Merge remote-tracking branch 'origin/dev.ssa' into merge

Merging from dev.ssa back into master.

Contains complete SSA backends for arm, arm64, 386, amd64p32.
Work in progress for PPC64.

Change-Id: Ifd7075e3ec6f88f776e29f8c7fd55830328897fd

src/cmd/asm/internal/asm/operand_test.go

@@ -17,6 +17,7 @@ import (
 
 func setArch(goarch string) (*arch.Arch, *obj.Link) {
 	os.Setenv("GOOS", "linux") // obj can handle this OS for all architectures.
+	os.Setenv("GOARCH", goarch)
 	architecture := arch.Set(goarch)
 	if architecture == nil {
 		panic("asm: unrecognized architecture " + goarch)

src/cmd/compile/internal/amd64/ssa.go

@@ -239,89 +239,87 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		}
 		opregreg(v.Op.Asm(), r, gc.SSARegNum(v.Args[1]))
 
-	case ssa.OpAMD64DIVQ, ssa.OpAMD64DIVL, ssa.OpAMD64DIVW,
-		ssa.OpAMD64DIVQU, ssa.OpAMD64DIVLU, ssa.OpAMD64DIVWU,
-		ssa.OpAMD64MODQ, ssa.OpAMD64MODL, ssa.OpAMD64MODW,
-		ssa.OpAMD64MODQU, ssa.OpAMD64MODLU, ssa.OpAMD64MODWU:
+	case ssa.OpAMD64DIVQU, ssa.OpAMD64DIVLU, ssa.OpAMD64DIVWU:
+		// Arg[0] (the dividend) is in AX.
+		// Arg[1] (the divisor) can be in any other register.
+		// Result[0] (the quotient) is in AX.
+		// Result[1] (the remainder) is in DX.
+		r := gc.SSARegNum(v.Args[1])
 
-		// Arg[0] is already in AX as it's the only register we allow
-		// and AX is the only output
-		x := gc.SSARegNum(v.Args[1])
+		// Zero extend dividend.
+		c := gc.Prog(x86.AXORL)
+		c.From.Type = obj.TYPE_REG
+		c.From.Reg = x86.REG_DX
+		c.To.Type = obj.TYPE_REG
+		c.To.Reg = x86.REG_DX
 
-		// CPU faults upon signed overflow, which occurs when most
-		// negative int is divided by -1.
-		var j *obj.Prog
-		if v.Op == ssa.OpAMD64DIVQ || v.Op == ssa.OpAMD64DIVL ||
-			v.Op == ssa.OpAMD64DIVW || v.Op == ssa.OpAMD64MODQ ||
-			v.Op == ssa.OpAMD64MODL || v.Op == ssa.OpAMD64MODW {
+		// Issue divide.
+		p := gc.Prog(v.Op.Asm())
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = r
 
+	case ssa.OpAMD64DIVQ, ssa.OpAMD64DIVL, ssa.OpAMD64DIVW:
+		// Arg[0] (the dividend) is in AX.
+		// Arg[1] (the divisor) can be in any other register.
+		// Result[0] (the quotient) is in AX.
+		// Result[1] (the remainder) is in DX.
+		r := gc.SSARegNum(v.Args[1])
+
+		// CPU faults upon signed overflow, which occurs when the most
+		// negative int is divided by -1. Handle divide by -1 as a special case.
 		var c *obj.Prog
 		switch v.Op {
-			case ssa.OpAMD64DIVQ, ssa.OpAMD64MODQ:
+		case ssa.OpAMD64DIVQ:
 			c = gc.Prog(x86.ACMPQ)
-				j = gc.Prog(x86.AJEQ)
-				// go ahead and sign extend to save doing it later
-				gc.Prog(x86.ACQO)
-
-			case ssa.OpAMD64DIVL, ssa.OpAMD64MODL:
+		case ssa.OpAMD64DIVL:
 			c = gc.Prog(x86.ACMPL)
-				j = gc.Prog(x86.AJEQ)
-				gc.Prog(x86.ACDQ)
-
-			case ssa.OpAMD64DIVW, ssa.OpAMD64MODW:
+		case ssa.OpAMD64DIVW:
 			c = gc.Prog(x86.ACMPW)
-				j = gc.Prog(x86.AJEQ)
-				gc.Prog(x86.ACWD)
 		}
 		c.From.Type = obj.TYPE_REG
-			c.From.Reg = x
+		c.From.Reg = r
 		c.To.Type = obj.TYPE_CONST
 		c.To.Offset = -1
+		j1 := gc.Prog(x86.AJEQ)
+		j1.To.Type = obj.TYPE_BRANCH
 
-			j.To.Type = obj.TYPE_BRANCH
-
-		}
-
-		// for unsigned ints, we sign extend by setting DX = 0
-		// signed ints were sign extended above
-		if v.Op == ssa.OpAMD64DIVQU || v.Op == ssa.OpAMD64MODQU ||
-			v.Op == ssa.OpAMD64DIVLU || v.Op == ssa.OpAMD64MODLU ||
-			v.Op == ssa.OpAMD64DIVWU || v.Op == ssa.OpAMD64MODWU {
-			c := gc.Prog(x86.AXORQ)
-			c.From.Type = obj.TYPE_REG
-			c.From.Reg = x86.REG_DX
-			c.To.Type = obj.TYPE_REG
-			c.To.Reg = x86.REG_DX
+		// Sign extend dividend.
+		switch v.Op {
+		case ssa.OpAMD64DIVQ:
+			gc.Prog(x86.ACQO)
+		case ssa.OpAMD64DIVL:
+			gc.Prog(x86.ACDQ)
+		case ssa.OpAMD64DIVW:
+			gc.Prog(x86.ACWD)
 		}
 
+		// Issue divide.
 		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
-		p.From.Reg = x
+		p.From.Reg = r
 
-		// signed division, rest of the check for -1 case
-		if j != nil {
+		// Skip over -1 fixup code.
 		j2 := gc.Prog(obj.AJMP)
 		j2.To.Type = obj.TYPE_BRANCH
 
