cmd/internal/obj/arm64: support more atomic instructions

LDADDALD(64-bit) and LDADDALW(32-bit) are already supported.
This CL adds supports of LDADDALH(16-bit) and LDADDALB(8-bit).

Change-Id: I4eac61adcec226d618dfce88618a2b98f5f1afe7
Reviewed-on: https://go-review.googlesource.com/132135
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/cmd/asm/internal/arch/arm64.go

@@ -79,7 +79,7 @@ func IsARM64STLXR(op obj.As) bool {
 		arm64.ALDANDB, arm64.ALDANDH, arm64.ALDANDW, arm64.ALDANDD,
 		arm64.ALDEORB, arm64.ALDEORH, arm64.ALDEORW, arm64.ALDEORD,
 		arm64.ALDORB, arm64.ALDORH, arm64.ALDORW, arm64.ALDORD,
-		arm64.ALDADDALD, arm64.ALDADDALW:
+		arm64.ALDADDALD, arm64.ALDADDALW, arm64.ALDADDALH, arm64.ALDADDALB:
 		return true
 	}
 	return false

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

@@ -622,7 +622,9 @@ again:
 	LDORB	R5, (R6), R7                         // c7302538
 	LDORB	R5, (RSP), R7                        // e7332538
 	LDADDALD	R2, (R1), R3                 // 2300e2f8
-	LDADDALW	R5, (R4), R6                 // 8600e5b8
+	LDADDALW	R2, (R1), R3                 // 2300e2b8
+	LDADDALH	R2, (R1), R3                 // 2300e278
+	LDADDALB	R2, (R1), R3                 // 2300e238
 
 // RET
 //

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

@@ -594,8 +594,10 @@ const (
 	AHVC
 	AIC
 	AISB
-	ALDADDALD
+	ALDADDALB
+	ALDADDALH
 	ALDADDALW
+	ALDADDALD
 	ALDADDB
 	ALDADDH
 	ALDADDW

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

@@ -95,8 +95,10 @@ var Anames = []string{
 	"HVC",
 	"IC",
 	"ISB",
-	"LDADDALD",
+	"LDADDALB",
+	"LDADDALH",
 	"LDADDALW",
+	"LDADDALD",
 	"LDADDB",
 	"LDADDH",
 	"LDADDW",

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

@@ -779,7 +779,7 @@ func span7(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
 	}
 
 	c := ctxt7{ctxt: ctxt, newprog: newprog, cursym: cursym, autosize: int32(p.To.Offset & 0xffffffff), extrasize: int32(p.To.Offset >> 32)}
-	p.To.Offset &= 0xffffffff  // extrasize is no longer needed
+	p.To.Offset &= 0xffffffff // extrasize is no longer needed
 
 	bflag := 1
 	pc := int64(0)
@@ -2023,8 +2023,10 @@ func buildop(ctxt *obj.Link) {
 			oprangeset(ASWPALB, t)
 			oprangeset(ASWPALH, t)
 			oprangeset(ASWPALW, t)
-			oprangeset(ALDADDALD, t)
+			oprangeset(ALDADDALB, t)
+			oprangeset(ALDADDALH, t)
 			oprangeset(ALDADDALW, t)
+			oprangeset(ALDADDALD, t)
 			oprangeset(ALDADDB, t)
 			oprangeset(ALDADDH, t)
 			oprangeset(ALDADDW, t)
@@ -3406,9 +3408,9 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			o1 = 3 << 30
 		case ASWPW, ASWPALW, ALDADDALW, ALDADDW, ALDANDW, ALDEORW, ALDORW: // 32-bit
 			o1 = 2 << 30
-		case ASWPH, ASWPALH, ALDADDH, ALDANDH, ALDEORH, ALDORH: // 16-bit
+		case ASWPH, ASWPALH, ALDADDALH, ALDADDH, ALDANDH, ALDEORH, ALDORH: // 16-bit
 			o1 = 1 << 30
-		case ASWPB, ASWPALB, ALDADDB, ALDANDB, ALDEORB, ALDORB: // 8-bit
+		case ASWPB, ASWPALB, ALDADDALB, ALDADDB, ALDANDB, ALDEORB, ALDORB: // 8-bit
 			o1 = 0 << 30
 		default:
 			c.ctxt.Diag("illegal instruction: %v\n", p)
@@ -3416,7 +3418,7 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		switch p.As {
 		case ASWPD, ASWPW, ASWPH, ASWPB, ASWPALD, ASWPALW, ASWPALH, ASWPALB:
 			o1 |= 0x20 << 10
-		case ALDADDALD, ALDADDALW, ALDADDD, ALDADDW, ALDADDH, ALDADDB:
+		case ALDADDALD, ALDADDALW, ALDADDALH, ALDADDALB, ALDADDD, ALDADDW, ALDADDH, ALDADDB:
 			o1 |= 0x00 << 10
 		case ALDANDD, ALDANDW, ALDANDH, ALDANDB:
 			o1 |= 0x04 << 10
@@ -3426,7 +3428,7 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			o1 |= 0x0c << 10
 		}
 		switch p.As {
-		case ALDADDALD, ALDADDALW, ASWPALD, ASWPALW, ASWPALH, ASWPALB:
+		case ALDADDALD, ALDADDALW, ALDADDALH, ALDADDALB, ASWPALD, ASWPALW, ASWPALH, ASWPALB:
 			o1 |= 3 << 22
 		}
 		o1 |= 0x1c1<<21 | uint32(rs&31)<<16 | uint32(rb&31)<<5 | uint32(rt&31)