bogus asm on alpha compiling big bison output

[Jeffrey A Law]

  In message <199802280357.WAA18457@xkey.com>you write:
  > egcs-1.0.1, Redhat 5.0, built it myself with no funny options, LX164
  > 
  > I have a rather large bison parser which works OK on our other systems.
  > On RH5.0/alpha, I get:
  > 
  > [lindahl@centurion001 mplc]$ g++-egcs
  > -I/home/localtmp/lindahl/Legion/src/Compiler/mplc
  > -I/home/localtmp/lindahl/Legion/obj/alpha_linux/g++
  > -I/home/localtmp/lindahl/Legion/include -w -fpic -DGNU -Dalpha_linux
  > -I/home/localtmp/lindahl/Legion/include/mplc/local/alpha_linux/gnu/g++
  > -g -c /home/localtmp/lindahl/Legion/obj/alpha_linux/g++/parser.tab.c
  > -o /home/localtmp/lindahl/Legion/obj/alpha_linux/g++/parser.o
  > /usr/lib/bison.simple: In function `int yyparse()':
  > /usr/lib/bison.simple:692: internal error--unrecognizable insn:
  > (insn 112872 112869 112873 (set (reg:DI 5 $5)
  >         (plus:DI (reg:DI 15 $15)
  >             (const_int 45248))) -1 (nil)
  >     (nil))
  > /usr/lib/bison.simple:692: confused by earlier errors, bailing out
  > 
  > If I turn on -O, the optimizer is not bounded in time and space
  > (cough), I killed it when it hit 100 megabytes of swap. This input
file is only 14,000 lines.
  > 
  > I have a much smaller grammar that compiles successfully.
  > 
  > Do you have any suggestions for reducing this to a bug report you
  > can actually do something with?
This bug was fixed in either egcs-1.0.2 or egcs-1.0.3 (the internal error).
We also made some changes to reduce memory consumption, but I don't know
if they would effect your testcase or not.

jeff