cmd/compile,cmd/internal/obj/ppc64: make math.Abs,math.Copysign instrinsics on ppc64x

This adds support for math Abs, Copysign to be instrinsics on ppc64x.

New instruction FCPSGN is added to generate fcpsgn. Some new
rules are added to improve the int<->float conversions that are
generated mainly due to the Float64bits and Float64frombits in
the math package. PPC64.rules is also modified as suggested
in the review for CL 63290.

Improvements:
benchmark                           old ns/op     new ns/op     delta
BenchmarkAbs-16                   1.12          0.69          -38.39%
BenchmarkCopysign-16              1.30          0.93          -28.46%
BenchmarkNextafter32-16           9.34          8.05          -13.81%
BenchmarkFrexp-16                 8.81          7.60          -13.73%

Others that used Copysign also saw smaller improvements.

I attempted to make this work using rules since that
seems to be preferred, but due to the use of Float64bits and
Float64frombits in these functions, several rules had to be added and
even then not all cases were matched. Using rules became too
complicated and seemed too fragile for these.

Updates #21390

Change-Id: Ia265da9a18355e08000818a4fba1a40e9e031995
Reviewed-on: https://go-review.googlesource.com/67130
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/asm/internal/asm/testdata/ppc64.s

@@ -734,6 +734,9 @@ label1:
 	POPCNTW	R1,R2
 	POPCNTB R1,R2
 
+//	Copysign
+	FCPSGN F1,F2,F3
+
 //	Random number generator, X-form
 //	DARN  L,RT produces
 //	darn  RT,L

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

@@ -2219,6 +2219,24 @@ var linuxPPC64LETests = []*asmTest{
 		pos: []string{"\tROTL\t"},
 	},
 
+	{
+		fn: `
+                func f12(a, b float64) float64 {
+                        return math.Copysign(a, b)
+                }
+                `,
+		pos: []string{"\tFCPSGN\t"},
+	},
+
+	{
+		fn: `
+                func f13(a float64) float64 {
+                        return math.Abs(a)
+                }
+                `,
+		pos: []string{"\tFABS\t"},
+	},
+
 	{
 		// check that stack store is optimized away
 		fn: `

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

@@ -2806,6 +2806,16 @@ func init() {
 			return s.newValue1(ssa.OpRound, types.Types[TFLOAT64], args[0])
 		},
 		sys.S390X)
+	addF("math", "Abs",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue1(ssa.OpAbs, types.Types[TFLOAT64], args[0])
+		},
+		sys.PPC64)
+	addF("math", "Copysign",
+		func(s *state, n *Node, args []*ssa.Value) *ssa.Value {
+			return s.newValue2(ssa.OpCopysign, types.Types[TFLOAT64], args[0], args[1])
+		},
+		sys.PPC64)
 
 	/******** math/bits ********/
 	addF("math/bits", "TrailingZeros64",

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

@@ -530,7 +530,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		ssa.OpPPC64SRAD, ssa.OpPPC64SRAW, ssa.OpPPC64SRD, ssa.OpPPC64SRW, ssa.OpPPC64SLD, ssa.OpPPC64SLW,
 		ssa.OpPPC64ROTL, ssa.OpPPC64ROTLW,
 		ssa.OpPPC64MULHD, ssa.OpPPC64MULHW, ssa.OpPPC64MULHDU, ssa.OpPPC64MULHWU,
-		ssa.OpPPC64FMUL, ssa.OpPPC64FMULS, ssa.OpPPC64FDIV, ssa.OpPPC64FDIVS,
+		ssa.OpPPC64FMUL, ssa.OpPPC64FMULS, ssa.OpPPC64FDIV, ssa.OpPPC64FDIVS, ssa.OpPPC64FCPSGN,
 		ssa.OpPPC64AND, ssa.OpPPC64OR, ssa.OpPPC64ANDN, ssa.OpPPC64ORN, ssa.OpPPC64NOR, ssa.OpPPC64XOR, ssa.OpPPC64EQV:
 		r := v.Reg()
 		r1 := v.Args[0].Reg()
@@ -581,7 +581,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = ppc64.REGTMP // Ignored; this is for the carry effect.
 
-	case ssa.OpPPC64NEG, ssa.OpPPC64FNEG, ssa.OpPPC64FSQRT, ssa.OpPPC64FSQRTS, ssa.OpPPC64FFLOOR, ssa.OpPPC64FTRUNC, ssa.OpPPC64FCEIL, ssa.OpPPC64FCTIDZ, ssa.OpPPC64FCTIWZ, ssa.OpPPC64FCFID, ssa.OpPPC64FCFIDS, ssa.OpPPC64FRSP, ssa.OpPPC64CNTLZD, ssa.OpPPC64CNTLZW, ssa.OpPPC64POPCNTD, ssa.OpPPC64POPCNTW, ssa.OpPPC64POPCNTB, ssa.OpPPC64MFVSRD, ssa.OpPPC64MTVSRD:
+	case ssa.OpPPC64NEG, ssa.OpPPC64FNEG, ssa.OpPPC64FSQRT, ssa.OpPPC64FSQRTS, ssa.OpPPC64FFLOOR, ssa.OpPPC64FTRUNC, ssa.OpPPC64FCEIL, ssa.OpPPC64FCTIDZ, ssa.OpPPC64FCTIWZ, ssa.OpPPC64FCFID, ssa.OpPPC64FCFIDS, ssa.OpPPC64FRSP, ssa.OpPPC64CNTLZD, ssa.OpPPC64CNTLZW, ssa.OpPPC64POPCNTD, ssa.OpPPC64POPCNTW, ssa.OpPPC64POPCNTB, ssa.OpPPC64MFVSRD, ssa.OpPPC64MTVSRD, ssa.OpPPC64FABS, ssa.OpPPC64FNABS:
 		r := v.Reg()
 		p := s.Prog(v.Op.Asm())
 		p.To.Type = obj.TYPE_REG

src/cmd/compile/internal/ssa/gen/PPC64.rules

@@ -70,12 +70,6 @@
 (Cvt32Fto64F x) -> x // Note x will have the wrong type for patterns dependent on Float32/Float64
 (Cvt64Fto32F x) -> (FRSP x)
 
-(MOVDload [off] {sym} ptr (FMOVDstore [off] {sym} ptr x _)) -> (MFVSRD x)
-(FMOVDload [off] {sym} ptr (MOVDstore [off] {sym} ptr x _)) -> (MTVSRD x)
-
-(FMOVDstore [off] {sym} ptr (MTVSRD x) mem) -> (MOVDstore [off] {sym} ptr x mem)
-(MOVDstore [off] {sym} ptr (MFVSRD x) mem) -> (FMOVDstore [off] {sym} ptr x mem)
-
 (Round32F x) -> (LoweredRound32F x)
 (Round64F x) -> (LoweredRound64F x)
 
@@ -83,6 +77,8 @@
 (Floor x) -> (FFLOOR x)
 (Ceil x) -> (FCEIL x)
 (Trunc x) -> (FTRUNC x)
+(Copysign x y) -> (FCPSGN y x)
+(Abs x) -> (FABS x)
 
 // Lowering constants
 (Const8   [val]) -> (MOVDconst [val])
@@ -94,6 +90,13 @@
 (ConstNil) -> (MOVDconst [0])
 (ConstBool [b]) -> (MOVDconst [b])
 
+// Constant folding
+(FABS (FMOVDconst [x])) -> (FMOVDconst [f2i(math.Abs(i2f(x)))])
+(FSQRT (FMOVDconst [x])) -> (FMOVDconst [f2i(math.Sqrt(i2f(x)))])
+(FFLOOR (FMOVDconst [x])) -> (FMOVDconst [f2i(math.Floor(i2f(x)))])
+(FCEIL (FMOVDconst [x])) -> (FMOVDconst [f2i(math.Ceil(i2f(x)))])
+(FTRUNC (FMOVDconst [x])) -> (FMOVDconst [f2i(math.Trunc(i2f(x)))])
+
 // Rotate generation with const shift
 (ADD (SLDconst x [c]) (SRDconst x [d])) && d == 64-c -> (ROTLconst [c] x)
 ( OR (SLDconst x [c]) (SRDconst x [d])) && d == 64-c -> (ROTLconst [c] x)
@@ -734,6 +737,20 @@
 
 (ADDconst [c] (MOVDaddr [d] {sym} x)) -> (MOVDaddr [c+d] {sym} x)
 
+// Use register moves instead of stores and loads to move int<->float values
+// Common with math Float64bits, Float64frombits
+(MOVDload [off] {sym} ptr (FMOVDstore [off] {sym} ptr x _)) -> (MFVSRD x)
+(FMOVDload [off] {sym} ptr (MOVDstore [off] {sym} ptr x _)) -> (MTVSRD x)
+
+(FMOVDstore [off] {sym} ptr (MTVSRD x) mem) -> (MOVDstore [off] {sym} ptr x mem)
+(MOVDstore [off] {sym} ptr (MFVSRD x) mem) -> (FMOVDstore [off] {sym} ptr x mem)
+
+(MTVSRD (MOVDconst [c])) -> (FMOVDconst [c])
+(MFVSRD (FMOVDconst [c])) -> (MOVDconst [c])
+
+(MTVSRD x:(MOVDload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (FMOVDload [off] {sym} ptr mem)
+(MFVSRD x:(FMOVDload [off] {sym} ptr mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVDload [off] {sym} ptr mem)
+
 // Fold offsets for stores.
 (MOVDstore [off1] {sym} (ADDconst [off2] x) val mem) && is16Bit(off1+off2) -> (MOVDstore [off1+off2] {sym} x val mem)
 (MOVWstore [off1] {sym} (ADDconst [off2] x) val mem) && is16Bit(off1+off2) -> (MOVWstore [off1+off2] {sym} x val mem)
@@ -896,6 +913,10 @@
 (AND (MOVDconst [c]) x:(MOVBZload _ _)) -> (ANDconst [c&0xFF] x)
 (AND x:(MOVBZload _ _) (MOVDconst [c])) -> (ANDconst [c&0xFF] x)
 
+// floating point negative abs
+(FNEG (FABS x)) -> (FNABS x)
+(FNEG (FNABS x)) -> (FABS x)
+
 // floating-point fused multiply-add/sub
 (FADD (FMUL x y) z) -> (FMADD x y z)
 (FSUB (FMUL x y) z) -> (FMSUB x y z)

src/cmd/compile/internal/ssa/gen/PPC64Ops.go

@@ -249,6 +249,9 @@ func init() {
 		{name: "FFLOOR", argLength: 1, reg: fp11, asm: "FRIM"},                              // floor(arg0), float64
 		{name: "FCEIL", argLength: 1, reg: fp11, asm: "FRIP"},                               // ceil(arg0), float64
 		{name: "FTRUNC", argLength: 1, reg: fp11, asm: "FRIZ"},                              // trunc(arg0), float64
+		{name: "FABS", argLength: 1, reg: fp11, asm: "FABS"},                                // abs(arg0), float64
+		{name: "FNABS", argLength: 1, reg: fp11, asm: "FNABS"},                              // -abs(arg0), float64
+		{name: "FCPSGN", argLength: 2, reg: fp21, asm: "FCPSGN"},                            // copysign arg0 -> arg1, float64
 
 		{name: "ORconst", argLength: 1, reg: gp11, asm: "OR", aux: "Int64"},                                                                                     // arg0|aux
 		{name: "XORconst", argLength: 1, reg: gp11, asm: "XOR", aux: "Int64"},                                                                                   // arg0^aux

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

@@ -273,6 +273,10 @@ var genericOps = []opData{
 	{name: "Trunc", argLength: 1}, // round arg0 toward 0
 	{name: "Round", argLength: 1}, // round arg0 to nearest, ties away from 0
 
+	// Modify the sign bit
+	{name: "Abs", argLength: 1},      // absolute value arg0
+	{name: "Copysign", argLength: 2}, // copy sign from arg0 to arg1
+
 	// Data movement, max argument length for Phi is indefinite so just pick
 	// a really large number
 	{name: "Phi", argLength: -1}, // select an argument based on which predecessor block we came from

