Revert "compile: prefer an AND instead of SHR+SHL instructions"

This reverts CL 194297.

Reason for revert: introduced register allocation failures on PPC64LE builders.

Updates #33826
Updates #32781
Updates #34468

Change-Id: I7d0b55df8cdf8e7d2277f1814299b083c2692e48
Reviewed-on: https://go-review.googlesource.com/c/go/+/196957
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>

7 files changed, 195 insertions, 527 deletions

diff --git a/src/cmd/compile/internal/ssa/gen/ARM64.rules b/src/cmd/compile/internal/ssa/gen/ARM64.rules
index ddb504531c..d4b47bfb0b 100644
--- a/src/cmd/compile/internal/ssa/gen/ARM64.rules
+++ b/src/cmd/compile/internal/ssa/gen/ARM64.rules
@@ -1863,8 +1863,9 @@
 (XORshiftLL <t> [c] (UBFX [bfc] x) x2) && c < 32 && t.Size() == 4 && bfc == armBFAuxInt(32-c, c)
 	-> (EXTRWconst [32-c] x2 x)
 
-// Rewrite special pairs of shifts to AND.
-// On ARM64 the bitmask can fit into an instruction.
+// Generic rules rewrite certain AND to a pair of shifts.
+// However, on ARM64 the bitmask can fit into an instruction.
+// Rewrite it back to AND.
 (SRLconst [c] (SLLconst [c] x)) && 0 < c && c < 64 -> (ANDconst [1<<uint(64-c)-1] x) // mask out high bits
 (SLLconst [c] (SRLconst [c] x)) && 0 < c && c < 64 -> (ANDconst [^(1<<uint(c)-1)] x) // mask out low bits
 
@@ -1970,8 +1971,6 @@
 	-> (BFXIL [bfc] y x)
 (ORshiftLL [sc] (UBFX [bfc] x) (SRLconst [sc] y)) && sc == getARM64BFwidth(bfc)
 	-> (BFXIL [bfc] y x)
-(ORshiftRL [rc] (ANDconst [ac] y) (SLLconst [lc] x)) && lc < rc && ac == ^((1<<uint(64-rc)-1))
-	-> (BFXIL [armBFAuxInt(rc-lc, 64-rc)] y x)
 
 // do combined loads
 // little endian loads
diff --git a/src/cmd/compile/internal/ssa/gen/S390X.rules b/src/cmd/compile/internal/ssa/gen/S390X.rules
index 6d8ce4b107..98bf875f80 100644
--- a/src/cmd/compile/internal/ssa/gen/S390X.rules
+++ b/src/cmd/compile/internal/ssa/gen/S390X.rules
@@ -701,8 +701,6 @@
 // may need to be reworked when NIHH/OIHH are added
 (SRDconst [1] (SLDconst [1] (LGDR <t> x))) -> (LGDR <t> (LPDFR <x.Type> x))
 (LDGR <t> (SRDconst [1] (SLDconst [1] x))) -> (LPDFR (LDGR <t> x))
-(AND (MOVDconst [^(-1<<63)]) (LGDR <t> x)) -> (LGDR <t> (LPDFR <x.Type> x))
-(LDGR <t> (AND (MOVDconst [^(-1<<63)]) x)) -> (LPDFR (LDGR <t> x))
 (OR (MOVDconst [-1<<63]) (LGDR <t> x))     -> (LGDR <t> (LNDFR <x.Type> x))
 (LDGR <t> (OR (MOVDconst [-1<<63]) x))     -> (LNDFR (LDGR <t> x))
 
@@ -712,8 +710,6 @@
 // detect copysign
 (OR (SLDconst [63] (SRDconst [63] (LGDR x))) (LGDR (LPDFR <t> y))) -> (LGDR (CPSDR <t> y x))
 (OR (SLDconst [63] (SRDconst [63] (LGDR x))) (MOVDconst [c])) && c & -1<<63 == 0 -> (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [c]) x))
-(OR (AND (MOVDconst [-1<<63]) (LGDR x)) (LGDR (LPDFR <t> y))) -> (LGDR (CPSDR <t> y x))
-(OR (AND (MOVDconst [-1<<63]) (LGDR x)) (MOVDconst [c])) && c & -1<<63 == 0 -> (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [c]) x))
 (CPSDR y (FMOVDconst [c])) && c & -1<<63 == 0 -> (LPDFR y)
 (CPSDR y (FMOVDconst [c])) && c & -1<<63 != 0 -> (LNDFR y)
 
diff --git a/src/cmd/compile/internal/ssa/gen/generic.rules b/src/cmd/compile/internal/ssa/gen/generic.rules
index 8696464a70..ef5d7a63ff 100644
--- a/src/cmd/compile/internal/ssa/gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/gen/generic.rules
@@ -542,6 +542,14 @@
 (Slicemask (Const64 [x])) && x > 0 -> (Const64 [-1])
 (Slicemask (Const64 [0]))          -> (Const64 [0])
 
+// Rewrite AND of consts as shifts if possible, slightly faster for 64 bit operands
+// leading zeros can be shifted left, then right
+(And64 <t> (Const64 [y]) x) && nlz(y) + nto(y) == 64 && nto(y) >= 32
+  -> (Rsh64Ux64 (Lsh64x64 <t> x (Const64 <t> [nlz(y)])) (Const64 <t> [nlz(y)]))
+// trailing zeros can be shifted right, then left
+(And64 <t> (Const64 [y]) x) && nlo(y) + ntz(y) == 64 && ntz(y) >= 32
+  -> (Lsh64x64 (Rsh64Ux64 <t> x (Const64 <t> [ntz(y)])) (Const64 <t> [ntz(y)]))
+
 // simplifications often used for lengths.  e.g. len(s[i:i+5])==5
 (Sub(64|32|16|8) (Add(64|32|16|8) x y) x) -> y
 (Sub(64|32|16|8) (Add(64|32|16|8) x y) y) -> x
