vectorizer cost model enhancement

From-SVN: r202980

gcc/ChangeLog

@@ -1,3 +1,24 @@
+2013-09-27  Xinliang David Li  <davidxl@google.com>
+
+	* opts.c (finish_options): Adjust parameters
+	according to vect cost model.
+	(common_handle_option): Set dynamic vect cost
+	model for FDO.
+	targhooks.c (default_add_stmt_cost): Compute stmt cost
+	unconditionally.
+	* tree-vect-loop.c (vect_estimate_min_profitable_iters):
+	Use helper function.
+	* tree-vectorizer.h (unlimited_cost_model): New function.
+	* tree-vect-slp.c (vect_slp_analyze_bb_1): Use helper function.
+	* tree-vect-data-refs.c (vect_peeling_hash_insert): Use helper
+	function.
+	(vect_enhance_data_refs_alignment): Ditto.
+	* flag-types.h: New enum.
+	* common/config/i386/i386-common.c (ix86_option_init_struct):
+	No need to initialize vect_cost_model flag.
+	* config/i386/i386.c (ix86_add_stmt_cost): Compute stmt cost
+	unconditionally.
+
 2013-09-27  Diego Novillo  <dnovillo@google.com>
 
 	* gimple.h (enum ssa_mode): Remove.

gcc/common.opt

@@ -2278,13 +2278,33 @@ ftree-slp-vectorize
 Common Report Var(flag_tree_slp_vectorize) Optimization
 Enable basic block vectorization (SLP) on trees
 
+fvect-cost-model=
+Common Joined RejectNegative Enum(vect_cost_model) Var(flag_vect_cost_model) Init(VECT_COST_MODEL_DEFAULT)
+Specifies the cost model for vectorization
+ 
+Enum
+Name(vect_cost_model) Type(enum vect_cost_model) UnknownError(unknown vectorizer cost model %qs)
+
+EnumValue
+Enum(vect_cost_model) String(unlimited) Value(VECT_COST_MODEL_UNLIMITED)
+
+EnumValue
+Enum(vect_cost_model) String(dynamic) Value(VECT_COST_MODEL_DYNAMIC)
+
+EnumValue
+Enum(vect_cost_model) String(cheap) Value(VECT_COST_MODEL_CHEAP)
+
 fvect-cost-model
-Common Report Var(flag_vect_cost_model) Optimization
-Enable use of cost model in vectorization
+Common RejectNegative Alias(fvect-cost-model=,dynamic)
+Enables the dynamic vectorizer cost model.  Preserved for backward compatibility.
+
+fno-vect-cost-model
+Common RejectNegative Alias(fvect-cost-model=,unlimited)
+Enables the unlimited vectorizer cost model.  Preserved for backward compatibility.
 
 ftree-vect-loop-version
-Common Report Var(flag_tree_vect_loop_version) Init(1) Optimization
-Enable loop versioning when doing loop vectorization on trees
+Common Ignore
+Does nothing. Preserved for backward compatibility.
 
 ftree-scev-cprop
 Common Report Var(flag_tree_scev_cprop) Init(1) Optimization

gcc/common/config/i386/i386-common.c

@@ -811,7 +811,6 @@ ix86_option_init_struct (struct gcc_options *opts)
 
   opts->x_flag_pcc_struct_return = 2;
   opts->x_flag_asynchronous_unwind_tables = 2;
-  opts->x_flag_vect_cost_model = 1;
 }
 
 /* On the x86 -fsplit-stack and -fstack-protector both use the same

gcc/config/i386/i386.c

@@ -42782,20 +42782,17 @@ ix86_add_stmt_cost (void *data, int count, enum vect_cost_for_stmt kind,
   unsigned *cost = (unsigned *) data;
   unsigned retval = 0;
 
-  if (flag_vect_cost_model)
-    {
-      tree vectype = stmt_info ? stmt_vectype (stmt_info) : NULL_TREE;
-      int stmt_cost = ix86_builtin_vectorization_cost (kind, vectype, misalign);
+  tree vectype = stmt_info ? stmt_vectype (stmt_info) : NULL_TREE;
+  int stmt_cost = ix86_builtin_vectorization_cost (kind, vectype, misalign);
 
-      /* Statements in an inner loop relative to the loop being
-	 vectorized are weighted more heavily.  The value here is
-	 arbitrary and could potentially be improved with analysis.  */
-      if (where == vect_body && stmt_info && stmt_in_inner_loop_p (stmt_info))
-	count *= 50;  /* FIXME.  */
+  /* Statements in an inner loop relative to the loop being
+     vectorized are weighted more heavily.  The value here is
+      arbitrary and could potentially be improved with analysis.  */
+  if (where == vect_body && stmt_info && stmt_in_inner_loop_p (stmt_info))
+    count *= 50;  /* FIXME.  */
 
