cmd/compile: optimize multiplication on loong64

Previously, multiplication on loong64 architecture was performed using
MULV and MULHVU instructions to calculate the low 64-bit and high
64-bit of a multiplication respectively. However, in most cases, only
the low 64-bits are needed. This commit enalbes only computating the low
64-bit result with the MULV instruction.

Reduce the binary size slightly.

file      before    after     Δ       %
addr2line 2833777   2833849   +72     +0.003%
asm       5267499   5266963   -536    -0.010%
buildid   2579706   2579402   -304    -0.012%
cgo       4798260   4797444   -816    -0.017%
compile   25247419  25175030  -72389  -0.287%
cover     4973091   4972027   -1064   -0.021%
dist      3631013   3565653   -65360  -1.800%
doc       4076036   4074004   -2032   -0.050%
fix       3496378   3496066   -312    -0.009%
link      6984102   6983214   -888    -0.013%
nm        2743820   2743516   -304    -0.011%
objdump   4277171   4277035   -136    -0.003%
pack      2379248   2378872   -376    -0.016%
pprof     14419090  14419874  +784    +0.005%
test2json 2684386   2684018   -368    -0.014%
trace     13640018  13631034  -8984   -0.066%
vet       7748918   7752630   +3712   +0.048%
go        15643850  15638098  -5752   -0.037%
total     127423782 127268729 -155053 -0.122%

Change-Id: Ifce4a9a3ed1d03c170681e39cb6f3541db9882dc
Reviewed-on: https://go-review.googlesource.com/c/go/+/472775
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Run-TryBot: Wayne Zuo <wdvxdr@golangcn.org>
Reviewed-by: David Chase <drchase@google.com>

5 files changed, 192 insertions, 309 deletions

diff --git a/src/cmd/compile/internal/loong64/ssa.go b/src/cmd/compile/internal/loong64/ssa.go
index 81ea25781e..54afacce7f 100644
--- a/src/cmd/compile/internal/loong64/ssa.go
+++ b/src/cmd/compile/internal/loong64/ssa.go
@@ -140,7 +140,8 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		ssa.OpLOONG64MULF,
 		ssa.OpLOONG64MULD,
 		ssa.OpLOONG64DIVF,
-		ssa.OpLOONG64DIVD:
+		ssa.OpLOONG64DIVD,
+		ssa.OpLOONG64MULV, ssa.OpLOONG64MULHV, ssa.OpLOONG64MULHVU:
 		p := s.Prog(v.Op.Asm())
 		p.From.Type = obj.TYPE_REG
 		p.From.Reg = v.Args[1].Reg()
@@ -174,32 +175,6 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		p.Reg = v.Args[0].Reg()
 		p.To.Type = obj.TYPE_REG
 		p.To.Reg = v.Reg()
-	case ssa.OpLOONG64MULV:
-		p := s.Prog(loong64.AMULV)
-		p.From.Type = obj.TYPE_REG
-		p.From.Reg = v.Args[1].Reg()
-		p.Reg = v.Args[0].Reg()
-		p.To.Type = obj.TYPE_REG
-		p.To.Reg = v.Reg1()
-		p1 := s.Prog(loong64.AMULHV)
-		p1.From.Type = obj.TYPE_REG
-		p1.From.Reg = v.Args[1].Reg()
-		p1.Reg = v.Args[0].Reg()
-		p1.To.Type = obj.TYPE_REG
-		p1.To.Reg = v.Reg0()
-	case ssa.OpLOONG64MULVU:
-		p := s.Prog(loong64.AMULV)
-		p.From.Type = obj.TYPE_REG
-		p.From.Reg = v.Args[1].Reg()
-		p.Reg = v.Args[0].Reg()
-		p.To.Type = obj.TYPE_REG
-		p.To.Reg = v.Reg1()
-		p1 := s.Prog(loong64.AMULHVU)
-		p1.From.Type = obj.TYPE_REG
-		p1.From.Reg = v.Args[1].Reg()
-		p1.Reg = v.Args[0].Reg()
-		p1.To.Type = obj.TYPE_REG
-		p1.To.Reg = v.Reg0()
 	case ssa.OpLOONG64DIVV:
 		p := s.Prog(loong64.ADIVV)
 		p.From.Type = obj.TYPE_REG
diff --git a/src/cmd/compile/internal/ssa/_gen/LOONG64.rules b/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
index e034fb2654..e68baf07f6 100644
--- a/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
+++ b/src/cmd/compile/internal/ssa/_gen/LOONG64.rules
@@ -8,16 +8,17 @@
 (Sub(Ptr|64|32|16|8) ...) => (SUBV ...)
