[dev.ssa] cmd/compile: implement more 64-bit ops on 386

add/sub/mul, plus constant input variants.

Change-Id: I1c8006727c4fdf73558da0e646e7d1fa130ed773
Reviewed-on: https://go-review.googlesource.com/25006
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

6 files changed, 369 insertions, 29 deletions

diff --git a/src/cmd/compile/internal/ssa/gen/386.rules b/src/cmd/compile/internal/ssa/gen/386.rules
index 9a3b16cd52..6569533b77 100644
--- a/src/cmd/compile/internal/ssa/gen/386.rules
+++ b/src/cmd/compile/internal/ssa/gen/386.rules
@@ -20,12 +20,17 @@
 (Sub32F x y) -> (SUBSS x y)
 (Sub64F x y) -> (SUBSD x y)
 
+(Sub32carry x y) -> (SUBLcarry x y)
+(Sub32withcarry x y c) -> (SBBL x y c)
+
 (Mul32  x y) -> (MULL  x y)
 (Mul16  x y) -> (MULL  x y)
 (Mul8   x y) -> (MULL  x y)
 (Mul32F x y) -> (MULSS x y)
 (Mul64F x y) -> (MULSD x y)
 
+(Mul32uhilo x y) -> (MULLQU x y)
+
 (Div32F x y) -> (DIVSS x y)
 (Div64F x y) -> (DIVSD x y)
 
@@ -216,7 +221,7 @@
 (Neq32F x y) -> (SETNEF (UCOMISS x y))
 
 // Lowering loads
-(Load <t> ptr mem) && is32BitInt(t) -> (MOVLload ptr mem)
+(Load <t> ptr mem) && (is32BitInt(t) || isPtr(t)) -> (MOVLload ptr mem)
 (Load <t> ptr mem) && is16BitInt(t) -> (MOVWload ptr mem)
 (Load <t> ptr mem) && (t.IsBoolean() || is8BitInt(t)) -> (MOVBload ptr mem)
 (Load <t> ptr mem) && is32BitFloat(t) -> (MOVSSload ptr mem)
@@ -383,9 +388,15 @@
 // fold constants into instructions
 (ADDL x (MOVLconst [c])) -> (ADDLconst [c] x)
 (ADDL (MOVLconst [c]) x) -> (ADDLconst [c] x)
+(ADDLcarry x (MOVLconst [c])) -> (ADDLconstcarry [c] x)
+(ADDLcarry (MOVLconst [c]) x) -> (ADDLconstcarry [c] x)
+(ADCL x (MOVLconst [c]) f) -> (ADCLconst [c] x f)
+(ADCL (MOVLconst [c]) x f) -> (ADCLconst [c] x f)
 
 (SUBL x (MOVLconst [c])) -> (SUBLconst x [c])
 (SUBL (MOVLconst [c]) x) -> (NEGL (SUBLconst <v.Type> x [c]))
+(SUBLcarry x (MOVLconst [c])) -> (SUBLconstcarry [c] x)
+(SBBL x (MOVLconst [c]) f) -> (SBBLconst [c] x f)
 
 (MULL x (MOVLconst [c])) -> (MULLconst [c] x)
 (MULL (MOVLconst [c]) x) -> (MULLconst [c] x)
diff --git a/src/cmd/compile/internal/ssa/gen/386Ops.go b/src/cmd/compile/internal/ssa/gen/386Ops.go
index f2a8edf456..68bcfa9649 100644
--- a/src/cmd/compile/internal/ssa/gen/386Ops.go
+++ b/src/cmd/compile/internal/ssa/gen/386Ops.go
@@ -99,17 +99,20 @@ func init() {
 		gp11nf    = regInfo{inputs: []regMask{gpsp}, outputs: gponly} // nf: no flags clobbered
 		gp11sb    = regInfo{inputs: []regMask{gpspsb}, outputs: gponly}
 		gp21      = regInfo{inputs: []regMask{gp, gp}, outputs: gponly, clobbers: flags}
+		gp11carry = regInfo{inputs: []regMask{gp}, outputs: []regMask{flags, gp}}
 		gp21carry = regInfo{inputs: []regMask{gp, gp}, outputs: []regMask{flags, gp}}
+		gp1carry1 = regInfo{inputs: []regMask{gp, flags}, outputs: gponly}
 		gp2carry1 = regInfo{inputs: []regMask{gp, gp, flags}, outputs: gponly}
 		gp21sp    = regInfo{inputs: []regMask{gpsp, gp}, outputs: gponly, clobbers: flags}
 		gp21sb    = regInfo{inputs: []regMask{gpspsb, gpsp}, outputs: gponly}
 		gp21shift = regInfo{inputs: []regMask{gp, cx}, outputs: []regMask{gp}, clobbers: flags}
 		gp11div   = regInfo{inputs: []regMask{ax, gpsp &^ dx}, outputs: []regMask{ax},
 			clobbers: dx | flags}
-		gp11hmul = regInfo{inputs: []regMask{ax, gpsp}, outputs: []regMask{dx},
+		gp21hmul = regInfo{inputs: []regMask{ax, gpsp}, outputs: []regMask{dx},
 			clobbers: ax | flags}
 		gp11mod = regInfo{inputs: []regMask{ax, gpsp &^ dx}, outputs: []regMask{dx},
 			clobbers: ax | flags}
+		gp21mul = regInfo{inputs: []regMask{ax, gpsp}, outputs: []regMask{dx, ax}, clobbers: flags}
 
