cmd/compile: optimize shift pairs and masks on s390x

Optimize combinations of left and right shifts by a constant value
into a 'rotate then insert selected bits [into zero]' instruction.
Use the same instruction for contiguous masks since it has some
benefits over 'and immediate' (not restricted to 32-bits, does not
overwrite source register).

To keep the complexity of this change under control I've only
implemented 64 bit operations for now.

There are a lot more optimizations that can be done with this
instruction family. However, since their function overlaps with other
instructions we need to be somewhat careful not to break existing
optimization rules by creating optimization dead ends. This is
particularly true of the load/store merging rules which contain lots
of zero extensions and shifts.

This CL does interfere with the store merging rules when an operand
is shifted left before it is stored:

  binary.BigEndian.PutUint64(b, x << 1)

This is unfortunate but it's not critical and somewhat complex so
I plan to fix that in a follow up CL.

file      before    after     Δ       %
addr2line 4117446   4117282   -164    -0.004%
api       4945184   4942752   -2432   -0.049%
asm       4998079   4991891   -6188   -0.124%
buildid   2685158   2684074   -1084   -0.040%
cgo       4553732   4553394   -338    -0.007%
compile   19294446  19245070  -49376  -0.256%
cover     4897105   4891319   -5786   -0.118%
dist      3544389   3542785   -1604   -0.045%
doc       3926795   3927617   +822    +0.021%
fix       3302958   3293868   -9090   -0.275%
link      6546274   6543456   -2818   -0.043%
nm        4102021   4100825   -1196   -0.029%
objdump   4542431   4548483   +6052   +0.133%
pack      2482465   2416389   -66076  -2.662%
pprof     13366541  13363915  -2626   -0.020%
test2json 2829007   2761515   -67492  -2.386%
trace     10216164  10219684  +3520   +0.034%
vet       6773956   6773572   -384    -0.006%
total     107124151 106917891 -206260 -0.193%

Change-Id: I7591cce41e06867ba10a745daae9333513062746
Reviewed-on: https://go-review.googlesource.com/c/go/+/233317
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Michael Munday <mike.munday@ibm.com>

1 files changed, 593 insertions, 251 deletions

diff --git a/src/cmd/compile/internal/ssa/rewriteS390X.go b/src/cmd/compile/internal/ssa/rewriteS390X.go
index 8c3c61d584..d66113d111 100644
--- a/src/cmd/compile/internal/ssa/rewriteS390X.go
+++ b/src/cmd/compile/internal/ssa/rewriteS390X.go
@@ -699,6 +699,8 @@ func rewriteValueS390X(v *Value) bool {
 		return rewriteValueS390X_OpS390XORconst(v)
 	case OpS390XORload:
 		return rewriteValueS390X_OpS390XORload(v)
+	case OpS390XRISBGZ:
+		return rewriteValueS390X_OpS390XRISBGZ(v)
 	case OpS390XRLL:
 		return rewriteValueS390X_OpS390XRLL(v)
 	case OpS390XRLLG:
@@ -5272,9 +5274,8 @@ func rewriteValueS390X_OpS390XADD(v *Value) bool {
 		}
 		break
 	}
-	// match: (ADD (SLDconst x [c]) (SRDconst x [d]))
-	// cond: d == 64-c
-	// result: (RLLGconst [c] x)
+	// match: (ADD (SLDconst x [c]) (SRDconst x [64-c]))
+	// result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)})
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			if v_0.Op != OpS390XSLDconst {
@@ -5282,15 +5283,11 @@ func rewriteValueS390X_OpS390XADD(v *Value) bool {
 			}
 			c := auxIntToInt8(v_0.AuxInt)
 			x := v_0.Args[0]
-			if v_1.Op != OpS390XSRDconst {
+			if v_1.Op != OpS390XSRDconst || auxIntToInt8(v_1.AuxInt) != 64-c || x != v_1.Args[0] {
 				continue
 			}
-			d := auxIntToInt8(v_1.AuxInt)
-			if x != v_1.Args[0] || !(d == 64-c) {
-				continue
-			}
-			v.reset(OpS390XRLLGconst)
-			v.AuxInt = int8ToAuxInt(c)
+			v.reset(OpS390XRISBGZ)
+			v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c))
 			v.AddArg(x)
 			return true
 		}
@@ -5470,9 +5467,8 @@ func rewriteValueS390X_OpS390XADDW(v *Value) bool {
 		}
 		break
 	}
-	// match: (ADDW (SLWconst x [c]) (SRWconst x [d]))
-	// cond: d == 32-c
-	// result: (RLLconst [c] x)
+	// match: (ADDW (SLWconst x [c]) (SRWconst x [32-c]))
+	// result: (RLLconst x [c])
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			if v_0.Op != OpS390XSLWconst {
@@ -5480,11 +5476,7 @@ func rewriteValueS390X_OpS390XADDW(v *Value) bool {
 			}
 			c := auxIntToInt8(v_0.AuxInt)
 			x := v_0.Args[0]
-			if v_1.Op != OpS390XSRWconst {
-				continue
-			}
-			d := auxIntToInt8(v_1.AuxInt)