diff --git a/src/cmd/compile/internal/ssa/rewriteARM64.go b/src/cmd/compile/internal/ssa/rewriteARM64.go
index 06019ebbd8..fc677726dc 100644
--- a/src/cmd/compile/internal/ssa/rewriteARM64.go
+++ b/src/cmd/compile/internal/ssa/rewriteARM64.go
@@ -28467,33 +28467,6 @@ func rewriteValueARM64_OpARM64ORshiftRL_0(v *Value) bool {
 		v.AddArg(y)
 		return true
 	}
-	// match: (ORshiftRL [rc] (ANDconst [ac] y) (SLLconst [lc] x))
-	// cond: lc < rc && ac == ^((1<<uint(64-rc)-1))
-	// result: (BFXIL [armBFAuxInt(rc-lc, 64-rc)] y x)
-	for {
-		rc := v.AuxInt
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpARM64ANDconst {
-			break
-		}
-		ac := v_0.AuxInt
-		y := v_0.Args[0]
-		v_1 := v.Args[1]
-		if v_1.Op != OpARM64SLLconst {
-			break
-		}
-		lc := v_1.AuxInt
-		x := v_1.Args[0]
-		if !(lc < rc && ac == ^(1<<uint(64-rc)-1)) {
-			break
-		}
-		v.reset(OpARM64BFXIL)
-		v.AuxInt = armBFAuxInt(rc-lc, 64-rc)
-		v.AddArg(y)
-		v.AddArg(x)
-		return true
-	}
 	return false
 }
 func rewriteValueARM64_OpARM64RORWconst_0(v *Value) bool {
diff --git a/src/cmd/compile/internal/ssa/rewriteS390X.go b/src/cmd/compile/internal/ssa/rewriteS390X.go
index b22a5fe359..264bf255ce 100644
--- a/src/cmd/compile/internal/ssa/rewriteS390X.go
+++ b/src/cmd/compile/internal/ssa/rewriteS390X.go
@@ -664,7 +664,7 @@ func rewriteValueS390X(v *Value) bool {
 	case OpS390XNOTW:
 		return rewriteValueS390X_OpS390XNOTW_0(v)
 	case OpS390XOR:
-		return rewriteValueS390X_OpS390XOR_0(v) || rewriteValueS390X_OpS390XOR_10(v) || rewriteValueS390X_OpS390XOR_20(v) || rewriteValueS390X_OpS390XOR_30(v) || rewriteValueS390X_OpS390XOR_40(v) || rewriteValueS390X_OpS390XOR_50(v) || rewriteValueS390X_OpS390XOR_60(v) || rewriteValueS390X_OpS390XOR_70(v) || rewriteValueS390X_OpS390XOR_80(v) || rewriteValueS390X_OpS390XOR_90(v) || rewriteValueS390X_OpS390XOR_100(v) || rewriteValueS390X_OpS390XOR_110(v) || rewriteValueS390X_OpS390XOR_120(v) || rewriteValueS390X_OpS390XOR_130(v) || rewriteValueS390X_OpS390XOR_140(v) || rewriteValueS390X_OpS390XOR_150(v) || rewriteValueS390X_OpS390XOR_160(v)
+		return rewriteValueS390X_OpS390XOR_0(v) || rewriteValueS390X_OpS390XOR_10(v) || rewriteValueS390X_OpS390XOR_20(v) || rewriteValueS390X_OpS390XOR_30(v) || rewriteValueS390X_OpS390XOR_40(v) || rewriteValueS390X_OpS390XOR_50(v) || rewriteValueS390X_OpS390XOR_60(v) || rewriteValueS390X_OpS390XOR_70(v) || rewriteValueS390X_OpS390XOR_80(v) || rewriteValueS390X_OpS390XOR_90(v) || rewriteValueS390X_OpS390XOR_100(v) || rewriteValueS390X_OpS390XOR_110(v) || rewriteValueS390X_OpS390XOR_120(v) || rewriteValueS390X_OpS390XOR_130(v) || rewriteValueS390X_OpS390XOR_140(v) || rewriteValueS390X_OpS390XOR_150(v)
 	case OpS390XORW:
 		return rewriteValueS390X_OpS390XORW_0(v) || rewriteValueS390X_OpS390XORW_10(v) || rewriteValueS390X_OpS390XORW_20(v) || rewriteValueS390X_OpS390XORW_30(v) || rewriteValueS390X_OpS390XORW_40(v) || rewriteValueS390X_OpS390XORW_50(v) || rewriteValueS390X_OpS390XORW_60(v) || rewriteValueS390X_OpS390XORW_70(v) || rewriteValueS390X_OpS390XORW_80(v) || rewriteValueS390X_OpS390XORW_90(v)
 	case OpS390XORWconst:
@@ -7904,57 +7904,6 @@ func rewriteValueS390X_OpS390XAND_0(v *Value) bool {
 	return false
 }
 func rewriteValueS390X_OpS390XAND_10(v *Value) bool {
-	b := v.Block
-	// match: (AND (MOVDconst [^(-1<<63)]) (LGDR <t> x))
-	// cond:
-	// result: (LGDR <t> (LPDFR <x.Type> x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_0.AuxInt != ^(-1 << 63) {
-			break
-		}
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XLGDR {
-			break
-		}
-		t := v_1.Type
-		x := v_1.Args[0]
-		v.reset(OpS390XLGDR)
-		v.Type = t
-		v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (AND (LGDR <t> x) (MOVDconst [^(-1<<63)]))
-	// cond:
-	// result: (LGDR <t> (LPDFR <x.Type> x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XLGDR {
-			break
-		}
-		t := v_0.Type
-		x := v_0.Args[0]
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_1.AuxInt != ^(-1 << 63) {
-			break
-		}
-		v.reset(OpS390XLGDR)
-		v.Type = t
-		v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
