* Global CSE and constant/copy propagation.

* Makefile.in (OBJS): Add gcse.o
        (STAGESTUFF): Add *.gcse.
        (gcse.o): Add dependencies.
        (mostlyclean): Remove *.gcse and */*.gcse.
        * gcse.c: New file.
        * loop.c (loop_optimize): Move call to init_alias_analysis.
        * recog.c (validate_replace_src): New function.
        * toplev.c (gcse_dump): New global variable.
        (flag_gcse, gcse_time): Likewise.
        (compile_file): Initialize gcse_time and clean out the gcse dump
        file if necessary.
        (rest_of_compilation): Call gcse_main as requested.  Dump RTL
        after gcse if requested.
        (main): Enable gcse for -O2 and above.  Handle -dG.  Enable gcse
        dumps for -da.
        * gcc.texi: Add gcse related internal documentation.
        * invoke.texi: Note new command line options for gcse.
        * tm.texi: Document AVOID_CCMODE_COPIES.
        * mips.h (AVOID_CCMODE_COPIES): Define.

Co-Authored-By: Jeffrey A Law <law@cygnus.com>

From-SVN: r19901

gcc/ChangeLog

@@ -1,3 +1,27 @@
+Wed May 20 01:11:02 1998  Doug Evans (devans@cygnus.com)
+			  Jeff Law (law@cygnus.com)
+
+	* Global CSE and constant/copy propagation.
+	* Makefile.in (OBJS): Add gcse.o
+	(STAGESTUFF): Add *.gcse.
+	(gcse.o): Add dependencies.
+	(mostlyclean): Remove *.gcse and */*.gcse.
+	* gcse.c: New file.
+	* loop.c (loop_optimize): Move call to init_alias_analysis.
+	* recog.c (validate_replace_src): New function.
+	* toplev.c (gcse_dump): New global variable.
+	(flag_gcse, gcse_time): Likewise.
+	(compile_file): Initialize gcse_time and clean out the gcse dump
+	file if necessary.
+	(rest_of_compilation): Call gcse_main as requested.  Dump RTL
+	after gcse if requested.
+	(main): Enable gcse for -O2 and above.  Handle -dG.  Enable gcse
+	dumps for -da.
+	* gcc.texi: Add gcse related internal documentation.
+	* invoke.texi: Note new command line options for gcse.
+	* tm.texi: Document AVOID_CCMODE_COPIES.
+	* mips.h (AVOID_CCMODE_COPIES): Define.
+
 Tue May 19 22:31:20 1998  Jeffrey A Law  (law@cygnus.com)
 
 	* Makefile.in (deduced.h): Only run scan-types if $(SYSTEM_HEADER_DIR)

gcc/Makefile.in

@@ -652,7 +652,7 @@ OBJS = toplev.o version.o tree.o print-tree.o stor-layout.o fold-const.o \
  varasm.o rtl.o print-rtl.o rtlanal.o emit-rtl.o genrtl.o real.o regmove.o \
  dbxout.o sdbout.o dwarfout.o dwarf2out.o xcoffout.o bitmap.o alias.o \
  integrate.o jump.o cse.o loop.o unroll.o flow.o stupid.o combine.o \
- regclass.o local-alloc.o global.o reload.o reload1.o caller-save.o \
+ regclass.o local-alloc.o global.o reload.o reload1.o caller-save.o gcse.o \
  insn-peep.o reorg.o $(SCHED_PREFIX)sched.o final.o recog.o reg-stack.o \
  insn-opinit.o insn-recog.o insn-extract.o insn-output.o insn-emit.o \
  profile.o insn-attrtab.o $(out_object_file) getpwd.o $(EXTRA_OBJS) convert.o
@@ -685,7 +685,7 @@ STAGESTUFF = *$(objext) insn-flags.h insn-config.h insn-codes.h \
  specs collect2$(exeext) $(USE_COLLECT2) underscore.c \
  gcov$(exeext) *.bp \
  *.greg *.lreg *.combine *.flow *.cse *.jump *.rtl *.tree *.loop \
- *.dbr *.jump2 *.sched *.cse2 *.sched2 *.stack \
+ *.dbr *.jump2 *.sched *.cse2 *.sched2 *.stack *.gcse \
  *.[si] \
  $(LANG_STAGESTUFF)
 
