiling "toplev.c". If I repeat this process without the -O, ie, CFLAGS= in the Makefile, and make make stage1 make CC=stage1/gcc CFLAGS=-Bstage1/ rm -fr stage1 make stage2 make CC=stage2/gcc CFLAGS=-Bstage2/ Then I don't get any core dump and the *.o files compare ok with the ones in stage2/*.o My interpretation of this is that the "stage1/gcc -O" phase generates bad code which makes stage2/cc1 core dump on toplev.c.