 	// match: (AND (MOVDconst [c]) (MOVDconst [d]))
 	// cond:
 	// result: (MOVDconst [c&d])
@@ -10597,53 +10546,6 @@ func rewriteValueS390X_OpS390XLDGR_0(v *Value) bool {
 		v.AddArg(v0)
 		return true
 	}
-	// match: (LDGR <t> (AND (MOVDconst [^(-1<<63)]) x))
-	// cond:
-	// result: (LPDFR (LDGR <t> x))
-	for {
-		t := v.Type
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XAND {
-			break
-		}
-		x := v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_0_0.AuxInt != ^(-1 << 63) {
-			break
-		}
-		v.reset(OpS390XLPDFR)
-		v0 := b.NewValue0(v.Pos, OpS390XLDGR, t)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (LDGR <t> (AND x (MOVDconst [^(-1<<63)])))
-	// cond:
-	// result: (LPDFR (LDGR <t> x))
-	for {
-		t := v.Type
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XAND {
-			break
-		}
-		_ = v_0.Args[1]
-		x := v_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		if v_0_1.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_0_1.AuxInt != ^(-1 << 63) {
-			break
-		}
-		v.reset(OpS390XLPDFR)
-		v0 := b.NewValue0(v.Pos, OpS390XLDGR, t)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
 	// match: (LDGR <t> (OR (MOVDconst [-1<<63]) x))
 	// cond:
 	// result: (LNDFR (LDGR <t> x))
@@ -22925,318 +22827,7 @@ func rewriteValueS390X_OpS390XOR_0(v *Value) bool {
 }
 func rewriteValueS390X_OpS390XOR_10(v *Value) bool {
 	b := v.Block
-	// match: (OR (AND (MOVDconst [-1<<63]) (LGDR x)) (LGDR (LPDFR <t> y)))
-	// cond:
-	// result: (LGDR (CPSDR <t> y x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XAND {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_0_0.AuxInt != -1<<63 {
-			break
-		}
-		v_0_1 := v_0.Args[1]
-		if v_0_1.Op != OpS390XLGDR {
-			break
-		}
-		x := v_0_1.Args[0]
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XLGDR {
-			break
-		}
-		v_1_0 := v_1.Args[0]
-		if v_1_0.Op != OpS390XLPDFR {
-			break
-		}
-		t := v_1_0.Type
-		y := v_1_0.Args[0]
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t)
-		v0.AddArg(y)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (OR (AND (LGDR x) (MOVDconst [-1<<63])) (LGDR (LPDFR <t> y)))
-	// cond:
-	// result: (LGDR (CPSDR <t> y x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XAND {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XLGDR {
-			break
-		}
-		x := v_0_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		if v_0_1.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_0_1.AuxInt != -1<<63 {
-			break
-		}
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XLGDR {
-			break
-		}
-		v_1_0 := v_1.Args[0]
-		if v_1_0.Op != OpS390XLPDFR {
-			break
-		}
-		t := v_1_0.Type
-		y := v_1_0.Args[0]
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t)
-		v0.AddArg(y)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (OR (LGDR (LPDFR <t> y)) (AND (MOVDconst [-1<<63]) (LGDR x)))
-	// cond:
-	// result: (LGDR (CPSDR <t> y x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XLGDR {
-			break
-		}
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XLPDFR {
-			break
-		}
-		t := v_0_0.Type
-		y := v_0_0.Args[0]
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XAND {
-			break
-		}
-		_ = v_1.Args[1]
-		v_1_0 := v_1.Args[0]
-		if v_1_0.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_1_0.AuxInt != -1<<63 {
-			break
-		}
-		v_1_1 := v_1.Args[1]
-		if v_1_1.Op != OpS390XLGDR {
-			break
-		}
-		x := v_1_1.Args[0]
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t)
-		v0.AddArg(y)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (OR (LGDR (LPDFR <t> y)) (AND (LGDR x) (MOVDconst [-1<<63])))
-	// cond:
-	// result: (LGDR (CPSDR <t> y x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XLGDR {
-			break
-		}
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XLPDFR {
-			break
-		}
-		t := v_0_0.Type
-		y := v_0_0.Args[0]
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XAND {
-			break
-		}
-		_ = v_1.Args[1]
-		v_1_0 := v_1.Args[0]
-		if v_1_0.Op != OpS390XLGDR {
-			break
-		}
-		x := v_1_0.Args[0]
-		v_1_1 := v_1.Args[1]
-		if v_1_1.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_1_1.AuxInt != -1<<63 {