@@ -1449,6 +1449,8 @@ stupid.o : stupid.c $(CONFIG_H) system.h $(RTL_H) regs.h hard-reg-set.h flags.h
 
 cse.o : cse.c $(CONFIG_H) system.h $(RTL_H) regs.h hard-reg-set.h flags.h \
    real.h insn-config.h insn-codes.h $(RECOG_H) expr.h
+gcse.o : gcse.c $(CONFIG_H) system.h $(RTL_H) regs.h hard-reg-set.h flags.h \
+   real.h insn-config.h insn-codes.h $(RECOG_H) expr.h basic-block.h
 profile.o : profile.c $(CONFIG_H) system.h $(RTL_H) flags.h insn-flags.h \
    gcov-io.h $(TREE_H) output.h regs.h toplev.h
 loop.o : loop.c $(CONFIG_H) system.h $(RTL_H) flags.h loop.h insn-config.h \
@@ -2159,10 +2161,11 @@ mostlyclean: lang.mostlyclean
 	-rm -f */stamp-* */tmp-*
 # Delete debugging dump files.
 	-rm -f *.greg *.lreg *.combine *.flow *.cse *.jump *.rtl *.tree *.loop
-	-rm -f *.dbr *.jump2 *.sched *.cse2 *.sched2 *.stack *.addressof *.regmove *.mach *.bp
+	-rm -f *.dbr *.jump2 *.sched *.cse2 *.sched2 *.stack *.addressof
+	-rm -f *.regmove *.mach *.bp *.gcse
 	-rm -f */*.greg */*.lreg */*.combine */*.flow */*.cse */*.jump */*.rtl
 	-rm -f */*.tree */*.loop */*.dbr */*.jump2 */*.sched */*.cse2
-	-rm -f */*.sched2 */*.stack */*.regmove
+	-rm -f */*.sched2 */*.stack */*.regmove */*.gcse
 # Delete some files made during installation.
 	-rm -f specs gfloat.h float.h-* enquire SYSCALLS.c.X SYSCALLS.c
 	-rm -f collect collect2 mips-tfile mips-tdump alloca.s

gcc/config/mips/mips.h

@@ -3561,6 +3561,13 @@ while (0)
   (((mips_cpu == PROCESSOR_R4000 || mips_cpu == PROCESSOR_R6000) ? 6 : 4) \
    + memory_move_secondary_cost ((MODE), (CLASS), (TO_P)))
 
+/* Define if copies to/from condition code registers should be avoided.
+
+   This is needed for the MIPS because reload_outcc is not complete;
+   it needs to handle cases where the source is a general or another
+   condition code register.  */
+#define AVOID_CCMODE_COPIES
+
 /* A C expression for the cost of a branch instruction.  A value of
    1 is the default; other values are interpreted relative to that.  */
 

gcc/gcc.texi

@@ -3302,6 +3302,22 @@ The option @samp{-ds} causes a debugging dump of the RTL code after
 this pass.  This dump file's name is made by appending @samp{.cse} to
 the input file name.
 
+@cindex global common subexpression elimination
+@cindex constant propagation
+@cindex copy propagation
+@item               
+Global common subexpression elimination.  This pass performs GCSE
+using Morel-Renvoise Partial Redundancy Elimination, with the exception
+that it does not try to move invariants out of loops - that is left to
+the loop optimization pass.  This pass also performs global constant
+and copy propagation.
+
+The source file for this pass is gcse.c.
+
+The option @samp{-dG} causes a debugging dump of the RTL code after
+this pass.  This dump file's name is made by appending @samp{.gcse} to
+the input file name.
+
 @cindex loop optimization
 @cindex code motion
 @cindex strength-reduction

