profiledbootstrap broken on powerpc-apple-darwin
Mike Stump
mrs@apple.com
Wed Sep 8 23:31:00 GMT 2004
On Sep 8, 2004, at 2:26 PM, Caroline Tice wrote:
> I'm getting the following failures doing a profiled bootstrap using
> FSF mainline on a G4 running powerpc-apple-darwin. Normal bootstrap
> works fine, and profiled bootstrap appears to work on x86 running
> Linux. I believe this has been broken since last Friday. Does
> anybody have any ideas as to who/what broke this?
> stage1/xgcc -Bstage1/ -B/usr/local/powerpc-apple-darwin7.0.0/bin/ -g
> -O2 -mdynamic-no-pic -fprofile-generate -DIN_GCC -W -Wall
> -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic
> -Wno-long-long -Wno-variadic-macros -Wold-style-definition -Werror
> -fno-common -DHAVE_CONFIG_H -DGENERATOR_FILE -o gengtype \
> gengtype.o gengtype-lex.o gengtype-yacc.o
> ../build-powerpc-apple-darwin7.0.0/libiberty/libiberty.a
> ./gengtype
> make[2]: *** [s-gtype] Bus error
Ah! Cool, someone else is seeing this... I was seeing it with a patch
that should not have failed to bootstrap. I wonder if someone killed
binary compatibility between stage1/stage[^1].... :-(
*** config/darwin.h.~1~ Wed Dec 10 17:22:30 2003
--- config/darwin.h Thu Jan 22 18:03:20 2004
*************** objc_section_init (void) \
*** 712,717 ****
--- 712,729 ----
#define GLOBAL_ASM_OP ".globl "
#define TARGET_ASM_GLOBALIZE_LABEL darwin_globalize_label
+ #define ASM_OUTPUT_ALIGNED_BSS(FILE, DECL, NAME, SIZE, ALIGNMENT)
\
+ do {
\
+ unsigned int localsize = SIZE;
\
+ if (localsize == 0)
\
+ ++localsize;
\
+ if (FILE)
\
+ fprintf (FILE, "\t.zerofill __DATA,__bss,");
\
+ assemble_name (FILE, NAME);
\
+ fprintf (FILE, ",%u,%d\n", localsize,
\
+ floor_log2 (ALIGNMENT / BITS_PER_UNIT));
\
+ } while (0)
+
/* Emit an assembler directive to set visibility for a symbol. Used
to support visibility attribute and Darwin's private extern
feature. */
--------------
is the patch, just in case someone else wants a simpler testcase... In
time, I'll have to tack this down if I want my patch in... While I
think this may be the same problem, it might not be...
More information about the Gcc-patches
mailing list