 		gp2flags = regInfo{inputs: []regMask{gpsp, gpsp}, outputs: flagsonly}
 		gp1flags = regInfo{inputs: []regMask{gpsp}, outputs: flagsonly}
@@ -174,20 +177,29 @@ func init() {
 		{name: "ADDLconst", argLength: 1, reg: gp11sp, asm: "ADDL", aux: "Int32", typ: "UInt32"}, // arg0 + auxint
 
 		{name: "ADDLcarry", argLength: 2, reg: gp21carry, asm: "ADDL", commutative: true, resultInArg0: true}, // arg0 + arg1, generates <carry,result> pair
+		{name: "ADDLconstcarry", argLength: 1, reg: gp11carry, asm: "ADDL", aux: "Int32", resultInArg0: true}, // arg0 + auxint, generates <carry,result> pair
 		{name: "ADCL", argLength: 3, reg: gp2carry1, asm: "ADCL", commutative: true, resultInArg0: true},      // arg0+arg1+carry(arg2), where arg2 is flags
+		{name: "ADCLconst", argLength: 2, reg: gp1carry1, asm: "ADCL", aux: "Int32", resultInArg0: true},      // arg0+auxint+carry(arg1), where arg1 is flags
 
 		{name: "SUBL", argLength: 2, reg: gp21, asm: "SUBL", resultInArg0: true},                    // arg0 - arg1
 		{name: "SUBLconst", argLength: 1, reg: gp11, asm: "SUBL", aux: "Int32", resultInArg0: true}, // arg0 - auxint
 
+		{name: "SUBLcarry", argLength: 2, reg: gp21carry, asm: "SUBL", resultInArg0: true},                    // arg0-arg1, generates <borrow,result> pair
+		{name: "SUBLconstcarry", argLength: 1, reg: gp11carry, asm: "SUBL", aux: "Int32", resultInArg0: true}, // arg0-auxint, generates <borrow,result> pair
+		{name: "SBBL", argLength: 3, reg: gp2carry1, asm: "SBBL", resultInArg0: true},                         // arg0-arg1-borrow(arg2), where arg2 is flags
+		{name: "SBBLconst", argLength: 2, reg: gp1carry1, asm: "SBBL", aux: "Int32", resultInArg0: true},      // arg0-auxint-borrow(arg1), where arg1 is flags
+
 		{name: "MULL", argLength: 2, reg: gp21, asm: "IMULL", commutative: true, resultInArg0: true}, // arg0 * arg1
 		{name: "MULLconst", argLength: 1, reg: gp11, asm: "IMULL", aux: "Int32", resultInArg0: true}, // arg0 * auxint
 