-			if x != v_1.Args[0] || !(d == 32-c) {
+			if v_1.Op != OpS390XSRWconst || auxIntToInt8(v_1.AuxInt) != 32-c || x != v_1.Args[0] {
 				continue
 			}
 			v.reset(OpS390XRLLconst)
@@ -5844,8 +5836,8 @@ func rewriteValueS390X_OpS390XAND(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
 	// match: (AND x (MOVDconst [c]))
-	// cond: is32Bit(c) && c < 0
-	// result: (ANDconst [c] x)
+	// cond: s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil
+	// result: (RISBGZ x {*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c))})
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			x := v_0
@@ -5853,19 +5845,19 @@ func rewriteValueS390X_OpS390XAND(v *Value) bool {
 				continue
 			}
 			c := auxIntToInt64(v_1.AuxInt)
-			if !(is32Bit(c) && c < 0) {
+			if !(s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)) != nil) {
 				continue
 			}
-			v.reset(OpS390XANDconst)
-			v.AuxInt = int64ToAuxInt(c)
+			v.reset(OpS390XRISBGZ)
+			v.Aux = s390xRotateParamsToAux(*s390x.NewRotateParams(0, 63, 0).OutMerge(uint64(c)))
 			v.AddArg(x)
 			return true
 		}
 		break
 	}
 	// match: (AND x (MOVDconst [c]))
-	// cond: is32Bit(c) && c >= 0
-	// result: (MOVWZreg (ANDWconst <typ.UInt32> [int32(c)] x))
+	// cond: is32Bit(c) && c < 0
+	// result: (ANDconst [c] x)
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			x := v_0
@@ -5873,72 +5865,32 @@ func rewriteValueS390X_OpS390XAND(v *Value) bool {
 				continue
 			}
 			c := auxIntToInt64(v_1.AuxInt)
-			if !(is32Bit(c) && c >= 0) {
-				continue
-			}
-			v.reset(OpS390XMOVWZreg)
-			v0 := b.NewValue0(v.Pos, OpS390XANDWconst, typ.UInt32)
-			v0.AuxInt = int32ToAuxInt(int32(c))
-			v0.AddArg(x)
-			v.AddArg(v0)
-			return true
-		}
-		break
-	}
-	// match: (AND x (MOVDconst [0xFF]))
-	// result: (MOVBZreg 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 != OpS390XMOVDconst || auxIntToInt64(v_1.AuxInt) != 0xFF {
+			if !(is32Bit(c) && c < 0) {
 				continue
 			}
-			v.reset(OpS390XMOVBZreg)
+			v.reset(OpS390XANDconst)
+			v.AuxInt = int64ToAuxInt(c)
 			v.AddArg(x)
 			return true
 		}
 		break
 	}
-	// match: (AND x (MOVDconst [0xFFFF]))
-	// result: (MOVHZreg x)
+	// match: (AND x (MOVDconst [c]))
+	// cond: is32Bit(c) && c >= 0
+	// result: (MOVWZreg (ANDWconst <typ.UInt32> [int32(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 != OpS390XMOVDconst || auxIntToInt64(v_1.AuxInt) != 0xFFFF {
+			if v_1.Op != OpS390XMOVDconst {
 				continue
 			}
-			v.reset(OpS390XMOVHZreg)
-			v.AddArg(x)
-			return true
-		}
-		break
-	}
-	// match: (AND x (MOVDconst [0xFFFFFFFF]))
-	// result: (MOVWZreg 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 != OpS390XMOVDconst || auxIntToInt64(v_1.AuxInt) != 0xFFFFFFFF {
+			c := auxIntToInt64(v_1.AuxInt)
+			if !(is32Bit(c) && c >= 0) {
 				continue
 			}
 			v.reset(OpS390XMOVWZreg)
-			v.AddArg(x)
-			return true
-		}
-		break
-	}
-	// match: (AND (MOVDconst [^(-1<<63)]) (LGDR <t> x))
-	// result: (LGDR <t> (LPDFR <x.Type> x))
-	for {
-		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
-			if v_0.Op != OpS390XMOVDconst || auxIntToInt64(v_0.AuxInt) != ^(-1<<63) || v_1.Op != OpS390XLGDR {
-				continue
-			}
-			t := v_1.Type
-			x := v_1.Args[0]
-			v.reset(OpS390XLGDR)
-			v.Type = t
-			v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type)
+			v0 := b.NewValue0(v.Pos, OpS390XANDWconst, typ.UInt32)
+			v0.AuxInt = int32ToAuxInt(int32(c))
 			v0.AddArg(x)
 			v.AddArg(v0)
 			return true
@@ -6103,10 +6055,10 @@ func rewriteValueS390X_OpS390XANDWconst(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
-	// match: (ANDWconst [0xFF] x)
+	// match: (ANDWconst [0x00ff] x)
 	// result: (MOVBZreg x)
 	for {
-		if auxIntToInt32(v.AuxInt) != 0xFF {
+		if auxIntToInt32(v.AuxInt) != 0x00ff {
 			break
 		}
 		x := v_0
@@ -6114,10 +6066,10 @@ func rewriteValueS390X_OpS390XANDWconst(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
-	// match: (ANDWconst [0xFFFF] x)
+	// match: (ANDWconst [0xffff] x)
 	// result: (MOVHZreg x)
 	for {
-		if auxIntToInt32(v.AuxInt) != 0xFFFF {
+		if auxIntToInt32(v.AuxInt) != 0xffff {
 			break
 		}
 		x := v_0
@@ -6515,6 +6467,21 @@ func rewriteValueS390X_OpS390XCMPUconst(v *Value) bool {
 		v.reset(OpS390XFlagLT)
 		return true
 	}
+	// match: (CMPUconst (RISBGZ x {r}) [c])
+	// cond: r.OutMask() < uint64(uint32(c))
+	// result: (FlagLT)
+	for {
+		c := auxIntToInt32(v.AuxInt)
+		if v_0.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_0.Aux)
+		if !(r.OutMask() < uint64(uint32(c))) {
+			break
+		}
+		v.reset(OpS390XFlagLT)
+		return true
+	}
 	// match: (CMPUconst (MOVWZreg x) [c])
 	// result: (CMPWUconst x [c])
 	for {
@@ -7152,6 +7119,21 @@ func rewriteValueS390X_OpS390XCMPconst(v *Value) bool {
 		v.reset(OpS390XFlagGT)
 		return true
 	}
+	// match: (CMPconst (RISBGZ x {r}) [c])
+	// cond: c > 0 && r.OutMask() < uint64(c)
+	// result: (FlagLT)
+	for {
+		c := auxIntToInt32(v.AuxInt)
+		if v_0.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_0.Aux)
+		if !(c > 0 && r.OutMask() < uint64(c)) {
+			break
+		}
+		v.reset(OpS390XFlagLT)
+		return true
+	}
 	// match: (CMPconst (MOVWreg x) [c])
 	// result: (CMPWconst x [c])
 	for {
@@ -7684,47 +7666,25 @@ func rewriteValueS390X_OpS390XFNEGS(v *Value) bool {
 func rewriteValueS390X_OpS390XLDGR(v *Value) bool {
 	v_0 := v.Args[0]
 	b := v.Block
-	// match: (LDGR <t> (SRDconst [1] (SLDconst [1] x)))
+	// match: (LDGR <t> (RISBGZ x {r}))
+	// cond: r == s390x.NewRotateParams(1, 63, 0)
 	// result: (LPDFR (LDGR <t> x))
 	for {
 		t := v.Type
-		if v_0.Op != OpS390XSRDconst || auxIntToInt8(v_0.AuxInt) != 1 {
+		if v_0.Op != OpS390XRISBGZ {
 			break
 		}
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XSLDconst || auxIntToInt8(v_0_0.AuxInt) != 1 {
+		r := auxToS390xRotateParams(v_0.Aux)
+		x := v_0.Args[0]
+		if !(r == s390x.NewRotateParams(1, 63, 0)) {
 			break
 		}
-		x := v_0_0.Args[0]
 		v.reset(OpS390XLPDFR)
 		v0 := b.NewValue0(v.Pos, OpS390XLDGR, t)
 		v0.AddArg(x)
 		v.AddArg(v0)
 		return true
 	}
-	// match: (LDGR <t> (AND (MOVDconst [^(-1<<63)]) x))
-	// result: (LPDFR (LDGR <t> x))
-	for {
-		t := v.Type
-		if v_0.Op != OpS390XAND {
-			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 != OpS390XMOVDconst || auxIntToInt64(v_0_0.AuxInt) != ^(-1<<63) {
-				continue
-			}
-			x := v_0_1
-			v.reset(OpS390XLPDFR)
-			v0 := b.NewValue0(v.Pos, OpS390XLDGR, t)
-			v0.AddArg(x)
-			v.AddArg(v0)
-			return true
-		}
-		break
-	}
 	// match: (LDGR <t> (OR (MOVDconst [-1<<63]) x))
 	// result: (LNDFR (LDGR <t> x))
 	for {
@@ -8309,6 +8269,23 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value) bool {
 		v.copyOf(x)
 		return true
 	}
+	// match: (MOVBZreg (RISBGZ x {r}))
+	// cond: r.OutMerge(0x000000ff) != nil
+	// result: (RISBGZ x {*r.OutMerge(0x000000ff)})
+	for {
+		if v_0.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_0.Aux)
+		x := v_0.Args[0]
+		if !(r.OutMerge(0x000000ff) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(*r.OutMerge(0x000000ff))
+		v.AddArg(x)
+		return true
+	}
 	// match: (MOVBZreg (ANDWconst [m] x))
 	// result: (MOVWZreg (ANDWconst <typ.UInt32> [int32( uint8(m))] x))
 	for {
@@ -9697,6 +9674,23 @@ func rewriteValueS390X_OpS390XMOVHZreg(v *Value) bool {
 		v.AuxInt = int64ToAuxInt(int64(uint16(c)))
 		return true
 	}
+	// match: (MOVHZreg (RISBGZ x {r}))
+	// cond: r.OutMerge(0x0000ffff) != nil
+	// result: (RISBGZ x {*r.OutMerge(0x0000ffff)})
+	for {
+		if v_0.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_0.Aux)
+		x := v_0.Args[0]
+		if !(r.OutMerge(0x0000ffff) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(*r.OutMerge(0x0000ffff))
+		v.AddArg(x)
+		return true
+	}
 	// match: (MOVHZreg (ANDWconst [m] x))
 	// result: (MOVWZreg (ANDWconst <typ.UInt32> [int32(uint16(m))] x))
 	for {
@@ -10547,6 +10541,23 @@ func rewriteValueS390X_OpS390XMOVWZreg(v *Value) bool {
 		v.AuxInt = int64ToAuxInt(int64(uint32(c)))
 		return true
 	}
+	// match: (MOVWZreg (RISBGZ x {r}))
+	// cond: r.OutMerge(0xffffffff) != nil
+	// result: (RISBGZ x {*r.OutMerge(0xffffffff)})
+	for {
+		if v_0.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_0.Aux)
+		x := v_0.Args[0]
+		if !(r.OutMerge(0xffffffff) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(*r.OutMerge(0xffffffff))
+		v.AddArg(x)
+		return true
+	}
 	return false
 }
 func rewriteValueS390X_OpS390XMOVWload(v *Value) bool {
@@ -11622,9 +11633,8 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool {
 		}
 		break
 	}
-	// match: ( OR (SLDconst x [c]) (SRDconst x [d]))
-	// cond: d == 64-c
-	// result: (RLLGconst [c] x)
+	// match: (OR (SLDconst x [c]) (SRDconst x [64-c]))
+	// result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)})
