https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #10 from Oleg Endo ---
(In reply to Rich Felker from comment #9)
> I think it's actually just a matter of removing the patterns for generating
> bsrf, but I may be mistaken. Generating jsr should be what happens "by
> default" in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #9 from Rich Felker ---
I think it's actually just a matter of removing the patterns for generating
bsrf, but I may be mistaken. Generating jsr should be what happens "by default"
in some sense if GCC just has to load the address,
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #8 from Oleg Endo ---
Converting the FDPIC from bsrf back to jsr sounds like quite some work.
However, I think chances of success are higher of it does the same thing as the
normal PIC code.
Do you know what the main reason was to
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #7 from Rich Felker ---
I think all the FDPIC work done to use bsrf like this was probably a mistake.
It ends up greatly enlarging functions that make a lot of such calls, for
example soft-float that does it for each floating point
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
Oleg Endo changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Last reconfirmed|
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #5 from Zach van Rijn ---
Created attachment 45546
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45546=edit
All files produced by -O2 -da
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #4 from Zach van Rijn ---
The error can be reproduced at `O1` optimization level with both
(strictly both) of the following options:
./cc -c mintest.c -O1 -freorder-blocks-algorithm=stc -ftree-pre
Changing to
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #3 from Zach van Rijn ---
Created attachment 45545
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45545=edit
Tarball containing intermediate asm (with -dp) for each of 5 cases.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #2 from Oleg Endo ---
You can compile the code with the '-dp' option to see which insn patterns make
up the asm code. The pattern names will be emitted as comments in the asm
output.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89012
--- Comment #1 from Rich Felker ---
Binutils version should not be relevant; the bug here is before anything even
gets to binutils. It looks like one of the RTL patterns used for calling libgcc
bitshift functions from FDPIC was inteded to be
10 matches
Mail list logo