-			var n *obj.Prog
-			if v.Op == ssa.OpAMD64DIVQ || v.Op == ssa.OpAMD64DIVL ||
-				v.Op == ssa.OpAMD64DIVW {
-				// n * -1 = -n
-				n = gc.Prog(x86.ANEGQ)
-				n.To.Type = obj.TYPE_REG
-				n.To.Reg = x86.REG_AX
-			} else {
+		// Issue -1 fixup code.
+		// n / -1 = -n
+		n1 := gc.Prog(x86.ANEGQ)
+		n1.To.Type = obj.TYPE_REG
+		n1.To.Reg = x86.REG_AX
+
 		// n % -1 == 0
-				n = gc.Prog(x86.AXORQ)
-				n.From.Type = obj.TYPE_REG
-				n.From.Reg = x86.REG_DX
-				n.To.Type = obj.TYPE_REG
-				n.To.Reg = x86.REG_DX
-			}
+		n2 := gc.Prog(x86.AXORL)
+		n2.From.Type = obj.TYPE_REG
+		n2.From.Reg = x86.REG_DX
+		n2.To.Type = obj.TYPE_REG
+		n2.To.Reg = x86.REG_DX
+
+		// TODO(khr): issue only the -1 fixup code we need.
+		// For instance, if only the quotient is used, no point in zeroing the remainder.
 
-			j.To.Val = n
+		j1.To.Val = n1
 		j2.To.Val = s.Pc()
-		}
 
 	case ssa.OpAMD64HMULQ, ssa.OpAMD64HMULL, ssa.OpAMD64HMULW, ssa.OpAMD64HMULB,
 		ssa.OpAMD64HMULQU, ssa.OpAMD64HMULLU, ssa.OpAMD64HMULWU, ssa.OpAMD64HMULBU:
@@ -500,8 +498,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		gc.AddAux(&p.From, v)
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = gc.SSARegNum(v)
-	case ssa.OpAMD64LEAQ:
-		p := gc.Prog(x86.ALEAQ)
+	case ssa.OpAMD64LEAQ, ssa.OpAMD64LEAL:
+		p := gc.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_MEM
 		p.From.Reg = gc.SSARegNum(v.Args[0])
 		gc.AddAux(&p.From, v)
@@ -705,7 +703,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.To.Sym = gc.Linksym(gc.Pkglookup("duffcopy", gc.Runtimepkg))
 		p.To.Offset = v.AuxInt
 
-	case ssa.OpCopy, ssa.OpAMD64MOVQconvert: // TODO: use MOVQreg for reg->reg copies instead of OpCopy?
+	case ssa.OpCopy, ssa.OpAMD64MOVQconvert, ssa.OpAMD64MOVLconvert: // TODO: use MOVQreg for reg->reg copies instead of OpCopy?
 		if v.Type.IsMemory() {
 			return
 		}
@@ -754,27 +752,14 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 			p.To.Name = obj.NAME_AUTO
 		}
 	case ssa.OpPhi:
-		// just check to make sure regalloc and stackalloc did it right
-		if v.Type.IsMemory() {
-			return
-		}
-		f := v.Block.Func
-		loc := f.RegAlloc[v.ID]
-		for _, a := range v.Args {
-			if aloc := f.RegAlloc[a.ID]; aloc != loc { // TODO: .Equal() instead?
-				v.Fatalf("phi arg at different location than phi: %v @ %v, but arg %v @ %v\n%s\n", v, loc, a, aloc, v.Block.Func)
-			}
-		}
+		gc.CheckLoweredPhi(v)
 	case ssa.OpInitMem:
 		// memory arg needs no code
 	case ssa.OpArg:
 		// input args need no code
 	case ssa.OpAMD64LoweredGetClosurePtr:
-		// Output is hardwired to DX only,
-		// and DX contains the closure pointer on
-		// closure entry, and this "instruction"
-		// is scheduled to the very beginning
-		// of the entry block.
+		// Closure pointer is DX.
+		gc.CheckLoweredGetClosurePtr(v)
 	case ssa.OpAMD64LoweredGetG:
 		r := gc.SSARegNum(v)
 		// See the comments in cmd/internal/obj/x86/obj6.go
@@ -871,6 +856,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.To.Reg = gc.SSARegNum(v)
 	case ssa.OpSP, ssa.OpSB:
 		// nothing to do
+	case ssa.OpSelect0, ssa.OpSelect1:
+		// nothing to do
 	case ssa.OpAMD64SETEQ, ssa.OpAMD64SETNE,
 		ssa.OpAMD64SETL, ssa.OpAMD64SETLE,
 		ssa.OpAMD64SETG, ssa.OpAMD64SETGE,

src/cmd/compile/internal/arm/prog.go

@@ -79,6 +79,8 @@ var progtable = [arm.ALAST & obj.AMask]obj.ProgInfo{
 	arm.AMULF & obj.AMask:  {Flags: gc.SizeF | gc.LeftRead | RightRdwr},
 	arm.ASUBD & obj.AMask:  {Flags: gc.SizeD | gc.LeftRead | RightRdwr},
 	arm.ASUBF & obj.AMask:  {Flags: gc.SizeF | gc.LeftRead | RightRdwr},
+	arm.ANEGD & obj.AMask:  {Flags: gc.SizeD | gc.LeftRead | RightRdwr},
+	arm.ANEGF & obj.AMask:  {Flags: gc.SizeF | gc.LeftRead | RightRdwr},
 	arm.ASQRTD & obj.AMask: {Flags: gc.SizeD | gc.LeftRead | RightRdwr},
 
 	// Conversions.

src/cmd/compile/internal/arm64/galign.go

@@ -6,6 +6,7 @@ package arm64
 
 import (
 	"cmd/compile/internal/gc"
+	"cmd/compile/internal/ssa"
 	"cmd/internal/obj/arm64"
 )
 