gcc/invoke.texi

@@ -149,7 +149,7 @@ in the following sections.
 -fcaller-saves  -fcse-follow-jumps  -fcse-skip-blocks
 -fdelayed-branch   -fexpensive-optimizations
 -ffast-math  -ffloat-store  -fforce-addr  -fforce-mem
--ffunction-sections  -finline-functions
+-ffunction-sections  -fgcse -finline-functions
 -fkeep-inline-functions  -fno-default-inline
 -fno-defer-pop  -fno-function-cse
 -fno-inline  -fno-peephole  -fomit-frame-pointer -fregmove
@@ -1986,6 +1986,8 @@ Dump after purging ADDRESSOF, to @file{@var{file}.addressof}.
 Dump after flow analysis, to @file{@var{file}.flow}.
 @item g
 Dump after global register allocation, to @file{@var{file}.greg}.
+@item G      
+Dump after GCSE, to @file{@var{file}.gcse}.
 @item j
 Dump after first jump optimization, to @file{@var{file}.jump}.
 @item J
@@ -2299,6 +2301,10 @@ performed.
 @item -frerun-loop-opt
 Run the loop optimizer twice.
 
+@item -fgcse
+Perform a global common subexpression elimination pass.
+This pass also performs global constant and copy propagation.
+
 @item -fexpensive-optimizations
 Perform a number of minor optimizations that are relatively expensive.
 

gcc/loop.c

@@ -400,7 +400,6 @@ loop_optimize (f, dumpfile, unroll_p)
   loop_dump_stream = dumpfile;
 
   init_recog_no_volatile ();
-  init_alias_analysis ();
 
   max_reg_before_loop = max_reg_num ();
 
@@ -477,6 +476,13 @@ loop_optimize (f, dumpfile, unroll_p)
      function.  */
   reg_scan (f, max_reg_num (), 1);
 
+  /* This must occur after reg_scan so that registers created by gcse
+     will have entries in the register tables.
+
+     We could have added a call to reg_scan after gcse_main in toplev.c,
+     but moving this call to init_alias_analysis is more efficient.  */
+  init_alias_analysis ();
+
   /* See if we went too far.  */
   if (get_max_uid () > max_uid_for_loop)
     abort ();

gcc/recog.c

@@ -537,6 +537,25 @@ validate_replace_rtx (from, to, insn)
   validate_replace_rtx_1 (&PATTERN (insn), from, to, insn);
   return apply_change_group ();
 }
+
+/* Try replacing every occurrence of FROM in INSN with TO, avoiding
+   SET_DESTs.  After all changes have been made, validate by seeing if
+   INSN is still valid.  */
+
+int
+validate_replace_src (from, to, insn)
+     rtx from, to, insn;
+{
+  if ((GET_CODE (insn) != INSN && GET_CODE (insn) != JUMP_INSN)
+      || GET_CODE (PATTERN (insn)) != SET)
+    abort ();
+
+  validate_replace_rtx_1 (&SET_SRC (PATTERN (insn)), from, to, insn);
+  if (GET_CODE (SET_DEST (PATTERN (insn))) == MEM)
+    validate_replace_rtx_1 (&XEXP (SET_DEST (PATTERN (insn)), 0),
+			    from, to, insn);
+  return apply_change_group ();
+}
 
 #ifdef HAVE_cc0
 /* Return 1 if the insn using CC0 set by INSN does not contain

gcc/tm.texi

@@ -1505,6 +1505,12 @@ accessibility of the value in a narrower mode.
 You should define this macro to return nonzero in as many cases as
 possible since doing so will allow GNU CC to perform better register
 allocation.
+
+@findex AVOID_CCMODE_COPIES
+@item AVOID_CCMODE_COPIES
+Define this macro if the compiler should avoid copies to/from @code{CCmode}
+registers.  You should only define this macro if support fo copying to/from
+@code{CCmode} is incomplete.
 @end table
 
 @node Leaf Functions

gcc/toplev.c

@@ -267,6 +267,7 @@ int rtl_dump_and_exit = 0;
 int jump_opt_dump = 0;
 int addressof_dump = 0;
 int cse_dump = 0;
+int gcse_dump = 0;
 int loop_dump = 0;
 int cse2_dump = 0;
 int branch_prob_dump = 0;
@@ -538,6 +539,10 @@ int flag_volatile_global;
 
 int flag_syntax_only = 0;
 
+/* Nonzero means perform global cse.  */
+
+static int flag_gcse;
+
 /* Nonzero means to rerun cse after loop optimization.  This increases
    compilation time about 20% and picks up a few more common expressions.  */
 
@@ -738,6 +743,7 @@ struct { char *string; int *variable; int on_value;} f_options[] =
   {"pcc-struct-return", &flag_pcc_struct_return, 1},
   {"reg-struct-return", &flag_pcc_struct_return, 0},
   {"delayed-branch", &flag_delayed_branch, 1},
+  {"gcse", &flag_gcse, 1},
   {"rerun-cse-after-loop", &flag_rerun_cse_after_loop, 1},
   {"rerun-loop-opt", &flag_rerun_loop_opt, 1},
   {"pretend-float", &flag_pretend_float, 1},
@@ -977,6 +983,7 @@ int varconst_time;
 int integration_time;
 int jump_time;
 int cse_time;
+int gcse_time;
 int loop_time;
 int cse2_time;
 int branch_prob_time;
@@ -2274,6 +2281,7 @@ compile_file (name)
   integration_time = 0;
   jump_time = 0;
   cse_time = 0;
+  gcse_time = 0;
   loop_time = 0;
   cse2_time = 0;
   branch_prob_time = 0;
@@ -2360,6 +2368,8 @@ compile_file (name)
   if (dbr_sched_dump)
     clean_dump_file (".dbr");
 #endif
+  if (gcse_dump)
+    clean_dump_file (".gcse");
 #ifdef STACK_REGS
   if (stack_reg_dump)
     clean_dump_file (".stack");
@@ -2840,6 +2850,7 @@ compile_file (name)
       print_time ("integration", integration_time);
       print_time ("jump", jump_time);
       print_time ("cse", cse_time);
+      print_time ("gcse", gcse_time);
       print_time ("loop", loop_time);
       print_time ("cse2", cse2_time);
       print_time ("branch-prob", branch_prob_time);
@@ -3257,6 +3268,18 @@ rest_of_compilation (decl)
   if (addressof_dump)
     dump_rtl (".addressof", decl, print_rtl, insns);
   
+  /* Perform global cse.  */
+
+  if (optimize > 0 && flag_gcse)
+    {
+      if (gcse_dump)
+	open_dump_file (".gcse", IDENTIFIER_POINTER (DECL_NAME (decl)));
+      
+      TIMEVAR (gcse_time, gcse_main (insns, rtl_dump_file));
+
+      if (gcse_dump)
+	close_dump_file (print_rtl, insns);
+    }
   /* Move constant computations out of loops.  */
 
   if (optimize > 0)
@@ -3801,6 +3824,7 @@ main (argc, argv, envp)
     {
       flag_cse_follow_jumps = 1;
       flag_cse_skip_blocks = 1;
+      flag_gcse = 1;
       flag_expensive_optimizations = 1;
       flag_strength_reduce = 1;
       flag_rerun_cse_after_loop = 1;
@@ -3879,6 +3903,7 @@ main (argc, argv, envp)
 		    regmove_dump = 1;
  		    rtl_dump = 1;
  		    cse_dump = 1, cse2_dump = 1;
+		    gcse_dump = 1;
  		    sched_dump = 1;
  		    sched2_dump = 1;
 #ifdef STACK_REGS
@@ -3911,6 +3936,9 @@ main (argc, argv, envp)
 		  case 'g':
 		    global_reg_dump = 1;
 		    break;
+		  case 'G':
+		    gcse_dump = 1;
+		    break;
 		  case 'j':
 		    jump_opt_dump = 1;
 		    break;