cmd/internal/obj/arm64: add SWPALD/SWPALW/SWPALH/SWPALB

Those new instructions have acquire/release semantics, besides
normal atomic SWPD/SWPW/SWPH/SWPB.

Change-Id: I24821a4d21aebc342897ae52903aef612c8d8a4a
Reviewed-on: https://go-review.googlesource.com/128476
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

@@ -74,6 +74,7 @@ func IsARM64STLXR(op obj.As) bool {
 		arm64.ASTXRB, arm64.ASTXRH, arm64.ASTXRW, arm64.ASTXR,
 		arm64.ASTXP, arm64.ASTXPW, arm64.ASTLXP, arm64.ASTLXPW,
 		arm64.ASWPB, arm64.ASWPH, arm64.ASWPW, arm64.ASWPD,
+		arm64.ASWPALB, arm64.ASWPALH, arm64.ASWPALW, arm64.ASWPALD,
 		arm64.ALDADDB, arm64.ALDADDH, arm64.ALDADDW, arm64.ALDADDD,
 		arm64.ALDANDB, arm64.ALDANDH, arm64.ALDANDW, arm64.ALDANDD,
 		arm64.ALDEORB, arm64.ALDEORH, arm64.ALDEORW, arm64.ALDEORD,

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

@@ -572,6 +572,14 @@ again:
 	SWPH	R5, (RSP), R7                        // e7832578
 	SWPB	R5, (R6), R7                         // c7802538
 	SWPB	R5, (RSP), R7                        // e7832538
+	SWPALD	R5, (R6), R7                         // c780e5f8
+	SWPALD	R5, (RSP), R7                        // e783e5f8
+	SWPALW	R5, (R6), R7                         // c780e5b8
+	SWPALW	R5, (RSP), R7                        // e783e5b8
+	SWPALH	R5, (R6), R7                         // c780e578
+	SWPALH	R5, (RSP), R7                        // e783e578
+	SWPALB	R5, (R6), R7                         // c780e538
+	SWPALB	R5, (RSP), R7                        // e783e538
 	LDADDD	R5, (R6), R7                         // c70025f8
 	LDADDD	R5, (RSP), R7                        // e70325f8
 	LDADDW	R5, (R6), R7                         // c70025b8

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

@@ -774,9 +774,13 @@ const (
 	AMOVPSW
 	AMOVPW
 	ASWPD
+	ASWPALD
 	ASWPW
+	ASWPALW
 	ASWPH
+	ASWPALH
 	ASWPB
+	ASWPALB
 	ABEQ
 	ABNE
 	ABCS

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

@@ -275,9 +275,13 @@ var Anames = []string{
 	"MOVPSW",
 	"MOVPW",
 	"SWPD",
+	"SWPALD",
 	"SWPW",
+	"SWPALW",
 	"SWPH",
+	"SWPALH",
 	"SWPB",
+	"SWPALB",
 	"BEQ",
 	"BNE",
 	"BCS",

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

@@ -2008,9 +2008,13 @@ func buildop(ctxt *obj.Link) {
 			oprangeset(AMOVZW, t)
 
 		case ASWPD:
+			oprangeset(ASWPALD, t)
 			oprangeset(ASWPB, t)
 			oprangeset(ASWPH, t)
 			oprangeset(ASWPW, t)
+			oprangeset(ASWPALB, t)
+			oprangeset(ASWPALH, t)
+			oprangeset(ASWPALW, t)
 			oprangeset(ALDADDALD, t)
 			oprangeset(ALDADDALW, t)
 			oprangeset(ALDADDB, t)
@@ -3383,19 +3387,19 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		rt := p.RegTo2
 		rb := p.To.Reg
 		switch p.As {
-		case ASWPD, ALDADDALD, ALDADDD, ALDANDD, ALDEORD, ALDORD: // 64-bit
+		case ASWPD, ASWPALD, ALDADDALD, ALDADDD, ALDANDD, ALDEORD, ALDORD: // 64-bit
 			o1 = 3 << 30
-		case ASWPW, ALDADDALW, ALDADDW, ALDANDW, ALDEORW, ALDORW: // 32-bit
+		case ASWPW, ASWPALW, ALDADDALW, ALDADDW, ALDANDW, ALDEORW, ALDORW: // 32-bit
 			o1 = 2 << 30
-		case ASWPH, ALDADDH, ALDANDH, ALDEORH, ALDORH: // 16-bit
+		case ASWPH, ASWPALH, ALDADDH, ALDANDH, ALDEORH, ALDORH: // 16-bit
 			o1 = 1 << 30
-		case ASWPB, ALDADDB, ALDANDB, ALDEORB, ALDORB: // 8-bit
+		case ASWPB, ASWPALB, ALDADDB, ALDANDB, ALDEORB, ALDORB: // 8-bit
 			o1 = 0 << 30
 		default:
 			c.ctxt.Diag("illegal instruction: %v\n", p)
 		}
 		switch p.As {
-		case ASWPD, ASWPW, ASWPH, ASWPB:
+		case ASWPD, ASWPW, ASWPH, ASWPB, ASWPALD, ASWPALW, ASWPALH, ASWPALB:
 			o1 |= 0x20 << 10
 		case ALDADDALD, ALDADDALW, ALDADDD, ALDADDW, ALDADDH, ALDADDB:
 			o1 |= 0x00 << 10
@@ -3407,7 +3411,7 @@ func (c *ctxt7) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			o1 |= 0x0c << 10
 		}
 		switch p.As {
-		case ALDADDALD, ALDADDALW:
+		case ALDADDALD, ALDADDALW, ASWPALD, ASWPALW, ASWPALH, ASWPALB:
 			o1 |= 3 << 22
 		}
 		o1 |= 0x1c1<<21 | uint32(rs&31)<<16 | uint32(rb&31)<<5 | uint32(rt&31)