@@ -61,6 +62,11 @@ func Main() {
 	gc.Thearch.Doregbits = doregbits
 	gc.Thearch.Regnames = regnames
 
+	gc.Thearch.SSARegToReg = ssaRegToReg
+	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}
+	gc.Thearch.SSAGenValue = ssaGenValue
+	gc.Thearch.SSAGenBlock = ssaGenBlock
+
 	gc.Main()
 	gc.Exit(0)
 }

src/cmd/compile/internal/arm64/prog.go

@@ -44,24 +44,37 @@ var progtable = [arm64.ALAST & obj.AMask]obj.ProgInfo{
 	// Integer
 	arm64.AADD & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.ASUB & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
-	arm64.ANEG & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.ANEG & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite}, // why RegRead? revisit once the old backend gone
 	arm64.AAND & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.AORR & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.AEOR & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.ABIC & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AMVN & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RightWrite},
 	arm64.AMUL & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AMULW & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.ASMULL & obj.AMask: {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.AUMULL & obj.AMask: {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
-	arm64.ASMULH & obj.AMask: {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
-	arm64.AUMULH & obj.AMask: {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.ASMULH & obj.AMask: {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AUMULH & obj.AMask: {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.ASDIV & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.AUDIV & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.ASDIVW & obj.AMask: {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AUDIVW & obj.AMask: {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AREM & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AUREM & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AREMW & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.AUREMW & obj.AMask: {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.ALSL & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.ALSR & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.AASR & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.ACMP & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead},
+	arm64.ACMPW & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RegRead},
 	arm64.AADC & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite | gc.UseCarry},
 	arm64.AROR & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	arm64.ARORW & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	arm64.AADDS & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite | gc.SetCarry},
+	arm64.ACSET & obj.AMask:  {Flags: gc.SizeQ | gc.RightWrite},
+	arm64.ACSEL & obj.AMask:  {Flags: gc.SizeQ | gc.RegRead | gc.RightWrite},
 
 	// Floating point.
 	arm64.AFADDD & obj.AMask:  {Flags: gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite},

src/cmd/compile/internal/gc/builtin.go

@@ -95,14 +95,15 @@ const runtimeimport = "" +
 	"4div\x00\x03\n\x00\n\x00\x01\n\x00\t\x11uint64div\x00\x03\x14\x00\x14\x00\x01\x14\x00\t\x0fint64" +
 	"mod\x00\x03\n\x00\n\x00\x01\n\x00\t\x11uint64mod\x00\x03\x14\x00\x14\x00\x01\x14\x00\t\x1bfloat6" +
 	"4toint64\x00\x01\x1a\x00\x01\n\x00\t\x1dfloat64touint64\x00\x01\x1a\x00\x01\x14\x00\t" +
-	"\x1bint64tofloat64\x00\x01\n\x00\x01\x1a\x00\t\x1duint64tofloat64\x00" +
-	"\x01\x14\x00\x01\x1a\x00\t\x19complex128div\x00\x04\x1e\vnum·2\x00\x00\x1e\vden·" +
-	"3\x00\x00\x02\x1e\vquo·1\x00\x00\t\x19racefuncenter\x00\x01\x16d\x00\t\x17race" +
-	"funcexit\x00\x00\x00\t\x0fraceread\x00\x01\x16d\x00\t\x11racewrite\x00\x01\x16" +
-	"d\x00\t\x19racereadrange\x00\x04\x16\raddr·1\x00d\x16\rsize·2\x00" +
-	"d\x00\t\x1bracewriterange\x00\x04\x16\x94\x03\x00d\x16\x96\x03\x00d\x00\t\x0fmsanrea" +
-	"d\x00\x04\x16\x94\x03\x00d\x16\x96\x03\x00d\x00\t\x11msanwrite\x00\x04\x16\x94\x03\x00d\x16\x96\x03\x00d\x00\v\xf4" +
-	"\x01\x02\v\x00\x01\x00\n$$\n"
+	"\x1dfloat64touint32\x00\x01\x1a\x00\x01\x12\x00\t\x1bint64tofloat64\x00" +
+	"\x01\n\x00\x01\x1a\x00\t\x1duint64tofloat64\x00\x01\x14\x00\x01\x1a\x00\t\x1duint32to" +
+	"float64\x00\x01\x12\x00\x01\x1a\x00\t\x19complex128div\x00\x04\x1e\vnum·2\x00" +
+	"\x00\x1e\vden·3\x00\x00\x02\x1e\vquo·1\x00\x00\t\x19racefuncenter\x00\x01\x16" +
+	"d\x00\t\x17racefuncexit\x00\x00\x00\t\x0fraceread\x00\x01\x16d\x00\t\x11race" +
+	"write\x00\x01\x16d\x00\t\x19racereadrange\x00\x04\x16\raddr·1\x00d\x16\r" +
+	"size·2\x00d\x00\t\x1bracewriterange\x00\x04\x16\x98\x03\x00d\x16\x9a\x03\x00d\x00\t" +
+	"\x0fmsanread\x00\x04\x16\x98\x03\x00d\x16\x9a\x03\x00d\x00\t\x11msanwrite\x00\x04\x16\x98\x03\x00d" +
+	"\x16\x9a\x03\x00d\x00\v\xf8\x01\x02\v\x00\x01\x00\n$$\n"
 
 const unsafeimport = "" +
 	"cn\x00\x03v1\x01\vunsafe\x00\x05\r\rPointer\x00\x16\x00\t\x0fOffsetof\x00\x01" +

src/cmd/compile/internal/gc/builtin/runtime.go

@@ -150,8 +150,10 @@ func int64mod(int64, int64) int64
 func uint64mod(uint64, uint64) uint64
 func float64toint64(float64) int64
 func float64touint64(float64) uint64
+func float64touint32(float64) uint32
 func int64tofloat64(int64) float64
 func uint64tofloat64(uint64) float64
+func uint32tofloat64(uint32) float64
 
 func complex128div(num complex128, den complex128) (quo complex128)
 