-      retval = (unsigned) (count * stmt_cost);
-      cost[where] += retval;
-    }
+  retval = (unsigned) (count * stmt_cost);
+  cost[where] += retval;
 
   return retval;
 }

gcc/doc/invoke.texi

@@ -423,7 +423,7 @@ Objective-C and Objective-C++ Dialects}.
 -ftree-parallelize-loops=@var{n} -ftree-pre -ftree-partial-pre -ftree-pta @gol
 -ftree-reassoc -ftree-sink -ftree-slsr -ftree-sra @gol
 -ftree-switch-conversion -ftree-tail-merge -ftree-ter @gol
--ftree-vect-loop-version -ftree-vectorize -ftree-vrp @gol
+-ftree-vectorize -ftree-vrp @gol
 -funit-at-a-time -funroll-all-loops -funroll-loops @gol
 -funsafe-loop-optimizations -funsafe-math-optimizations -funswitch-loops @gol
 -fvariable-expansion-in-unroller -fvect-cost-model -fvpt -fweb @gol
@@ -6770,7 +6770,7 @@ optimizations designed to reduce code size.
 @option{-Os} disables the following optimization flags:
 @gccoptlist{-falign-functions  -falign-jumps  -falign-loops @gol
 -falign-labels  -freorder-blocks  -freorder-blocks-and-partition @gol
--fprefetch-loop-arrays  -ftree-vect-loop-version}
+-fprefetch-loop-arrays}
 
 @item -Ofast
 @opindex Ofast
@@ -8025,19 +8025,20 @@ Perform loop vectorization on trees. This flag is enabled by default at
 Perform basic block vectorization on trees. This flag is enabled by default at
 @option{-O3} and when @option{-ftree-vectorize} is enabled.
 
-@item -ftree-vect-loop-version
-@opindex ftree-vect-loop-version
-Perform loop versioning when doing loop vectorization on trees.  When a loop
-appears to be vectorizable except that data alignment or data dependence cannot
-be determined at compile time, then vectorized and non-vectorized versions of
-the loop are generated along with run-time checks for alignment or dependence
-to control which version is executed.  This option is enabled by default
-except at level @option{-Os} where it is disabled.
-
-@item -fvect-cost-model
+@item -fvect-cost-model=@var{model}
 @opindex fvect-cost-model
-Enable cost model for vectorization.  This option is enabled by default at
-@option{-O3}.
+Alter the cost model used for vectorization.  The @var{model} argument
+should be one of @code{unlimited}, @code{dynamic} or @code{cheap}.
+With the @code{unlimited} model the vectorized code-path is assumed
+to be profitable while with the @code{dynamic} model a runtime check
+will guard the vectorized code-path to enable it only for iteration
+counts that will likely execute faster than when executing the original
+scalar loop.  The @code{cheap} model will disable vectorization of
+loops where doing so would be cost prohibitive for example due to
+required runtime checks for data dependence or alignment but otherwise
+is equal to the @code{dynamic} model.
+The default cost model depends on other optimization flags and is
+either @code{dynamic} or @code{cheap}.
 
 @item -ftree-vrp
 @opindex ftree-vrp
@@ -9443,13 +9444,11 @@ constraints.  The default value is 0.
 
 @item vect-max-version-for-alignment-checks
 The maximum number of run-time checks that can be performed when
-doing loop versioning for alignment in the vectorizer.  See option
-@option{-ftree-vect-loop-version} for more information.
+doing loop versioning for alignment in the vectorizer. 
 
 @item vect-max-version-for-alias-checks
 The maximum number of run-time checks that can be performed when
-doing loop versioning for alias in the vectorizer.  See option
-@option{-ftree-vect-loop-version} for more information.
+doing loop versioning for alias in the vectorizer. 
 
 @item vect-max-peeling-for-alignment
 The maximum number of loop peels to enhance access alignment

gcc/flag-types.h

@@ -191,6 +191,15 @@ enum fp_contract_mode {
   FP_CONTRACT_FAST = 2
 };
 
+/* Vectorizer cost-model.  */
+enum vect_cost_model {
+  VECT_COST_MODEL_UNLIMITED = 0,
+  VECT_COST_MODEL_CHEAP = 1,
+  VECT_COST_MODEL_DYNAMIC = 2,
+  VECT_COST_MODEL_DEFAULT = 3
+};
+
+
 /* Different instrumentation modes.  */
 enum sanitize_code {
   /* AddressSanitizer.  */

gcc/opts.c

@@ -486,6 +486,7 @@ static const struct default_options default_options_table[] =
     { OPT_LEVELS_2_PLUS, OPT_falign_labels, NULL, 1 },
     { OPT_LEVELS_2_PLUS, OPT_falign_functions, NULL, 1 },
     { OPT_LEVELS_2_PLUS, OPT_ftree_tail_merge, NULL, 1 },
+    { OPT_LEVELS_2_PLUS, OPT_fvect_cost_model_, NULL, VECT_COST_MODEL_CHEAP },
     { OPT_LEVELS_2_PLUS_SPEED_ONLY, OPT_foptimize_strlen, NULL, 1 },
     { OPT_LEVELS_2_PLUS, OPT_fhoist_adjacent_loads, NULL, 1 },
 
@@ -500,7 +501,7 @@ static const struct default_options default_options_table[] =
     { OPT_LEVELS_3_PLUS, OPT_fgcse_after_reload, NULL, 1 },
     { OPT_LEVELS_3_PLUS, OPT_ftree_loop_vectorize, NULL, 1 },
     { OPT_LEVELS_3_PLUS, OPT_ftree_slp_vectorize, NULL, 1 },
-    { OPT_LEVELS_3_PLUS, OPT_fvect_cost_model, NULL, 1 },
+    { OPT_LEVELS_3_PLUS, OPT_fvect_cost_model_, NULL, VECT_COST_MODEL_DYNAMIC },
     { OPT_LEVELS_3_PLUS, OPT_fipa_cp_clone, NULL, 1 },
     { OPT_LEVELS_3_PLUS, OPT_ftree_partial_pre, NULL, 1 },
 
@@ -825,6 +826,17 @@ finish_options (struct gcc_options *opts, struct gcc_options *opts_set,
 	}
     }
 
+  /* Tune vectorization related parametees according to cost model.  */
+  if (opts->x_flag_vect_cost_model == VECT_COST_MODEL_CHEAP)
+    {
+      maybe_set_param_value (PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS,
+            6, opts->x_param_values, opts_set->x_param_values);
+      maybe_set_param_value (PARAM_VECT_MAX_VERSION_FOR_ALIGNMENT_CHECKS,
+            0, opts->x_param_values, opts_set->x_param_values);
+      maybe_set_param_value (PARAM_VECT_MAX_PEELING_FOR_ALIGNMENT,
+            0, opts->x_param_values, opts_set->x_param_values);
+    }
+
   /* Set PARAM_MAX_STORES_TO_SINK to 0 if either vectorization or if-conversion
      is disabled.  */
   if ((!opts->x_flag_tree_loop_vectorize && !opts->x_flag_tree_slp_vectorize)
@@ -1669,7 +1681,7 @@ common_handle_option (struct gcc_options *opts,
           && !opts_set->x_flag_tree_vectorize)
 	opts->x_flag_tree_slp_vectorize = value;
       if (!opts_set->x_flag_vect_cost_model)
-	opts->x_flag_vect_cost_model = value;
+	opts->x_flag_vect_cost_model = VECT_COST_MODEL_DYNAMIC;
       if (!opts_set->x_flag_tree_loop_distribute_patterns)
 	opts->x_flag_tree_loop_distribute_patterns = value;
       /* Indirect call profiling should do all useful transformations

gcc/targhooks.c

@@ -1057,20 +1057,17 @@ default_add_stmt_cost (void *data, int count, enum vect_cost_for_stmt kind,
   unsigned *cost = (unsigned *) data;
   unsigned retval = 0;
 
-  if (flag_vect_cost_model)
-    {
-      tree vectype = stmt_info ? stmt_vectype (stmt_info) : NULL_TREE;
-      int stmt_cost = default_builtin_vectorization_cost (kind, vectype,
+  tree vectype = stmt_info ? stmt_vectype (stmt_info) : NULL_TREE;
+  int stmt_cost = default_builtin_vectorization_cost (kind, vectype,
 							  misalign);
-      /* Statements in an inner loop relative to the loop being
-	 vectorized are weighted more heavily.  The value here is
-	 arbitrary and could potentially be improved with analysis.  */
-      if (where == vect_body && stmt_info && stmt_in_inner_loop_p (stmt_info))
-	count *= 50;  /* FIXME.  */
+   /* Statements in an inner loop relative to the loop being
+      vectorized are weighted more heavily.  The value here is
+      arbitrary and could potentially be improved with analysis.  */
+  if (where == vect_body && stmt_info && stmt_in_inner_loop_p (stmt_info))
+    count *= 50;  /* FIXME.  */
 