-		{name: "HMULL", argLength: 2, reg: gp11hmul, asm: "IMULL"}, // (arg0 * arg1) >> width
-		{name: "HMULW", argLength: 2, reg: gp11hmul, asm: "IMULW"}, // (arg0 * arg1) >> width
-		{name: "HMULB", argLength: 2, reg: gp11hmul, asm: "IMULB"}, // (arg0 * arg1) >> width
-		{name: "HMULLU", argLength: 2, reg: gp11hmul, asm: "MULL"}, // (arg0 * arg1) >> width
-		{name: "HMULWU", argLength: 2, reg: gp11hmul, asm: "MULW"}, // (arg0 * arg1) >> width
-		{name: "HMULBU", argLength: 2, reg: gp11hmul, asm: "MULB"}, // (arg0 * arg1) >> width
+		{name: "HMULL", argLength: 2, reg: gp21hmul, asm: "IMULL"}, // (arg0 * arg1) >> width
+		{name: "HMULLU", argLength: 2, reg: gp21hmul, asm: "MULL"}, // (arg0 * arg1) >> width
+		{name: "HMULW", argLength: 2, reg: gp21hmul, asm: "IMULW"}, // (arg0 * arg1) >> width
+		{name: "HMULB", argLength: 2, reg: gp21hmul, asm: "IMULB"}, // (arg0 * arg1) >> width
+		{name: "HMULWU", argLength: 2, reg: gp21hmul, asm: "MULW"}, // (arg0 * arg1) >> width
+		{name: "HMULBU", argLength: 2, reg: gp21hmul, asm: "MULB"}, // (arg0 * arg1) >> width
+
+		{name: "MULLQU", argLength: 2, reg: gp21mul, asm: "MULL"}, // arg0 * arg1, high 32 in result[0], low 32 in result[1]
 
 		{name: "DIVL", argLength: 2, reg: gp11div, asm: "IDIVL"}, // arg0 / arg1
 		{name: "DIVW", argLength: 2, reg: gp11div, asm: "IDIVW"}, // arg0 / arg1
diff --git a/src/cmd/compile/internal/ssa/gen/AMD64Ops.go b/src/cmd/compile/internal/ssa/gen/AMD64Ops.go
index c87172e8e3..c63e6edeb9 100644
--- a/src/cmd/compile/internal/ssa/gen/AMD64Ops.go
+++ b/src/cmd/compile/internal/ssa/gen/AMD64Ops.go
@@ -121,7 +121,7 @@ func init() {
 		gp21shift = regInfo{inputs: []regMask{gp, cx}, outputs: []regMask{gp}, clobbers: flags}
 		gp11div   = regInfo{inputs: []regMask{ax, gpsp &^ dx}, outputs: []regMask{ax, dx},
 			clobbers: flags}
-		gp11hmul = regInfo{inputs: []regMask{ax, gpsp}, outputs: []regMask{dx},
+		gp21hmul = regInfo{inputs: []regMask{ax, gpsp}, outputs: []regMask{dx},
 			clobbers: ax | flags}
 
 		gp2flags = regInfo{inputs: []regMask{gpsp, gpsp}, outputs: flagsonly}
@@ -201,14 +201,14 @@ func init() {
 		{name: "MULQconst", argLength: 1, reg: gp11, asm: "IMULQ", aux: "Int64", resultInArg0: true}, // arg0 * auxint
 		{name: "MULLconst", argLength: 1, reg: gp11, asm: "IMULL", aux: "Int32", resultInArg0: true}, // arg0 * auxint
 
-		{name: "HMULQ", argLength: 2, reg: gp11hmul, asm: "IMULQ"}, // (arg0 * arg1) >> width
-		{name: "HMULL", argLength: 2, reg: gp11hmul, asm: "IMULL"}, // (arg0 * arg1) >> width
-		{name: "HMULW", argLength: 2, reg: gp11hmul, asm: "IMULW"}, // (arg0 * arg1) >> width
-		{name: "HMULB", argLength: 2, reg: gp11hmul, asm: "IMULB"}, // (arg0 * arg1) >> width
-		{name: "HMULQU", argLength: 2, reg: gp11hmul, asm: "MULQ"}, // (arg0 * arg1) >> width
-		{name: "HMULLU", argLength: 2, reg: gp11hmul, asm: "MULL"}, // (arg0 * arg1) >> width
-		{name: "HMULWU", argLength: 2, reg: gp11hmul, asm: "MULW"}, // (arg0 * arg1) >> width
-		{name: "HMULBU", argLength: 2, reg: gp11hmul, asm: "MULB"}, // (arg0 * arg1) >> width
+		{name: "HMULQ", argLength: 2, reg: gp21hmul, asm: "IMULQ"}, // (arg0 * arg1) >> width
+		{name: "HMULL", argLength: 2, reg: gp21hmul, asm: "IMULL"}, // (arg0 * arg1) >> width
+		{name: "HMULW", argLength: 2, reg: gp21hmul, asm: "IMULW"}, // (arg0 * arg1) >> width
+		{name: "HMULB", argLength: 2, reg: gp21hmul, asm: "IMULB"}, // (arg0 * arg1) >> width
+		{name: "HMULQU", argLength: 2, reg: gp21hmul, asm: "MULQ"}, // (arg0 * arg1) >> width
+		{name: "HMULLU", argLength: 2, reg: gp21hmul, asm: "MULL"}, // (arg0 * arg1) >> width
+		{name: "HMULWU", argLength: 2, reg: gp21hmul, asm: "MULW"}, // (arg0 * arg1) >> width
+		{name: "HMULBU", argLength: 2, reg: gp21hmul, asm: "MULB"}, // (arg0 * arg1) >> width
 
 		{name: "AVGQU", argLength: 2, reg: gp21, commutative: true, resultInArg0: true}, // (arg0 + arg1) / 2 as unsigned, all 64 result bits
 
diff --git a/src/cmd/compile/internal/ssa/opGen.go b/src/cmd/compile/internal/ssa/opGen.go
index 76baa9ea5e..47cfda86b5 100644
--- a/src/cmd/compile/internal/ssa/opGen.go
+++ b/src/cmd/compile/internal/ssa/opGen.go
@@ -172,17 +172,24 @@ const (
 	Op386ADDL
 	Op386ADDLconst
 	Op386ADDLcarry
+	Op386ADDLconstcarry
 	Op386ADCL
+	Op386ADCLconst
 	Op386SUBL
 	Op386SUBLconst
+	Op386SUBLcarry
+	Op386SUBLconstcarry
+	Op386SBBL
+	Op386SBBLconst
 	Op386MULL
 	Op386MULLconst
 	Op386HMULL
+	Op386HMULLU
 	Op386HMULW
 	Op386HMULB
-	Op386HMULLU
 	Op386HMULWU
 	Op386HMULBU
+	Op386MULLQU
 	Op386DIVL
 	Op386DIVW
 	Op386DIVLU
@@ -1504,6 +1511,22 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:         "ADDLconstcarry",
+		auxType:      auxInt32,
+		argLen:       1,
+		resultInArg0: true,
+		asm:          x86.AADDL,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 239}, // AX CX DX BX BP SI DI
+			},
+			outputs: []outputInfo{
+				{0, 131072}, // FLAGS
+				{1, 239},    // AX CX DX BX BP SI DI
+			},
+		},
+	},
+	{
 		name:         "ADCL",
 		argLen:       3,
 		commutative:  true,
@@ -1521,6 +1544,22 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:         "ADCLconst",
+		auxType:      auxInt32,
+		argLen:       2,
+		resultInArg0: true,
+		asm:          x86.AADCL,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{1, 131072}, // FLAGS
+				{0, 239},    // AX CX DX BX BP SI DI
+			},
+			outputs: []outputInfo{
+				{0, 239}, // AX CX DX BX BP SI DI
+			},
+		},
+	},
+	{
 		name:         "SUBL",
 		argLen:       2,
 		resultInArg0: true,
@@ -1553,6 +1592,70 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:         "SUBLcarry",
+		argLen:       2,
+		resultInArg0: true,
+		asm:          x86.ASUBL,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 239}, // AX CX DX BX BP SI DI
+				{1, 239}, // AX CX DX BX BP SI DI
+			},
+			outputs: []outputInfo{
+				{0, 131072}, // FLAGS
+				{1, 239},    // AX CX DX BX BP SI DI
+			},
+		},
+	},
+	{
+		name:         "SUBLconstcarry",
+		auxType:      auxInt32,
+		argLen:       1,
+		resultInArg0: true,
+		asm:          x86.ASUBL,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 239}, // AX CX DX BX BP SI DI
+			},
+			outputs: []outputInfo{
+				{0, 131072}, // FLAGS
+				{1, 239},    // AX CX DX BX BP SI DI
+			},
+		},
+	},
+	{
+		name:         "SBBL",
+		argLen:       3,
+		resultInArg0: true,
+		asm:          x86.ASBBL,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{2, 131072}, // FLAGS
+				{0, 239},    // AX CX DX BX BP SI DI
+				{1, 239},    // AX CX DX BX BP SI DI
+			},
+			outputs: []outputInfo{
+				{0, 239}, // AX CX DX BX BP SI DI
+			},
+		},
+	},
+	{
+		name:         "SBBLconst",
+		auxType:      auxInt32,
+		argLen:       2,
+		resultInArg0: true,
+		asm:          x86.ASBBL,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{1, 131072}, // FLAGS
+				{0, 239},    // AX CX DX BX BP SI DI
+			},
+			outputs: []outputInfo{
+				{0, 239}, // AX CX DX BX BP SI DI
+			},
+		},
+	},
+	{
 		name:         "MULL",
 		argLen:       2,
 		commutative:  true,
@@ -1601,9 +1704,9 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
-		name:   "HMULW",
+		name:   "HMULLU",
 		argLen: 2,
-		asm:    x86.AIMULW,
+		asm:    x86.AMULL,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{0, 1},   // AX
@@ -1616,9 +1719,9 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
-		name:   "HMULB",
+		name:   "HMULW",
 		argLen: 2,
-		asm:    x86.AIMULB,
+		asm:    x86.AIMULW,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{0, 1},   // AX
@@ -1631,9 +1734,9 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
-		name:   "HMULLU",
+		name:   "HMULB",
 		argLen: 2,
-		asm:    x86.AMULL,