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			if v_0.Op != OpS390XSLDconst {
@@ -11632,15 +11642,11 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool {
 			}
 			c := auxIntToInt8(v_0.AuxInt)
 			x := v_0.Args[0]
-			if v_1.Op != OpS390XSRDconst {
-				continue
-			}
-			d := auxIntToInt8(v_1.AuxInt)
-			if x != v_1.Args[0] || !(d == 64-c) {
+			if v_1.Op != OpS390XSRDconst || auxIntToInt8(v_1.AuxInt) != 64-c || x != v_1.Args[0] {
 				continue
 			}
-			v.reset(OpS390XRLLGconst)
-			v.AuxInt = int8ToAuxInt(c)
+			v.reset(OpS390XRISBGZ)
+			v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c))
 			v.AddArg(x)
 			return true
 		}
@@ -11664,22 +11670,20 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool {
 		}
 		break
 	}
-	// match: (OR (SLDconst [63] (SRDconst [63] (LGDR x))) (LGDR (LPDFR <t> y)))
+	// match: (OR (RISBGZ (LGDR x) {r}) (LGDR (LPDFR <t> y)))
+	// cond: r == s390x.NewRotateParams(0, 0, 0)
 	// result: (LGDR (CPSDR <t> y x))
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
-			if v_0.Op != OpS390XSLDconst || auxIntToInt8(v_0.AuxInt) != 63 {
+			if v_0.Op != OpS390XRISBGZ {
 				continue
 			}
+			r := auxToS390xRotateParams(v_0.Aux)
 			v_0_0 := v_0.Args[0]
-			if v_0_0.Op != OpS390XSRDconst || auxIntToInt8(v_0_0.AuxInt) != 63 {
-				continue
-			}
-			v_0_0_0 := v_0_0.Args[0]
-			if v_0_0_0.Op != OpS390XLGDR {
+			if v_0_0.Op != OpS390XLGDR {
 				continue
 			}
-			x := v_0_0_0.Args[0]
+			x := v_0_0.Args[0]
 			if v_1.Op != OpS390XLGDR {
 				continue
 			}
@@ -11689,6 +11693,9 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool {
 			}
 			t := v_1_0.Type
 			y := v_1_0.Args[0]
+			if !(r == s390x.NewRotateParams(0, 0, 0)) {
+				continue
+			}
 			v.reset(OpS390XLGDR)
 			v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t)
 			v0.AddArg2(y, x)
@@ -11697,28 +11704,25 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool {
 		}
 		break
 	}
-	// match: (OR (SLDconst [63] (SRDconst [63] (LGDR x))) (MOVDconst [c]))
-	// cond: c & -1<<63 == 0
+	// match: (OR (RISBGZ (LGDR x) {r}) (MOVDconst [c]))
+	// cond: c >= 0 && r == s390x.NewRotateParams(0, 0, 0)
 	// result: (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [math.Float64frombits(uint64(c))]) x))
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
-			if v_0.Op != OpS390XSLDconst || auxIntToInt8(v_0.AuxInt) != 63 {
+			if v_0.Op != OpS390XRISBGZ {
 				continue
 			}
+			r := auxToS390xRotateParams(v_0.Aux)
 			v_0_0 := v_0.Args[0]
-			if v_0_0.Op != OpS390XSRDconst || auxIntToInt8(v_0_0.AuxInt) != 63 {
+			if v_0_0.Op != OpS390XLGDR {
 				continue
 			}
-			v_0_0_0 := v_0_0.Args[0]
-			if v_0_0_0.Op != OpS390XLGDR {
-				continue
-			}
-			x := v_0_0_0.Args[0]
+			x := v_0_0.Args[0]
 			if v_1.Op != OpS390XMOVDconst {
 				continue
 			}
 			c := auxIntToInt64(v_1.AuxInt)
-			if !(c&-1<<63 == 0) {
+			if !(c >= 0 && r == s390x.NewRotateParams(0, 0, 0)) {
 				continue
 			}
 			v.reset(OpS390XLGDR)
@@ -11731,73 +11735,6 @@ func rewriteValueS390X_OpS390XOR(v *Value) bool {
 		}
 		break
 	}
-	// match: (OR (AND (MOVDconst [-1<<63]) (LGDR x)) (LGDR (LPDFR <t> y)))
-	// result: (LGDR (CPSDR <t> y x))
-	for {
-		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
-			if v_0.Op != OpS390XAND {
-				continue
-			}
-			_ = v_0.Args[1]
-			v_0_0 := v_0.Args[0]
-			v_0_1 := v_0.Args[1]
-			for _i1 := 0; _i1 <= 1; _i1, v_0_0, v_0_1 = _i1+1, v_0_1, v_0_0 {
-				if v_0_0.Op != OpS390XMOVDconst || auxIntToInt64(v_0_0.AuxInt) != -1<<63 || v_0_1.Op != OpS390XLGDR {
-					continue
-				}
-				x := v_0_1.Args[0]
-				if v_1.Op != OpS390XLGDR {
-					continue
-				}
-				v_1_0 := v_1.Args[0]
-				if v_1_0.Op != OpS390XLPDFR {
-					continue
-				}
-				t := v_1_0.Type
-				y := v_1_0.Args[0]
-				v.reset(OpS390XLGDR)
-				v0 := b.NewValue0(v.Pos, OpS390XCPSDR, t)
-				v0.AddArg2(y, x)
-				v.AddArg(v0)
-				return true
-			}
-		}
-		break
-	}
-	// match: (OR (AND (MOVDconst [-1<<63]) (LGDR x)) (MOVDconst [c]))
-	// cond: c & -1<<63 == 0
-	// result: (LGDR (CPSDR <x.Type> (FMOVDconst <x.Type> [math.Float64frombits(uint64(c))]) x))
-	for {
-		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
-			if v_0.Op != OpS390XAND {
-				continue
-			}
-			_ = v_0.Args[1]
-			v_0_0 := v_0.Args[0]
-			v_0_1 := v_0.Args[1]
-			for _i1 := 0; _i1 <= 1; _i1, v_0_0, v_0_1 = _i1+1, v_0_1, v_0_0 {
