ia64 scheduling

Brad Lucier lucier@math.purdue.edu
Tue Dec 26 12:11:00 GMT 2000


Bernd:

Thanks for your reply.

It's clear that I'm not the person to do this.  On alphaev6-unknown-linux-gnu,
after changing schedule_insns to

      timevar_push (TV_SCHED2);
      open_dump_file (DFI_sched2, decl);

      /* Do control and data sched analysis again,
         and write some more of the results to dump file.  */

      schedule_ebbs (rtl_dump_file);

      close_dump_file (DFI_sched2, print_rtl_with_bb, insns);
      timevar_pop (TV_SCHED2);

          timevar_push (TV_FLOW);
          find_basic_blocks (insns, max_reg_num (), rtl_dump_file);
          timevar_pop (TV_FLOW);

I get with with --enable-checking=yes even before stage1 completes:

./xgcc -B./ -B/export/u10/egcs-test2/alphaev6-unknown-linux-gnu/bin/ -isystem /export/u10/egcs-test2/alphaev6-unknown-linux-gnu/include -O2   -DIN_GCC    -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -isystem ./include  -fPIC -g1  -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED  -I. -I. -I../../gcc -I../../gcc/. -I../../gcc/config -I../../gcc/../include  -DL_eprintf -c ../../gcc/libgcc2.c -o libgcc/./_eprintf.o
../../gcc/libgcc2.c: In function `__dummy':
../../gcc/libgcc2.c:1245: Return not followed by barrier
(jump_insn 14 9 16 (return) 183 {*return_internal} (nil)
    (nil))
../../gcc/libgcc2.c:1245: Internal compiler error in verify_flow_info, at flow.c:6875

and I'm stuck.

So I'll wait until someone else does this.

Brad

> 
> On Sat, 23 Dec 2000, Brad Lucier wrote:
> 
> > Bernd:
> >
> > Will the changes you're adding to allow scheduling over extended
> > basic blocks apply easily to architectures other then ia64, or will
> > extensive architecture-specific changes be needed to use them
> > on alpha or ia32?
> 
> In theory you can replace the second call to schedule_insns with a
> call to schedule_ebbs in rest_of_compilation, and it should work.
> You may need to add things like a call to find_basic_blocks afterwards,
> since it destroys the CFG.
> If you can get it to work I'd love to see whether it makes a difference.
> 
> 
> Bernd
> 



More information about the Gcc-patches mailing list