Add parameters to control histogram profiling and peeling

5 files changed, 28 insertions, 5 deletions

diff --git a/gcc/common.opt b/gcc/common.opt
index 5b1633cb5ac..ff44d84bcd5 100644
--- a/gcc/common.opt
+++ b/gcc/common.opt
@@ -2293,7 +2293,11 @@ Perform loop peeling.
 
 fuse-histograms-in-peeling
 Common Var(flag_use_histogram_in_peeling) Init(1) Optimization
-Use histograms in loop peelings.
+Use histograms in loop peeling heuristics.
+
+fpeel-loops-without-histogram
+Common Var(flag_peel_loops_without_histogram) Init(1) Optimization
+Peel also loop with no loop histogram.
 
 fpeephole
 Common Var(flag_no_peephole,0) Optimization
@@ -2392,6 +2396,10 @@ fprofile-reproducible=
 Common Joined RejectNegative Var(flag_profile_reproducible) Enum(profile_reproducibility) Init(PROFILE_REPRODUCIBILITY_SERIAL)
 -fprofile-reproducible=[serial|parallel-runs|multithreaded]	Control level of reproducibility of profile gathered by -fprofile-generate.
 
+fprofile-loops
+Common Var(flag_profile_loops) Init(1) Optimization
+Profile loop histograms.
+
 Enum
 Name(profile_update) Type(enum profile_update) UnknownError(unknown profile update method %qs)
 
diff --git a/gcc/gimple-loop-versioning-histograms.cc b/gcc/gimple-loop-versioning-histograms.cc
index 1e39aa66bd5..6abe4d3e9b1 100644
--- a/gcc/gimple-loop-versioning-histograms.cc
+++ b/gcc/gimple-loop-versioning-histograms.cc
@@ -118,7 +118,7 @@ pass_loop_histogram_versioning::execute (function *fn)
 	      best_iters = i;
 	    }
 	}
-      if (best_val <= 0 || best_val < loop->counters->sum * 9 / 10)
+      if (best_val <= 0 || best_val < loop->counters->sum * param_loop_versioning_histogram_prcnt / 100)
 	continue;
       if (dump_file)
 	fprintf (dump_file, "Loop %i has dominating number of iterations %i\n",
diff --git a/gcc/params.opt b/gcc/params.opt
index 72f6a474b4e..9f32d551ace 100644
--- a/gcc/params.opt
+++ b/gcc/params.opt
@@ -426,6 +426,10 @@ The maximum number of instructions in an inner loop that is being considered for
 Common Joined UInteger Var(param_loop_versioning_max_outer_insns) Init(100) Param Optimization
 The maximum number of instructions in an outer loop that is being considered for versioning, on top of the instructions in inner loops.
 
+-param=loop-versioning-histogram_prcnts=
+Common Joined UInteger Var(param_loop_versioning_histogram_prcnt) Init(30) Param Optimization
+Percentage of the dominating iteration count required for versioning.
+
 -param=lra-inheritance-ebb-probability-cutoff=
 Common Joined UInteger Var(param_lra_inheritance_ebb_probability_cutoff) Init(40) IntegerRange(0, 100) Param Optimization
 Minimal fall-through edge probability in percentage used to add BB to inheritance EBB in LRA.
@@ -890,6 +894,10 @@ Size of the linear portion of the histogram counter for profile feedback.
 Common Joined UInteger Var(param_profile_histogram_peel_prcnt) Init(6) Param Optimization
 Peels if it can eliminate at least this percentage per 1 peeling.
 
+-param=profile-histogram-peel-overall-prcnt=
+Common Joined UInteger Var(param_profile_histogram_peel_overall_prcnt) Init(30) Param Optimization
+Peels if it can eliminate at least this percentage of executions of the loop.
+
 -param=profile-func-internal-id=
 Common Joined UInteger Var(param_profile_func_internal_id) IntegerRange(0, 1) Param
 Use internal function id in profile lookup.
diff --git a/gcc/tree-ssa-loop-ivcanon.cc b/gcc/tree-ssa-loop-ivcanon.cc
index f0066d00ee8..5483a4aba9a 100644
--- a/gcc/tree-ssa-loop-ivcanon.cc
+++ b/gcc/tree-ssa-loop-ivcanon.cc
@@ -1122,17 +1122,22 @@ try_peel_loop (class loop *loop,
       gcov_type sum = loop->counters->sum;
       gcov_type rest = sum;
       gcov_type psum = 0;
+      gcov_type osum = 0;
       int good_percentage = param_profile_histogram_peel_prcnt;
+      int good_overall_percentage = param_profile_histogram_peel_overall_prcnt;
       for (int i = 0; i < (int)loop->counters->lin->length (); i++)
 	{
-	  psum += (*(loop->counters->lin))[i];
+	  gcov_type e = (*(loop->counters->lin))[i];
+	  psum += e;
+	  osum += e;
 	  if ((maxiter >= 0 && maxiter <= i)
 	      || (i >= param_max_peel_times - 1)
 	      || rest == 0)
 	    break;
 	  // iteration has enough cumulated in partial sum and itself has at
 	  // least 1 percent or we have complete peeling
-	  if ((100 * psum) / sum >= good_percentage || psum == rest)
+	  if (e && ((100 * psum) / sum >= good_percentage || psum == rest)
+	      && ((100 * osum) / sum) >= good_overall_percentage)
 	    {
 	      prcnt.safe_push (prcnt.last () + (100 * psum) / sum);
 	      good_peels.safe_push (i);
@@ -1152,7 +1157,7 @@ try_peel_loop (class loop *loop,
 	{
 	  if (dump_file)
 	    fprintf (dump_file,
-		     "Not peeling: found no good candidates in th ehistogram\n");
+		     "Not peeling: found no good candidates in the histogram\n");
 	  return false;
 	}
     }
diff --git a/gcc/value-prof.cc b/gcc/value-prof.cc
index 777ea47d063..f79dedd74fb 100644
--- a/gcc/value-prof.cc
+++ b/gcc/value-prof.cc
@@ -1950,6 +1950,8 @@ gimple_stringops_values_to_profile (gimple *gs, histogram_values *values)
 static void
 gimple_histogram_values_to_profile (function *fun, histogram_values *values)
 {
+  if (!flag_profile_loops)
+    return;
   for (auto loop : loops_list (fun, 0))
     {
       tree var;