cmd/compile: intrinsify math.RoundToEven on amd64

We already do this for floor/ceil, but RoundToEven was added later.
Intrinsify it also.

name           old time/op  new time/op  delta
RoundToEven-8  3.00ns ± 1%  0.68ns ± 2%  -77.34%  (p=0.000 n=10+10)

Change-Id: Ib158cbceb436c6725b2d9353a526c5c4be19bcad
Reviewed-on: https://go-review.googlesource.com/74852
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/compile/internal/amd64/ssa.go

@@ -855,8 +855,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 	case ssa.OpAMD64ROUNDSD:
 		p := s.Prog(v.Op.Asm())
 		val := v.AuxInt
-		// 1 means math.Floor, 2 Ceil, 3 Trunc
-		if val != 1 && val != 2 && val != 3 {
+		// 0 means math.RoundToEven, 1 Floor, 2 Ceil, 3 Trunc
+		if val != 0 && val != 1 && val != 2 && val != 3 {
 			v.Fatalf("Invalid rounding mode")
 		}
 		p.From.Offset = val

src/cmd/compile/internal/gc/ssa.go

@@ -2860,6 +2860,9 @@ func init() {
 			return s.variable(n, types.Types[TFLOAT64])
 		}
 	}
+	addF("math", "RoundToEven",
+		makeRoundAMD64(ssa.OpRoundToEven),
+		sys.AMD64)
 	addF("math", "Floor",
 		makeRoundAMD64(ssa.OpFloor),
 		sys.AMD64)

src/cmd/compile/internal/ssa/gen/AMD64.rules

@@ -113,9 +113,10 @@
 
 (Sqrt x) -> (SQRTSD x)
 
-(Floor x) -> (ROUNDSD [1] x)
-(Ceil x)  -> (ROUNDSD [2] x)
-(Trunc x) -> (ROUNDSD [3] x)
+(RoundToEven x) -> (ROUNDSD [0] x)
+(Floor x)	-> (ROUNDSD [1] x)
+(Ceil x)  	-> (ROUNDSD [2] x)
+(Trunc x) 	-> (ROUNDSD [3] x)
 
 // Lowering extension
 // Note: we always extend to 64 bits even though some ops don't need that many result bits.

src/cmd/compile/internal/ssa/rewriteAMD64.go

@@ -795,6 +795,8 @@ func rewriteValueAMD64(v *Value) bool {
 		return rewriteValueAMD64_OpRound32F_0(v)
 	case OpRound64F:
 		return rewriteValueAMD64_OpRound64F_0(v)
+	case OpRoundToEven:
+		return rewriteValueAMD64_OpRoundToEven_0(v)
 	case OpRsh16Ux16:
 		return rewriteValueAMD64_OpRsh16Ux16_0(v)
 	case OpRsh16Ux32:
@@ -45652,6 +45654,18 @@ func rewriteValueAMD64_OpRound64F_0(v *Value) bool {
 		return true
 	}
 }
+func rewriteValueAMD64_OpRoundToEven_0(v *Value) bool {
+	// match: (RoundToEven x)
+	// cond:
+	// result: (ROUNDSD [0] x)
+	for {
+		x := v.Args[0]
+		v.reset(OpAMD64ROUNDSD)
+		v.AuxInt = 0
+		v.AddArg(x)
+		return true
+	}
+}
 func rewriteValueAMD64_OpRsh16Ux16_0(v *Value) bool {
 	b := v.Block
 	_ = b