-				if v_0_0.Op != OpS390XMOVDconst || auxIntToInt64(v_0_0.AuxInt) != -1<<63 || v_0_1.Op != OpS390XLGDR {
-					continue
-				}
-				x := v_0_1.Args[0]
-				if v_1.Op != OpS390XMOVDconst {
-					continue
-				}
-				c := auxIntToInt64(v_1.AuxInt)
-				if !(c&-1<<63 == 0) {
-					continue
-				}
-				v.reset(OpS390XLGDR)
-				v0 := b.NewValue0(v.Pos, OpS390XCPSDR, x.Type)
-				v1 := b.NewValue0(v.Pos, OpS390XFMOVDconst, x.Type)
-				v1.AuxInt = float64ToAuxInt(math.Float64frombits(uint64(c)))
-				v0.AddArg2(v1, x)
-				v.AddArg(v0)
-				return true
-			}
-		}
-		break
-	}
 	// match: (OR (MOVDconst [c]) (MOVDconst [d]))
 	// result: (MOVDconst [c|d])
 	for {
@@ -12394,9 +12331,8 @@ func rewriteValueS390X_OpS390XORW(v *Value) bool {
 		}
 		break
 	}
-	// match: ( ORW (SLWconst x [c]) (SRWconst x [d]))
-	// cond: d == 32-c
-	// result: (RLLconst [c] x)
+	// match: (ORW (SLWconst x [c]) (SRWconst x [32-c]))
+	// result: (RLLconst x [c])
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			if v_0.Op != OpS390XSLWconst {
@@ -12404,11 +12340,7 @@ func rewriteValueS390X_OpS390XORW(v *Value) bool {
 			}
 			c := auxIntToInt8(v_0.AuxInt)
 			x := v_0.Args[0]
-			if v_1.Op != OpS390XSRWconst {
-				continue
-			}
-			d := auxIntToInt8(v_1.AuxInt)
-			if x != v_1.Args[0] || !(d == 32-c) {
+			if v_1.Op != OpS390XSRWconst || auxIntToInt8(v_1.AuxInt) != 32-c || x != v_1.Args[0] {
 				continue
 			}
 			v.reset(OpS390XRLLconst)
@@ -12980,6 +12912,221 @@ func rewriteValueS390X_OpS390XORload(v *Value) bool {
 	}
 	return false
 }
+func rewriteValueS390X_OpS390XRISBGZ(v *Value) bool {
+	v_0 := v.Args[0]
+	b := v.Block
+	// match: (RISBGZ (MOVWZreg x) {r})
+	// cond: r.InMerge(0xffffffff) != nil
+	// result: (RISBGZ x {*r.InMerge(0xffffffff)})
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XMOVWZreg {
+			break
+		}
+		x := v_0.Args[0]
+		if !(r.InMerge(0xffffffff) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(*r.InMerge(0xffffffff))
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ (MOVHZreg x) {r})
+	// cond: r.InMerge(0x0000ffff) != nil
+	// result: (RISBGZ x {*r.InMerge(0x0000ffff)})
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XMOVHZreg {
+			break
+		}
+		x := v_0.Args[0]
+		if !(r.InMerge(0x0000ffff) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(*r.InMerge(0x0000ffff))
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ (MOVBZreg x) {r})
+	// cond: r.InMerge(0x000000ff) != nil
+	// result: (RISBGZ x {*r.InMerge(0x000000ff)})
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XMOVBZreg {
+			break
+		}
+		x := v_0.Args[0]
+		if !(r.InMerge(0x000000ff) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(*r.InMerge(0x000000ff))
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ (SLDconst x [c]) {r})
+	// cond: r.InMerge(^uint64(0)<<c) != nil
+	// result: (RISBGZ x {(*r.InMerge(^uint64(0)<<c)).RotateLeft(c)})
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XSLDconst {
+			break
+		}
+		c := auxIntToInt8(v_0.AuxInt)
+		x := v_0.Args[0]
+		if !(r.InMerge(^uint64(0)<<c) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux((*r.InMerge(^uint64(0) << c)).RotateLeft(c))
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ (SRDconst x [c]) {r})
+	// cond: r.InMerge(^uint64(0)>>c) != nil
+	// result: (RISBGZ x {(*r.InMerge(^uint64(0)>>c)).RotateLeft(-c)})
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XSRDconst {
+			break
+		}
+		c := auxIntToInt8(v_0.AuxInt)
+		x := v_0.Args[0]
+		if !(r.InMerge(^uint64(0)>>c) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux((*r.InMerge(^uint64(0) >> c)).RotateLeft(-c))
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ (RISBGZ x {y}) {z})
+	// cond: z.InMerge(y.OutMask()) != nil
+	// result: (RISBGZ x {(*z.InMerge(y.OutMask())).RotateLeft(y.Amount)})
+	for {
+		z := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XRISBGZ {
+			break
+		}
+		y := auxToS390xRotateParams(v_0.Aux)
+		x := v_0.Args[0]
+		if !(z.InMerge(y.OutMask()) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux((*z.InMerge(y.OutMask())).RotateLeft(y.Amount))
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ x {r})
+	// cond: r.End == 63 && r.Start == -r.Amount&63
+	// result: (SRDconst x [-r.Amount&63])
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		x := v_0
+		if !(r.End == 63 && r.Start == -r.Amount&63) {
+			break
+		}