-      retval = (unsigned) (count * stmt_cost);
-      cost[where] += retval;
-    }
+  retval = (unsigned) (count * stmt_cost);
+  cost[where] += retval;
 
   return retval;
 }

gcc/tree-vect-data-refs.c

@@ -1115,7 +1115,7 @@ vect_peeling_hash_insert (loop_vec_info loop_vinfo, struct data_reference *dr,
       *new_slot = slot;
     }
 
-  if (!supportable_dr_alignment && !flag_vect_cost_model)
+  if (!supportable_dr_alignment && unlimited_cost_model ())
     slot->count += VECT_MAX_COST;
 }
 
@@ -1225,7 +1225,7 @@ vect_peeling_hash_choose_best_peeling (loop_vec_info loop_vinfo,
    res.peel_info.dr = NULL;
    res.body_cost_vec = stmt_vector_for_cost();
 
-   if (flag_vect_cost_model)
+   if (!unlimited_cost_model ())
      {
        res.inside_cost = INT_MAX;
        res.outside_cost = INT_MAX;
@@ -1454,7 +1454,7 @@ vect_enhance_data_refs_alignment (loop_vec_info loop_vinfo)
                  vectorization factor.
                  We do this automtically for cost model, since we calculate cost
                  for every peeling option.  */
-              if (!flag_vect_cost_model)
+              if (unlimited_cost_model ())
                 possible_npeel_number = vf /nelements;
 
               /* Handle the aligned case. We may decide to align some other
@@ -1462,7 +1462,7 @@ vect_enhance_data_refs_alignment (loop_vec_info loop_vinfo)
               if (DR_MISALIGNMENT (dr) == 0)
                 {
                   npeel_tmp = 0;
-                  if (!flag_vect_cost_model)
+                  if (unlimited_cost_model ())
                     possible_npeel_number++;
                 }
 
@@ -1795,16 +1795,14 @@ vect_enhance_data_refs_alignment (loop_vec_info loop_vinfo)
   /* (2) Versioning to force alignment.  */
 
   /* Try versioning if:
-     1) flag_tree_vect_loop_version is TRUE
-     2) optimize loop for speed
-     3) there is at least one unsupported misaligned data ref with an unknown
+     1) optimize loop for speed
+     2) there is at least one unsupported misaligned data ref with an unknown
         misalignment, and
-     4) all misaligned data refs with a known misalignment are supported, and
-     5) the number of runtime alignment checks is within reason.  */
+     3) all misaligned data refs with a known misalignment are supported, and
+     4) the number of runtime alignment checks is within reason.  */
 
   do_versioning =
-	flag_tree_vect_loop_version
-	&& optimize_loop_nest_for_speed_p (loop)
+	optimize_loop_nest_for_speed_p (loop)
 	&& (!loop->inner); /* FORNOW */
 
   if (do_versioning)

gcc/tree-vect-loop.c

@@ -2680,7 +2680,7 @@ vect_estimate_min_profitable_iters (loop_vec_info loop_vinfo,
   void *target_cost_data = LOOP_VINFO_TARGET_COST_DATA (loop_vinfo);
 
   /* Cost model disabled.  */
-  if (!flag_vect_cost_model)
+  if (unlimited_cost_model ())
     {
       dump_printf_loc (MSG_NOTE, vect_location, "cost model disabled.\n");
       *ret_min_profitable_niters = 0;

gcc/tree-vect-slp.c

@@ -2168,7 +2168,7 @@ vect_slp_analyze_bb_1 (basic_block bb)
     }
 
   /* Cost model: check if the vectorization is worthwhile.  */
-  if (flag_vect_cost_model
+  if (!unlimited_cost_model ()
       && !vect_bb_vectorization_profitable_p (bb_vinfo))
     {
       if (dump_enabled_p ())

gcc/tree-vectorizer.h

@@ -880,6 +880,14 @@ known_alignment_for_access_p (struct data_reference *data_ref_info)
   return (DR_MISALIGNMENT (data_ref_info) != -1);
 }
 
+
+/* Return true if the vect cost model is unlimited.  */
+static inline bool
+unlimited_cost_model ()
+{
+  return flag_vect_cost_model == VECT_COST_MODEL_UNLIMITED;
+}
+
 /* Source location */
 extern LOC vect_location;