 (Sub(32|64)F ...) => (SUB(F|D) ...)
 
-(Mul(64|32|16|8) x y) => (Select1 (MULVU x y))
+(Mul(64|32|16|8) ...) => (MULV ...)
 (Mul(32|64)F ...) => (MUL(F|D) ...)
-(Mul64uhilo ...) => (MULVU ...)
-(Select0 (Mul64uover x y)) => (Select1 <typ.UInt64> (MULVU x y))
-(Select1 (Mul64uover x y)) => (SGTU <typ.Bool> (Select0 <typ.UInt64> (MULVU x y)) (MOVVconst <typ.UInt64> [0]))
+(Select0 (Mul64uhilo x y)) => (MULHVU x y)
+(Select1 (Mul64uhilo x y)) => (MULV x y)
+(Select0 (Mul64uover x y)) => (MULV x y)
+(Select1 (Mul64uover x y)) => (SGTU <typ.Bool> (MULHVU x y) (MOVVconst <typ.UInt64> [0]))
 
-(Hmul64 x y) => (Select0 (MULV x y))
-(Hmul64u x y) => (Select0 (MULVU x y))
-(Hmul32 x y) => (SRAVconst (Select1 <typ.Int64> (MULV (SignExt32to64 x) (SignExt32to64 y))) [32])
-(Hmul32u x y) => (SRLVconst (Select1 <typ.UInt64> (MULVU (ZeroExt32to64 x) (ZeroExt32to64 y))) [32])
+(Hmul64 ...)  => (MULHV  ...)
+(Hmul64u ...) => (MULHVU ...)
+(Hmul32 x y)  => (SRAVconst (MULV (SignExt32to64 x) (SignExt32to64 y)) [32])
+(Hmul32u x y) => (SRLVconst (MULV (ZeroExt32to64 x) (ZeroExt32to64 y)) [32])
 
 (Div64 x y) => (Select1 (DIVV x y))
 (Div64u x y) => (Select1 (DIVVU x y))
@@ -591,10 +592,10 @@
 (SGTU (MOVVconst [c]) x) && is32Bit(c) => (SGTUconst [c] x)
 
 // mul by constant
-(Select1 (MULVU x (MOVVconst [-1]))) => (NEGV x)
-(Select1 (MULVU _ (MOVVconst [0]))) => (MOVVconst [0])
-(Select1 (MULVU x (MOVVconst [1]))) => x
-(Select1 (MULVU x (MOVVconst [c]))) && isPowerOfTwo64(c) => (SLLVconst [log64(c)] x)
+(MULV x (MOVVconst [-1])) => (NEGV x)
+(MULV _ (MOVVconst [0])) => (MOVVconst [0])
+(MULV x (MOVVconst [1])) => x
+(MULV x (MOVVconst [c])) && isPowerOfTwo64(c) => (SLLVconst [log64(c)] x)
 
 // div by constant
 (Select1 (DIVVU x (MOVVconst [1]))) => x
@@ -634,7 +635,7 @@
 (SLLVconst [c] (MOVVconst [d]))  => (MOVVconst [d<<uint64(c)])
 (SRLVconst [c] (MOVVconst [d]))  => (MOVVconst [int64(uint64(d)>>uint64(c))])
 (SRAVconst [c] (MOVVconst [d]))  => (MOVVconst [d>>uint64(c)])
-(Select1 (MULVU (MOVVconst [c]) (MOVVconst [d]))) => (MOVVconst [c*d])
+(MULV (MOVVconst [c]) (MOVVconst [d])) => (MOVVconst [c*d])
 (Select1 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [c/d])
 (Select1 (DIVVU (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [int64(uint64(c)/uint64(d))])
 (Select0 (DIVV  (MOVVconst [c]) (MOVVconst [d]))) && d != 0 => (MOVVconst [c%d])   // mod
diff --git a/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go b/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go
index b104660767..b593d3d41a 100644
--- a/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go
+++ b/src/cmd/compile/internal/ssa/_gen/LOONG64Ops.go
@@ -163,10 +163,11 @@ func init() {
 		{name: "SUBV", argLength: 2, reg: gp21, asm: "SUBVU"},                      // arg0 - arg1
 		{name: "SUBVconst", argLength: 1, reg: gp11, asm: "SUBVU", aux: "Int64"},   // arg0 - auxInt
 
-		{name: "MULV", argLength: 2, reg: gp22, resultNotInArgs: true, commutative: true, typ: "(Int64,Int64)"},    // arg0 * arg1, signed
-		{name: "MULVU", argLength: 2, reg: gp22, resultNotInArgs: true, commutative: true, typ: "(UInt64,UInt64)"}, // arg0 * arg1, unsigned
-		{name: "DIVV", argLength: 2, reg: gp22, resultNotInArgs: true, typ: "(Int64,Int64)"},                       // arg0 / arg1, signed
-		{name: "DIVVU", argLength: 2, reg: gp22, resultNotInArgs: true, typ: "(UInt64,UInt64)"},                    // arg0 / arg1, unsigned
+		{name: "MULV", argLength: 2, reg: gp21, asm: "MULV", commutative: true, typ: "Int64"},      // arg0 * arg1
+		{name: "MULHV", argLength: 2, reg: gp21, asm: "MULHV", commutative: true, typ: "Int64"},    // (arg0 * arg1) >> 64, signed
+		{name: "MULHVU", argLength: 2, reg: gp21, asm: "MULHVU", commutative: true, typ: "UInt64"}, // (arg0 * arg1) >> 64, unsigned
+		{name: "DIVV", argLength: 2, reg: gp22, resultNotInArgs: true, typ: "(Int64,Int64)"},       // arg0 / arg1, signed
+		{name: "DIVVU", argLength: 2, reg: gp22, resultNotInArgs: true, typ: "(UInt64,UInt64)"},    // arg0 / arg1, unsigned
 
 		{name: "ADDF", argLength: 2, reg: fp21, asm: "ADDF", commutative: true}, // arg0 + arg1
 		{name: "ADDD", argLength: 2, reg: fp21, asm: "ADDD", commutative: true}, // arg0 + arg1
diff --git a/src/cmd/compile/internal/ssa/opGen.go b/src/cmd/compile/internal/ssa/opGen.go
index 26ed4e552f..4a24012b1d 100644
--- a/src/cmd/compile/internal/ssa/opGen.go
+++ b/src/cmd/compile/internal/ssa/opGen.go
@@ -1716,7 +1716,8 @@ const (
 	OpLOONG64SUBV
 	OpLOONG64SUBVconst
 	OpLOONG64MULV
-	OpLOONG64MULVU
+	OpLOONG64MULHV
+	OpLOONG64MULHVU
 	OpLOONG64DIVV
 	OpLOONG64DIVVU
 	OpLOONG64ADDF
@@ -22912,34 +22913,47 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
-		name:            "MULV",
-		argLen:          2,
-		commutative:     true,
-		resultNotInArgs: true,
+		name:        "MULV",
+		argLen:      2,
+		commutative: true,
+		asm:         loong64.AMULV,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 1072496632}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
-				{1, 1072496632}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
+				{0, 1072693240}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
+				{1, 1072693240}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
 			},
 			outputs: []outputInfo{
 				{0, 1070596088}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R23 R24 R25 R26 R27 R28 R29 R31
-				{1, 1070596088}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R23 R24 R25 R26 R27 R28 R29 R31
 			},
 		},
 	},
 	{
-		name:            "MULVU",
-		argLen:          2,
-		commutative:     true,
-		resultNotInArgs: true,
+		name:        "MULHV",
+		argLen:      2,
+		commutative: true,
+		asm:         loong64.AMULHV,
 		reg: regInfo{
 			inputs: []inputInfo{
-				{0, 1072496632}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
-				{1, 1072496632}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
+				{0, 1072693240}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
+				{1, 1072693240}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
+			},
+			outputs: []outputInfo{
+				{0, 1070596088}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R23 R24 R25 R26 R27 R28 R29 R31
+			},
+		},
+	},
+	{
+		name:        "MULHVU",
+		argLen:      2,
+		commutative: true,
+		asm:         loong64.AMULHVU,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 1072693240}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
+				{1, 1072693240}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 g R23 R24 R25 R26 R27 R28 R29 R31
 			},
 			outputs: []outputInfo{
 				{0, 1070596088}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R23 R24 R25 R26 R27 R28 R29 R31
-				{1, 1070596088}, // R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R23 R24 R25 R26 R27 R28 R29 R31
 			},
 		},
 	},
diff --git a/src/cmd/compile/internal/ssa/rewriteLOONG64.go b/src/cmd/compile/internal/ssa/rewriteLOONG64.go
index fdd19bf167..8eacc1fda7 100644
--- a/src/cmd/compile/internal/ssa/rewriteLOONG64.go
+++ b/src/cmd/compile/internal/ssa/rewriteLOONG64.go
@@ -201,9 +201,11 @@ func rewriteValueLOONG64(v *Value) bool {
 	case OpHmul32u:
 		return rewriteValueLOONG64_OpHmul32u(v)
 	case OpHmul64:
-		return rewriteValueLOONG64_OpHmul64(v)
+		v.Op = OpLOONG64MULHV
+		return true
 	case OpHmul64u:
-		return rewriteValueLOONG64_OpHmul64u(v)
+		v.Op = OpLOONG64MULHVU
+		return true
 	case OpInterCall:
 		v.Op = OpLOONG64CALLinter
 		return true
@@ -285,6 +287,8 @@ func rewriteValueLOONG64(v *Value) bool {
 		return rewriteValueLOONG64_OpLOONG64MOVWstore(v)
 	case OpLOONG64MOVWstorezero:
 		return rewriteValueLOONG64_OpLOONG64MOVWstorezero(v)
+	case OpLOONG64MULV:
+		return rewriteValueLOONG64_OpLOONG64MULV(v)
 	case OpLOONG64NEGV:
 		return rewriteValueLOONG64_OpLOONG64NEGV(v)
 	case OpLOONG64NOR:
@@ -422,22 +426,23 @@ func rewriteValueLOONG64(v *Value) bool {
 	case OpMove:
 		return rewriteValueLOONG64_OpMove(v)
 	case OpMul16:
-		return rewriteValueLOONG64_OpMul16(v)
+		v.Op = OpLOONG64MULV
+		return true
 	case OpMul32:
-		return rewriteValueLOONG64_OpMul32(v)
+		v.Op = OpLOONG64MULV
+		return true
 	case OpMul32F:
 		v.Op = OpLOONG64MULF
 		return true
 	case OpMul64:
-		return rewriteValueLOONG64_OpMul64(v)
+		v.Op = OpLOONG64MULV
+		return true
 	case OpMul64F:
 		v.Op = OpLOONG64MULD
 		return true
-	case OpMul64uhilo:
-		v.Op = OpLOONG64MULVU
-		return true
 	case OpMul8:
-		return rewriteValueLOONG64_OpMul8(v)
+		v.Op = OpLOONG64MULV
+		return true
 	case OpNeg16:
 		v.Op = OpLOONG64NEGV
 		return true
@@ -1228,20 +1233,18 @@ func rewriteValueLOONG64_OpHmul32(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
 	// match: (Hmul32 x y)
-	// result: (SRAVconst (Select1 <typ.Int64> (MULV (SignExt32to64 x) (SignExt32to64 y))) [32])
+	// result: (SRAVconst (MULV (SignExt32to64 x) (SignExt32to64 y)) [32])
 	for {
 		x := v_0
 		y := v_1
 		v.reset(OpLOONG64SRAVconst)
 		v.AuxInt = int64ToAuxInt(32)
-		v0 := b.NewValue0(v.Pos, OpSelect1, typ.Int64)
-		v1 := b.NewValue0(v.Pos, OpLOONG64MULV, types.NewTuple(typ.Int64, typ.Int64))
+		v0 := b.NewValue0(v.Pos, OpLOONG64MULV, typ.Int64)
+		v1 := b.NewValue0(v.Pos, OpSignExt32to64, typ.Int64)
+		v1.AddArg(x)
 		v2 := b.NewValue0(v.Pos, OpSignExt32to64, typ.Int64)
-		v2.AddArg(x)
-		v3 := b.NewValue0(v.Pos, OpSignExt32to64, typ.Int64)
-		v3.AddArg(y)
-		v1.AddArg2(v2, v3)
-		v0.AddArg(v1)
+		v2.AddArg(y)
+		v0.AddArg2(v1, v2)
 		v.AddArg(v0)
 		return true
 	}
@@ -1252,54 +1255,18 @@ func rewriteValueLOONG64_OpHmul32u(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
 	// match: (Hmul32u x y)
-	// result: (SRLVconst (Select1 <typ.UInt64> (MULVU (ZeroExt32to64 x) (ZeroExt32to64 y))) [32])
+	// result: (SRLVconst (MULV (ZeroExt32to64 x) (ZeroExt32to64 y)) [32])
 	for {
 		x := v_0
 		y := v_1
 		v.reset(OpLOONG64SRLVconst)
 		v.AuxInt = int64ToAuxInt(32)
-		v0 := b.NewValue0(v.Pos, OpSelect1, typ.UInt64)
-		v1 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
+		v0 := b.NewValue0(v.Pos, OpLOONG64MULV, typ.Int64)
+		v1 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.UInt64)
+		v1.AddArg(x)
 		v2 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.UInt64)
-		v2.AddArg(x)
-		v3 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.UInt64)
-		v3.AddArg(y)
-		v1.AddArg2(v2, v3)
-		v0.AddArg(v1)
-		v.AddArg(v0)
-		return true
-	}
-}
-func rewriteValueLOONG64_OpHmul64(v *Value) bool {
-	v_1 := v.Args[1]
-	v_0 := v.Args[0]
-	b := v.Block
-	typ := &b.Func.Config.Types
-	// match: (Hmul64 x y)
-	// result: (Select0 (MULV x y))
-	for {
-		x := v_0
-		y := v_1
-		v.reset(OpSelect0)
-		v0 := b.NewValue0(v.Pos, OpLOONG64MULV, types.NewTuple(typ.Int64, typ.Int64))
-		v0.AddArg2(x, y)
-		v.AddArg(v0)
-		return true
-	}
-}
-func rewriteValueLOONG64_OpHmul64u(v *Value) bool {
-	v_1 := v.Args[1]
-	v_0 := v.Args[0]
-	b := v.Block
-	typ := &b.Func.Config.Types
-	// match: (Hmul64u x y)
-	// result: (Select0 (MULVU x y))
-	for {
-		x := v_0
-		y := v_1
-		v.reset(OpSelect0)
-		v0 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
-		v0.AddArg2(x, y)
+		v2.AddArg(y)
+		v0.AddArg2(v1, v2)
 		v.AddArg(v0)
 		return true
 	}
@@ -3272,6 +3239,89 @@ func rewriteValueLOONG64_OpLOONG64MOVWstorezero(v *Value) bool {
 	}
 	return false
 }
+func rewriteValueLOONG64_OpLOONG64MULV(v *Value) bool {
+	v_1 := v.Args[1]
+	v_0 := v.Args[0]
+	// match: (MULV x (MOVVconst [-1]))
+	// result: (NEGV x)
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			x := v_0
+			if v_1.Op != OpLOONG64MOVVconst || auxIntToInt64(v_1.AuxInt) != -1 {
+				continue
+			}
+			v.reset(OpLOONG64NEGV)
+			v.AddArg(x)
+			return true
+		}
+		break
+	}
+	// match: (MULV _ (MOVVconst [0]))
+	// result: (MOVVconst [0])
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			if v_1.Op != OpLOONG64MOVVconst || auxIntToInt64(v_1.AuxInt) != 0 {
+				continue
+			}
+			v.reset(OpLOONG64MOVVconst)
+			v.AuxInt = int64ToAuxInt(0)
+			return true
+		}
+		break
+	}
+	// match: (MULV x (MOVVconst [1]))
+	// result: x
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			x := v_0
+			if v_1.Op != OpLOONG64MOVVconst || auxIntToInt64(v_1.AuxInt) != 1 {
+				continue
+			}
+			v.copyOf(x)
+			return true
+		}
+		break
+	}
+	// match: (MULV x (MOVVconst [c]))
+	// cond: isPowerOfTwo64(c)
+	// result: (SLLVconst [log64(c)] x)
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			x := v_0
+			if v_1.Op != OpLOONG64MOVVconst {
+				continue
+			}
+			c := auxIntToInt64(v_1.AuxInt)
+			if !(isPowerOfTwo64(c)) {
+				continue
+			}
+			v.reset(OpLOONG64SLLVconst)
+			v.AuxInt = int64ToAuxInt(log64(c))
+			v.AddArg(x)
+			return true
+		}
+		break
+	}
+	// match: (MULV (MOVVconst [c]) (MOVVconst [d]))
+	// result: (MOVVconst [c*d])
+	for {
+		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
+			if v_0.Op != OpLOONG64MOVVconst {
+				continue
+			}
+			c := auxIntToInt64(v_0.AuxInt)
+			if v_1.Op != OpLOONG64MOVVconst {
+				continue
+			}
+			d := auxIntToInt64(v_1.AuxInt)
+			v.reset(OpLOONG64MOVVconst)
+			v.AuxInt = int64ToAuxInt(c * d)
+			return true
+		}
+		break
+	}
+	return false
+}
 func rewriteValueLOONG64_OpLOONG64NEGV(v *Value) bool {
 	v_0 := v.Args[0]
 	// match: (NEGV (MOVVconst [c]))
@@ -5659,74 +5709,6 @@ func rewriteValueLOONG64_OpMove(v *Value) bool {
 	}
 	return false
 }
-func rewriteValueLOONG64_OpMul16(v *Value) bool {
-	v_1 := v.Args[1]
-	v_0 := v.Args[0]
-	b := v.Block
-	typ := &b.Func.Config.Types
-	// match: (Mul16 x y)
-	// result: (Select1 (MULVU x y))
-	for {
-		x := v_0
-		y := v_1
-		v.reset(OpSelect1)
-		v0 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
-		v0.AddArg2(x, y)
-		v.AddArg(v0)
-		return true
-	}
-}
-func rewriteValueLOONG64_OpMul32(v *Value) bool {
-	v_1 := v.Args[1]
-	v_0 := v.Args[0]
-	b := v.Block
-	typ := &b.Func.Config.Types
-	// match: (Mul32 x y)
-	// result: (Select1 (MULVU x y))
-	for {
-		x := v_0
-		y := v_1
-		v.reset(OpSelect1)
-		v0 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
-		v0.AddArg2(x, y)
-		v.AddArg(v0)
-		return true
-	}
-}
-func rewriteValueLOONG64_OpMul64(v *Value) bool {
-	v_1 := v.Args[1]
-	v_0 := v.Args[0]
-	b := v.Block
-	typ := &b.Func.Config.Types
-	// match: (Mul64 x y)
-	// result: (Select1 (MULVU x y))
-	for {
-		x := v_0
-		y := v_1
-		v.reset(OpSelect1)
-		v0 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
-		v0.AddArg2(x, y)
-		v.AddArg(v0)
-		return true
-	}
-}
-func rewriteValueLOONG64_OpMul8(v *Value) bool {
-	v_1 := v.Args[1]
-	v_0 := v.Args[0]
-	b := v.Block
-	typ := &b.Func.Config.Types
-	// match: (Mul8 x y)
-	// result: (Select1 (MULVU x y))
-	for {
-		x := v_0
-		y := v_1
-		v.reset(OpSelect1)
-		v0 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
-		v0.AddArg2(x, y)
-		v.AddArg(v0)
-		return true
-	}
-}
 func rewriteValueLOONG64_OpNeq16(v *Value) bool {
 	v_1 := v.Args[1]
 	v_0 := v.Args[0]
@@ -6881,20 +6863,28 @@ func rewriteValueLOONG64_OpRsh8x8(v *Value) bool {
 func rewriteValueLOONG64_OpSelect0(v *Value) bool {
 	v_0 := v.Args[0]
 	b := v.Block
-	typ := &b.Func.Config.Types
+	// match: (Select0 (Mul64uhilo x y))
+	// result: (MULHVU x y)
+	for {
+		if v_0.Op != OpMul64uhilo {
+			break
+		}
+		y := v_0.Args[1]
+		x := v_0.Args[0]
+		v.reset(OpLOONG64MULHVU)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (Select0 (Mul64uover x y))
-	// result: (Select1 <typ.UInt64> (MULVU x y))
+	// result: (MULV x y)
 	for {
 		if v_0.Op != OpMul64uover {
 			break
 		}
 		y := v_0.Args[1]
 		x := v_0.Args[0]
-		v.reset(OpSelect1)
-		v.Type = typ.UInt64
-		v0 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
-		v0.AddArg2(x, y)
-		v.AddArg(v0)
+		v.reset(OpLOONG64MULV)
+		v.AddArg2(x, y)
 		return true
 	}
 	// match: (Select0 <t> (Add64carry x y c))
@@ -7022,8 +7012,20 @@ func rewriteValueLOONG64_OpSelect1(v *Value) bool {
 	v_0 := v.Args[0]
 	b := v.Block
 	typ := &b.Func.Config.Types
+	// match: (Select1 (Mul64uhilo x y))
+	// result: (MULV x y)
+	for {
+		if v_0.Op != OpMul64uhilo {
+			break
+		}
+		y := v_0.Args[1]
+		x := v_0.Args[0]
+		v.reset(OpLOONG64MULV)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (Select1 (Mul64uover x y))
-	// result: (SGTU <typ.Bool> (Select0 <typ.UInt64> (MULVU x y)) (MOVVconst <typ.UInt64> [0]))
+	// result: (SGTU <typ.Bool> (MULHVU x y) (MOVVconst <typ.UInt64> [0]))
 	for {
 		if v_0.Op != OpMul64uover {
 			break
@@ -7032,13 +7034,11 @@ func rewriteValueLOONG64_OpSelect1(v *Value) bool {
 		x := v_0.Args[0]
 		v.reset(OpLOONG64SGTU)
 		v.Type = typ.Bool
-		v0 := b.NewValue0(v.Pos, OpSelect0, typ.UInt64)
-		v1 := b.NewValue0(v.Pos, OpLOONG64MULVU, types.NewTuple(typ.UInt64, typ.UInt64))
-		v1.AddArg2(x, y)
-		v0.AddArg(v1)
-		v2 := b.NewValue0(v.Pos, OpLOONG64MOVVconst, typ.UInt64)
-		v2.AuxInt = int64ToAuxInt(0)
-		v.AddArg2(v0, v2)
+		v0 := b.NewValue0(v.Pos, OpLOONG64MULHVU, typ.UInt64)
+		v0.AddArg2(x, y)
+		v1 := b.NewValue0(v.Pos, OpLOONG64MOVVconst, typ.UInt64)
+		v1.AuxInt = int64ToAuxInt(0)
+		v.AddArg2(v0, v1)
 		return true
 	}
 	// match: (Select1 <t> (Add64carry x y c))
@@ -7085,90 +7085,6 @@ func rewriteValueLOONG64_OpSelect1(v *Value) bool {
 		v.AddArg2(v0, v2)
 		return true
 	}
-	// match: (Select1 (MULVU x (MOVVconst [-1])))
-	// result: (NEGV x)
-	for {
-		if v_0.Op != OpLOONG64MULVU {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		for _i0 := 0; _i0 <= 1; _i0, v_0_0, v_0_1 = _i0+1, v_0_1, v_0_0 {
-			x := v_0_0
-			if v_0_1.Op != OpLOONG64MOVVconst || auxIntToInt64(v_0_1.AuxInt) != -1 {
-				continue
-			}
-			v.reset(OpLOONG64NEGV)
-			v.AddArg(x)
-			return true
-		}
-		break
-	}
-	// match: (Select1 (MULVU _ (MOVVconst [0])))
-	// result: (MOVVconst [0])
-	for {
-		if v_0.Op != OpLOONG64MULVU {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		for _i0 := 0; _i0 <= 1; _i0, v_0_0, v_0_1 = _i0+1, v_0_1, v_0_0 {
-			if v_0_1.Op != OpLOONG64MOVVconst || auxIntToInt64(v_0_1.AuxInt) != 0 {
-				continue
-			}
-			v.reset(OpLOONG64MOVVconst)
-			v.AuxInt = int64ToAuxInt(0)
-			return true
-		}
-		break
-	}
-	// match: (Select1 (MULVU x (MOVVconst [1])))
-	// result: x
-	for {
-		if v_0.Op != OpLOONG64MULVU {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		for _i0 := 0; _i0 <= 1; _i0, v_0_0, v_0_1 = _i0+1, v_0_1, v_0_0 {
-			x := v_0_0
-			if v_0_1.Op != OpLOONG64MOVVconst || auxIntToInt64(v_0_1.AuxInt) != 1 {
-				continue
-			}
-			v.copyOf(x)
-			return true
-		}
-		break
-	}
-	// match: (Select1 (MULVU x (MOVVconst [c])))
-	// cond: isPowerOfTwo64(c)
-	// result: (SLLVconst [log64(c)] x)
-	for {
-		if v_0.Op != OpLOONG64MULVU {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		for _i0 := 0; _i0 <= 1; _i0, v_0_0, v_0_1 = _i0+1, v_0_1, v_0_0 {
-			x := v_0_0
-			if v_0_1.Op != OpLOONG64MOVVconst {
-				continue
-			}
-			c := auxIntToInt64(v_0_1.AuxInt)
-			if !(isPowerOfTwo64(c)) {
-				continue
-			}
-			v.reset(OpLOONG64SLLVconst)
-			v.AuxInt = int64ToAuxInt(log64(c))
-			v.AddArg(x)
-			return true
-		}
-		break
-	}
 	// match: (Select1 (DIVVU x (MOVVconst [1])))
 	// result: x
 	for {
@@ -7206,30 +7122,6 @@ func rewriteValueLOONG64_OpSelect1(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
-	// match: (Select1 (MULVU (MOVVconst [c]) (MOVVconst [d])))
-	// result: (MOVVconst [c*d])
-	for {
-		if v_0.Op != OpLOONG64MULVU {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		for _i0 := 0; _i0 <= 1; _i0, v_0_0, v_0_1 = _i0+1, v_0_1, v_0_0 {
-			if v_0_0.Op != OpLOONG64MOVVconst {
-				continue
-			}
-			c := auxIntToInt64(v_0_0.AuxInt)
-			if v_0_1.Op != OpLOONG64MOVVconst {
-				continue
-			}
-			d := auxIntToInt64(v_0_1.AuxInt)
-			v.reset(OpLOONG64MOVVconst)
-			v.AuxInt = int64ToAuxInt(c * d)
-			return true
-		}
-		break
-	}
 	// match: (Select1 (DIVV (MOVVconst [c]) (MOVVconst [d])))
 	// cond: d != 0
 	// result: (MOVVconst [c/d])