src/cmd/compile/internal/gc/testdata/string_ssa.go

@@ -110,6 +110,67 @@ func testSmallIndexType() {
 	}
 }
 
+//go:noinline
+func testInt64Index_ssa(s string, i int64) byte {
+	return s[i]
+}
+
+//go:noinline
+func testInt64Slice_ssa(s string, i, j int64) string {
+	return s[i:j]
+}
+
+func testInt64Index() {
+	tests := []struct {
+		i int64
+		j int64
+		b byte
+		s string
+	}{
+		{0, 5, 'B', "Below"},
+		{5, 10, 'E', "Exact"},
+		{10, 15, 'A', "Above"},
+	}
+
+	str := "BelowExactAbove"
+	for i, t := range tests {
+		if got := testInt64Index_ssa(str, t.i); got != t.b {
+			println("#", i, "got ", got, ", wanted", t.b)
+			failed = true
+		}
+		if got := testInt64Slice_ssa(str, t.i, t.j); got != t.s {
+			println("#", i, "got ", got, ", wanted", t.s)
+			failed = true
+		}
+	}
+}
+
+func testInt64IndexPanic() {
+	defer func() {
+		if r := recover(); r != nil {
+			println("paniced as expected")
+		}
+	}()
+
+	str := "foobar"
+	println("got ", testInt64Index_ssa(str, 1<<32+1))
+	println("expected to panic, but didn't")
+	failed = true
+}
+
+func testInt64SlicePanic() {
+	defer func() {
+		if r := recover(); r != nil {
+			println("paniced as expected")
+		}
+	}()
+
+	str := "foobar"
+	println("got ", testInt64Slice_ssa(str, 1<<32, 1<<32+1))
+	println("expected to panic, but didn't")
+	failed = true
+}
+
 //go:noinline
 func testStringElem_ssa(s string, i int) byte {
 	return s[i]
@@ -153,6 +214,9 @@ func main() {
 	testSmallIndexType()
 	testStringElem()
 	testStringElemConst()
+	testInt64Index()
+	testInt64IndexPanic()
+	testInt64SlicePanic()
 
 	if failed {
 		panic("failed")

src/cmd/compile/internal/gc/type.go

@@ -1207,6 +1207,7 @@ func (t *Type) ChanDir() ChanDir {
 func (t *Type) IsMemory() bool { return false }
 func (t *Type) IsFlags() bool  { return false }
 func (t *Type) IsVoid() bool   { return false }
+func (t *Type) IsTuple() bool  { return false }
 
 // IsUntyped reports whether t is an untyped type.
 func (t *Type) IsUntyped() bool {

src/cmd/compile/internal/gc/walk.go

@@ -1094,12 +1094,45 @@ opswitch:
 
 			if n.Type.IsFloat() {
 				if n.Left.Type.Etype == TINT64 {
-					n = mkcall("int64tofloat64", n.Type, init, conv(n.Left, Types[TINT64]))
+					n = conv(mkcall("int64tofloat64", Types[TFLOAT64], init, conv(n.Left, Types[TINT64])), n.Type)
 					break
 				}
 
 				if n.Left.Type.Etype == TUINT64 {
-					n = mkcall("uint64tofloat64", n.Type, init, conv(n.Left, Types[TUINT64]))
+					n = conv(mkcall("uint64tofloat64", Types[TFLOAT64], init, conv(n.Left, Types[TUINT64])), n.Type)
+					break
+				}
+			}
+		}
+
+		if Thearch.LinkArch.Family == sys.I386 {
+			if n.Left.Type.IsFloat() {
+				if n.Type.Etype == TINT64 {
+					n = mkcall("float64toint64", n.Type, init, conv(n.Left, Types[TFLOAT64]))
+					break
+				}
+
+				if n.Type.Etype == TUINT64 {
+					n = mkcall("float64touint64", n.Type, init, conv(n.Left, Types[TFLOAT64]))
+					break
+				}
+				if n.Type.Etype == TUINT32 || n.Type.Etype == TUINTPTR {
+					n = mkcall("float64touint32", n.Type, init, conv(n.Left, Types[TFLOAT64]))
+					break
+				}
+			}
+			if n.Type.IsFloat() {
+				if n.Left.Type.Etype == TINT64 {
+					n = conv(mkcall("int64tofloat64", Types[TFLOAT64], init, conv(n.Left, Types[TINT64])), n.Type)
+					break
+				}
+
+				if n.Left.Type.Etype == TUINT64 {
+					n = conv(mkcall("uint64tofloat64", Types[TFLOAT64], init, conv(n.Left, Types[TUINT64])), n.Type)
+					break
+				}
+				if n.Left.Type.Etype == TUINT32 || n.Left.Type.Etype == TUINTPTR {
+					n = conv(mkcall("uint32tofloat64", Types[TFLOAT64], init, conv(n.Left, Types[TUINT32])), n.Type)
 					break
 				}
 			}
@@ -3303,6 +3336,7 @@ func samecheap(a *Node, b *Node) bool {
 // The result of walkrotate MUST be assigned back to n, e.g.
 // 	n.Left = walkrotate(n.Left)
 func walkrotate(n *Node) *Node {
+	//TODO: enable LROT on ARM64 once the old backend is gone
 	if Thearch.LinkArch.InFamily(sys.MIPS64, sys.ARM64, sys.PPC64) {
 		return n
 	}
@@ -3496,16 +3530,6 @@ func walkdiv(n *Node, init *Nodes) *Node {
 			goto ret
 		}
 
-		// TODO(zhongwei) Test shows that TUINT8, TINT8, TUINT16 and TINT16's "quick division" method
-		// on current arm64 backend is slower than hardware div instruction on ARM64 due to unnecessary
-		// data movement between registers. It could be enabled when generated code is good enough.
-		if Thearch.LinkArch.Family == sys.ARM64 {
-			switch Simtype[nl.Type.Etype] {
-			case TUINT8, TINT8, TUINT16, TINT16:
-				return n
-			}
-		}
-
 		switch Simtype[nl.Type.Etype] {
 		default:
 			return n

src/cmd/compile/internal/ppc64/galign.go

@@ -66,6 +66,11 @@ func Main() {
 	gc.Thearch.Doregbits = doregbits
 	gc.Thearch.Regnames = regnames
 
