cmd/compile/internal: generate ADDZE on PPC64

This usage shows up in quite a few places, and helps reduce
register pressure in several complex cryto functions by
removing a MOVD $0,... instruction.

Change-Id: I9444ea8f9d19bfd68fb71ea8dc34e109681b3802
Reviewed-on: https://go-review.googlesource.com/c/go/+/571055
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Paul Murphy <murp@ibm.com>

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

@@ -710,6 +710,13 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = r
 
+	case ssa.OpPPC64ADDZE:
+		p := s.Prog(v.Op.Asm())
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = v.Args[0].Reg()
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = v.Reg0()
+
 	case ssa.OpPPC64ADDZEzero, ssa.OpPPC64SUBZEzero:
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG

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

@@ -107,6 +107,7 @@
 (ADDE x y (Select1 <typ.UInt64> (ADDCconst (MOVDconst [0]) [-1]))) => (ADDC x y)
 // Fold transfer of CA -> GPR -> CA. Note 2 uses when feeding into a chained Add64carry.
 (Select1 (ADDCconst n:(ADDZEzero x) [-1])) && n.Uses <= 2 => x
+(ADDE (MOVDconst [0]) y c) => (ADDZE y c)
 
 // Borrowing subtraction.
 (Select0 (Sub64borrow x y c)) =>                 (Select0 <typ.UInt64> (SUBE x y (Select1 <typ.UInt64> (SUBCconst c [0]))))

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

@@ -139,6 +139,7 @@ func init() {
 		xergp       = regInfo{inputs: []regMask{xer}, outputs: []regMask{gp}, clobbers: xer}
 		gp11cxer    = regInfo{inputs: []regMask{gp | sp | sb}, outputs: []regMask{gp}, clobbers: xer}
 		gp11xer     = regInfo{inputs: []regMask{gp | sp | sb}, outputs: []regMask{gp, xer}}
+		gp1xer1xer  = regInfo{inputs: []regMask{gp | sp | sb, xer}, outputs: []regMask{gp, xer}, clobbers: xer}
 		gp21        = regInfo{inputs: []regMask{gp | sp | sb, gp | sp | sb}, outputs: []regMask{gp}}
 		gp21a0      = regInfo{inputs: []regMask{gp, gp | sp | sb}, outputs: []regMask{gp}}
 		gp21cxer    = regInfo{inputs: []regMask{gp | sp | sb, gp | sp | sb}, outputs: []regMask{gp}, clobbers: xer}
@@ -227,6 +228,7 @@ func init() {
 		{name: "ADDCconst", argLength: 1, reg: gp11xer, asm: "ADDC", typ: "(UInt64, UInt64)", aux: "Int64"},    // arg0 + imm16 -> out, CA
 		{name: "SUBCconst", argLength: 1, reg: gp11xer, asm: "SUBC", typ: "(UInt64, UInt64)", aux: "Int64"},    // imm16 - arg0 -> out, CA
 		{name: "ADDE", argLength: 3, reg: gp2xer1xer, asm: "ADDE", typ: "(UInt64, UInt64)", commutative: true}, // arg0 + arg1 + CA (arg2) -> out, CA
+		{name: "ADDZE", argLength: 2, reg: gp1xer1xer, asm: "ADDZE", typ: "(UInt64, UInt64)"},                  // arg0 + CA (arg1) -> out, CA
 		{name: "SUBE", argLength: 3, reg: gp2xer1xer, asm: "SUBE", typ: "(UInt64, UInt64)"},                    // arg0 - arg1 - CA (arg2) -> out, CA
 		{name: "ADDZEzero", argLength: 1, reg: xergp, asm: "ADDZE", typ: "UInt64"},                             // CA (arg0) + $0 -> out
 		{name: "SUBZEzero", argLength: 1, reg: xergp, asm: "SUBZE", typ: "UInt64"},                             // $0 - CA (arg0) -> out

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

@@ -2146,6 +2146,7 @@ const (
 	OpPPC64ADDCconst
 	OpPPC64SUBCconst
 	OpPPC64ADDE
+	OpPPC64ADDZE
 	OpPPC64SUBE
 	OpPPC64ADDZEzero
 	OpPPC64SUBZEzero
@@ -28842,6 +28843,22 @@ var opcodeTable = [...]opInfo{
 			},
 		},
 	},
+	{
+		name:   "ADDZE",
+		argLen: 2,
+		asm:    ppc64.AADDZE,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{1, 9223372036854775808}, // XER
+				{0, 1073733630},          // SP SB R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29
+			},
+			clobbers: 9223372036854775808, // XER
+			outputs: []outputInfo{
+				{1, 9223372036854775808}, // XER
+				{0, 1073733624},          // R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29
+			},
+		},
+	},
 	{
 		name:   "SUBE",
 		argLen: 3,

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

@@ -4003,6 +4003,21 @@ func rewriteValuePPC64_OpPPC64ADDE(v *Value) bool {
 		v.AddArg2(x, y)
 		return true
 	}
+	// match: (ADDE (MOVDconst [0]) y c)
+	// result: (ADDZE y c)
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			if v_0.Op != OpPPC64MOVDconst || auxIntToInt64(v_0.AuxInt) != 0 {
+				continue
+			}
+			y := v_1
+			c := v_2
+			v.reset(OpPPC64ADDZE)
+			v.AddArg2(y, c)
+			return true
+		}
+		break
+	}
 	return false
 }
 func rewriteValuePPC64_OpPPC64ADDconst(v *Value) bool {

test/codegen/mathbits.go

@@ -516,6 +516,7 @@ func Add64R(x, y, ci uint64) uint64 {
 	r, _ := bits.Add64(x, y, ci)
 	return r
 }
+
 func Add64M(p, q, r *[3]uint64) {
 	var c uint64
 	r[0], c = bits.Add64(p[0], q[0], c)
@@ -527,6 +528,15 @@ func Add64M(p, q, r *[3]uint64) {
 	r[2], c = bits.Add64(p[2], q[2], c)
 }
 
+func Add64M0(p, q, r *[3]uint64) {
+	var c uint64
+	r[0], c = bits.Add64(p[0], q[0], 0)
+	// ppc64x: -"ADDC", -"ADDE", "ADDZE\tR[1-9]"
+	r[1], c = bits.Add64(p[1], 0, c)
+	// ppc64x: -"ADDC", "ADDE", -"ADDZE"
+	r[2], c = bits.Add64(p[2], p[2], c)
+}
+
 func Add64MSaveC(p, q, r, c *[2]uint64) {
 	// ppc64x: "ADDC\tR", "ADDZE"
 	r[0], c[0] = bits.Add64(p[0], q[0], 0)