cmd/internal/obj/arm: support more ARM VFP instructions

Add support of more ARM VFP instructions in the assembler.
They were introduced in ARM VFPv4.

"FMULAF/FMULAD   Fm, Fn, Fd": Fd = Fd + Fn*Fm
"FNMULAF/FNMULAD Fm, Fn, Fd": Fd = -(Fd + Fn*Fm)
"FMULSF/FMULSD   Fm, Fn, Fd": Fd = Fd - Fn*Fm
"FNMULSF/FNMULSD Fm, Fn, Fd": Fd = -(Fd - Fn*Fm)

The multiplication results are not rounded.

Change-Id: Id9cc52fd8e1b9a708103cd1e514c85a9e1cb3f47
Reviewed-on: https://go-review.googlesource.com/62550Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>

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

@@ -43,6 +43,14 @@ TEXT errors(SB),$0
 	NMULAF	F0, F1             // ERROR "illegal combination"
 	NMULSD	F0, F1             // ERROR "illegal combination"
 	NMULSF	F0, F1             // ERROR "illegal combination"
+	FMULAD	F0, F1             // ERROR "illegal combination"
+	FMULAF	F0, F1             // ERROR "illegal combination"
+	FMULSD	F0, F1             // ERROR "illegal combination"
+	FMULSF	F0, F1             // ERROR "illegal combination"
+	FNMULAD	F0, F1             // ERROR "illegal combination"
+	FNMULAF	F0, F1             // ERROR "illegal combination"
+	FNMULSD	F0, F1             // ERROR "illegal combination"
+	FNMULSF	F0, F1             // ERROR "illegal combination"
 	NEGF	F0, F1, F2         // ERROR "illegal combination"
 	NEGD	F0, F1, F2         // ERROR "illegal combination"
 	ABSF	F0, F1, F2         // ERROR "illegal combination"

src/cmd/asm/internal/asm/testdata/armv6.s

@@ -30,6 +30,14 @@ TEXT	foo(SB), DUPOK|NOSPLIT, $0
 	NMULAD	F5, F6, F7    // 057b16ee
 	NMULSF	F5, F6, F7    // 457a16ee
 	NMULSD	F5, F6, F7    // 457b16ee
+	FMULAF  F5, F6, F7    // 057aa6ee
+	FMULAD  F5, F6, F7    // 057ba6ee
+	FMULSF  F5, F6, F7    // 457aa6ee
+	FMULSD  F5, F6, F7    // 457ba6ee
+	FNMULAF F5, F6, F7    // 457a96ee
+	FNMULAD F5, F6, F7    // 457b96ee
+	FNMULSF F5, F6, F7    // 057a96ee
+	FNMULSD F5, F6, F7    // 057b96ee
 	DIVF	F0, F1, F2    // 002a81ee
 	DIVD.EQ	F3, F4, F5    // 035b840e
 	DIVF.NE	F0, F2        // 002a821e

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

@@ -240,6 +240,14 @@ const (
 	AMULSD
 	ANMULSF
 	ANMULSD
+	AFMULAF
+	AFMULAD
+	AFNMULAF
+	AFNMULAD
+	AFMULSF
+	AFMULSD
+	AFNMULSF
+	AFNMULSD
 	ADIVF
 	ADIVD
 	ASQRTF

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

@@ -63,6 +63,14 @@ var Anames = []string{
 	"MULSD",
 	"NMULSF",
 	"NMULSD",
+	"FMULAF",
+	"FMULAD",
+	"FNMULAF",
+	"FNMULAD",
+	"FMULSF",
+	"FMULSD",
+	"FNMULSF",
+	"FNMULSD",
 	"DIVF",
 	"DIVD",
 	"SQRTF",

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

@@ -1661,6 +1661,14 @@ func buildop(ctxt *obj.Link) {
 			opset(ANMULAD, r0)
 			opset(ANMULSF, r0)
 			opset(ANMULSD, r0)
+			opset(AFMULAF, r0)
+			opset(AFMULAD, r0)
+			opset(AFMULSF, r0)
+			opset(AFMULSD, r0)
+			opset(AFNMULAF, r0)
+			opset(AFNMULAD, r0)
+			opset(AFNMULSF, r0)
+			opset(AFNMULSD, r0)
 			opset(ADIVF, r0)
 			opset(ADIVD, r0)
 
@@ -2270,7 +2278,8 @@ func (c *ctxt5) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		r := int(p.Reg)
 		if r == 0 {
 			switch p.As {
-			case AMULAD, AMULAF, AMULSF, AMULSD, ANMULAF, ANMULAD, ANMULSF, ANMULSD:
+			case AMULAD, AMULAF, AMULSF, AMULSD, ANMULAF, ANMULAD, ANMULSF, ANMULSD,
+				AFMULAD, AFMULAF, AFMULSF, AFMULSD, AFNMULAF, AFNMULAD, AFNMULSF, AFNMULSD:
 				c.ctxt.Diag("illegal combination: %v", p)
 			default:
 				r = rt
@@ -2905,6 +2914,22 @@ func (c *ctxt5) oprrr(p *obj.Prog, a obj.As, sc int) uint32 {
 		return o | 0xe<<24 | 0x1<<20 | 0xb<<8 | 0x4<<4
 	case ANMULSF:
 		return o | 0xe<<24 | 0x1<<20 | 0xa<<8 | 0x4<<4
+	case AFMULAD:
+		return o | 0xe<<24 | 0xa<<20 | 0xb<<8
+	case AFMULAF:
+		return o | 0xe<<24 | 0xa<<20 | 0xa<<8
+	case AFMULSD:
+		return o | 0xe<<24 | 0xa<<20 | 0xb<<8 | 0x4<<4
+	case AFMULSF:
+		return o | 0xe<<24 | 0xa<<20 | 0xa<<8 | 0x4<<4
+	case AFNMULAD:
+		return o | 0xe<<24 | 0x9<<20 | 0xb<<8 | 0x4<<4
+	case AFNMULAF:
+		return o | 0xe<<24 | 0x9<<20 | 0xa<<8 | 0x4<<4
+	case AFNMULSD:
+		return o | 0xe<<24 | 0x9<<20 | 0xb<<8
+	case AFNMULSF:
+		return o | 0xe<<24 | 0x9<<20 | 0xa<<8
 	case ADIVD:
 		return o | 0xe<<24 | 0x8<<20 | 0xb<<8 | 0<<4
 	case ADIVF: