cmd/internal/obj/arm: add BFC/BFI to arm's assembler

BFC (Bit Field Clear) and BFI (Bit Field Insert) were
introduced in ARMv6T2, and the compiler can use them
to do further optimization.

Change-Id: I5a3fbcd2c2400c9bf4b939da6366c854c744c27f
Reviewed-on: https://go-review.googlesource.com/72891
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

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

@@ -122,10 +122,11 @@ func IsARMMRC(op obj.As) bool {
 	return false
 }
 
-// IsARMBFX reports whether the op is arm.BFX or arm.BFXU
+// IsARMBFX reports whether the op (as defined by an arm.A* constant) is one the
+// BFX-like instructions which are in the form of "op $width, $LSB, (Reg,) Reg".
 func IsARMBFX(op obj.As) bool {
 	switch op {
-	case arm.ABFX, arm.ABFXU:
+	case arm.ABFX, arm.ABFXU, arm.ABFC, arm.ABFI:
 		return true
 	}
 	return false

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

@@ -1029,11 +1029,14 @@ jmp_label_3:
 	SWI	$65535         // ffff00ef
 	SWI	               // 000000ef
 
-// BFX/BFXU
+// BFX/BFXU/BFC/BFI
 	BFX	$16, $8, R1, R2 // BFX $16, R1, $8, R2   // 5124afe7
 	BFX	$29, $2, R8                              // 5881bce7
 	BFXU	$16, $8, R1, R2 // BFXU $16, R1, $8, R2  // 5124efe7
 	BFXU	$29, $2, R8                              // 5881fce7
+	BFC	$29, $2, R8                              // 1f81dee7
+	BFI	$29, $2, R8                              // 1881dee7
+	BFI	$16, $8, R1, R2 // BFI $16, R1, $8, R2   // 1124d7e7
 
 // synthetic arithmatic
 	ADD	$0xffffffaa, R2, R3 // ADD $4294967210, R2, R3   // 55b0e0e30b3082e0

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

@@ -149,6 +149,7 @@ TEXT errors(SB),$0
 	BFX	$-2, $4, R2, R3    // ERROR "wrong width or LSB"
 	BFXU	$4, R2, R5, R2     // ERROR "missing or wrong LSB"
 	BFXU	$4, R2, R5         // ERROR "missing or wrong LSB"
+	BFC	$12, $8, R2, R3    // ERROR "illegal combination"
 	MOVB	R0>>8, R2          // ERROR "illegal shift"
 	MOVH	R0<<16, R2         // ERROR "illegal shift"
 	MOVBS	R0->8, R2          // ERROR "illegal shift"

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

@@ -320,6 +320,8 @@ const (
 
 	ABFX
 	ABFXU
+	ABFC
+	ABFI
 
 	AMULWT
 	AMULWB

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

@@ -131,6 +131,8 @@ var Anames = []string{
 	"XTAHU",
 	"BFX",
 	"BFXU",
+	"BFC",
+	"BFI",
 	"MULWT",
 	"MULWB",
 	"MULBB",

src/cmd/internal/obj/arm/asm5.go

@@ -69,7 +69,7 @@ type Optab struct {
 	param    int16
 	flag     int8
 	pcrelsiz uint8
-	scond    uint8  // optional flags accepted by the instruction
+	scond    uint8 // optional flags accepted by the instruction
 }
 
 type Opcross [32][2][32]uint8
@@ -1679,6 +1679,8 @@ func buildop(ctxt *obj.Link) {
 
 		case ABFX:
 			opset(ABFXU, r0)
+			opset(ABFC, r0)
+			opset(ABFI, r0)
 
 		case ACLZ:
 			opset(AREV, r0)
@@ -2033,12 +2035,14 @@ func (c *ctxt5) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		r := int(p.Reg)
 		o1 |= (uint32(rf)&15)<<8 | (uint32(r)&15)<<0 | (uint32(rt)&15)<<16 | (uint32(rt2)&15)<<12
 
-	case 18: /* BFX/BFXU */
+	case 18: /* BFX/BFXU/BFC/BFI */
 		o1 = c.oprrr(p, p.As, int(p.Scond))
 		rt := int(p.To.Reg)
 		r := int(p.Reg)
 		if r == 0 {
 			r = rt
+		} else if p.As == ABFC { // only "BFC $width, $lsb, Reg" is accepted, p.Reg must be 0
+			c.ctxt.Diag("illegal combination: %v", p)
 		}
 		if p.GetFrom3() == nil || p.GetFrom3().Type != obj.TYPE_CONST {
 			c.ctxt.Diag("%v: missing or wrong LSB", p)
@@ -2046,10 +2050,17 @@ func (c *ctxt5) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		}
 		lsb := p.GetFrom3().Offset
 		width := p.From.Offset
-		if lsb < 0 || lsb > 31 || width <= 0 || (lsb+width) > 31 {
+		if lsb < 0 || lsb > 31 || width <= 0 || (lsb+width) > 32 {
 			c.ctxt.Diag("%v: wrong width or LSB", p)
 		}
-		o1 |= (uint32(r)&15)<<0 | (uint32(rt)&15)<<12 | uint32(lsb)<<7 | uint32(width-1)<<16
+		switch p.As {
+		case ABFX, ABFXU: // (width-1) is encoded
+			o1 |= (uint32(r)&15)<<0 | (uint32(rt)&15)<<12 | uint32(lsb)<<7 | uint32(width-1)<<16
+		case ABFC, ABFI: // MSB is encoded
+			o1 |= (uint32(r)&15)<<0 | (uint32(rt)&15)<<12 | uint32(lsb)<<7 | uint32(lsb+width-1)<<16
+		default:
+			c.ctxt.Diag("illegal combination: %v", p)
+		}
 
 	case 20: /* mov/movb/movbu R,O(R) */
 		c.aclass(&p.To)
@@ -3022,6 +3033,12 @@ func (c *ctxt5) oprrr(p *obj.Prog, a obj.As, sc int) uint32 {
 	case ABFXU:
 		return o | 0x3f<<21 | 0x5<<4
 
+	case ABFC:
+		return o | 0x3e<<21 | 0x1f
+
+	case ABFI:
+		return o | 0x3e<<21 | 0x1<<4
+
 	case AXTAB:
 		return o | 0x6a<<20 | 0x7<<4