-fprofile-arcs and -fbranch-probabilities

["G. Sumner Hayes" <zumner@collegium.adsl.net.cmu.edu>]

Are -fprofile-arcs and -fbranch-probabilities supposed to be stable
in egcs-1.0, or are they still under development?

Compiling with -fprofile-arcs works fine and running the resulting
executable generates many .da files; trying to compile the results
with -fbranch-probabilities gives mixed results.  Some files compile
fine.  Some have warnings (shown below).  Some cause cc1 to get fatal
signal 6 and die (shown below).  If this is expected, fine; if not,
any ideas what's causing it?

Warnings are often of this form:
../bbox.c: In function `Destroy_Bounding_Slabs':
../bbox.c:1909: warning: Arc profiling: some arc counts were bad.
../bbox.c: In function `Destroy_BBox_Tree':
../bbox.c:1909: warning: Arc profiling: some arc counts were bad.
../bbox.c: In function `Create_Rayinfo':
../bbox.c:1909: warning: Arc profiling: some arc counts were bad.
../bbox.c: In function `create_bbox_node':
../bbox.c:1909: warning: Arc profiling: some arc counts were bad.
../bbox.c: In function `calc_bbox':
../bbox.c:1909: warning: Arc profiling: some arc counts were bad.
../bbox.c: At top level:
../bbox.c:1909: warning: .da file contents not exhausted

Some files die with errors like the following:
../boxes.c: In function `Intersect_Box':
../boxes.c:528: internal error--unrecognizable insn:
(jump_insn/s 92 91 94 (set (pc)
        (if_then_else (le (cc0)
                (const_int 10000))
            (label_ref 157)
            (pc))) -1 (insn_list:REG_DEP_ANTI 82 (insn_list:REG_DEP_ANTI 84 (insn_list 89 (insn_list 88 (nil)))))
    (expr_list:REG_BR_PROB (const_int 5809)
        (nil)))
gcc: Internal compiler error: program cc1 got fatal signal 6


Thanks for your time,

  Sumner

-- 
rage, rage against the dying of the light