public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* Move EXTRACT_LAST_REDUCTION costing to vectorizable_condition
@ 2019-11-21 12:08 Richard Sandiford
  2019-11-21 14:02 ` Richard Biener
  0 siblings, 1 reply; 2+ messages in thread
From: Richard Sandiford @ 2019-11-21 12:08 UTC (permalink / raw)
  To: gcc-patches

gcc.target/aarch64/sve/clastb_[57].c started failing after the increase
in the cost of vec_to_scalar (r278452).  The problem is that we were
double-counting the cost of the CLASTB: once in vect_model_reduction_cost
as a vec_to_scalar and once in vectorizable_condition as a plain
vector_stmt.

Based on the TODO above vect_model_reduction_cost, I think the
preferred long-term direction is for vectorizable_* to cost these
things itself, so that's what the patch does (for this one case only).

Tested on aarch64-linux-gnu and x86_64-linux-gnu.  OK to install?

Richard


2019-11-21  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
	* tree-vect-stmts.c (vect_model_simple_cost): Take an optional
	vect_cost_for_stmt.
	(vectorizable_condition): Calculate the cost of EXTRACT_LAST_REDUCTION
	here rather than...
	* tree-vect-loop.c (vect_model_reduction_cost): ...here.

Index: gcc/tree-vect-stmts.c
===================================================================
--- gcc/tree-vect-stmts.c	2019-11-16 10:29:21.211212189 +0000
+++ gcc/tree-vect-stmts.c	2019-11-21 12:02:44.058240392 +0000
@@ -860,7 +860,8 @@ vect_model_simple_cost (stmt_vec_info st
 			enum vect_def_type *dt,
 			int ndts,
 			slp_tree node,
-			stmt_vector_for_cost *cost_vec)
+			stmt_vector_for_cost *cost_vec,
+			vect_cost_for_stmt kind = vector_stmt)
 {
   int inside_cost = 0, prologue_cost = 0;
 
@@ -907,7 +908,7 @@ vect_model_simple_cost (stmt_vec_info st
     }
 
   /* Pass the inside-of-loop statements to the target-specific cost model.  */
-  inside_cost += record_stmt_cost (cost_vec, ncopies, vector_stmt,
+  inside_cost += record_stmt_cost (cost_vec, ncopies, kind,
 				   stmt_info, 0, vect_body);
 
   if (dump_enabled_p ())
@@ -10081,15 +10082,18 @@ vectorizable_condition (stmt_vec_info st
 		return false;
 	    }
 	}
-      if (expand_vec_cond_expr_p (vectype, comp_vectype,
-				     cond_code))
-	{
-	  STMT_VINFO_TYPE (stmt_info) = condition_vec_info_type;
-	  vect_model_simple_cost (stmt_info, ncopies, dts, ndts, slp_node,
-				  cost_vec);
-	  return true;
-	}
-      return false;
+
+      vect_cost_for_stmt kind = vector_stmt;
+      if (reduction_type == EXTRACT_LAST_REDUCTION)
+	/* Count one reduction-like operation per vector.  */
+	kind = vec_to_scalar;
+      else if (!expand_vec_cond_expr_p (vectype, comp_vectype, cond_code))
+	return false;
+
+      STMT_VINFO_TYPE (stmt_info) = condition_vec_info_type;
+      vect_model_simple_cost (stmt_info, ncopies, dts, ndts, slp_node,
+			      cost_vec, kind);
+      return true;
     }
 
   /* Transform.  */
Index: gcc/tree-vect-loop.c
===================================================================
--- gcc/tree-vect-loop.c	2019-11-19 16:25:24.882645490 +0000
+++ gcc/tree-vect-loop.c	2019-11-21 12:02:44.050240447 +0000
@@ -3924,8 +3924,11 @@ vect_model_reduction_cost (stmt_vec_info
 
   code = gimple_assign_rhs_code (orig_stmt_info->stmt);
 
-  if (reduction_type == EXTRACT_LAST_REDUCTION
-      || reduction_type == FOLD_LEFT_REDUCTION)
+  if (reduction_type == EXTRACT_LAST_REDUCTION)
+    /* No extra instructions are needed in the prologue.  The loop body
+       operations are costed in vectorizable_condition.  */
+    inside_cost = 0;
+  else if (reduction_type == FOLD_LEFT_REDUCTION)
     {
       /* No extra instructions needed in the prologue.  */
       prologue_cost = 0;

^ permalink raw reply	[flat|nested] 2+ messages in thread
end of thread, other threads:[~2019-11-21 14:02 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-11-21 12:08 Move EXTRACT_LAST_REDUCTION costing to vectorizable_condition Richard Sandiford
2019-11-21 14:02 ` Richard Biener

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).