math: optimize Ceil, Floor and Trunc on s390x

Use the FIDBR instruction to round floating-point numbers to integers.

name   old time/op  new time/op  delta
Ceil   14.1ns ± 0%   3.0ns ± 0%  -78.89%  (p=0.000 n=10+10)
Floor  6.42ns ± 0%  3.03ns ± 0%  -52.80%  (p=0.000 n=10+10)
Trunc  6.67ns ± 0%  3.03ns ± 0%  -54.57%   (p=0.000 n=10+9)

Change-Id: I3b416f6d0bccaaa9b547de86356471365862399c
Reviewed-on: https://go-review.googlesource.com/27827Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

src/math/floor_s390x.s

+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#include "textflag.h"
+
+// func Floor(x float64) float64
+TEXT ·Floor(SB),NOSPLIT,$0
+	FMOVD	x+0(FP), F0
+	FIDBR	$7, F0, F0
+	FMOVD	F0, ret+8(FP)
+	RET
+
+// func Ceil(x float64) float64
+TEXT ·Ceil(SB),NOSPLIT,$0
+	FMOVD	x+0(FP), F0
+	FIDBR	$6, F0, F0
+	FMOVD	F0, ret+8(FP)
+	RET
+
+// func Trunc(x float64) float64
+TEXT ·Trunc(SB),NOSPLIT,$0
+	FMOVD	x+0(FP), F0
+	FIDBR	$5, F0, F0
+	FMOVD	F0, ret+8(FP)
+	RET

src/math/stubs_s390x.s

@@ -25,15 +25,6 @@ TEXT ·Expm1(SB),NOSPLIT,$0
 TEXT ·Exp(SB),NOSPLIT,$0
 	BR ·exp(SB)
 
-TEXT ·Floor(SB),NOSPLIT,$0
-	BR ·floor(SB)
-
-TEXT ·Ceil(SB),NOSPLIT,$0
-	BR ·ceil(SB)
-
-TEXT ·Trunc(SB),NOSPLIT,$0
-	BR ·trunc(SB)
-
 TEXT ·Frexp(SB),NOSPLIT,$0
 	BR ·frexp(SB)