+		v.reset(OpS390XSRDconst)
+		v.AuxInt = int8ToAuxInt(-r.Amount & 63)
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ x {r})
+	// cond: r.Start == 0 && r.End == 63-r.Amount
+	// result: (SLDconst x [r.Amount])
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		x := v_0
+		if !(r.Start == 0 && r.End == 63-r.Amount) {
+			break
+		}
+		v.reset(OpS390XSLDconst)
+		v.AuxInt = int8ToAuxInt(r.Amount)
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ (SRADconst x [c]) {r})
+	// cond: r.Start == r.End && (r.Start+r.Amount)&63 <= c
+	// result: (RISBGZ x {s390x.NewRotateParams(r.Start, r.Start, -r.Start&63)})
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XSRADconst {
+			break
+		}
+		c := auxIntToInt8(v_0.AuxInt)
+		x := v_0.Args[0]
+		if !(r.Start == r.End && (r.Start+r.Amount)&63 <= c) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(r.Start, r.Start, -r.Start&63))
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ x {r})
+	// cond: r == s390x.NewRotateParams(56, 63, 0)
+	// result: (MOVBZreg x)
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		x := v_0
+		if !(r == s390x.NewRotateParams(56, 63, 0)) {
+			break
+		}
+		v.reset(OpS390XMOVBZreg)
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ x {r})
+	// cond: r == s390x.NewRotateParams(48, 63, 0)
+	// result: (MOVHZreg x)
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		x := v_0
+		if !(r == s390x.NewRotateParams(48, 63, 0)) {
+			break
+		}
+		v.reset(OpS390XMOVHZreg)
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ x {r})
+	// cond: r == s390x.NewRotateParams(32, 63, 0)
+	// result: (MOVWZreg x)
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		x := v_0
+		if !(r == s390x.NewRotateParams(32, 63, 0)) {
+			break
+		}
+		v.reset(OpS390XMOVWZreg)
+		v.AddArg(x)
+		return true
+	}
+	// match: (RISBGZ (LGDR <t> x) {r})
+	// cond: r == s390x.NewRotateParams(1, 63, 0)
+	// result: (LGDR <t> (LPDFR <x.Type> x))
+	for {
+		r := auxToS390xRotateParams(v.Aux)
+		if v_0.Op != OpS390XLGDR {
+			break
+		}
+		t := v_0.Type
+		x := v_0.Args[0]
+		if !(r == s390x.NewRotateParams(1, 63, 0)) {
+			break
+		}
+		v.reset(OpS390XLGDR)
+		v.Type = t
+		v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type)
+		v0.AddArg(x)
+		v.AddArg(v0)
+		return true
+	}
+	return false
+}
 func rewriteValueS390X_OpS390XRLL(v *Value) bool {
 	v_1 := v.Args[1]
 	v_0 := v.Args[0]
@@ -13002,15 +13149,15 @@ func rewriteValueS390X_OpS390XRLLG(v *Value) bool {
 	v_1 := v.Args[1]
 	v_0 := v.Args[0]
 	// match: (RLLG x (MOVDconst [c]))
-	// result: (RLLGconst x [int8(c&63)])
+	// result: (RISBGZ x {s390x.NewRotateParams(0, 63, int8(c&63))})
 	for {
 		x := v_0
 		if v_1.Op != OpS390XMOVDconst {
 			break
 		}
 		c := auxIntToInt64(v_1.AuxInt)
-		v.reset(OpS390XRLLGconst)
-		v.AuxInt = int8ToAuxInt(int8(c & 63))
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, int8(c&63)))
 		v.AddArg(x)
 		return true
 	}
@@ -13034,6 +13181,23 @@ func rewriteValueS390X_OpS390XSLD(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
+	// match: (SLD x (RISBGZ y {r}))
+	// cond: r.Amount == 0 && r.OutMask()&63 == 63
+	// result: (SLD x y)
+	for {
+		x := v_0
+		if v_1.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_1.Aux)
+		y := v_1.Args[0]
+		if !(r.Amount == 0 && r.OutMask()&63 == 63) {
+			break
+		}
+		v.reset(OpS390XSLD)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (SLD x (AND (MOVDconst [c]) y))
 	// result: (SLD x (ANDWconst <typ.UInt32> [int32(c&63)] y))
 	for {
@@ -13152,6 +13316,38 @@ func rewriteValueS390X_OpS390XSLD(v *Value) bool {
 }
 func rewriteValueS390X_OpS390XSLDconst(v *Value) bool {
 	v_0 := v.Args[0]
+	// match: (SLDconst (SRDconst x [c]) [d])
+	// result: (RISBGZ x {s390x.NewRotateParams(max8(0, c-d), 63-d, (d-c)&63)})
+	for {
+		d := auxIntToInt8(v.AuxInt)
+		if v_0.Op != OpS390XSRDconst {
+			break
+		}
+		c := auxIntToInt8(v_0.AuxInt)
+		x := v_0.Args[0]
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(max8(0, c-d), 63-d, (d-c)&63))
+		v.AddArg(x)
+		return true
+	}
+	// match: (SLDconst (RISBGZ x {r}) [c])
+	// cond: s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil
+	// result: (RISBGZ x {(*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount)})
+	for {
+		c := auxIntToInt8(v.AuxInt)
+		if v_0.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_0.Aux)
+		x := v_0.Args[0]
+		if !(s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask()) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux((*s390x.NewRotateParams(0, 63-c, c).InMerge(r.OutMask())).RotateLeft(r.Amount))
+		v.AddArg(x)
+		return true
+	}
 	// match: (SLDconst x [0])
 	// result: x
 	for {
@@ -13170,18 +13366,54 @@ func rewriteValueS390X_OpS390XSLW(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
 	// match: (SLW x (MOVDconst [c]))
-	// result: (SLWconst x [int8(c&63)])
+	// cond: c&32 == 0
+	// result: (SLWconst x [int8(c&31)])
 	for {
 		x := v_0
 		if v_1.Op != OpS390XMOVDconst {
 			break
 		}
 		c := auxIntToInt64(v_1.AuxInt)
+		if !(c&32 == 0) {
+			break
+		}
 		v.reset(OpS390XSLWconst)
-		v.AuxInt = int8ToAuxInt(int8(c & 63))
+		v.AuxInt = int8ToAuxInt(int8(c & 31))
 		v.AddArg(x)
 		return true
 	}
+	// match: (SLW _ (MOVDconst [c]))
+	// cond: c&32 != 0
+	// result: (MOVDconst [0])
+	for {
+		if v_1.Op != OpS390XMOVDconst {
+			break
+		}
+		c := auxIntToInt64(v_1.AuxInt)
+		if !(c&32 != 0) {
+			break
+		}
+		v.reset(OpS390XMOVDconst)
+		v.AuxInt = int64ToAuxInt(0)
+		return true
+	}
+	// match: (SLW x (RISBGZ y {r}))
+	// cond: r.Amount == 0 && r.OutMask()&63 == 63
+	// result: (SLW x y)
+	for {
+		x := v_0
+		if v_1.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_1.Aux)
+		y := v_1.Args[0]
+		if !(r.Amount == 0 && r.OutMask()&63 == 63) {
+			break
+		}
+		v.reset(OpS390XSLW)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (SLW x (AND (MOVDconst [c]) y))
 	// result: (SLW x (ANDWconst <typ.UInt32> [int32(c&63)] y))
 	for {
@@ -13330,6 +13562,23 @@ func rewriteValueS390X_OpS390XSRAD(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
+	// match: (SRAD x (RISBGZ y {r}))
+	// cond: r.Amount == 0 && r.OutMask()&63 == 63
+	// result: (SRAD x y)
+	for {
+		x := v_0
+		if v_1.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_1.Aux)
+		y := v_1.Args[0]
+		if !(r.Amount == 0 && r.OutMask()&63 == 63) {
+			break
+		}
+		v.reset(OpS390XSRAD)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (SRAD x (AND (MOVDconst [c]) y))
 	// result: (SRAD x (ANDWconst <typ.UInt32> [int32(c&63)] y))
 	for {
@@ -13478,18 +13727,56 @@ func rewriteValueS390X_OpS390XSRAW(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
 	// match: (SRAW x (MOVDconst [c]))
-	// result: (SRAWconst x [int8(c&63)])
+	// cond: c&32 == 0
+	// result: (SRAWconst x [int8(c&31)])
 	for {
 		x := v_0
 		if v_1.Op != OpS390XMOVDconst {
 			break
 		}
 		c := auxIntToInt64(v_1.AuxInt)
+		if !(c&32 == 0) {
+			break
+		}
 		v.reset(OpS390XSRAWconst)
-		v.AuxInt = int8ToAuxInt(int8(c & 63))
+		v.AuxInt = int8ToAuxInt(int8(c & 31))
 		v.AddArg(x)
 		return true
 	}
+	// match: (SRAW x (MOVDconst [c]))
+	// cond: c&32 != 0
+	// result: (SRAWconst x [31])
+	for {
+		x := v_0
+		if v_1.Op != OpS390XMOVDconst {
+			break
+		}
+		c := auxIntToInt64(v_1.AuxInt)
+		if !(c&32 != 0) {
+			break
+		}
+		v.reset(OpS390XSRAWconst)
+		v.AuxInt = int8ToAuxInt(31)
+		v.AddArg(x)
+		return true
+	}
+	// match: (SRAW x (RISBGZ y {r}))
+	// cond: r.Amount == 0 && r.OutMask()&63 == 63
+	// result: (SRAW x y)
+	for {
+		x := v_0
+		if v_1.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_1.Aux)
+		y := v_1.Args[0]
+		if !(r.Amount == 0 && r.OutMask()&63 == 63) {
+			break
+		}
+		v.reset(OpS390XSRAW)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (SRAW x (AND (MOVDconst [c]) y))
 	// result: (SRAW x (ANDWconst <typ.UInt32> [int32(c&63)] y))
 	for {
@@ -13650,6 +13937,23 @@ func rewriteValueS390X_OpS390XSRD(v *Value) bool {
 		v.AddArg(x)
 		return true
 	}
+	// match: (SRD x (RISBGZ y {r}))
+	// cond: r.Amount == 0 && r.OutMask()&63 == 63
+	// result: (SRD x y)
+	for {
+		x := v_0
+		if v_1.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_1.Aux)
+		y := v_1.Args[0]
+		if !(r.Amount == 0 && r.OutMask()&63 == 63) {
+			break
+		}
+		v.reset(OpS390XSRD)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (SRD x (AND (MOVDconst [c]) y))
 	// result: (SRD x (ANDWconst <typ.UInt32> [int32(c&63)] y))
 	for {
@@ -13768,24 +14072,36 @@ func rewriteValueS390X_OpS390XSRD(v *Value) bool {
 }
 func rewriteValueS390X_OpS390XSRDconst(v *Value) bool {
 	v_0 := v.Args[0]
-	b := v.Block
-	// match: (SRDconst [1] (SLDconst [1] (LGDR <t> x)))
-	// result: (LGDR <t> (LPDFR <x.Type> x))
+	// match: (SRDconst (SLDconst x [c]) [d])
+	// result: (RISBGZ x {s390x.NewRotateParams(d, min8(63, 63-c+d), (c-d)&63)})
 	for {
-		if auxIntToInt8(v.AuxInt) != 1 || v_0.Op != OpS390XSLDconst || auxIntToInt8(v_0.AuxInt) != 1 {
+		d := auxIntToInt8(v.AuxInt)
+		if v_0.Op != OpS390XSLDconst {
 			break
 		}
-		v_0_0 := v_0.Args[0]
-		if v_0_0.Op != OpS390XLGDR {
+		c := auxIntToInt8(v_0.AuxInt)
+		x := v_0.Args[0]
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(d, min8(63, 63-c+d), (c-d)&63))
+		v.AddArg(x)
+		return true
+	}
+	// match: (SRDconst (RISBGZ x {r}) [c])
+	// cond: s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil
+	// result: (RISBGZ x {(*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount)})
+	for {
+		c := auxIntToInt8(v.AuxInt)
+		if v_0.Op != OpS390XRISBGZ {
 			break
 		}
-		t := v_0_0.Type
-		x := v_0_0.Args[0]
-		v.reset(OpS390XLGDR)
-		v.Type = t
-		v0 := b.NewValue0(v.Pos, OpS390XLPDFR, x.Type)
-		v0.AddArg(x)
-		v.AddArg(v0)
+		r := auxToS390xRotateParams(v_0.Aux)
+		x := v_0.Args[0]
+		if !(s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask()) != nil) {
+			break
+		}
+		v.reset(OpS390XRISBGZ)
+		v.Aux = s390xRotateParamsToAux((*s390x.NewRotateParams(c, 63, -c&63).InMerge(r.OutMask())).RotateLeft(r.Amount))
+		v.AddArg(x)
 		return true
 	}
 	// match: (SRDconst x [0])
@@ -13806,18 +14122,54 @@ func rewriteValueS390X_OpS390XSRW(v *Value) bool {
 	b := v.Block
 	typ := &b.Func.Config.Types
 	// match: (SRW x (MOVDconst [c]))
-	// result: (SRWconst x [int8(c&63)])
+	// cond: c&32 == 0
+	// result: (SRWconst x [int8(c&31)])
 	for {
 		x := v_0
 		if v_1.Op != OpS390XMOVDconst {
 			break
 		}
 		c := auxIntToInt64(v_1.AuxInt)
+		if !(c&32 == 0) {
+			break
+		}
 		v.reset(OpS390XSRWconst)
-		v.AuxInt = int8ToAuxInt(int8(c & 63))
+		v.AuxInt = int8ToAuxInt(int8(c & 31))
 		v.AddArg(x)
 		return true
 	}
+	// match: (SRW _ (MOVDconst [c]))
+	// cond: c&32 != 0
+	// result: (MOVDconst [0])
+	for {
+		if v_1.Op != OpS390XMOVDconst {
+			break
+		}
+		c := auxIntToInt64(v_1.AuxInt)
+		if !(c&32 != 0) {
+			break
+		}
+		v.reset(OpS390XMOVDconst)
+		v.AuxInt = int64ToAuxInt(0)
+		return true
+	}
+	// match: (SRW x (RISBGZ y {r}))
+	// cond: r.Amount == 0 && r.OutMask()&63 == 63
+	// result: (SRW x y)
+	for {
+		x := v_0
+		if v_1.Op != OpS390XRISBGZ {
+			break
+		}
+		r := auxToS390xRotateParams(v_1.Aux)
+		y := v_1.Args[0]
+		if !(r.Amount == 0 && r.OutMask()&63 == 63) {
+			break
+		}
+		v.reset(OpS390XSRW)
+		v.AddArg2(x, y)
+		return true
+	}
 	// match: (SRW x (AND (MOVDconst [c]) y))
 	// result: (SRW x (ANDWconst <typ.UInt32> [int32(c&63)] y))
 	for {
@@ -14564,9 +14916,8 @@ func rewriteValueS390X_OpS390XXOR(v *Value) bool {
 		}
 		break
 	}
-	// match: (XOR (SLDconst x [c]) (SRDconst x [d]))
-	// cond: d == 64-c
-	// result: (RLLGconst [c] x)
+	// match: (XOR (SLDconst x [c]) (SRDconst x [64-c]))
+	// result: (RISBGZ x {s390x.NewRotateParams(0, 63, c)})
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			if v_0.Op != OpS390XSLDconst {
@@ -14574,15 +14925,11 @@ func rewriteValueS390X_OpS390XXOR(v *Value) bool {
 			}
 			c := auxIntToInt8(v_0.AuxInt)
 			x := v_0.Args[0]
-			if v_1.Op != OpS390XSRDconst {
-				continue
-			}
-			d := auxIntToInt8(v_1.AuxInt)
-			if x != v_1.Args[0] || !(d == 64-c) {
+			if v_1.Op != OpS390XSRDconst || auxIntToInt8(v_1.AuxInt) != 64-c || x != v_1.Args[0] {
 				continue
 			}
-			v.reset(OpS390XRLLGconst)
-			v.AuxInt = int8ToAuxInt(c)
+			v.reset(OpS390XRISBGZ)
+			v.Aux = s390xRotateParamsToAux(s390x.NewRotateParams(0, 63, c))
 			v.AddArg(x)
 			return true
 		}
@@ -14665,9 +15012,8 @@ func rewriteValueS390X_OpS390XXORW(v *Value) bool {
 		}
 		break
 	}
-	// match: (XORW (SLWconst x [c]) (SRWconst x [d]))
-	// cond: d == 32-c
-	// result: (RLLconst [c] x)
+	// match: (XORW (SLWconst x [c]) (SRWconst x [32-c]))
+	// result: (RLLconst x [c])
 	for {
 		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
 			if v_0.Op != OpS390XSLWconst {
@@ -14675,11 +15021,7 @@ func rewriteValueS390X_OpS390XXORW(v *Value) bool {
 			}
 			c := auxIntToInt8(v_0.AuxInt)
 			x := v_0.Args[0]
-			if v_1.Op != OpS390XSRWconst {
-				continue
-			}
-			d := auxIntToInt8(v_1.AuxInt)
-			if x != v_1.Args[0] || !(d == 32-c) {
+			if v_1.Op != OpS390XSRWconst || auxIntToInt8(v_1.AuxInt) != 32-c || x != v_1.Args[0] {
 				continue
 			}
 			v.reset(OpS390XRLLconst)