+		asm:    x86.AIMULB,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{0, 1},   // AX
@@ -1676,6 +1779,22 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "MULLQU",
+		argLen: 2,
+		asm:    x86.AMULL,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 1},   // AX
+				{1, 255}, // AX CX DX BX SP BP SI DI
+			},
+			clobbers: 131072, // FLAGS
+			outputs: []outputInfo{
+				{0, 4}, // DX
+				{1, 1}, // AX
+			},
+		},
+	},
+	{
 		name:   "DIVL",
 		argLen: 2,
 		asm:    x86.AIDIVL,
diff --git a/src/cmd/compile/internal/ssa/rewrite386.go b/src/cmd/compile/internal/ssa/rewrite386.go
index 59b44973de..f3f021493d 100644
--- a/src/cmd/compile/internal/ssa/rewrite386.go
+++ b/src/cmd/compile/internal/ssa/rewrite386.go
@@ -8,8 +8,12 @@ import "math"
 var _ = math.MinInt8 // in case not otherwise used
 func rewriteValue386(v *Value, config *Config) bool {
 	switch v.Op {
+	case Op386ADCL:
+		return rewriteValue386_Op386ADCL(v, config)
 	case Op386ADDL:
 		return rewriteValue386_Op386ADDL(v, config)
+	case Op386ADDLcarry:
+		return rewriteValue386_Op386ADDLcarry(v, config)
 	case Op386ADDLconst:
 		return rewriteValue386_Op386ADDLconst(v, config)
 	case Op386ANDL:
@@ -362,6 +366,8 @@ func rewriteValue386(v *Value, config *Config) bool {
 		return rewriteValue386_OpMul32(v, config)
 	case OpMul32F:
 		return rewriteValue386_OpMul32F(v, config)
+	case OpMul32uhilo:
+		return rewriteValue386_OpMul32uhilo(v, config)
 	case OpMul64F:
 		return rewriteValue386_OpMul64F(v, config)
 	case OpMul8:
@@ -466,6 +472,8 @@ func rewriteValue386(v *Value, config *Config) bool {
 		return rewriteValue386_Op386SARW(v, config)
 	case Op386SARWconst:
 		return rewriteValue386_Op386SARWconst(v, config)
+	case Op386SBBL:
+		return rewriteValue386_Op386SBBL(v, config)
 	case Op386SBBLcarrymask:
 		return rewriteValue386_Op386SBBLcarrymask(v, config)
 	case Op386SETA:
@@ -498,6 +506,8 @@ func rewriteValue386(v *Value, config *Config) bool {
 		return rewriteValue386_Op386SHRW(v, config)
 	case Op386SUBL:
 		return rewriteValue386_Op386SUBL(v, config)
+	case Op386SUBLcarry:
+		return rewriteValue386_Op386SUBLcarry(v, config)
 	case Op386SUBLconst:
 		return rewriteValue386_Op386SUBLconst(v, config)
 	case OpSignExt16to32:
@@ -518,6 +528,10 @@ func rewriteValue386(v *Value, config *Config) bool {
 		return rewriteValue386_OpSub32(v, config)
 	case OpSub32F:
 		return rewriteValue386_OpSub32F(v, config)
+	case OpSub32carry:
+		return rewriteValue386_OpSub32carry(v, config)
+	case OpSub32withcarry:
+		return rewriteValue386_OpSub32withcarry(v, config)
 	case OpSub64F:
 		return rewriteValue386_OpSub64F(v, config)
 	case OpSub8:
@@ -551,6 +565,45 @@ func rewriteValue386(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValue386_Op386ADCL(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (ADCL x (MOVLconst [c]) f)
+	// cond:
+	// result: (ADCLconst [c] x f)
+	for {
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != Op386MOVLconst {
+			break
+		}
+		c := v_1.AuxInt
+		f := v.Args[2]
+		v.reset(Op386ADCLconst)
+		v.AuxInt = c
+		v.AddArg(x)
+		v.AddArg(f)
+		return true
+	}
+	// match: (ADCL (MOVLconst [c]) x f)
+	// cond:
+	// result: (ADCLconst [c] x f)
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != Op386MOVLconst {
+			break
+		}
+		c := v_0.AuxInt
+		x := v.Args[1]
+		f := v.Args[2]
+		v.reset(Op386ADCLconst)
+		v.AuxInt = c
+		v.AddArg(x)
+		v.AddArg(f)
+		return true
+	}
+	return false
+}
 func rewriteValue386_Op386ADDL(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -787,6 +840,41 @@ func rewriteValue386_Op386ADDL(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValue386_Op386ADDLcarry(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (ADDLcarry x (MOVLconst [c]))
+	// cond:
+	// result: (ADDLconstcarry [c] x)
+	for {
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != Op386MOVLconst {
+			break
+		}
+		c := v_1.AuxInt
+		v.reset(Op386ADDLconstcarry)
+		v.AuxInt = c
+		v.AddArg(x)
+		return true
+	}
+	// match: (ADDLcarry (MOVLconst [c]) x)
+	// cond:
+	// result: (ADDLconstcarry [c] x)
+	for {
+		v_0 := v.Args[0]
+		if v_0.Op != Op386MOVLconst {
+			break
+		}
+		c := v_0.AuxInt
+		x := v.Args[1]
+		v.reset(Op386ADDLconstcarry)
+		v.AuxInt = c
+		v.AddArg(x)
+		return true
+	}
+	return false
+}
 func rewriteValue386_Op386ADDLconst(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -3799,13 +3887,13 @@ func rewriteValue386_OpLoad(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (Load <t> ptr mem)
-	// cond: is32BitInt(t)
+	// cond: (is32BitInt(t) || isPtr(t))
 	// result: (MOVLload ptr mem)
 	for {
 		t := v.Type
 		ptr := v.Args[0]
 		mem := v.Args[1]
-		if !(is32BitInt(t)) {
+		if !(is32BitInt(t) || isPtr(t)) {
 			break
 		}
 		v.reset(Op386MOVLload)
@@ -9205,6 +9293,21 @@ func rewriteValue386_OpMul32F(v *Value, config *Config) bool {
 		return true
 	}
 }
+func rewriteValue386_OpMul32uhilo(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (Mul32uhilo x y)
+	// cond:
+	// result: (MULLQU x y)
+	for {
+		x := v.Args[0]
+		y := v.Args[1]
+		v.reset(Op386MULLQU)
+		v.AddArg(x)
+		v.AddArg(y)
+		return true
+	}
+}
 func rewriteValue386_OpMul64F(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -10731,6 +10834,28 @@ func rewriteValue386_Op386SARWconst(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValue386_Op386SBBL(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (SBBL x (MOVLconst [c]) f)
+	// cond:
+	// result: (SBBLconst [c] x f)
+	for {
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != Op386MOVLconst {
+			break
+		}
+		c := v_1.AuxInt
+		f := v.Args[2]
+		v.reset(Op386SBBLconst)
+		v.AuxInt = c
+		v.AddArg(x)
+		v.AddArg(f)
+		return true
+	}
+	return false
+}
 func rewriteValue386_Op386SBBLcarrymask(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -11801,6 +11926,26 @@ func rewriteValue386_Op386SUBL(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValue386_Op386SUBLcarry(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (SUBLcarry x (MOVLconst [c]))
+	// cond:
+	// result: (SUBLconstcarry [c] x)
+	for {
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != Op386MOVLconst {
+			break
+		}
+		c := v_1.AuxInt
+		v.reset(Op386SUBLconstcarry)
+		v.AuxInt = c
+		v.AddArg(x)
+		return true
+	}
+	return false
+}
 func rewriteValue386_Op386SUBLconst(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -12035,6 +12180,38 @@ func rewriteValue386_OpSub32F(v *Value, config *Config) bool {
 		return true
 	}
 }
+func rewriteValue386_OpSub32carry(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (Sub32carry x y)
+	// cond:
+	// result: (SUBLcarry x y)
+	for {
+		x := v.Args[0]
+		y := v.Args[1]
+		v.reset(Op386SUBLcarry)
+		v.AddArg(x)
+		v.AddArg(y)
+		return true
+	}
+}
+func rewriteValue386_OpSub32withcarry(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (Sub32withcarry x y c)
+	// cond:
+	// result: (SBBL x y c)
+	for {
+		x := v.Args[0]
+		y := v.Args[1]
+		c := v.Args[2]
+		v.reset(Op386SBBL)
+		v.AddArg(x)
+		v.AddArg(y)
+		v.AddArg(c)
+		return true
+	}
+}
 func rewriteValue386_OpSub64F(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
diff --git a/src/cmd/compile/internal/x86/ssa.go b/src/cmd/compile/internal/x86/ssa.go
index 1c3baaa391..ab6410b1c3 100644
--- a/src/cmd/compile/internal/x86/ssa.go
+++ b/src/cmd/compile/internal/x86/ssa.go
@@ -178,21 +178,34 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		ssa.Op386ADDSS, ssa.Op386ADDSD, ssa.Op386SUBSS, ssa.Op386SUBSD,
 		ssa.Op386MULSS, ssa.Op386MULSD, ssa.Op386DIVSS, ssa.Op386DIVSD,
 		ssa.Op386PXOR,
-		ssa.Op386ADCL:
+		ssa.Op386ADCL,
+		ssa.Op386SBBL:
 		r := gc.SSARegNum(v)
 		if r != gc.SSARegNum(v.Args[0]) {
 			v.Fatalf("input[0] and output not in same register %s", v.LongString())
 		}
 		opregreg(v.Op.Asm(), r, gc.SSARegNum(v.Args[1]))
 
-	case ssa.Op386ADDLcarry:
-		// output 0 is carry, output 1 is the low 32 bits.
+	case ssa.Op386ADDLcarry, ssa.Op386SUBLcarry:
+		// output 0 is carry/borrow, output 1 is the low 32 bits.
 		r := gc.SSARegNum1(v)
 		if r != gc.SSARegNum(v.Args[0]) {
 			v.Fatalf("input[0] and output[1] not in same register %s", v.LongString())
 		}
 		opregreg(v.Op.Asm(), r, gc.SSARegNum(v.Args[1]))
 
+	case ssa.Op386ADDLconstcarry, ssa.Op386SUBLconstcarry:
+		// output 0 is carry/borrow, output 1 is the low 32 bits.
+		r := gc.SSARegNum1(v)
+		if r != gc.SSARegNum(v.Args[0]) {
+			v.Fatalf("input[0] and output[1] not in same register %s", v.LongString())
+		}
+		p := gc.Prog(v.Op.Asm())
+		p.From.Type = obj.TYPE_CONST
+		p.From.Offset = v.AuxInt
+		p.To.Type = obj.TYPE_REG
+		p.To.Reg = r
+
 	case ssa.Op386DIVL, ssa.Op386DIVW,
 		ssa.Op386DIVLU, ssa.Op386DIVWU,
 		ssa.Op386MODL, ssa.Op386MODW,
@@ -289,6 +302,12 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 			m.To.Reg = x86.REG_DX
 		}
 
+	case ssa.Op386MULLQU:
+		// AX * args[1], high 32 bits in DX (result[0]), low 32 bits in AX (result[1]).
+		p := gc.Prog(v.Op.Asm())
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = gc.SSARegNum(v.Args[1])
+
 	case ssa.Op386ADDLconst:
 		r := gc.SSARegNum(v)
 		a := gc.SSARegNum(v.Args[0])
@@ -336,6 +355,8 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		//p.From3.Reg = gc.SSARegNum(v.Args[0])
 
 	case ssa.Op386SUBLconst,
+		ssa.Op386ADCLconst,
+		ssa.Op386SBBLconst,
 		ssa.Op386ANDLconst,
 		ssa.Op386ORLconst,
 		ssa.Op386XORLconst,