+	gc.Thearch.SSARegToReg = ssaRegToReg
+	gc.Thearch.SSAMarkMoves = ssaMarkMoves
+	gc.Thearch.SSAGenValue = ssaGenValue
+	gc.Thearch.SSAGenBlock = ssaGenBlock
+
 	initvariants()
 	initproginfo()
 

src/cmd/compile/internal/ppc64/prog.go

@@ -42,22 +42,34 @@ var progtable = [ppc64.ALAST & obj.AMask]obj.ProgInfo{
 
 	// Integer
 	ppc64.AADD & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.AADDC & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ASUB & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.AADDME & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ANEG & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AAND & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.AANDN & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AOR & obj.AMask:     {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.AORN & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AXOR & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.AEQV & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AMULLD & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AMULLW & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AMULHD & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AMULHDU & obj.AMask: {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ADIVD & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ADIVDU & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.ADIVW & obj.AMask:   {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.ADIVWU & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ASLD & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ASRD & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ASRAD & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.ASLW & obj.AMask:    {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.ASRW & obj.AMask:    {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.ASRAW & obj.AMask:   {Flags: gc.SizeL | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.ACMP & obj.AMask:    {Flags: gc.SizeQ | gc.LeftRead | gc.RightRead},
 	ppc64.ACMPU & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RightRead},
+	ppc64.ACMPW & obj.AMask:   {Flags: gc.SizeL | gc.LeftRead | gc.RightRead},
+	ppc64.ACMPWU & obj.AMask:  {Flags: gc.SizeL | gc.LeftRead | gc.RightRead},
 	ppc64.ATD & obj.AMask:     {Flags: gc.SizeQ | gc.RightRead},
 
 	// Floating point.
@@ -70,11 +82,13 @@ var progtable = [ppc64.ALAST & obj.AMask]obj.ProgInfo{
 	ppc64.AFDIV & obj.AMask:   {Flags: gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AFDIVS & obj.AMask:  {Flags: gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AFCTIDZ & obj.AMask: {Flags: gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite},
+	ppc64.AFCTIWZ & obj.AMask: {Flags: gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AFCFID & obj.AMask:  {Flags: gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AFCFIDU & obj.AMask: {Flags: gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite},
 	ppc64.AFCMPU & obj.AMask:  {Flags: gc.SizeD | gc.LeftRead | gc.RightRead},
 	ppc64.AFRSP & obj.AMask:   {Flags: gc.SizeD | gc.LeftRead | gc.RightWrite | gc.Conv},
 	ppc64.AFSQRT & obj.AMask:  {Flags: gc.SizeD | gc.LeftRead | gc.RightWrite},
+	ppc64.AFNEG & obj.AMask:   {Flags: gc.SizeD | gc.LeftRead | gc.RightWrite},
 
 	// Moves
 	ppc64.AMOVB & obj.AMask:  {Flags: gc.SizeB | gc.LeftRead | gc.RightWrite | gc.Move | gc.Conv},
@@ -91,6 +105,8 @@ var progtable = [ppc64.ALAST & obj.AMask]obj.ProgInfo{
 	ppc64.AMOVD & obj.AMask:   {Flags: gc.SizeQ | gc.LeftRead | gc.RightWrite | gc.Move},
 	ppc64.AMOVDU & obj.AMask:  {Flags: gc.SizeQ | gc.LeftRead | gc.RightWrite | gc.Move | gc.PostInc},
 	ppc64.AFMOVS & obj.AMask:  {Flags: gc.SizeF | gc.LeftRead | gc.RightWrite | gc.Move | gc.Conv},
+	ppc64.AFMOVSX & obj.AMask: {Flags: gc.SizeF | gc.LeftRead | gc.RightWrite | gc.Move | gc.Conv},
+	ppc64.AFMOVSZ & obj.AMask: {Flags: gc.SizeF | gc.LeftRead | gc.RightWrite | gc.Move | gc.Conv},
 	ppc64.AFMOVD & obj.AMask:  {Flags: gc.SizeD | gc.LeftRead | gc.RightWrite | gc.Move},
 
 	// Jumps

src/cmd/compile/internal/ssa/config.go

@@ -20,11 +20,18 @@ type Config struct {
 	lowerBlock      func(*Block) bool          // lowering function
 	lowerValue      func(*Value, *Config) bool // lowering function
 	registers       []Register                 // machine registers
+	gpRegMask       regMask                    // general purpose integer register mask
+	fpRegMask       regMask                    // floating point register mask
+	FPReg           int8                       // register number of frame pointer, -1 if not used
+	hasGReg         bool                       // has hardware g register
 	fe              Frontend                   // callbacks into compiler frontend
 	HTML            *HTMLWriter                // html writer, for debugging
 	ctxt            *obj.Link                  // Generic arch information
 	optimize        bool                       // Do optimization
 	noDuffDevice    bool                       // Don't use Duff's device
+	nacl            bool                       // GOOS=nacl
+	use387          bool                       // GO386=387
+	NeedsFpScratch  bool                       // No direct move between GP and FP register sets
 	sparsePhiCutoff uint64                     // Sparse phi location algorithm used above this #blocks*#variables score
 	curFunc         *Func
 
@@ -106,6 +113,7 @@ type Frontend interface {
 	SplitSlice(LocalSlot) (LocalSlot, LocalSlot, LocalSlot)
 	SplitComplex(LocalSlot) (LocalSlot, LocalSlot)
 	SplitStruct(LocalSlot, int) LocalSlot
+	SplitInt64(LocalSlot) (LocalSlot, LocalSlot) // returns (hi, lo)
 