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

@@ -1407,6 +1407,9 @@ const (
 	OpPPC64FFLOOR
 	OpPPC64FCEIL
 	OpPPC64FTRUNC
+	OpPPC64FABS
+	OpPPC64FNABS
+	OpPPC64FCPSGN
 	OpPPC64ORconst
 	OpPPC64XORconst
 	OpPPC64ANDconst
@@ -1892,6 +1895,8 @@ const (
 	OpCeil
 	OpTrunc
 	OpRound
+	OpAbs
+	OpCopysign
 	OpPhi
 	OpCopy
 	OpConvert
@@ -18113,6 +18118,46 @@ var opcodeTable = [...]opInfo{
 			},
 		},
 	},
+	{
+		name:   "FABS",
+		argLen: 1,
+		asm:    ppc64.AFABS,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 576460743713488896}, // F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26
+			},
+			outputs: []outputInfo{
+				{0, 576460743713488896}, // F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26
+			},
+		},
+	},
+	{
+		name:   "FNABS",
+		argLen: 1,
+		asm:    ppc64.AFNABS,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 576460743713488896}, // F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26
+			},
+			outputs: []outputInfo{
+				{0, 576460743713488896}, // F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26
+			},
+		},
+	},
+	{
+		name:   "FCPSGN",
+		argLen: 2,
+		asm:    ppc64.AFCPSGN,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 576460743713488896}, // F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26
+				{1, 576460743713488896}, // F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26
+			},
+			outputs: []outputInfo{
+				{0, 576460743713488896}, // F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22 F23 F24 F25 F26
+			},
+		},
+	},
 	{
 		name:    "ORconst",
 		auxType: auxInt64,
@@ -23186,6 +23231,16 @@ var opcodeTable = [...]opInfo{
 		argLen:  1,
 		generic: true,
 	},
+	{
+		name:    "Abs",
+		argLen:  1,
+		generic: true,
+	},
+	{
+		name:    "Copysign",
+		argLen:  2,
+		generic: true,
+	},
 	{
 		name:    "Phi",
 		argLen:  -1,

src/cmd/compile/internal/ssa/rewritePPC64.go

@@ -15,6 +15,8 @@ var _ = types.TypeMem // in case not otherwise used
 
 func rewriteValuePPC64(v *Value) bool {
 	switch v.Op {
+	case OpAbs:
+		return rewriteValuePPC64_OpAbs_0(v)
 	case OpAdd16:
 		return rewriteValuePPC64_OpAdd16_0(v)
 	case OpAdd32:
@@ -103,6 +105,8 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpConstNil_0(v)
 	case OpConvert:
 		return rewriteValuePPC64_OpConvert_0(v)
+	case OpCopysign:
+		return rewriteValuePPC64_OpCopysign_0(v)
 	case OpCtz32:
 		return rewriteValuePPC64_OpCtz32_0(v)
 	case OpCtz64:
@@ -399,10 +403,16 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpPPC64CMPconst_0(v)
 	case OpPPC64Equal:
 		return rewriteValuePPC64_OpPPC64Equal_0(v)
+	case OpPPC64FABS:
+		return rewriteValuePPC64_OpPPC64FABS_0(v)
 	case OpPPC64FADD:
 		return rewriteValuePPC64_OpPPC64FADD_0(v)
 	case OpPPC64FADDS:
 		return rewriteValuePPC64_OpPPC64FADDS_0(v)
+	case OpPPC64FCEIL:
+		return rewriteValuePPC64_OpPPC64FCEIL_0(v)
+	case OpPPC64FFLOOR:
+		return rewriteValuePPC64_OpPPC64FFLOOR_0(v)
 	case OpPPC64FMOVDload:
 		return rewriteValuePPC64_OpPPC64FMOVDload_0(v)
 	case OpPPC64FMOVDstore:
@@ -411,10 +421,16 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpPPC64FMOVSload_0(v)
 	case OpPPC64FMOVSstore:
 		return rewriteValuePPC64_OpPPC64FMOVSstore_0(v)
+	case OpPPC64FNEG:
+		return rewriteValuePPC64_OpPPC64FNEG_0(v)
+	case OpPPC64FSQRT:
+		return rewriteValuePPC64_OpPPC64FSQRT_0(v)
 	case OpPPC64FSUB:
 		return rewriteValuePPC64_OpPPC64FSUB_0(v)
 	case OpPPC64FSUBS:
 		return rewriteValuePPC64_OpPPC64FSUBS_0(v)
+	case OpPPC64FTRUNC:
+		return rewriteValuePPC64_OpPPC64FTRUNC_0(v)
 	case OpPPC64GreaterEqual:
 		return rewriteValuePPC64_OpPPC64GreaterEqual_0(v)
 	case OpPPC64GreaterThan:
@@ -423,6 +439,8 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpPPC64LessEqual_0(v)
 	case OpPPC64LessThan:
 		return rewriteValuePPC64_OpPPC64LessThan_0(v)
+	case OpPPC64MFVSRD:
+		return rewriteValuePPC64_OpPPC64MFVSRD_0(v)
 	case OpPPC64MOVBZload:
 		return rewriteValuePPC64_OpPPC64MOVBZload_0(v)
 	case OpPPC64MOVBZreg:
@@ -463,6 +481,8 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpPPC64MOVWstore_0(v)
 	case OpPPC64MOVWstorezero:
 		return rewriteValuePPC64_OpPPC64MOVWstorezero_0(v)
+	case OpPPC64MTVSRD:
+		return rewriteValuePPC64_OpPPC64MTVSRD_0(v)
 	case OpPPC64MaskIfNotCarry:
 		return rewriteValuePPC64_OpPPC64MaskIfNotCarry_0(v)
 	case OpPPC64NotEqual:
@@ -628,6 +648,17 @@ func rewriteValuePPC64(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpAbs_0(v *Value) bool {
+	// match: (Abs x)
+	// cond:
+	// result: (FABS x)
+	for {
+		x := v.Args[0]
+		v.reset(OpPPC64FABS)
+		v.AddArg(x)
+		return true
+	}
+}
 func rewriteValuePPC64_OpAdd16_0(v *Value) bool {
 	// match: (Add16 x y)
 	// cond:
@@ -1258,6 +1289,20 @@ func rewriteValuePPC64_OpConvert_0(v *Value) bool {
 		return true
 	}
 }
+func rewriteValuePPC64_OpCopysign_0(v *Value) bool {
+	// match: (Copysign x y)
+	// cond:
+	// result: (FCPSGN y x)
+	for {
+		_ = v.Args[1]
+		x := v.Args[0]
+		y := v.Args[1]
+		v.reset(OpPPC64FCPSGN)
+		v.AddArg(y)
+		v.AddArg(x)
+		return true
+	}
+}
 func rewriteValuePPC64_OpCtz32_0(v *Value) bool {
 	b := v.Block
 	_ = b
@@ -6166,6 +6211,22 @@ func rewriteValuePPC64_OpPPC64Equal_0(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpPPC64FABS_0(v *Value) bool {
+	// match: (FABS (FMOVDconst [x]))
+	// cond:
+	// result: (FMOVDconst [f2i(math.Abs(i2f(x)))])
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FMOVDconst {
+			break
+		}
+		x := v_0.AuxInt
+		v.reset(OpPPC64FMOVDconst)
+		v.AuxInt = f2i(math.Abs(i2f(x)))
+		return true
+	}
+	return false
+}
 func rewriteValuePPC64_OpPPC64FADD_0(v *Value) bool {
 	// match: (FADD (FMUL x y) z)
 	// cond:
@@ -6248,6 +6309,38 @@ func rewriteValuePPC64_OpPPC64FADDS_0(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpPPC64FCEIL_0(v *Value) bool {
+	// match: (FCEIL (FMOVDconst [x]))
+	// cond:
+	// result: (FMOVDconst [f2i(math.Ceil(i2f(x)))])
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FMOVDconst {
+			break
+		}
+		x := v_0.AuxInt
+		v.reset(OpPPC64FMOVDconst)
+		v.AuxInt = f2i(math.Ceil(i2f(x)))
+		return true
+	}
+	return false
+}
+func rewriteValuePPC64_OpPPC64FFLOOR_0(v *Value) bool {
+	// match: (FFLOOR (FMOVDconst [x]))
+	// cond:
+	// result: (FMOVDconst [f2i(math.Floor(i2f(x)))])
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FMOVDconst {
+			break
+		}
+		x := v_0.AuxInt
+		v.reset(OpPPC64FMOVDconst)
+		v.AuxInt = f2i(math.Floor(i2f(x)))
+		return true
+	}
+	return false
+}
 func rewriteValuePPC64_OpPPC64FMOVDload_0(v *Value) bool {
 	// match: (FMOVDload [off] {sym} ptr (MOVDstore [off] {sym} ptr x _))
 	// cond:
@@ -6513,6 +6606,51 @@ func rewriteValuePPC64_OpPPC64FMOVSstore_0(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpPPC64FNEG_0(v *Value) bool {
+	// match: (FNEG (FABS x))
+	// cond:
+	// result: (FNABS x)
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FABS {
+			break
+		}
+		x := v_0.Args[0]
+		v.reset(OpPPC64FNABS)
+		v.AddArg(x)
+		return true
+	}
+	// match: (FNEG (FNABS x))
+	// cond:
+	// result: (FABS x)
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FNABS {
+			break
+		}
+		x := v_0.Args[0]
+		v.reset(OpPPC64FABS)
+		v.AddArg(x)
+		return true
+	}
+	return false
+}
+func rewriteValuePPC64_OpPPC64FSQRT_0(v *Value) bool {
+	// match: (FSQRT (FMOVDconst [x]))
+	// cond:
+	// result: (FMOVDconst [f2i(math.Sqrt(i2f(x)))])
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FMOVDconst {
+			break
+		}
+		x := v_0.AuxInt
+		v.reset(OpPPC64FMOVDconst)
+		v.AuxInt = f2i(math.Sqrt(i2f(x)))
+		return true
+	}
+	return false
+}
 func rewriteValuePPC64_OpPPC64FSUB_0(v *Value) bool {
 	// match: (FSUB (FMUL x y) z)
 	// cond:
@@ -6557,6 +6695,22 @@ func rewriteValuePPC64_OpPPC64FSUBS_0(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpPPC64FTRUNC_0(v *Value) bool {
+	// match: (FTRUNC (FMOVDconst [x]))
+	// cond:
+	// result: (FMOVDconst [f2i(math.Trunc(i2f(x)))])
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FMOVDconst {
+			break
+		}
+		x := v_0.AuxInt
+		v.reset(OpPPC64FMOVDconst)
+		v.AuxInt = f2i(math.Trunc(i2f(x)))
+		return true
+	}
+	return false
+}
 func rewriteValuePPC64_OpPPC64GreaterEqual_0(v *Value) bool {
 	// match: (GreaterEqual (FlagEQ))
 	// cond:
@@ -6765,6 +6919,52 @@ func rewriteValuePPC64_OpPPC64LessThan_0(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpPPC64MFVSRD_0(v *Value) bool {
+	b := v.Block
+	_ = b
+	typ := &b.Func.Config.Types
+	_ = typ
+	// match: (MFVSRD (FMOVDconst [c]))
+	// cond:
+	// result: (MOVDconst [c])
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64FMOVDconst {
+			break
+		}
+		c := v_0.AuxInt
+		v.reset(OpPPC64MOVDconst)
+		v.AuxInt = c
+		return true
+	}
+	// match: (MFVSRD x:(FMOVDload [off] {sym} ptr mem))
+	// cond: x.Uses == 1 && clobber(x)
+	// result: @x.Block (MOVDload [off] {sym} ptr mem)
+	for {
+		x := v.Args[0]
+		if x.Op != OpPPC64FMOVDload {
+			break
+		}
+		off := x.AuxInt
+		sym := x.Aux
+		_ = x.Args[1]
+		ptr := x.Args[0]
+		mem := x.Args[1]
+		if !(x.Uses == 1 && clobber(x)) {
+			break
+		}
+		b = x.Block
+		v0 := b.NewValue0(v.Pos, OpPPC64MOVDload, typ.Int64)
+		v.reset(OpCopy)
+		v.AddArg(v0)
+		v0.AuxInt = off
+		v0.Aux = sym
+		v0.AddArg(ptr)
+		v0.AddArg(mem)
+		return true
+	}
+	return false
+}
 func rewriteValuePPC64_OpPPC64MOVBZload_0(v *Value) bool {
 	// match: (MOVBZload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
 	// cond: canMergeSym(sym1,sym2)
@@ -8322,6 +8522,52 @@ func rewriteValuePPC64_OpPPC64MOVWstorezero_0(v *Value) bool {
 	}
 	return false
 }
+func rewriteValuePPC64_OpPPC64MTVSRD_0(v *Value) bool {
+	b := v.Block
+	_ = b
+	typ := &b.Func.Config.Types
+	_ = typ
+	// match: (MTVSRD (MOVDconst [c]))
+	// cond:
+	// result: (FMOVDconst [c])
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != OpPPC64MOVDconst {
+			break
+		}
+		c := v_0.AuxInt
+		v.reset(OpPPC64FMOVDconst)
+		v.AuxInt = c
+		return true
+	}
+	// match: (MTVSRD x:(MOVDload [off] {sym} ptr mem))
+	// cond: x.Uses == 1 && clobber(x)
+	// result: @x.Block (FMOVDload [off] {sym} ptr mem)
+	for {
+		x := v.Args[0]
+		if x.Op != OpPPC64MOVDload {
+			break
+		}
+		off := x.AuxInt
+		sym := x.Aux
+		_ = x.Args[1]
+		ptr := x.Args[0]
+		mem := x.Args[1]
+		if !(x.Uses == 1 && clobber(x)) {
+			break
+		}
+		b = x.Block
+		v0 := b.NewValue0(v.Pos, OpPPC64FMOVDload, typ.Float64)
+		v.reset(OpCopy)
+		v.AddArg(v0)
+		v0.AuxInt = off
+		v0.Aux = sym
+		v0.AddArg(ptr)
+		v0.AddArg(mem)
+		return true
+	}
+	return false
+}
 func rewriteValuePPC64_OpPPC64MaskIfNotCarry_0(v *Value) bool {
 	// match: (MaskIfNotCarry (ADDconstForCarry [c] (ANDconst [d] _)))
 	// cond: c < 0 && d > 0 && c + d < 0