-			break
-		}
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t)
-		v0.AddArg(y)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (OR (AND (MOVDconst [-1<<63]) (LGDR x)) (MOVDconst [c]))
-	// cond: c & -1<<63 == 0
-	// result: (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [c]) x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XAND {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_0_0.AuxInt != -1<<63 {
-			break
-		}
-		v_0_1 := v_0.Args[1]
-		if v_0_1.Op != OpS390XLGDR {
-			break
-		}
-		x := v_0_1.Args[0]
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XMOVDconst {
-			break
-		}
-		c := v_1.AuxInt
-		if !(c&-1<<63 == 0) {
-			break
-		}
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, x.Type)
-		v1 := b.NewValue0(v.Pos, OpS390XFMOVDconst, x.Type)
-		v1.AuxInt = c
-		v0.AddArg(v1)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (OR (AND (LGDR x) (MOVDconst [-1<<63])) (MOVDconst [c]))
-	// cond: c & -1<<63 == 0
-	// result: (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [c]) x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XAND {
-			break
-		}
-		_ = v_0.Args[1]
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XLGDR {
-			break
-		}
-		x := v_0_0.Args[0]
-		v_0_1 := v_0.Args[1]
-		if v_0_1.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_0_1.AuxInt != -1<<63 {
-			break
-		}
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XMOVDconst {
-			break
-		}
-		c := v_1.AuxInt
-		if !(c&-1<<63 == 0) {
-			break
-		}
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, x.Type)
-		v1 := b.NewValue0(v.Pos, OpS390XFMOVDconst, x.Type)
-		v1.AuxInt = c
-		v0.AddArg(v1)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (OR (MOVDconst [c]) (AND (MOVDconst [-1<<63]) (LGDR x)))
-	// cond: c & -1<<63 == 0
-	// result: (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [c]) x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XMOVDconst {
-			break
-		}
-		c := v_0.AuxInt
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XAND {
-			break
-		}
-		_ = v_1.Args[1]
-		v_1_0 := v_1.Args[0]
-		if v_1_0.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_1_0.AuxInt != -1<<63 {
-			break
-		}
-		v_1_1 := v_1.Args[1]
-		if v_1_1.Op != OpS390XLGDR {
-			break
-		}
-		x := v_1_1.Args[0]
-		if !(c&-1<<63 == 0) {
-			break
-		}
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, x.Type)
-		v1 := b.NewValue0(v.Pos, OpS390XFMOVDconst, x.Type)
-		v1.AuxInt = c
-		v0.AddArg(v1)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
-	// match: (OR (MOVDconst [c]) (AND (LGDR x) (MOVDconst [-1<<63])))
-	// cond: c & -1<<63 == 0
-	// result: (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [c]) x))
-	for {
-		_ = v.Args[1]
-		v_0 := v.Args[0]
-		if v_0.Op != OpS390XMOVDconst {
-			break
-		}
-		c := v_0.AuxInt
-		v_1 := v.Args[1]
-		if v_1.Op != OpS390XAND {
-			break
-		}
-		_ = v_1.Args[1]
-		v_1_0 := v_1.Args[0]
-		if v_1_0.Op != OpS390XLGDR {
-			break
-		}
-		x := v_1_0.Args[0]
-		v_1_1 := v_1.Args[1]
-		if v_1_1.Op != OpS390XMOVDconst {
-			break
-		}
-		if v_1_1.AuxInt != -1<<63 {
-			break
-		}
-		if !(c&-1<<63 == 0) {
-			break
-		}
-		v.reset(OpS390XLGDR)
-		v0 := b.NewValue0(v.Pos, OpS390XCPSDR, x.Type)
-		v1 := b.NewValue0(v.Pos, OpS390XFMOVDconst, x.Type)
-		v1.AuxInt = c
-		v0.AddArg(v1)
-		v0.AddArg(x)
-		v.AddArg(v0)
-		return true
-	}
+	typ := &b.Func.Config.Types
 	// match: (OR (MOVDconst [c]) (MOVDconst [d]))
 	// cond:
 	// result: (MOVDconst [c|d])
@@ -23275,11 +22866,6 @@ func rewriteValueS390X_OpS390XOR_10(v *Value) bool {
 		v.AuxInt = c | d
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_20(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR x x)
 	// cond:
 	// result: x
@@ -23543,6 +23129,11 @@ func rewriteValueS390X_OpS390XOR_20(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_20(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR sh:(SLDconst [16] x0:(MOVHZload [i0] {s} p mem)) x1:(MOVHZload [i1] {s} p mem))
 	// cond: i1 == i0+2 && p.Op != OpSB && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVWZload [i0] {s} p mem)