 	// Line returns a string describing the given line number.
 	Line(int32) string
@@ -128,29 +136,87 @@ func NewConfig(arch string, fe Frontend, ctxt *obj.Link, optimize bool) *Config
 		c.lowerBlock = rewriteBlockAMD64
 		c.lowerValue = rewriteValueAMD64
 		c.registers = registersAMD64[:]
-	case "386":
+		c.gpRegMask = gpRegMaskAMD64
+		c.fpRegMask = fpRegMaskAMD64
+		c.FPReg = framepointerRegAMD64
+		c.hasGReg = false
+	case "amd64p32":
 		c.IntSize = 4
 		c.PtrSize = 4
 		c.lowerBlock = rewriteBlockAMD64
-		c.lowerValue = rewriteValueAMD64 // TODO(khr): full 32-bit support
+		c.lowerValue = rewriteValueAMD64
+		c.registers = registersAMD64[:]
+		c.gpRegMask = gpRegMaskAMD64
+		c.fpRegMask = fpRegMaskAMD64
+		c.FPReg = framepointerRegAMD64
+		c.hasGReg = false
+		c.noDuffDevice = true
+	case "386":
+		c.IntSize = 4
+		c.PtrSize = 4
+		c.lowerBlock = rewriteBlock386
+		c.lowerValue = rewriteValue386
+		c.registers = registers386[:]
+		c.gpRegMask = gpRegMask386
+		c.fpRegMask = fpRegMask386
+		c.FPReg = framepointerReg386
+		c.hasGReg = false
 	case "arm":
 		c.IntSize = 4
 		c.PtrSize = 4
 		c.lowerBlock = rewriteBlockARM
 		c.lowerValue = rewriteValueARM
 		c.registers = registersARM[:]
+		c.gpRegMask = gpRegMaskARM
+		c.fpRegMask = fpRegMaskARM
+		c.FPReg = framepointerRegARM
+		c.hasGReg = true
+	case "arm64":
+		c.IntSize = 8
+		c.PtrSize = 8
+		c.lowerBlock = rewriteBlockARM64
+		c.lowerValue = rewriteValueARM64
+		c.registers = registersARM64[:]
+		c.gpRegMask = gpRegMaskARM64
+		c.fpRegMask = fpRegMaskARM64
+		c.FPReg = framepointerRegARM64
+		c.hasGReg = true
+	case "ppc64le":
+		c.IntSize = 8
+		c.PtrSize = 8
+		c.lowerBlock = rewriteBlockPPC64
+		c.lowerValue = rewriteValuePPC64
+		c.registers = registersPPC64[:]
+		c.gpRegMask = gpRegMaskPPC64
+		c.fpRegMask = fpRegMaskPPC64
+		c.FPReg = framepointerRegPPC64
+		c.noDuffDevice = true // TODO: Resolve PPC64 DuffDevice (has zero, but not copy)
+		c.NeedsFpScratch = true
+		c.hasGReg = true
 	default:
 		fe.Unimplementedf(0, "arch %s not implemented", arch)
 	}
 	c.ctxt = ctxt
 	c.optimize = optimize
+	c.nacl = obj.Getgoos() == "nacl"
 
-	// Don't use Duff's device on Plan 9, because floating
+	// Don't use Duff's device on Plan 9 AMD64, because floating
 	// point operations are not allowed in note handler.
-	if obj.Getgoos() == "plan9" {
+	if obj.Getgoos() == "plan9" && arch == "amd64" {
 		c.noDuffDevice = true
 	}
 
+	if c.nacl {
+		c.noDuffDevice = true // Don't use Duff's device on NaCl
+
+		// ARM assembler rewrites DIV/MOD to runtime calls, which
+		// clobber R12 on nacl
+		opcodeTable[OpARMDIV].reg.clobbers |= 1 << 12  // R12
+		opcodeTable[OpARMDIVU].reg.clobbers |= 1 << 12 // R12
+		opcodeTable[OpARMMOD].reg.clobbers |= 1 << 12  // R12
+		opcodeTable[OpARMMODU].reg.clobbers |= 1 << 12 // R12
+	}
+
 	// Assign IDs to preallocated values/blocks.
 	for i := range c.values {
 		c.values[i].ID = ID(i)
@@ -180,6 +246,11 @@ func NewConfig(arch string, fe Frontend, ctxt *obj.Link, optimize bool) *Config
 	return c
 }
 
+func (c *Config) Set387(b bool) {
+	c.NeedsFpScratch = b
+	c.use387 = b
+}
+
 func (c *Config) Frontend() Frontend      { return c.fe }
 func (c *Config) SparsePhiCutoff() uint64 { return c.sparsePhiCutoff }
 

src/cmd/compile/internal/ssa/cse.go

@@ -163,6 +163,29 @@ func cse(f *Func) {
 		}
 	}
 
+	// if we rewrite a tuple generator to a new one in a different block,
+	// copy its selectors to the new generator's block, so tuple generator
+	// and selectors stay together.
+	for _, b := range f.Blocks {
+		for _, v := range b.Values {
+			if rewrite[v.ID] != nil {
+				continue
+			}
+			if v.Op != OpSelect0 && v.Op != OpSelect1 {
+				continue
+			}
+			if !v.Args[0].Type.IsTuple() {
+				f.Fatalf("arg of tuple selector %s is not a tuple: %s", v.String(), v.Args[0].LongString())
+			}
+			t := rewrite[v.Args[0].ID]
+			if t != nil && t.Block != b {
+				// v.Args[0] is tuple generator, CSE'd into a different block as t, v is left behind
+				c := v.copyInto(t.Block)
+				rewrite[v.ID] = c
+			}
+		}
+	}
+
 	rewrites := int64(0)
 
