From ee8a9b7b507013b0431cabd314a245461c3b86b9 Mon Sep 17 00:00:00 2001
From: Joern Rennecke <joern.rennecke@embecosm.com>
Date: Wed, 10 Apr 2013 09:54:25 +0000
Subject: [PATCH] re PR tree-optimization/55524 (If fnma exists but not fms,
 convert_mult_to_fma should prefer to former over the latter.)

gcc:
2013-04-10  Joern Rennecke <joern.rennecke@embecosm.com>

        PR tree-optimization/55524
        * tree-ssa-math-opts.c
        (convert_mult_to_fma): Don't use an fms construct
        when we don't have an fms operation, but fnma, and it looks
        likely that we'll be able to use the latter.

gcc/testsuite:
2013-04-10  Joern Rennecke <joern.rennecke@embecosm.com>

        PR tree-optimization/55524
        * gcc.target/epiphany/fnma-1.c: New test.

From-SVN: r197668
---
 gcc/ChangeLog                              |  8 ++++++++
 gcc/testsuite/ChangeLog                    |  5 +++++
 gcc/testsuite/gcc.target/epiphany/fnma-1.c |  9 +++++++++
 gcc/tree-ssa-math-opts.c                   | 18 ++++++++++++++++++
 4 files changed, 40 insertions(+)
 create mode 100644 gcc/testsuite/gcc.target/epiphany/fnma-1.c

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index b8a67893a0a..98c6462794d 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,11 @@
+2013-04-10  Joern Rennecke <joern.rennecke@embecosm.com>
+
+	PR tree-optimization/55524
+	* tree-ssa-math-opts.c
+	(convert_mult_to_fma): Don't use an fms construct
+	when we don't have an fms operation, but fnma, and it looks
+	likely that we'll be able to use the latter.
+
 2013-04-10  Zhouyi Zhou <yizhouzhou@ict.ac.cn>
 
 	* cif-code.def (OVERWRITABLE): Correct the comment for overwritable
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 13585fc0758..aacd0094a09 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2013-04-10  Joern Rennecke <joern.rennecke@embecosm.com>
+
+	PR tree-optimization/55524
+	* gcc.target/epiphany/fnma-1.c: New test.
+
 2013-04-10  Zhouyi Zhou <yizhouzhou@ict.ac.cn>
 
 	* gcc.dg/tree-ssa/inline-11.c: New test
diff --git a/gcc/testsuite/gcc.target/epiphany/fnma-1.c b/gcc/testsuite/gcc.target/epiphany/fnma-1.c
new file mode 100644
index 00000000000..3155079f4a1
--- /dev/null
+++ b/gcc/testsuite/gcc.target/epiphany/fnma-1.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+/* { dg-final { scan-assembler-times "fmsub\[ \ta-zA-Z0-9\]*," 1 } } */
+
+float
+f (float ar, float ai, float br, float bi)
+{
+  return ar * br - ai * bi;
+}
diff --git a/gcc/tree-ssa-math-opts.c b/gcc/tree-ssa-math-opts.c
index 2140ced495b..e7e09f6a9de 100644
--- a/gcc/tree-ssa-math-opts.c
+++ b/gcc/tree-ssa-math-opts.c
@@ -2570,6 +2570,24 @@ convert_mult_to_fma (gimple mul_stmt, tree op1, tree op2)
 	  return false;
 	}
 
+      /* If the subtrahend (gimple_assign_rhs2 (use_stmt)) is computed
+	 by a MULT_EXPR that we'll visit later, we might be able to
+	 get a more profitable match with fnma.
+	 OTOH, if we don't, a negate / fma pair has likely lower latency
+	 that a mult / subtract pair.  */
+      if (use_code == MINUS_EXPR && !negate_p
+	  && gimple_assign_rhs1 (use_stmt) == result
+	  && optab_handler (fms_optab, TYPE_MODE (type)) == CODE_FOR_nothing
+	  && optab_handler (fnma_optab, TYPE_MODE (type)) != CODE_FOR_nothing)
+	{
+	  tree rhs2 = gimple_assign_rhs2 (use_stmt);
+	  gimple stmt2 = SSA_NAME_DEF_STMT (rhs2);
+
+	  if (has_single_use (rhs2)
+	      && gimple_assign_rhs_code (stmt2) == MULT_EXPR)
+	    return false;
+	}
+
       /* We can't handle a * b + a * b.  */
       if (gimple_assign_rhs1 (use_stmt) == gimple_assign_rhs2 (use_stmt))
 	return false;