src/cmd/internal/obj/ppc64/a.out.go

@@ -604,6 +604,8 @@ const (
 
 	ARFCI
 
+	AFCPSGN
+	AFCPSGNCC
 	/* optional on 32-bit */
 	AFRES
 	AFRESCC

src/cmd/internal/obj/ppc64/anames.go

@@ -230,6 +230,8 @@ var Anames = []string{
 	"SYSCALL",
 	"WORD",
 	"RFCI",
+	"FCPSGN",
+	"FCPSGNCC",
 	"FRES",
 	"FRESCC",
 	"FRIM",

src/cmd/internal/obj/ppc64/asm9.go

@@ -1619,6 +1619,8 @@ func buildop(ctxt *obj.Link) {
 			opset(AFADDS, r0)
 			opset(AFADDCC, r0)
 			opset(AFADDSCC, r0)
+			opset(AFCPSGN, r0)
+			opset(AFCPSGNCC, r0)
 			opset(AFDIV, r0)
 			opset(AFDIVS, r0)
 			opset(AFDIVCC, r0)
@@ -3756,6 +3758,10 @@ func (c *ctxt9) oprrr(a obj.As) uint32 {
 		return OPVCC(59, 30, 0, 0)
 	case AFNMSUBSCC:
 		return OPVCC(59, 30, 0, 1)
+	case AFCPSGN:
+		return OPVCC(63, 8, 0, 0)
+	case AFCPSGNCC:
+		return OPVCC(63, 8, 0, 1)
 	case AFRES:
 		return OPVCC(59, 24, 0, 0)
 	case AFRESCC: