[PATCH] Add capability to run several iterations of early optimizations

[Maxim Kuvyrkov <maxim@codesourcery.com>]

[-- Attachment #1: Type: text/plain, Size: 1483 bytes --]

The following patch adds new knob to make GCC perform several iterations of early optimizations and inlining.

This is for dont-care-about-compile-time-optimize-all-you-can scenarios.  Performing several iterations of optimizations does significantly improve code speed on a certain proprietary source base.  Some hand-tuning of the parameter value is required to get optimum performance.  Another good use for this option is for search and ad-hoc analysis of cases where GCC misses optimization opportunities.

With the default setting of '1', nothing is changed from the current status quo.

The patch was bootstrapped and regtested with 3 iterations set by default on i686-linux-gnu.  The only failures in regression testsuite were due to latent bugs in handling of EH information, which are being discussed in a different thread.

Performance impact on the standard benchmarks is not conclusive, there are improvements in SPEC2000 of up to 4% and regressions down to -2%, see [*].  SPEC2006 benchmarks will take another day or two to complete and I will update the spreadsheet then.  The benchmarks were run on a Core2 system for all combinations of {-m32/-m64}{-O2/-O3}.

Effect on compilation time is fairly predictable, about 10% compile time increase with 3 iterations.

OK for trunk?

[*] https://docs.google.com/spreadsheet/ccc?key=0AvK0Y-Pgj7bNdFBQMEJ6d3laeFdvdk9lQ1p0LUFkVFE&hl=en_US

Thank you,

--
Maxim Kuvyrkov
CodeSourcery / Mentor Graphics



[-- Attachment #2: fsf-gcc-iter-eipa.ChangeLog --]
[-- Type: application/octet-stream, Size: 429 bytes --]

2011-10-11  Maxim Kuvyrkov  <maxim@codesourcery.com>

	Add scheduling of several iterations of early IPA passes.

	* doc/invoke.texi (eipa-iterations): Document new parameter.
	* params.def (PARAM_EIPA_ITERATIONS): Define.
	* passes.c (init_optimization_passes): Schedule several iterations
	of pass_all_early_optimizations.
	* toplev.c (general_init, toplev_main): Move init_optimizations_passes
	after processing of arguments.

[-- Attachment #3: fsf-gcc-iter-eipa.patch --]
[-- Type: application/octet-stream, Size: 4741 bytes --]

commit a54b6cb0160d5413499e08a87da3a35eb6e29652
Author: Maxim Kuvyrkov <maxim@codesourcery.com>
Date:   Sat Oct 1 01:09:50 2011 -0700

    Add iterative optimization passes

diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index c92c6b2..1ebaf35 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -9060,6 +9060,12 @@ the parameter is reserved exclusively for debug insns created by
 @option{-fvar-tracking-assignments}, but debug insns may get
 (non-overlapping) uids above it if the reserved range is exhausted.
 
+@item eipa-iterations
+The pass scheduler will execute @option{eipa-iterations} iterations of
+early optimization passes before running interprocedural analysis.
+Running several iterations of optimization passes allows the compiler
+to provide thoroughly optimized code to the interprocedural analysis.
+
 @item ipa-sra-ptr-growth-factor
 IPA-SRA will replace a pointer to an aggregate with one or more new
 parameters only when their cumulative size is less or equal to
diff --git a/gcc/params.def b/gcc/params.def
index 5e49c48..9609919 100644
--- a/gcc/params.def
+++ b/gcc/params.def
@@ -861,6 +861,11 @@ DEFPARAM (PARAM_MIN_NONDEBUG_INSN_UID,
 	  "The minimum UID to be used for a nondebug insn",
 	  0, 1, 0)
 
+DEFPARAM (PARAM_EIPA_ITERATIONS,
+	  "eipa-iterations",
+	  "Number of iterations of early optimizations before IPA analysis",
+	  1, 1, 0)
+
 DEFPARAM (PARAM_IPA_SRA_PTR_GROWTH_FACTOR,
 	  "ipa-sra-ptr-growth-factor",
 	  "Maximum allowed growth of size of new parameters ipa-sra replaces "
diff --git a/gcc/passes.c b/gcc/passes.c
index 887007f..dcc3297 100644
--- a/gcc/passes.c
+++ b/gcc/passes.c
@@ -1191,6 +1191,7 @@ init_optimization_passes (void)
   NEXT_PASS (pass_early_local_passes);
     {
       struct opt_pass **p = &pass_early_local_passes.pass.sub;
+      int i;
       NEXT_PASS (pass_fixup_cfg);
       NEXT_PASS (pass_init_datastructures);
       NEXT_PASS (pass_expand_omp);
@@ -1199,12 +1200,29 @@ init_optimization_passes (void)
       NEXT_PASS (pass_build_ssa);
       NEXT_PASS (pass_lower_vector);
       NEXT_PASS (pass_early_warn_uninitialized);
-      NEXT_PASS (pass_rebuild_cgraph_edges);
-      NEXT_PASS (pass_inline_parameters);
-      NEXT_PASS (pass_early_inline);
-      NEXT_PASS (pass_all_early_optimizations);
+
+      /* Schedule PARAM_EIPA_ITERATIONS iterations of early optimizations.
+	 We run these passes at least once as pass_all_early_optimizations.  */
+      gcc_assert (PARAM_VALUE (PARAM_EIPA_ITERATIONS) >= 1);
+      for (i = 0; i < PARAM_VALUE (PARAM_EIPA_ITERATIONS); ++i)
 	{
-	  struct opt_pass **p = &pass_all_early_optimizations.pass.sub;
+	  struct opt_pass **q;
+
+	  NEXT_PASS (pass_rebuild_cgraph_edges);
+	  NEXT_PASS (pass_inline_parameters);
+	  NEXT_PASS (pass_early_inline);
+
+	  {
+	    struct opt_pass **r;
+	    r = p;
+	    NEXT_PASS (pass_all_early_optimizations);
+	    /* (*R) is the pointer to the newly allocated instance of
+	       pass_all_early_optimizations.  */
+	    q = &(*r)->next;
+	    /* Push to sub-pass.  */
+	    p = &(*r)->sub;
+	  }
+
 	  NEXT_PASS (pass_remove_cgraph_callee_edges);
 	  NEXT_PASS (pass_rename_ssa_copies);
 	  NEXT_PASS (pass_ccp);
@@ -1222,13 +1240,19 @@ init_optimization_passes (void)
 	  NEXT_PASS (pass_early_ipa_sra);
 	  NEXT_PASS (pass_tail_recursion);
 	  NEXT_PASS (pass_convert_switch);
-          NEXT_PASS (pass_cleanup_eh);
-          NEXT_PASS (pass_profile);
-          NEXT_PASS (pass_local_pure_const);
+	  NEXT_PASS (pass_cleanup_eh);
+	  if (i == PARAM_VALUE (PARAM_EIPA_ITERATIONS) - 1)
+	    /* Schedule pass_profile only for the last iteration.
+	       This pass assumes that it is run only once.  */
+	    NEXT_PASS (pass_profile);
+	  NEXT_PASS (pass_local_pure_const);
 	  /* Split functions creates parts that are not run through
 	     early optimizations again.  It is thus good idea to do this
 	     late.  */
-          NEXT_PASS (pass_split_functions);
+	  NEXT_PASS (pass_split_functions);
+
+	  /* Pop from sub-pass.  */
+	  p = q;
 	}
       NEXT_PASS (pass_release_ssa_names);
       NEXT_PASS (pass_rebuild_cgraph_edges);
diff --git a/gcc/toplev.c b/gcc/toplev.c
index ab6b5a4..a0ad17d 100644
--- a/gcc/toplev.c
+++ b/gcc/toplev.c
@@ -1224,7 +1224,6 @@ general_init (const char *argv0)
   /* This must be done after global_init_params but before argument
      processing.  */
   init_ggc_heuristics();
-  init_optimization_passes ();
   statistics_early_init ();
   finish_params ();
 }
@@ -1984,6 +1983,8 @@ toplev_main (int argc, char **argv)
 		  save_decoded_options, save_decoded_options_count,
 		  UNKNOWN_LOCATION, global_dc);
 
+  init_optimization_passes ();
+
   handle_common_deferred_options ();
 
   init_local_tick ();