cmd/asm: add rev64 instruction on ARM64

This change provides VREV64 instruction for AES-GCM implementation.

Change-Id: Icdf278862b03556388586f459964b025c47b8c19
Reviewed-on: https://go-review.googlesource.com/107696Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

src/cmd/asm/internal/asm/testdata/arm64.s

@@ -309,6 +309,8 @@ TEXT	foo(SB), DUPOK|NOSPLIT, $-8
 	VMOV	V8.B[0], V12.B[1]     // 0c05036e
 	VMOV	V8.B[7], V4.B[8]      // 043d116e
 	VREV32	V5.B16, V5.B16        // a508206e
+	VREV64	V2.S2, V3.S2          // 4308a00e
+	VREV64	V2.S4, V3.S4          // 4308a04e
 	VDUP	V19.S[0], V17.S4      // 7106044e
 //
 // B/BL

src/cmd/asm/internal/asm/testdata/arm64error.s

@@ -81,4 +81,6 @@ TEXT errors(SB),$0
 	VLD1.P	(R8)(R9<<2), [V2.B16]                            // ERROR "invalid extended register"
 	VST1.P	[V1.B16], (R8)(R9.UXTW)                          // ERROR "invalid extended register"
 	VST1.P	[V1.B16], (R8)(R9<<1)                            // ERROR "invalid extended register"
+	VREV64	V1.H4, V2.H8                                     // ERROR "invalid arrangement"
+	VREV64	V1.D1, V2.D1                                     // ERROR "invalid arrangement"
 	RET

src/cmd/internal/obj/arm64/a.out.go

@@ -881,6 +881,7 @@ const (
 	AVLD1
 	AVORR
 	AVREV32
+	AVREV64
 	AVST1
 	AVDUP
 	AVMOVS

src/cmd/internal/obj/arm64/anames.go

@@ -383,6 +383,7 @@ var Anames = []string{
 	"VLD1",
 	"VORR",
 	"VREV32",
+	"VREV64",
 	"VST1",
 	"VDUP",
 	"VMOVS",

src/cmd/internal/obj/arm64/asm7.go

@@ -2387,6 +2387,7 @@ func buildop(ctxt *obj.Link) {
 
 		case AVREV32:
 			oprangeset(AVRBIT, t)
+			oprangeset(AVREV64, t)
 
 		case ASHA1H,
 			AVCNT,
@@ -3964,8 +3965,7 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		rf := int((p.From.Reg) & 31)
 		rt := int((p.To.Reg) & 31)
 
-		Q := 0
-		size := 0
+		var Q, size uint32
 		switch af {
 		case ARNG_8B:
 			Q = 0
@@ -3979,6 +3979,12 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		case ARNG_8H:
 			Q = 1
 			size = 1
+		case ARNG_2S:
+			Q = 0
+			size = 2
+		case ARNG_4S:
+			Q = 1
+			size = 2
 		default:
 			c.ctxt.Diag("invalid arrangement: %v\n", p)
 		}
@@ -3987,6 +3993,10 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			c.ctxt.Diag("invalid arrangement: %v", p)
 		}
 
+		if p.As == AVREV32 && (af == ARNG_2S || af == ARNG_4S) {
+			c.ctxt.Diag("invalid arrangement: %v", p)
+		}
+
 		if p.As == AVMOV {
 			o1 |= uint32(rf&31) << 16
 		}
@@ -3995,7 +4005,7 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			size = 1
 		}
 
-		o1 |= (uint32(Q&1) << 30) | (uint32(size&3) << 22) | (uint32(rf&31) << 5) | uint32(rt&31)
+		o1 |= (Q&1) << 30 | (size&3) << 22 | uint32(rf&31) << 5 | uint32(rt&31)
 
 	case 84: /* vst1 [Vt1.<T>, Vt2.<T>, ...], (Rn) */
 		r := int(p.To.Reg)
@@ -5073,6 +5083,9 @@ func (c *ctxt7) oprrr(p *obj.Prog, a obj.As) uint32 {
 	case AVREV32:
 		return 11<<26 | 2<<24 | 1<<21 | 1<<11
 
+	case AVREV64:
+		return 3<<26 | 2<<24 | 1<<21 | 1<<11
+
 	case AVMOV:
 		return 7<<25 | 5<<21 | 7<<10
 

src/cmd/internal/obj/arm64/doc.go

@@ -258,6 +258,11 @@ Go Assembly for ARM64 Reference Manual
         <T> Is an arrangement specifier and can have the following values:
         B8, B16, H4, H8
 
+    VREV64: Reverse elements in 64-bit words (vector).
+      REV64 <Vn>.<T>, <Vd>.<T>
+        <T> Is an arrangement specifier and can have the following values:
+        B8, B16, H4, H8, S2, S4
+
     VSHL: Shift Left(immediate)
       VSHL 	$shift, <Vn>.<T>, <Vd>.<T>
         <T> is an arrangement specifier and can have the following values: