diff --git a/src/cmd/compile/internal/ssa/gen/generic.rules b/src/cmd/compile/internal/ssa/gen/generic.rules
index 15f629889a..12ca57ea6e 100644
--- a/src/cmd/compile/internal/ssa/gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/gen/generic.rules
@@ -1697,37 +1697,37 @@
 // Divisibility check for signed integers for power of two constant are simple mask.
 // However, we must match against the rewritten n%c == 0 -> n - c*(n/c) == 0 -> n == c*(n/c)
 // where n/c contains fixup code to handle signed n.
-(Eq8 n (Lsh8x64
+((Eq8|Neq8) n (Lsh8x64
   (Rsh8x64
     (Add8  <t> n (Rsh8Ux64  <t> (Rsh8x64  <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [kbar])))
     (Const64 <typ.UInt64> [k]))
 	(Const64 <typ.UInt64> [k]))
 ) && k > 0 && k < 7 && kbar == 8 - k
-  -> (Eq8 (And8 <t> n (Const8 <t> [int64(1<<uint(k)-1)])) (Const8 <t> [0]))
+  -> ((Eq8|Neq8) (And8 <t> n (Const8 <t> [int64(1<<uint(k)-1)])) (Const8 <t> [0]))
 
-(Eq16 n (Lsh16x64
+((Eq16|Neq16) n (Lsh16x64
   (Rsh16x64
     (Add16 <t> n (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [kbar])))
     (Const64 <typ.UInt64> [k]))
 	(Const64 <typ.UInt64> [k]))
 ) && k > 0 && k < 15 && kbar == 16 - k
-  -> (Eq16 (And16 <t> n (Const16 <t> [int64(1<<uint(k)-1)])) (Const16 <t> [0]))
+  -> ((Eq16|Neq16) (And16 <t> n (Const16 <t> [int64(1<<uint(k)-1)])) (Const16 <t> [0]))
 
-(Eq32 n (Lsh32x64
+((Eq32|Neq32) n (Lsh32x64
   (Rsh32x64
     (Add32 <t> n (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [kbar])))
     (Const64 <typ.UInt64> [k]))
 	(Const64 <typ.UInt64> [k]))
 ) && k > 0 && k < 31 && kbar == 32 - k
-  -> (Eq32 (And32 <t> n (Const32 <t> [int64(1<<uint(k)-1)])) (Const32 <t> [0]))
+  -> ((Eq32|Neq32) (And32 <t> n (Const32 <t> [int64(1<<uint(k)-1)])) (Const32 <t> [0]))
 
-(Eq64 n (Lsh64x64
+((Eq64|Neq64) n (Lsh64x64
   (Rsh64x64
     (Add64 <t> n (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [kbar])))
     (Const64 <typ.UInt64> [k]))
 	(Const64 <typ.UInt64> [k]))
 ) && k > 0 && k < 63 && kbar == 64 - k
-  -> (Eq64 (And64 <t> n (Const64 <t> [int64(1<<uint(k)-1)])) (Const64 <t> [0]))
+  -> ((Eq64|Neq64) (And64 <t> n (Const64 <t> [int64(1<<uint(k)-1)])) (Const64 <t> [0]))
 
 (Eq(8|16|32|64)  s:(Sub(8|16|32|64) x y) (Const(8|16|32|64) [0])) && s.Uses == 1 -> (Eq(8|16|32|64)  x y)
 (Neq(8|16|32|64) s:(Sub(8|16|32|64) x y) (Const(8|16|32|64) [0])) && s.Uses == 1 -> (Neq(8|16|32|64) x y)
diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go
index 4945d1974e..c104157cec 100644
--- a/src/cmd/compile/internal/ssa/rewriteAMD64.go
+++ b/src/cmd/compile/internal/ssa/rewriteAMD64.go
@@ -768,10 +768,10 @@ func rewriteValueAMD64(v *Value) bool {
 		return rewriteValueAMD64_OpEqB_0(v)
 	case OpEqPtr:
 		return rewriteValueAMD64_OpEqPtr_0(v)
-	case OpFloor:
-		return rewriteValueAMD64_OpFloor_0(v)
 	case OpFMA:
 		return rewriteValueAMD64_OpFMA_0(v)
+	case OpFloor:
+		return rewriteValueAMD64_OpFloor_0(v)
 	case OpGeq16:
 		return rewriteValueAMD64_OpGeq16_0(v)
 	case OpGeq16U:
@@ -52211,17 +52211,6 @@ func rewriteValueAMD64_OpEqPtr_0(v *Value) bool {
 		return true
 	}
 }
-func rewriteValueAMD64_OpFloor_0(v *Value) bool {
-	// match: (Floor x)
-	// result: (ROUNDSD [1] x)
-	for {
-		x := v.Args[0]
-		v.reset(OpAMD64ROUNDSD)
-		v.AuxInt = 1
-		v.AddArg(x)
-		return true
-	}
-}
 func rewriteValueAMD64_OpFMA_0(v *Value) bool {
 	// match: (FMA x y z)
 	// result: (VFMADD231SD z x y)
@@ -52236,6 +52225,17 @@ func rewriteValueAMD64_OpFMA_0(v *Value) bool {
 		return true
 	}
 }
+func rewriteValueAMD64_OpFloor_0(v *Value) bool {
+	// match: (Floor x)
+	// result: (ROUNDSD [1] x)
+	for {
+		x := v.Args[0]
+		v.reset(OpAMD64ROUNDSD)
+		v.AuxInt = 1
+		v.AddArg(x)
+		return true
+	}
+}
 func rewriteValueAMD64_OpGeq16_0(v *Value) bool {
 	b := v.Block
 	// match: (Geq16 x y)
diff --git a/src/cmd/compile/internal/ssa/rewriteARM64.go b/src/cmd/compile/internal/ssa/rewriteARM64.go
index c372865d0a..6ced6baeac 100644
--- a/src/cmd/compile/internal/ssa/rewriteARM64.go
+++ b/src/cmd/compile/internal/ssa/rewriteARM64.go
@@ -571,10 +571,10 @@ func rewriteValueARM64(v *Value) bool {
 		return rewriteValueARM64_OpEqB_0(v)
 	case OpEqPtr:
 		return rewriteValueARM64_OpEqPtr_0(v)
-	case OpFloor:
-		return rewriteValueARM64_OpFloor_0(v)
 	case OpFMA:
 		return rewriteValueARM64_OpFMA_0(v)
+	case OpFloor:
+		return rewriteValueARM64_OpFloor_0(v)
 	case OpGeq16:
 		return rewriteValueARM64_OpGeq16_0(v)
 	case OpGeq16U:
@@ -28573,16 +28573,6 @@ func rewriteValueARM64_OpEqPtr_0(v *Value) bool {
 		return true
 	}
 }
-func rewriteValueARM64_OpFloor_0(v *Value) bool {
-	// match: (Floor x)
-	// result: (FRINTMD x)
-	for {
-		x := v.Args[0]
-		v.reset(OpARM64FRINTMD)
-		v.AddArg(x)
-		return true
-	}
-}
 func rewriteValueARM64_OpFMA_0(v *Value) bool {
 	// match: (FMA x y z)
 	// result: (FMADDD z x y)
@@ -28597,6 +28587,16 @@ func rewriteValueARM64_OpFMA_0(v *Value) bool {
 		return true
 	}
 }
+func rewriteValueARM64_OpFloor_0(v *Value) bool {
+	// match: (Floor x)
+	// result: (FRINTMD x)
+	for {
+		x := v.Args[0]
+		v.reset(OpARM64FRINTMD)
+		v.AddArg(x)
+		return true
+	}
+}
 func rewriteValueARM64_OpGeq16_0(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
diff --git a/src/cmd/compile/internal/ssa/rewritePPC64.go b/src/cmd/compile/internal/ssa/rewritePPC64.go
index 4d282a6730..65f1f12f75 100644
--- a/src/cmd/compile/internal/ssa/rewritePPC64.go
+++ b/src/cmd/compile/internal/ssa/rewritePPC64.go
@@ -181,10 +181,10 @@ func rewriteValuePPC64(v *Value) bool {
 		return rewriteValuePPC64_OpEqB_0(v)
 	case OpEqPtr:
 		return rewriteValuePPC64_OpEqPtr_0(v)
-	case OpFloor:
-		return rewriteValuePPC64_OpFloor_0(v)
 	case OpFMA:
 		return rewriteValuePPC64_OpFMA_0(v)
+	case OpFloor:
+		return rewriteValuePPC64_OpFloor_0(v)
 	case OpGeq16:
 		return rewriteValuePPC64_OpGeq16_0(v)
 	case OpGeq16U:
@@ -1997,16 +1997,6 @@ func rewriteValuePPC64_OpEqPtr_0(v *Value) bool {
 		return true
 	}
 }
-func rewriteValuePPC64_OpFloor_0(v *Value) bool {
-	// match: (Floor x)
-	// result: (FFLOOR x)
-	for {
-		x := v.Args[0]
-		v.reset(OpPPC64FFLOOR)
-		v.AddArg(x)
-		return true
-	}
-}
 func rewriteValuePPC64_OpFMA_0(v *Value) bool {
 	// match: (FMA x y z)
 	// result: (FMADD x y z)
@@ -2021,6 +2011,16 @@ func rewriteValuePPC64_OpFMA_0(v *Value) bool {
 		return true
 	}
 }
+func rewriteValuePPC64_OpFloor_0(v *Value) bool {
+	// match: (Floor x)
+	// result: (FFLOOR x)
+	for {
+		x := v.Args[0]
+		v.reset(OpPPC64FFLOOR)
+		v.AddArg(x)
+		return true
+	}
+}
 func rewriteValuePPC64_OpGeq16_0(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
diff --git a/src/cmd/compile/internal/ssa/rewriteS390X.go b/src/cmd/compile/internal/ssa/rewriteS390X.go
index 429e3d5be0..5c3be6118b 100644
--- a/src/cmd/compile/internal/ssa/rewriteS390X.go
+++ b/src/cmd/compile/internal/ssa/rewriteS390X.go
@@ -166,10 +166,10 @@ func rewriteValueS390X(v *Value) bool {
 		return rewriteValueS390X_OpEqB_0(v)
 	case OpEqPtr:
 		return rewriteValueS390X_OpEqPtr_0(v)
-	case OpFloor:
-		return rewriteValueS390X_OpFloor_0(v)
 	case OpFMA:
 		return rewriteValueS390X_OpFMA_0(v)
+	case OpFloor:
+		return rewriteValueS390X_OpFloor_0(v)
 	case OpGeq16:
 		return rewriteValueS390X_OpGeq16_0(v)
 	case OpGeq16U:
@@ -1928,17 +1928,6 @@ func rewriteValueS390X_OpEqPtr_0(v *Value) bool {
 		return true
 	}
 }
-func rewriteValueS390X_OpFloor_0(v *Value) bool {
-	// match: (Floor x)
-	// result: (FIDBR [7] x)
-	for {
-		x := v.Args[0]
-		v.reset(OpS390XFIDBR)
-		v.AuxInt = 7
-		v.AddArg(x)
-		return true
-	}
-}
 func rewriteValueS390X_OpFMA_0(v *Value) bool {
 	// match: (FMA x y z)
 	// result: (FMADD z x y)
@@ -1953,6 +1942,17 @@ func rewriteValueS390X_OpFMA_0(v *Value) bool {
 		return true
 	}
 }
+func rewriteValueS390X_OpFloor_0(v *Value) bool {
+	// match: (Floor x)
+	// result: (FIDBR [7] x)
+	for {
+		x := v.Args[0]
+		v.reset(OpS390XFIDBR)
+		v.AuxInt = 7
+		v.AddArg(x)
+		return true
+	}
+}
 func rewriteValueS390X_OpGeq16_0(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index 7058e11215..d65d900843 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -34171,6 +34171,7 @@ func rewriteValuegeneric_OpNeg8_0(v *Value) bool {
 }
 func rewriteValuegeneric_OpNeq16_0(v *Value) bool {
 	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (Neq16 x x)
 	// result: (ConstBool [0])
 	for {
@@ -34328,6 +34329,281 @@ func rewriteValuegeneric_OpNeq16_0(v *Value) bool {
 		v.AuxInt = b2i(c != d)
 		return true
 	}
+	// match: (Neq16 n (Lsh16x64 (Rsh16x64 (Add16 <t> n (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 15 && kbar == 16 - k
+	// result: (Neq16 (And16 <t> n (Const16 <t> [int64(1<<uint(k)-1)])) (Const16 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh16x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh16x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd16 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		if n != v_1_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_1 := v_1_0_0.Args[1]
+		if v_1_0_0_1.Op != OpRsh16Ux64 || v_1_0_0_1.Type != t {
+			break
+		}
+		_ = v_1_0_0_1.Args[1]
+		v_1_0_0_1_0 := v_1_0_0_1.Args[0]
+		if v_1_0_0_1_0.Op != OpRsh16x64 || v_1_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_1_0.Args[1]
+		if n != v_1_0_0_1_0.Args[0] {
+			break
+		}
+		v_1_0_0_1_0_1 := v_1_0_0_1_0.Args[1]
+		if v_1_0_0_1_0_1.Op != OpConst64 || v_1_0_0_1_0_1.Type != typ.UInt64 || v_1_0_0_1_0_1.AuxInt != 15 {
+			break
+		}
+		v_1_0_0_1_1 := v_1_0_0_1.Args[1]
+		if v_1_0_0_1_1.Op != OpConst64 || v_1_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_1_1.AuxInt
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 15 && kbar == 16-k) {
+			break
+		}
+		v.reset(OpNeq16)
+		v0 := b.NewValue0(v.Pos, OpAnd16, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst16, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst16, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq16 n (Lsh16x64 (Rsh16x64 (Add16 <t> (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 15 && kbar == 16 - k
+	// result: (Neq16 (And16 <t> n (Const16 <t> [int64(1<<uint(k)-1)])) (Const16 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh16x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh16x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd16 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		v_1_0_0_0 := v_1_0_0.Args[0]
+		if v_1_0_0_0.Op != OpRsh16Ux64 || v_1_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0.Args[1]
+		v_1_0_0_0_0 := v_1_0_0_0.Args[0]
+		if v_1_0_0_0_0.Op != OpRsh16x64 || v_1_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0_0.Args[1]
+		if n != v_1_0_0_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_0_0_1 := v_1_0_0_0_0.Args[1]
+		if v_1_0_0_0_0_1.Op != OpConst64 || v_1_0_0_0_0_1.Type != typ.UInt64 || v_1_0_0_0_0_1.AuxInt != 15 {
+			break
+		}
+		v_1_0_0_0_1 := v_1_0_0_0.Args[1]
+		if v_1_0_0_0_1.Op != OpConst64 || v_1_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_0_1.AuxInt
+		if n != v_1_0_0.Args[1] {
+			break
+		}
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 15 && kbar == 16-k) {
+			break
+		}
+		v.reset(OpNeq16)
+		v0 := b.NewValue0(v.Pos, OpAnd16, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst16, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst16, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq16 (Lsh16x64 (Rsh16x64 (Add16 <t> n (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 15 && kbar == 16 - k
+	// result: (Neq16 (And16 <t> n (Const16 <t> [int64(1<<uint(k)-1)])) (Const16 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh16x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh16x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd16 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		if n != v_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_1 := v_0_0_0.Args[1]
+		if v_0_0_0_1.Op != OpRsh16Ux64 || v_0_0_0_1.Type != t {
+			break
+		}
+		_ = v_0_0_0_1.Args[1]
+		v_0_0_0_1_0 := v_0_0_0_1.Args[0]
+		if v_0_0_0_1_0.Op != OpRsh16x64 || v_0_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_1_0.Args[1]
+		if n != v_0_0_0_1_0.Args[0] {
+			break
+		}
+		v_0_0_0_1_0_1 := v_0_0_0_1_0.Args[1]
+		if v_0_0_0_1_0_1.Op != OpConst64 || v_0_0_0_1_0_1.Type != typ.UInt64 || v_0_0_0_1_0_1.AuxInt != 15 {
+			break
+		}
+		v_0_0_0_1_1 := v_0_0_0_1.Args[1]
+		if v_0_0_0_1_1.Op != OpConst64 || v_0_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_1_1.AuxInt
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 15 && kbar == 16-k) {
+			break
+		}
+		v.reset(OpNeq16)
+		v0 := b.NewValue0(v.Pos, OpAnd16, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst16, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst16, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	return false
+}
+func rewriteValuegeneric_OpNeq16_10(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
+	// match: (Neq16 (Lsh16x64 (Rsh16x64 (Add16 <t> (Rsh16Ux64 <t> (Rsh16x64 <t> n (Const64 <typ.UInt64> [15])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 15 && kbar == 16 - k
+	// result: (Neq16 (And16 <t> n (Const16 <t> [int64(1<<uint(k)-1)])) (Const16 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh16x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh16x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd16 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		v_0_0_0_0 := v_0_0_0.Args[0]
+		if v_0_0_0_0.Op != OpRsh16Ux64 || v_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0.Args[1]
+		v_0_0_0_0_0 := v_0_0_0_0.Args[0]