@@ -23639,11 +23230,6 @@ func rewriteValueS390X_OpS390XOR_20(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_30(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR sh:(SLDconst [32] x0:(MOVWZload [i0] {s} p mem)) x1:(MOVWZload [i1] {s} p mem))
 	// cond: i1 == i0+4 && p.Op != OpSB && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVDload [i0] {s} p mem)
@@ -24129,6 +23715,11 @@ func rewriteValueS390X_OpS390XOR_30(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_30(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR or:(OR y s1:(SLDconst [j1] x1:(MOVHZload [i1] {s} p mem))) s0:(SLDconst [j0] x0:(MOVHZload [i0] {s} p mem)))
 	// cond: i1 == i0+2 && j1 == j0-16 && j1 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVWZload [i0] {s} p mem)) y)
@@ -24245,11 +23836,6 @@ func rewriteValueS390X_OpS390XOR_30(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_40(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR x1:(MOVBZloadidx [i1] {s} idx p mem) sh:(SLDconst [8] x0:(MOVBZloadidx [i0] {s} p idx mem)))
 	// cond: i1 == i0+1 && p.Op != OpSB && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVHZloadidx [i0] {s} p idx mem)
@@ -24674,6 +24260,11 @@ func rewriteValueS390X_OpS390XOR_40(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_40(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR x1:(MOVHZloadidx [i1] {s} idx p mem) sh:(SLDconst [16] x0:(MOVHZloadidx [i0] {s} p idx mem)))
 	// cond: i1 == i0+2 && p.Op != OpSB && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVWZloadidx [i0] {s} p idx mem)
@@ -24780,11 +24371,6 @@ func rewriteValueS390X_OpS390XOR_40(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_50(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR x1:(MOVHZloadidx [i1] {s} idx p mem) sh:(SLDconst [16] x0:(MOVHZloadidx [i0] {s} idx p mem)))
 	// cond: i1 == i0+2 && p.Op != OpSB && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVWZloadidx [i0] {s} p idx mem)
@@ -25209,6 +24795,11 @@ func rewriteValueS390X_OpS390XOR_50(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_50(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR x1:(MOVWZloadidx [i1] {s} idx p mem) sh:(SLDconst [32] x0:(MOVWZloadidx [i0] {s} idx p mem)))
 	// cond: i1 == i0+4 && p.Op != OpSB && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVDloadidx [i0] {s} p idx mem)
@@ -25315,11 +24906,6 @@ func rewriteValueS390X_OpS390XOR_50(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_60(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR sh:(SLDconst [32] x0:(MOVWZloadidx [i0] {s} idx p mem)) x1:(MOVWZloadidx [i1] {s} p idx mem))
 	// cond: i1 == i0+4 && p.Op != OpSB && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVDloadidx [i0] {s} p idx mem)
@@ -25815,6 +25401,11 @@ func rewriteValueS390X_OpS390XOR_60(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_60(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} idx p mem)) or:(OR y s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} p idx mem))))
 	// cond: i1 == i0+1 && j1 == j0-8 && j1 % 16 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVHZloadidx [i0] {s} p idx mem)) y)
@@ -25951,11 +25542,6 @@ func rewriteValueS390X_OpS390XOR_60(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_70(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} idx p mem)) or:(OR y s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} idx p mem))))
 	// cond: i1 == i0+1 && j1 == j0-8 && j1 % 16 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVHZloadidx [i0] {s} p idx mem)) y)
@@ -26496,6 +26082,11 @@ func rewriteValueS390X_OpS390XOR_70(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_70(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR or:(OR y s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} idx p mem))) s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} idx p mem)))
 	// cond: i1 == i0+1 && j1 == j0-8 && j1 % 16 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVHZloadidx [i0] {s} p idx mem)) y)
@@ -26631,11 +26222,6 @@ func rewriteValueS390X_OpS390XOR_70(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_80(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR s0:(SLDconst [j0] x0:(MOVHZloadidx [i0] {s} idx p mem)) or:(OR s1:(SLDconst [j1] x1:(MOVHZloadidx [i1] {s} p idx mem)) y))
 	// cond: i1 == i0+2 && j1 == j0-16 && j1 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVWZloadidx [i0] {s} p idx mem)) y)
@@ -27176,6 +26762,11 @@ func rewriteValueS390X_OpS390XOR_80(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_80(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR or:(OR s1:(SLDconst [j1] x1:(MOVHZloadidx [i1] {s} idx p mem)) y) s0:(SLDconst [j0] x0:(MOVHZloadidx [i0] {s} p idx mem)))
 	// cond: i1 == i0+2 && j1 == j0-16 && j1 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVWZloadidx [i0] {s} p idx mem)) y)
@@ -27311,11 +26902,6 @@ func rewriteValueS390X_OpS390XOR_80(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_90(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR or:(OR y s1:(SLDconst [j1] x1:(MOVHZloadidx [i1] {s} idx p mem))) s0:(SLDconst [j0] x0:(MOVHZloadidx [i0] {s} p idx mem)))
 	// cond: i1 == i0+2 && j1 == j0-16 && j1 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j1] (MOVWZloadidx [i0] {s} p idx mem)) y)
@@ -27812,6 +27398,11 @@ func rewriteValueS390X_OpS390XOR_90(v *Value) bool {
 		v0.AddArg(v1)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_90(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR sh:(SLDconst [16] r1:(MOVHZreg x1:(MOVHBRload [i1] {s} p mem))) r0:(MOVHZreg x0:(MOVHBRload [i0] {s} p mem)))
 	// cond: i1 == i0+2 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVWZreg (MOVWBRload [i0] {s} p mem))
@@ -27926,11 +27517,6 @@ func rewriteValueS390X_OpS390XOR_90(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_100(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR sh:(SLDconst [32] r1:(MOVWZreg x1:(MOVWBRload [i1] {s} p mem))) r0:(MOVWZreg x0:(MOVWBRload [i0] {s} p mem)))
 	// cond: i1 == i0+4 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVDBRload [i0] {s} p mem)
@@ -28462,6 +28048,11 @@ func rewriteValueS390X_OpS390XOR_100(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_100(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR or:(OR y s0:(SLDconst [j0] r0:(MOVHZreg x0:(MOVHBRload [i0] {s} p mem)))) s1:(SLDconst [j1] r1:(MOVHZreg x1:(MOVHBRload [i1] {s} p mem))))
 	// cond: i1 == i0+2 && j1 == j0+16 && j0 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVWZreg (MOVWBRload [i0] {s} p mem))) y)
@@ -28590,11 +28181,6 @@ func rewriteValueS390X_OpS390XOR_100(v *Value) bool {
 		v0.AddArg(v1)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_110(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR x0:(MOVBZloadidx [i0] {s} idx p mem) sh:(SLDconst [8] x1:(MOVBZloadidx [i1] {s} p idx mem)))
 	// cond: p.Op != OpSB && i1 == i0+1 && x0.Uses == 1 && x1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))
@@ -29043,6 +28629,11 @@ func rewriteValueS390X_OpS390XOR_110(v *Value) bool {
 		v0.AddArg(v1)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_110(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} idx p mem)) sh:(SLDconst [16] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} p idx mem))))
 	// cond: i1 == i0+2 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVWZreg (MOVWBRloadidx [i0] {s} p idx mem))
@@ -29169,11 +28760,6 @@ func rewriteValueS390X_OpS390XOR_110(v *Value) bool {
 		v0.AddArg(v1)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_120(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} idx p mem)) sh:(SLDconst [16] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} idx p mem))))
 	// cond: i1 == i0+2 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVWZreg (MOVWBRloadidx [i0] {s} p idx mem))
@@ -29672,6 +29258,11 @@ func rewriteValueS390X_OpS390XOR_120(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_120(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR r0:(MOVWZreg x0:(MOVWBRloadidx [i0] {s} idx p mem)) sh:(SLDconst [32] r1:(MOVWZreg x1:(MOVWBRloadidx [i1] {s} idx p mem))))
 	// cond: i1 == i0+4 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVDBRloadidx [i0] {s} p idx mem)
@@ -29794,11 +29385,6 @@ func rewriteValueS390X_OpS390XOR_120(v *Value) bool {
 		v0.AddArg(mem)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_130(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR sh:(SLDconst [32] r1:(MOVWZreg x1:(MOVWBRloadidx [i1] {s} idx p mem))) r0:(MOVWZreg x0:(MOVWBRloadidx [i0] {s} p idx mem)))
 	// cond: i1 == i0+4 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && sh.Uses == 1 && mergePoint(b,x0,x1) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(sh)
 	// result: @mergePoint(b,x0,x1) (MOVDBRloadidx [i0] {s} p idx mem)
@@ -30328,6 +29914,11 @@ func rewriteValueS390X_OpS390XOR_130(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_130(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} idx p mem)) or:(OR y s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} p idx mem))))
 	// cond: p.Op != OpSB && i1 == i0+1 && j1 == j0+8 && j0 % 16 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))) y)
@@ -30468,11 +30059,6 @@ func rewriteValueS390X_OpS390XOR_130(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_140(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} idx p mem)) or:(OR y s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} idx p mem))))
 	// cond: p.Op != OpSB && i1 == i0+1 && j1 == j0+8 && j0 % 16 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))) y)
@@ -31029,6 +30615,11 @@ func rewriteValueS390X_OpS390XOR_140(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_140(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR or:(OR y s0:(SLDconst [j0] x0:(MOVBZloadidx [i0] {s} idx p mem))) s1:(SLDconst [j1] x1:(MOVBZloadidx [i1] {s} idx p mem)))
 	// cond: p.Op != OpSB && i1 == i0+1 && j1 == j0+8 && j0 % 16 == 0 && x0.Uses == 1 && x1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVHZreg (MOVHBRloadidx [i0] {s} p idx mem))) y)
@@ -31176,11 +30767,6 @@ func rewriteValueS390X_OpS390XOR_140(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_150(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR s1:(SLDconst [j1] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} idx p mem))) or:(OR s0:(SLDconst [j0] r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} p idx mem))) y))
 	// cond: i1 == i0+2 && j1 == j0+16 && j0 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVWZreg (MOVWBRloadidx [i0] {s} p idx mem))) y)
@@ -31801,6 +31387,11 @@ func rewriteValueS390X_OpS390XOR_150(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
+	return false
+}
+func rewriteValueS390X_OpS390XOR_150(v *Value) bool {
+	b := v.Block
+	typ := &b.Func.Config.Types
 	// match: (OR or:(OR s0:(SLDconst [j0] r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} idx p mem))) y) s1:(SLDconst [j1] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} p idx mem))))
 	// cond: i1 == i0+2 && j1 == j0+16 && j0 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVWZreg (MOVWBRloadidx [i0] {s} p idx mem))) y)
@@ -31956,11 +31547,6 @@ func rewriteValueS390X_OpS390XOR_150(v *Value) bool {
 		v0.AddArg(y)
 		return true
 	}
-	return false
-}
-func rewriteValueS390X_OpS390XOR_160(v *Value) bool {
-	b := v.Block
-	typ := &b.Func.Config.Types
 	// match: (OR or:(OR y s0:(SLDconst [j0] r0:(MOVHZreg x0:(MOVHBRloadidx [i0] {s} idx p mem)))) s1:(SLDconst [j1] r1:(MOVHZreg x1:(MOVHBRloadidx [i1] {s} p idx mem))))
 	// cond: i1 == i0+2 && j1 == j0+16 && j0 % 32 == 0 && x0.Uses == 1 && x1.Uses == 1 && r0.Uses == 1 && r1.Uses == 1 && s0.Uses == 1 && s1.Uses == 1 && or.Uses == 1 && mergePoint(b,x0,x1,y) != nil && clobber(x0) && clobber(x1) && clobber(r0) && clobber(r1) && clobber(s0) && clobber(s1) && clobber(or)
 	// result: @mergePoint(b,x0,x1,y) (OR <v.Type> (SLDconst <v.Type> [j0] (MOVWZreg (MOVWBRloadidx [i0] {s} p idx mem))) y)
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index a2d091d3d6..8aa07d20db 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -5735,6 +5735,112 @@ func rewriteValuegeneric_OpAnd64_10(v *Value) bool {
 		v.AddArg(y)