 	// Apply substitutions

src/cmd/compile/internal/ssa/deadstore.go

@@ -89,7 +89,7 @@ func dse(f *Func) {
 				} else {
 					// zero addr mem
 					sz := v.Args[0].Type.ElemType().Size()
-					if v.AuxInt != sz {
+					if SizeAndAlign(v.AuxInt).Size() != sz {
 						f.Fatalf("mismatched zero/store sizes: %d and %d [%s]",
 							v.AuxInt, sz, v.LongString())
 					}

src/cmd/compile/internal/ssa/decompose.go

@@ -25,6 +25,22 @@ func decomposeBuiltIn(f *Func) {
 	for _, name := range f.Names {
 		t := name.Type
 		switch {
+		case t.IsInteger() && t.Size() == 8 && f.Config.IntSize == 4:
+			var elemType Type
+			if t.IsSigned() {
+				elemType = f.Config.fe.TypeInt32()
+			} else {
+				elemType = f.Config.fe.TypeUInt32()
+			}
+			hiName, loName := f.Config.fe.SplitInt64(name)
+			newNames = append(newNames, hiName, loName)
+			for _, v := range f.NamedValues[name] {
+				hi := v.Block.NewValue1(v.Line, OpInt64Hi, elemType, v)
+				lo := v.Block.NewValue1(v.Line, OpInt64Lo, f.Config.fe.TypeUInt32(), v)
+				f.NamedValues[hiName] = append(f.NamedValues[hiName], hi)
+				f.NamedValues[loName] = append(f.NamedValues[loName], lo)
+			}
+			delete(f.NamedValues, name)
 		case t.IsComplex():
 			var elemType Type
 			if t.Size() == 16 {
@@ -78,6 +94,8 @@ func decomposeBuiltIn(f *Func) {
 				f.NamedValues[dataName] = append(f.NamedValues[dataName], data)
 			}
 			delete(f.NamedValues, name)
+		case t.IsFloat():
+			// floats are never decomposed, even ones bigger than IntSize
 		case t.Size() > f.Config.IntSize:
 			f.Unimplementedf("undecomposed named type %s %s", name, t)
 		default:
@@ -88,8 +106,13 @@ func decomposeBuiltIn(f *Func) {
 }
 
 func decomposeBuiltInPhi(v *Value) {
-	// TODO: decompose 64-bit ops on 32-bit archs?
 	switch {
+	case v.Type.IsInteger() && v.Type.Size() == 8 && v.Block.Func.Config.IntSize == 4:
+		if v.Block.Func.Config.arch == "amd64p32" {
+			// Even though ints are 32 bits, we have 64-bit ops.
+			break
+		}
+		decomposeInt64Phi(v)
 	case v.Type.IsComplex():
 		decomposeComplexPhi(v)
 	case v.Type.IsString():
@@ -98,6 +121,8 @@ func decomposeBuiltInPhi(v *Value) {
 		decomposeSlicePhi(v)
 	case v.Type.IsInterface():
 		decomposeInterfacePhi(v)
+	case v.Type.IsFloat():
+		// floats are never decomposed, even ones bigger than IntSize
 	case v.Type.Size() > v.Block.Func.Config.IntSize:
 		v.Unimplementedf("undecomposed type %s", v.Type)
 	}
@@ -138,6 +163,26 @@ func decomposeSlicePhi(v *Value) {
 	v.AddArg(cap)
 }
 
+func decomposeInt64Phi(v *Value) {
+	fe := v.Block.Func.Config.fe
+	var partType Type
+	if v.Type.IsSigned() {
+		partType = fe.TypeInt32()
+	} else {
+		partType = fe.TypeUInt32()
+	}
+
+	hi := v.Block.NewValue0(v.Line, OpPhi, partType)
+	lo := v.Block.NewValue0(v.Line, OpPhi, fe.TypeUInt32())
+	for _, a := range v.Args {
+		hi.AddArg(a.Block.NewValue1(v.Line, OpInt64Hi, partType, a))
+		lo.AddArg(a.Block.NewValue1(v.Line, OpInt64Lo, fe.TypeUInt32(), a))
+	}
+	v.reset(OpInt64Make)
+	v.AddArg(hi)
+	v.AddArg(lo)
+}
+
 func decomposeComplexPhi(v *Value) {
 	fe := v.Block.Func.Config.fe
 	var partType Type

src/cmd/compile/internal/ssa/export_test.go

@@ -49,6 +49,12 @@ func (d DummyFrontend) SplitComplex(s LocalSlot) (LocalSlot, LocalSlot) {
 	}
 	return LocalSlot{s.N, d.TypeFloat32(), s.Off}, LocalSlot{s.N, d.TypeFloat32(), s.Off + 4}
 }
+func (d DummyFrontend) SplitInt64(s LocalSlot) (LocalSlot, LocalSlot) {
+	if s.Type.IsSigned() {
+		return LocalSlot{s.N, d.TypeInt32(), s.Off + 4}, LocalSlot{s.N, d.TypeUInt32(), s.Off}
+	}
+	return LocalSlot{s.N, d.TypeUInt32(), s.Off + 4}, LocalSlot{s.N, d.TypeUInt32(), s.Off}
+}
 func (d DummyFrontend) SplitStruct(s LocalSlot, i int) LocalSlot {
 	return LocalSlot{s.N, s.Type.FieldType(i), s.Off + s.Type.FieldOff(i)}
 }