+		if v_0_0_0_0_0.Op != OpRsh16x64 || v_0_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0_0.Args[1]
+		if n != v_0_0_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_0_0_1 := v_0_0_0_0_0.Args[1]
+		if v_0_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_0_1.Type != typ.UInt64 || v_0_0_0_0_0_1.AuxInt != 15 {
+			break
+		}
+		v_0_0_0_0_1 := v_0_0_0_0.Args[1]
+		if v_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_0_1.AuxInt
+		if n != v_0_0_0.Args[1] {
+			break
+		}
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 15 && kbar == 16-k) {
+			break
+		}
+		v.reset(OpNeq16)
+		v0 := b.NewValue0(v.Pos, OpAnd16, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst16, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst16, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Neq16 s:(Sub16 x y) (Const16 [0]))
 	// cond: s.Uses == 1
 	// result: (Neq16 x y)
@@ -34404,10 +34680,6 @@ func rewriteValuegeneric_OpNeq16_0(v *Value) bool {
 		v.AddArg(v2)
 		return true
 	}
-	return false
-}
-func rewriteValuegeneric_OpNeq16_10(v *Value) bool {
-	b := v.Block
 	// match: (Neq16 (And16 <t> (Const16 <t> [y]) x) (Const16 <t> [y]))
 	// cond: isPowerOfTwo(y)
 	// result: (Eq16 (And16 <t> x (Const16 <t> [y])) (Const16 <t> [0]))
@@ -34509,6 +34781,7 @@ func rewriteValuegeneric_OpNeq16_10(v *Value) bool {
 }
 func rewriteValuegeneric_OpNeq32_0(v *Value) bool {
 	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (Neq32 x x)
 	// result: (ConstBool [0])
 	for {
@@ -34666,6 +34939,281 @@ func rewriteValuegeneric_OpNeq32_0(v *Value) bool {
 		v.AuxInt = b2i(c != d)
 		return true
 	}
+	// match: (Neq32 n (Lsh32x64 (Rsh32x64 (Add32 <t> n (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 31 && kbar == 32 - k
+	// result: (Neq32 (And32 <t> n (Const32 <t> [int64(1<<uint(k)-1)])) (Const32 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh32x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh32x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd32 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		if n != v_1_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_1 := v_1_0_0.Args[1]
+		if v_1_0_0_1.Op != OpRsh32Ux64 || v_1_0_0_1.Type != t {
+			break
+		}
+		_ = v_1_0_0_1.Args[1]
+		v_1_0_0_1_0 := v_1_0_0_1.Args[0]
+		if v_1_0_0_1_0.Op != OpRsh32x64 || v_1_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_1_0.Args[1]
+		if n != v_1_0_0_1_0.Args[0] {
+			break
+		}
+		v_1_0_0_1_0_1 := v_1_0_0_1_0.Args[1]
+		if v_1_0_0_1_0_1.Op != OpConst64 || v_1_0_0_1_0_1.Type != typ.UInt64 || v_1_0_0_1_0_1.AuxInt != 31 {
+			break
+		}
+		v_1_0_0_1_1 := v_1_0_0_1.Args[1]
+		if v_1_0_0_1_1.Op != OpConst64 || v_1_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_1_1.AuxInt
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 31 && kbar == 32-k) {
+			break
+		}
+		v.reset(OpNeq32)
+		v0 := b.NewValue0(v.Pos, OpAnd32, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst32, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst32, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq32 n (Lsh32x64 (Rsh32x64 (Add32 <t> (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 31 && kbar == 32 - k
+	// result: (Neq32 (And32 <t> n (Const32 <t> [int64(1<<uint(k)-1)])) (Const32 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh32x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh32x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd32 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		v_1_0_0_0 := v_1_0_0.Args[0]
+		if v_1_0_0_0.Op != OpRsh32Ux64 || v_1_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0.Args[1]
+		v_1_0_0_0_0 := v_1_0_0_0.Args[0]
+		if v_1_0_0_0_0.Op != OpRsh32x64 || v_1_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0_0.Args[1]
+		if n != v_1_0_0_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_0_0_1 := v_1_0_0_0_0.Args[1]
+		if v_1_0_0_0_0_1.Op != OpConst64 || v_1_0_0_0_0_1.Type != typ.UInt64 || v_1_0_0_0_0_1.AuxInt != 31 {
+			break
+		}
+		v_1_0_0_0_1 := v_1_0_0_0.Args[1]
+		if v_1_0_0_0_1.Op != OpConst64 || v_1_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_0_1.AuxInt
+		if n != v_1_0_0.Args[1] {
+			break
+		}
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 31 && kbar == 32-k) {
+			break
+		}
+		v.reset(OpNeq32)
+		v0 := b.NewValue0(v.Pos, OpAnd32, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst32, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst32, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq32 (Lsh32x64 (Rsh32x64 (Add32 <t> n (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 31 && kbar == 32 - k
+	// result: (Neq32 (And32 <t> n (Const32 <t> [int64(1<<uint(k)-1)])) (Const32 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh32x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh32x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd32 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		if n != v_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_1 := v_0_0_0.Args[1]
+		if v_0_0_0_1.Op != OpRsh32Ux64 || v_0_0_0_1.Type != t {
+			break
+		}
+		_ = v_0_0_0_1.Args[1]
+		v_0_0_0_1_0 := v_0_0_0_1.Args[0]
+		if v_0_0_0_1_0.Op != OpRsh32x64 || v_0_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_1_0.Args[1]
+		if n != v_0_0_0_1_0.Args[0] {
+			break
+		}
+		v_0_0_0_1_0_1 := v_0_0_0_1_0.Args[1]
+		if v_0_0_0_1_0_1.Op != OpConst64 || v_0_0_0_1_0_1.Type != typ.UInt64 || v_0_0_0_1_0_1.AuxInt != 31 {
+			break
+		}
+		v_0_0_0_1_1 := v_0_0_0_1.Args[1]
+		if v_0_0_0_1_1.Op != OpConst64 || v_0_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_1_1.AuxInt
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 31 && kbar == 32-k) {
+			break
+		}
+		v.reset(OpNeq32)
+		v0 := b.NewValue0(v.Pos, OpAnd32, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst32, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst32, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	return false
+}
+func rewriteValuegeneric_OpNeq32_10(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
+	// match: (Neq32 (Lsh32x64 (Rsh32x64 (Add32 <t> (Rsh32Ux64 <t> (Rsh32x64 <t> n (Const64 <typ.UInt64> [31])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 31 && kbar == 32 - k
+	// result: (Neq32 (And32 <t> n (Const32 <t> [int64(1<<uint(k)-1)])) (Const32 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh32x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh32x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd32 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		v_0_0_0_0 := v_0_0_0.Args[0]
+		if v_0_0_0_0.Op != OpRsh32Ux64 || v_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0.Args[1]
+		v_0_0_0_0_0 := v_0_0_0_0.Args[0]
+		if v_0_0_0_0_0.Op != OpRsh32x64 || v_0_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0_0.Args[1]
+		if n != v_0_0_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_0_0_1 := v_0_0_0_0_0.Args[1]
+		if v_0_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_0_1.Type != typ.UInt64 || v_0_0_0_0_0_1.AuxInt != 31 {
+			break
+		}
+		v_0_0_0_0_1 := v_0_0_0_0.Args[1]
+		if v_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_0_1.AuxInt
+		if n != v_0_0_0.Args[1] {
+			break
+		}
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 31 && kbar == 32-k) {
+			break
+		}
+		v.reset(OpNeq32)
+		v0 := b.NewValue0(v.Pos, OpAnd32, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst32, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst32, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Neq32 s:(Sub32 x y) (Const32 [0]))
 	// cond: s.Uses == 1
 	// result: (Neq32 x y)
@@ -34742,10 +35290,6 @@ func rewriteValuegeneric_OpNeq32_0(v *Value) bool {
 		v.AddArg(v2)
 		return true
 	}
-	return false
-}
-func rewriteValuegeneric_OpNeq32_10(v *Value) bool {
-	b := v.Block
 	// match: (Neq32 (And32 <t> (Const32 <t> [y]) x) (Const32 <t> [y]))
 	// cond: isPowerOfTwo(y)
 	// result: (Eq32 (And32 <t> x (Const32 <t> [y])) (Const32 <t> [0]))
@@ -34886,6 +35430,7 @@ func rewriteValuegeneric_OpNeq32F_0(v *Value) bool {
 }
 func rewriteValuegeneric_OpNeq64_0(v *Value) bool {
 	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (Neq64 x x)
 	// result: (ConstBool [0])
 	for {
@@ -35043,6 +35588,281 @@ func rewriteValuegeneric_OpNeq64_0(v *Value) bool {
 		v.AuxInt = b2i(c != d)
 		return true
 	}
+	// match: (Neq64 n (Lsh64x64 (Rsh64x64 (Add64 <t> n (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 63 && kbar == 64 - k
+	// result: (Neq64 (And64 <t> n (Const64 <t> [int64(1<<uint(k)-1)])) (Const64 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh64x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh64x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd64 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		if n != v_1_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_1 := v_1_0_0.Args[1]
+		if v_1_0_0_1.Op != OpRsh64Ux64 || v_1_0_0_1.Type != t {
+			break
+		}
+		_ = v_1_0_0_1.Args[1]
+		v_1_0_0_1_0 := v_1_0_0_1.Args[0]
+		if v_1_0_0_1_0.Op != OpRsh64x64 || v_1_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_1_0.Args[1]
+		if n != v_1_0_0_1_0.Args[0] {
+			break
+		}
+		v_1_0_0_1_0_1 := v_1_0_0_1_0.Args[1]
+		if v_1_0_0_1_0_1.Op != OpConst64 || v_1_0_0_1_0_1.Type != typ.UInt64 || v_1_0_0_1_0_1.AuxInt != 63 {
+			break
+		}
+		v_1_0_0_1_1 := v_1_0_0_1.Args[1]
+		if v_1_0_0_1_1.Op != OpConst64 || v_1_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_1_1.AuxInt
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 63 && kbar == 64-k) {
+			break
+		}
+		v.reset(OpNeq64)
+		v0 := b.NewValue0(v.Pos, OpAnd64, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq64 n (Lsh64x64 (Rsh64x64 (Add64 <t> (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 63 && kbar == 64 - k
+	// result: (Neq64 (And64 <t> n (Const64 <t> [int64(1<<uint(k)-1)])) (Const64 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh64x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh64x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd64 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		v_1_0_0_0 := v_1_0_0.Args[0]
+		if v_1_0_0_0.Op != OpRsh64Ux64 || v_1_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0.Args[1]
+		v_1_0_0_0_0 := v_1_0_0_0.Args[0]
+		if v_1_0_0_0_0.Op != OpRsh64x64 || v_1_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0_0.Args[1]
+		if n != v_1_0_0_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_0_0_1 := v_1_0_0_0_0.Args[1]
+		if v_1_0_0_0_0_1.Op != OpConst64 || v_1_0_0_0_0_1.Type != typ.UInt64 || v_1_0_0_0_0_1.AuxInt != 63 {
+			break
+		}
+		v_1_0_0_0_1 := v_1_0_0_0.Args[1]
+		if v_1_0_0_0_1.Op != OpConst64 || v_1_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_0_1.AuxInt
+		if n != v_1_0_0.Args[1] {
+			break
+		}
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 63 && kbar == 64-k) {
+			break
+		}
+		v.reset(OpNeq64)
+		v0 := b.NewValue0(v.Pos, OpAnd64, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq64 (Lsh64x64 (Rsh64x64 (Add64 <t> n (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 63 && kbar == 64 - k
+	// result: (Neq64 (And64 <t> n (Const64 <t> [int64(1<<uint(k)-1)])) (Const64 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh64x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh64x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd64 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		if n != v_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_1 := v_0_0_0.Args[1]
+		if v_0_0_0_1.Op != OpRsh64Ux64 || v_0_0_0_1.Type != t {
+			break
+		}
+		_ = v_0_0_0_1.Args[1]
+		v_0_0_0_1_0 := v_0_0_0_1.Args[0]
+		if v_0_0_0_1_0.Op != OpRsh64x64 || v_0_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_1_0.Args[1]
+		if n != v_0_0_0_1_0.Args[0] {
+			break
+		}
+		v_0_0_0_1_0_1 := v_0_0_0_1_0.Args[1]
+		if v_0_0_0_1_0_1.Op != OpConst64 || v_0_0_0_1_0_1.Type != typ.UInt64 || v_0_0_0_1_0_1.AuxInt != 63 {
+			break
+		}
+		v_0_0_0_1_1 := v_0_0_0_1.Args[1]
+		if v_0_0_0_1_1.Op != OpConst64 || v_0_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_1_1.AuxInt
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 63 && kbar == 64-k) {
+			break
+		}
+		v.reset(OpNeq64)
+		v0 := b.NewValue0(v.Pos, OpAnd64, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	return false
+}
+func rewriteValuegeneric_OpNeq64_10(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
+	// match: (Neq64 (Lsh64x64 (Rsh64x64 (Add64 <t> (Rsh64Ux64 <t> (Rsh64x64 <t> n (Const64 <typ.UInt64> [63])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 63 && kbar == 64 - k
+	// result: (Neq64 (And64 <t> n (Const64 <t> [int64(1<<uint(k)-1)])) (Const64 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh64x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh64x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd64 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		v_0_0_0_0 := v_0_0_0.Args[0]
+		if v_0_0_0_0.Op != OpRsh64Ux64 || v_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0.Args[1]
+		v_0_0_0_0_0 := v_0_0_0_0.Args[0]
+		if v_0_0_0_0_0.Op != OpRsh64x64 || v_0_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0_0.Args[1]
+		if n != v_0_0_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_0_0_1 := v_0_0_0_0_0.Args[1]
+		if v_0_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_0_1.Type != typ.UInt64 || v_0_0_0_0_0_1.AuxInt != 63 {
+			break
+		}
+		v_0_0_0_0_1 := v_0_0_0_0.Args[1]
+		if v_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_0_1.AuxInt
+		if n != v_0_0_0.Args[1] {
+			break
+		}
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 63 && kbar == 64-k) {
+			break
+		}
+		v.reset(OpNeq64)
+		v0 := b.NewValue0(v.Pos, OpAnd64, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Neq64 s:(Sub64 x y) (Const64 [0]))
 	// cond: s.Uses == 1
 	// result: (Neq64 x y)
@@ -35119,10 +35939,6 @@ func rewriteValuegeneric_OpNeq64_0(v *Value) bool {
 		v.AddArg(v2)
 		return true
 	}
-	return false
-}
-func rewriteValuegeneric_OpNeq64_10(v *Value) bool {
-	b := v.Block
 	// match: (Neq64 (And64 <t> (Const64 <t> [y]) x) (Const64 <t> [y]))
 	// cond: isPowerOfTwo(y)
 	// result: (Eq64 (And64 <t> x (Const64 <t> [y])) (Const64 <t> [0]))
@@ -35263,6 +36079,7 @@ func rewriteValuegeneric_OpNeq64F_0(v *Value) bool {
 }
 func rewriteValuegeneric_OpNeq8_0(v *Value) bool {
 	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (Neq8 x x)
 	// result: (ConstBool [0])
 	for {
@@ -35420,6 +36237,281 @@ func rewriteValuegeneric_OpNeq8_0(v *Value) bool {
 		v.AuxInt = b2i(c != d)
 		return true
 	}
+	// match: (Neq8 n (Lsh8x64 (Rsh8x64 (Add8 <t> n (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 7 && kbar == 8 - k
+	// result: (Neq8 (And8 <t> n (Const8 <t> [int64(1<<uint(k)-1)])) (Const8 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh8x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh8x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd8 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		if n != v_1_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_1 := v_1_0_0.Args[1]
+		if v_1_0_0_1.Op != OpRsh8Ux64 || v_1_0_0_1.Type != t {
+			break
+		}
+		_ = v_1_0_0_1.Args[1]
+		v_1_0_0_1_0 := v_1_0_0_1.Args[0]
+		if v_1_0_0_1_0.Op != OpRsh8x64 || v_1_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_1_0.Args[1]
+		if n != v_1_0_0_1_0.Args[0] {
+			break
+		}
+		v_1_0_0_1_0_1 := v_1_0_0_1_0.Args[1]
+		if v_1_0_0_1_0_1.Op != OpConst64 || v_1_0_0_1_0_1.Type != typ.UInt64 || v_1_0_0_1_0_1.AuxInt != 7 {
+			break
+		}
+		v_1_0_0_1_1 := v_1_0_0_1.Args[1]
+		if v_1_0_0_1_1.Op != OpConst64 || v_1_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_1_1.AuxInt
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 7 && kbar == 8-k) {
+			break
+		}
+		v.reset(OpNeq8)
+		v0 := b.NewValue0(v.Pos, OpAnd8, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst8, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst8, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq8 n (Lsh8x64 (Rsh8x64 (Add8 <t> (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])))
+	// cond: k > 0 && k < 7 && kbar == 8 - k
+	// result: (Neq8 (And8 <t> n (Const8 <t> [int64(1<<uint(k)-1)])) (Const8 <t> [0]))
+	for {
+		_ = v.Args[1]
+		n := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpLsh8x64 {
+			break
+		}
+		_ = v_1.Args[1]
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpRsh8x64 {
+			break
+		}
+		_ = v_1_0.Args[1]
+		v_1_0_0 := v_1_0.Args[0]
+		if v_1_0_0.Op != OpAdd8 {
+			break
+		}
+		t := v_1_0_0.Type
+		_ = v_1_0_0.Args[1]
+		v_1_0_0_0 := v_1_0_0.Args[0]
+		if v_1_0_0_0.Op != OpRsh8Ux64 || v_1_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0.Args[1]
+		v_1_0_0_0_0 := v_1_0_0_0.Args[0]
+		if v_1_0_0_0_0.Op != OpRsh8x64 || v_1_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_1_0_0_0_0.Args[1]
+		if n != v_1_0_0_0_0.Args[0] {
+			break
+		}
+		v_1_0_0_0_0_1 := v_1_0_0_0_0.Args[1]
+		if v_1_0_0_0_0_1.Op != OpConst64 || v_1_0_0_0_0_1.Type != typ.UInt64 || v_1_0_0_0_0_1.AuxInt != 7 {
+			break
+		}
+		v_1_0_0_0_1 := v_1_0_0_0.Args[1]
+		if v_1_0_0_0_1.Op != OpConst64 || v_1_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_1_0_0_0_1.AuxInt
+		if n != v_1_0_0.Args[1] {
+			break
+		}
+		v_1_0_1 := v_1_0.Args[1]
+		if v_1_0_1.Op != OpConst64 || v_1_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_1_0_1.AuxInt
+		v_1_1 := v_1.Args[1]
+		if v_1_1.Op != OpConst64 || v_1_1.Type != typ.UInt64 || v_1_1.AuxInt != k || !(k > 0 && k < 7 && kbar == 8-k) {
+			break
+		}
+		v.reset(OpNeq8)
+		v0 := b.NewValue0(v.Pos, OpAnd8, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst8, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst8, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	// match: (Neq8 (Lsh8x64 (Rsh8x64 (Add8 <t> n (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [kbar]))) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 7 && kbar == 8 - k
+	// result: (Neq8 (And8 <t> n (Const8 <t> [int64(1<<uint(k)-1)])) (Const8 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh8x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh8x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd8 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		if n != v_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_1 := v_0_0_0.Args[1]
+		if v_0_0_0_1.Op != OpRsh8Ux64 || v_0_0_0_1.Type != t {
+			break
+		}
+		_ = v_0_0_0_1.Args[1]
+		v_0_0_0_1_0 := v_0_0_0_1.Args[0]
+		if v_0_0_0_1_0.Op != OpRsh8x64 || v_0_0_0_1_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_1_0.Args[1]
+		if n != v_0_0_0_1_0.Args[0] {
+			break
+		}
+		v_0_0_0_1_0_1 := v_0_0_0_1_0.Args[1]
+		if v_0_0_0_1_0_1.Op != OpConst64 || v_0_0_0_1_0_1.Type != typ.UInt64 || v_0_0_0_1_0_1.AuxInt != 7 {
+			break
+		}
+		v_0_0_0_1_1 := v_0_0_0_1.Args[1]
+		if v_0_0_0_1_1.Op != OpConst64 || v_0_0_0_1_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_1_1.AuxInt
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 7 && kbar == 8-k) {
+			break
+		}
+		v.reset(OpNeq8)
+		v0 := b.NewValue0(v.Pos, OpAnd8, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst8, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst8, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
+	return false
+}
+func rewriteValuegeneric_OpNeq8_10(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
+	// match: (Neq8 (Lsh8x64 (Rsh8x64 (Add8 <t> (Rsh8Ux64 <t> (Rsh8x64 <t> n (Const64 <typ.UInt64> [ 7])) (Const64 <typ.UInt64> [kbar])) n) (Const64 <typ.UInt64> [k])) (Const64 <typ.UInt64> [k])) n)
+	// cond: k > 0 && k < 7 && kbar == 8 - k
+	// result: (Neq8 (And8 <t> n (Const8 <t> [int64(1<<uint(k)-1)])) (Const8 <t> [0]))
+	for {
+		n := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpLsh8x64 {
+			break
+		}
+		_ = v_0.Args[1]
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpRsh8x64 {
+			break
+		}
+		_ = v_0_0.Args[1]
+		v_0_0_0 := v_0_0.Args[0]
+		if v_0_0_0.Op != OpAdd8 {
+			break
+		}
+		t := v_0_0_0.Type
+		_ = v_0_0_0.Args[1]
+		v_0_0_0_0 := v_0_0_0.Args[0]
+		if v_0_0_0_0.Op != OpRsh8Ux64 || v_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0.Args[1]
+		v_0_0_0_0_0 := v_0_0_0_0.Args[0]
+		if v_0_0_0_0_0.Op != OpRsh8x64 || v_0_0_0_0_0.Type != t {
+			break
+		}
+		_ = v_0_0_0_0_0.Args[1]
+		if n != v_0_0_0_0_0.Args[0] {
+			break
+		}
+		v_0_0_0_0_0_1 := v_0_0_0_0_0.Args[1]
+		if v_0_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_0_1.Type != typ.UInt64 || v_0_0_0_0_0_1.AuxInt != 7 {
+			break
+		}
+		v_0_0_0_0_1 := v_0_0_0_0.Args[1]
+		if v_0_0_0_0_1.Op != OpConst64 || v_0_0_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		kbar := v_0_0_0_0_1.AuxInt
+		if n != v_0_0_0.Args[1] {
+			break
+		}
+		v_0_0_1 := v_0_0.Args[1]
+		if v_0_0_1.Op != OpConst64 || v_0_0_1.Type != typ.UInt64 {
+			break
+		}
+		k := v_0_0_1.AuxInt
+		v_0_1 := v_0.Args[1]
+		if v_0_1.Op != OpConst64 || v_0_1.Type != typ.UInt64 || v_0_1.AuxInt != k || !(k > 0 && k < 7 && kbar == 8-k) {
+			break
+		}
+		v.reset(OpNeq8)
+		v0 := b.NewValue0(v.Pos, OpAnd8, t)
+		v0.AddArg(n)
+		v1 := b.NewValue0(v.Pos, OpConst8, t)
+		v1.AuxInt = int64(1<<uint(k) - 1)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst8, t)
+		v2.AuxInt = 0
+		v.AddArg(v2)
+		return true
+	}
 	// match: (Neq8 s:(Sub8 x y) (Const8 [0]))
 	// cond: s.Uses == 1
 	// result: (Neq8 x y)
@@ -35496,10 +36588,6 @@ func rewriteValuegeneric_OpNeq8_0(v *Value) bool {
 		v.AddArg(v2)
 		return true
 	}
-	return false
-}
-func rewriteValuegeneric_OpNeq8_10(v *Value) bool {
-	b := v.Block
 	// match: (Neq8 (And8 <t> (Const8 <t> [y]) x) (Const8 <t> [y]))
 	// cond: isPowerOfTwo(y)
 	// result: (Eq8 (And8 <t> x (Const8 <t> [y])) (Const8 <t> [0]))
diff --git a/test/codegen/arithmetic.go b/test/codegen/arithmetic.go
index dcbc6d3f97..a076664e8e 100644
--- a/test/codegen/arithmetic.go
+++ b/test/codegen/arithmetic.go
@@ -188,14 +188,24 @@ func Pow2Mods(n1 uint, n2 int) (uint, int) {
 }
 
 // Check that signed divisibility checks get converted to AND on low bits
-func Pow2DivisibleSigned(n int) bool {
+func Pow2DivisibleSigned(n1, n2 int) (bool, bool) {
 	// 386:"TESTL\t[$]63",-"DIVL",-"SHRL"
 	// amd64:"TESTQ\t[$]63",-"DIVQ",-"SHRQ"
 	// arm:"AND\t[$]63",-".*udiv",-"SRA"
 	// arm64:"AND\t[$]63",-"UDIV",-"ASR"
 	// ppc64:"ANDCC\t[$]63",-"SRAD"
 	// ppc64le:"ANDCC\t[$]63",-"SRAD"
-	return n%64 == 0 // signed
+	a := n1%64 == 0 // signed divisible
+
+	// 386:"TESTL\t[$]63",-"DIVL",-"SHRL"
+	// amd64:"TESTQ\t[$]63",-"DIVQ",-"SHRQ"
+	// arm:"AND\t[$]63",-".*udiv",-"SRA"
+	// arm64:"AND\t[$]63",-"UDIV",-"ASR"
+	// ppc64:"ANDCC\t[$]63",-"SRAD"
+	// ppc64le:"ANDCC\t[$]63",-"SRAD"
+	b := n2%64 != 0 // signed indivisible
+
+	return a, b
 }
 
 // Check that constant modulo divs get turned into MULs