 		return true
 	}
+	// match: (And64 <t> (Const64 [y]) x)
+	// cond: nlz(y) + nto(y) == 64 && nto(y) >= 32
+	// result: (Rsh64Ux64 (Lsh64x64 <t> x (Const64 <t> [nlz(y)])) (Const64 <t> [nlz(y)]))
+	for {
+		t := v.Type
+		x := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst64 {
+			break
+		}
+		y := v_0.AuxInt
+		if !(nlz(y)+nto(y) == 64 && nto(y) >= 32) {
+			break
+		}
+		v.reset(OpRsh64Ux64)
+		v0 := b.NewValue0(v.Pos, OpLsh64x64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = nlz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = nlz(y)
+		v.AddArg(v2)
+		return true
+	}
+	// match: (And64 <t> x (Const64 [y]))
+	// cond: nlz(y) + nto(y) == 64 && nto(y) >= 32
+	// result: (Rsh64Ux64 (Lsh64x64 <t> x (Const64 <t> [nlz(y)])) (Const64 <t> [nlz(y)]))
+	for {
+		t := v.Type
+		_ = v.Args[1]
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst64 {
+			break
+		}
+		y := v_1.AuxInt
+		if !(nlz(y)+nto(y) == 64 && nto(y) >= 32) {
+			break
+		}
+		v.reset(OpRsh64Ux64)
+		v0 := b.NewValue0(v.Pos, OpLsh64x64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = nlz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = nlz(y)
+		v.AddArg(v2)
+		return true
+	}
+	// match: (And64 <t> (Const64 [y]) x)
+	// cond: nlo(y) + ntz(y) == 64 && ntz(y) >= 32
+	// result: (Lsh64x64 (Rsh64Ux64 <t> x (Const64 <t> [ntz(y)])) (Const64 <t> [ntz(y)]))
+	for {
+		t := v.Type
+		x := v.Args[1]
+		v_0 := v.Args[0]
+		if v_0.Op != OpConst64 {
+			break
+		}
+		y := v_0.AuxInt
+		if !(nlo(y)+ntz(y) == 64 && ntz(y) >= 32) {
+			break
+		}
+		v.reset(OpLsh64x64)
+		v0 := b.NewValue0(v.Pos, OpRsh64Ux64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = ntz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = ntz(y)
+		v.AddArg(v2)
+		return true
+	}
+	// match: (And64 <t> x (Const64 [y]))
+	// cond: nlo(y) + ntz(y) == 64 && ntz(y) >= 32
+	// result: (Lsh64x64 (Rsh64Ux64 <t> x (Const64 <t> [ntz(y)])) (Const64 <t> [ntz(y)]))
+	for {
+		t := v.Type
+		_ = v.Args[1]
+		x := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpConst64 {
+			break
+		}
+		y := v_1.AuxInt
+		if !(nlo(y)+ntz(y) == 64 && ntz(y) >= 32) {
+			break
+		}
+		v.reset(OpLsh64x64)
+		v0 := b.NewValue0(v.Pos, OpRsh64Ux64, t)
+		v0.AddArg(x)
+		v1 := b.NewValue0(v.Pos, OpConst64, t)
+		v1.AuxInt = ntz(y)
+		v0.AddArg(v1)
+		v.AddArg(v0)
+		v2 := b.NewValue0(v.Pos, OpConst64, t)
+		v2.AuxInt = ntz(y)
+		v.AddArg(v2)
+		return true
+	}
 	// match: (And64 (And64 i:(Const64 <t>) z) x)
 	// cond: (z.Op != OpConst64 && x.Op != OpConst64)
 	// result: (And64 i (And64 <t> z x))
@@ -5761,6 +5867,10 @@ func rewriteValuegeneric_OpAnd64_10(v *Value) bool {
 		v.AddArg(v0)
 		return true
 	}
+	return false
+}
+func rewriteValuegeneric_OpAnd64_20(v *Value) bool {
+	b := v.Block
 	// match: (And64 (And64 z i:(Const64 <t>)) x)
 	// cond: (z.Op != OpConst64 && x.Op != OpConst64)
 	// result: (And64 i (And64 <t> z x))
@@ -5874,10 +5984,6 @@ func rewriteValuegeneric_OpAnd64_10(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
-	return false
-}
-func rewriteValuegeneric_OpAnd64_20(v *Value) bool {
-	b := v.Block
 	// match: (And64 (Const64 <t> [c]) (And64 x (Const64 <t> [d])))
 	// cond:
 	// result: (And64 (Const64 <t> [c&d]) x)
diff --git a/test/codegen/math.go b/test/codegen/math.go
index 8aa652595b..36252710d1 100644
--- a/test/codegen/math.go
+++ b/test/codegen/math.go
@@ -81,7 +81,7 @@ func abs32(x float32) float32 {
 
 // Check that it's using integer registers
 func copysign(a, b, c float64) {
-	// amd64:"BTRQ\t[$]63","ANDQ","ORQ"
+	// amd64:"BTRQ\t[$]63","SHRQ\t[$]63","SHLQ\t[$]63","ORQ"
 	// s390x:"CPSDR",-"MOVD"         (no integer load/store)
 	// ppc64:"FCPSGN"
 	// ppc64le:"FCPSGN"
@@ -100,7 +100,7 @@ func copysign(a, b, c float64) {
 	// s390x:"LNDFR\t",-"MOVD\t"     (no integer load/store)
 	sink64[2] = math.Float64frombits(math.Float64bits(a) | 1<<63)
 
-	// amd64:"ANDQ","ORQ"
+	// amd64:-"SHLQ\t[$]1",-"SHRQ\t[$]1","SHRQ\t[$]63","SHLQ\t[$]63","ORQ"
 	// s390x:"CPSDR\t",-"MOVD\t"     (no integer load/store)
 	// ppc64:"FCPSGN"
 	// ppc64le:"FCPSGN"