src/cmd/compile/internal/ssa/flagalloc.go

@@ -4,8 +4,6 @@
 
 package ssa
 
-const flagRegMask = regMask(1) << 33 // TODO: arch-specific
-
 // flagalloc allocates the flag register among all the flag-generating
 // instructions. Flag values are recomputed if they need to be
 // spilled/restored.
@@ -33,7 +31,7 @@ func flagalloc(f *Func) {
 				if v == flag {
 					flag = nil
 				}
-				if opcodeTable[v.Op].reg.clobbers&flagRegMask != 0 {
+				if opcodeTable[v.Op].clobberFlags {
 					flag = nil
 				}
 				for _, a := range v.Args {
@@ -97,7 +95,7 @@ func flagalloc(f *Func) {
 					continue
 				}
 				// Recalculate a
-				c := a.copyInto(b)
+				c := copyFlags(a, b)
 				// Update v.
 				v.SetArg(i, c)
 				// Remember the most-recently computed flag value.
@@ -105,7 +103,7 @@ func flagalloc(f *Func) {
 			}
 			// Issue v.
 			b.Values = append(b.Values, v)
-			if opcodeTable[v.Op].reg.clobbers&flagRegMask != 0 {
+			if opcodeTable[v.Op].clobberFlags {
 				flag = nil
 			}
 			if v.Type.IsFlags() {
@@ -121,7 +119,7 @@ func flagalloc(f *Func) {
 		if v := end[b.ID]; v != nil && v != flag {
 			// Need to reissue flag generator for use by
 			// subsequent blocks.
-			_ = v.copyInto(b)
+			copyFlags(v, b)
 			// Note: this flag generator is not properly linked up
 			// with the flag users. This breaks the SSA representation.
 			// We could fix up the users with another pass, but for now
@@ -135,3 +133,19 @@ func flagalloc(f *Func) {
 		b.FlagsLiveAtEnd = end[b.ID] != nil
 	}
 }
+
+// copyFlags copies v (flag generator) into b, returns the copy.
+// If v's arg is also flags, copy recursively.
+func copyFlags(v *Value, b *Block) *Value {
+	flagsArgs := make(map[int]*Value)
+	for i, a := range v.Args {
+		if a.Type.IsFlags() || a.Type.IsTuple() {
+			flagsArgs[i] = copyFlags(a, b)
+		}
+	}
+	c := v.copyInto(b)
+	for i, a := range flagsArgs {
+		c.SetArg(i, a)
+	}
+	return c
+}

src/cmd/compile/internal/ssa/gen/dec64Ops.go

+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build ignore
+
+package main
+
+var dec64Ops = []opData{}
+
+var dec64Blocks = []blockData{}
+
+func init() {
+	archs = append(archs, arch{
+		name:    "dec64",
+		ops:     dec64Ops,
+		blocks:  dec64Blocks,
+		generic: true,
+	})
+}

src/cmd/compile/internal/ssa/gen/generic.rules

@@ -67,6 +67,12 @@
         (Const32F [f2i(float64(i2f32(c) * i2f32(d)))])
 (Mul64F (Const64F [c]) (Const64F [d])) -> (Const64F [f2i(i2f(c) * i2f(d))])
 
+// Convert x * -1 to -x. The front-end catches some but not all of these.
+(Mul8  (Const8  [-1]) x) -> (Neg8  x)
+(Mul16 (Const16 [-1]) x) -> (Neg16 x)
+(Mul32 (Const32 [-1]) x) -> (Neg32 x)
+(Mul64 (Const64 [-1]) x) -> (Neg64 x)
+
 (Mod8  (Const8  [c]) (Const8  [d])) && d != 0 -> (Const8  [int64(int8(c % d))])
 (Mod16 (Const16 [c]) (Const16 [d])) && d != 0 -> (Const16 [int64(int16(c % d))])
 (Mod32 (Const32 [c]) (Const32 [d])) && d != 0 -> (Const32 [int64(int32(c % d))])
@@ -625,8 +631,10 @@
         (Store [t.FieldType(0).Size()] dst f0 mem))))
 
 // un-SSAable values use mem->mem copies
-(Store [size] dst (Load <t> src mem) mem) && !config.fe.CanSSA(t) -> (Move [size] dst src mem)
-(Store [size] dst (Load <t> src mem) (VarDef {x} mem)) && !config.fe.CanSSA(t) -> (Move [size] dst src (VarDef {x} mem))
+(Store [size] dst (Load <t> src mem) mem) && !config.fe.CanSSA(t) ->
+	(Move [MakeSizeAndAlign(size, t.Alignment()).Int64()] dst src mem)
+(Store [size] dst (Load <t> src mem) (VarDef {x} mem)) && !config.fe.CanSSA(t) ->
+	(Move [MakeSizeAndAlign(size, t.Alignment()).Int64()] dst src (VarDef {x} mem))
 
 // string ops
 // Decomposing StringMake and lowering of StringPtr and StringLen
@@ -832,3 +840,23 @@
   -> (Sub64 x (Mul64 <t> (Div64  <t> x (Const64 <t> [c])) (Const64 <t> [c])))
 (Mod64u <t> x (Const64 [c])) && x.Op != OpConst64 && umagic64ok(c)
   -> (Sub64 x (Mul64 <t> (Div64u <t> x (Const64 <t> [c])) (Const64 <t> [c])))
+
+// floating point optimizations
+(Add32F x (Const32F [0])) -> x
+(Add32F (Const32F [0]) x) -> x
+(Add64F x (Const64F [0])) -> x
+(Add64F (Const64F [0]) x) -> x
+(Sub32F x (Const32F [0])) -> x
+(Sub64F x (Const64F [0])) -> x
+(Mul32F x (Const32F [f2i(1)])) -> x
+(Mul32F (Const32F [f2i(1)]) x) -> x
+(Mul64F x (Const64F [f2i(1)])) -> x
+(Mul64F (Const64F [f2i(1)]) x) -> x
+(Mul32F x (Const32F [f2i(-1)])) -> (Neg32F x)
+(Mul32F (Const32F [f2i(-1)]) x) -> (Neg32F x)
+(Mul64F x (Const64F [f2i(-1)])) -> (Neg64F x)
+(Mul64F (Const64F [f2i(-1)]) x) -> (Neg64F x)
+(Div32F x (Const32F [f2i(1)])) -> x
+(Div64F x (Const64F [f2i(1)])) -> x
+(Div32F x (Const32F [f2i(-1)])) -> (Neg32F x)
+(Div64F x (Const64F [f2i(-1)])) -> (Neg32F x)

src/cmd/compile/internal/ssa/html.go

@@ -359,7 +359,7 @@ func (v *Value) LongHTML() string {
 	}
 	r := v.Block.Func.RegAlloc
 	if int(v.ID) < len(r) && r[v.ID] != nil {
-		s += " : " + r[v.ID].Name()
+		s += " : " + html.EscapeString(r[v.ID].Name())
 	}
 	s += "</span>"
 	return s