[release-branch.go1.8] cmd/compile: fix type propagation through s390x SSA rules

This CL fixes two issues:

1. Load ops were initially always lowered to unsigned loads, even
   for signed types. This was fine by itself however LoadReg ops
   (used to re-load spilled values) were lowered to signed loads
   for signed types. This meant that spills could invalidate
   optimizations that assumed the original unsigned load.

2. Types were not always being maintained correctly through rules
   designed to eliminate unnecessary zero and sign extensions.

Updates #18906 and fixes #18958 (backport of CL 36256 to 1.8).

Change-Id: Id44953b0f644cad047e8474edbd24e8a344ca9a7
Reviewed-on: https://go-review.googlesource.com/36350
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

6 files changed, 297 insertions, 202 deletions

diff --git a/src/cmd/compile/internal/s390x/ssa.go b/src/cmd/compile/internal/s390x/ssa.go
index e2d3c2849b..a9dafa862e 100644
--- a/src/cmd/compile/internal/s390x/ssa.go
+++ b/src/cmd/compile/internal/s390x/ssa.go
@@ -424,7 +424,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.To.Type = obj.TYPE_MEM
 		p.To.Reg = v.Args[0].Reg()
 		gc.AddAux2(&p.To, v, sc.Off())
-	case ssa.OpCopy, ssa.OpS390XMOVDconvert:
+	case ssa.OpCopy, ssa.OpS390XMOVDconvert, ssa.OpS390XMOVDreg:
 		if v.Type.IsMemory() {
 			return
 		}
@@ -433,6 +433,11 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		if x != y {
 			opregreg(moveByType(v.Type), y, x)
 		}
+	case ssa.OpS390XMOVDnop:
+		if v.Reg() != v.Args[0].Reg() {
+			v.Fatalf("input[0] and output not in same register %s", v.LongString())
+		}
+		// nothing to do
 	case ssa.OpLoadReg:
 		if v.Type.IsFlags() {
 			v.Fatalf("load flags not implemented: %v", v.LongString())
diff --git a/src/cmd/compile/internal/ssa/gen/S390X.rules b/src/cmd/compile/internal/ssa/gen/S390X.rules
index be0d581fe0..c26515c569 100644
--- a/src/cmd/compile/internal/ssa/gen/S390X.rules
+++ b/src/cmd/compile/internal/ssa/gen/S390X.rules
@@ -312,9 +312,12 @@
 
 // Lowering loads
 (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) -> (MOVDload ptr mem)
-(Load <t> ptr mem) && is32BitInt(t) -> (MOVWZload ptr mem)
-(Load <t> ptr mem) && is16BitInt(t) -> (MOVHZload ptr mem)
-(Load <t> ptr mem) && (t.IsBoolean() || is8BitInt(t)) -> (MOVBZload ptr mem)
+(Load <t> ptr mem) && is32BitInt(t) && isSigned(t) -> (MOVWload ptr mem)
+(Load <t> ptr mem) && is32BitInt(t) && !isSigned(t) -> (MOVWZload ptr mem)
+(Load <t> ptr mem) && is16BitInt(t) && isSigned(t) -> (MOVHload ptr mem)
+(Load <t> ptr mem) && is16BitInt(t) && !isSigned(t) -> (MOVHZload ptr mem)
+(Load <t> ptr mem) && is8BitInt(t) && isSigned(t) -> (MOVBload ptr mem)
+(Load <t> ptr mem) && (t.IsBoolean() || (is8BitInt(t) && !isSigned(t))) -> (MOVBZload ptr mem)
 (Load <t> ptr mem) && is32BitFloat(t) -> (FMOVSload ptr mem)
 (Load <t> ptr mem) && is64BitFloat(t) -> (FMOVDload ptr mem)
 
@@ -445,16 +448,20 @@
 // ***************************
 // TODO: Should the optimizations be a separate pass?
 
+// if a register move has only 1 use, just use the same register without emitting instruction
+// MOVDnop doesn't emit instruction, only for ensuring the type.
+(MOVDreg x) && x.Uses == 1 -> (MOVDnop x)
+
 // Fold sign extensions into conditional moves of constants.
 // Designed to remove the MOVBZreg inserted by the If lowering.
-(MOVBZreg x:(MOVDLT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDLE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDEQ (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDNE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGTnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGEnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
+(MOVBZreg x:(MOVDLT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDLE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDEQ (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDNE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGTnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGEnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
 
 // Fold boolean tests into blocks.
 (NE (CMPWconst [0] (MOVDLT (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (LT cmp yes no)
@@ -572,46 +579,46 @@
 (MOVDNE x y (InvertFlags cmp)) -> (MOVDNE x y cmp)
 
 // don't extend after proper load
-(MOVBreg x:(MOVBload _ _)) -> x
-(MOVBZreg x:(MOVBZload _ _)) -> x
-(MOVHreg x:(MOVBload _ _)) -> x
-(MOVHreg x:(MOVBZload _ _)) -> x
-(MOVHreg x:(MOVHload _ _)) -> x
-(MOVHZreg x:(MOVBZload _ _)) -> x
-(MOVHZreg x:(MOVHZload _ _)) -> x
-(MOVWreg x:(MOVBload _ _)) -> x
-(MOVWreg x:(MOVBZload _ _)) -> x
-(MOVWreg x:(MOVHload _ _)) -> x
-(MOVWreg x:(MOVHZload _ _)) -> x
-(MOVWreg x:(MOVWload _ _)) -> x
-(MOVWZreg x:(MOVBZload _ _)) -> x
-(MOVWZreg x:(MOVHZload _ _)) -> x
-(MOVWZreg x:(MOVWZload _ _)) -> x
+(MOVBreg x:(MOVBload _ _)) -> (MOVDreg x)
+(MOVBZreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBload _ _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVHreg x:(MOVHload _ _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVHZload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHZload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVWload _ _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVHZload _ _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVWZload _ _)) -> (MOVDreg x)
 
 // don't extend if argument is already extended
-(MOVBreg x:(Arg <t>)) && is8BitInt(t) && isSigned(t) -> x
-(MOVBZreg x:(Arg <t>)) && is8BitInt(t) && !isSigned(t) -> x
-(MOVHreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && isSigned(t) -> x
-(MOVHZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && !isSigned(t) -> x
-(MOVWreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && isSigned(t) -> x
-(MOVWZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && !isSigned(t) -> x
+(MOVBreg x:(Arg <t>)) && is8BitInt(t) && isSigned(t) -> (MOVDreg x)
+(MOVBZreg x:(Arg <t>)) && is8BitInt(t) && !isSigned(t) -> (MOVDreg x)
+(MOVHreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && isSigned(t) -> (MOVDreg x)
+(MOVHZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && !isSigned(t) -> (MOVDreg x)
+(MOVWreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && isSigned(t) -> (MOVDreg x)
+(MOVWZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && !isSigned(t) -> (MOVDreg x)
 
 // fold double extensions
-(MOVBreg x:(MOVBreg _)) -> x
-(MOVBZreg x:(MOVBZreg _)) -> x
-(MOVHreg x:(MOVBreg _)) -> x
-(MOVHreg x:(MOVBZreg _)) -> x
-(MOVHreg x:(MOVHreg _)) -> x
-(MOVHZreg x:(MOVBZreg _)) -> x
-(MOVHZreg x:(MOVHZreg _)) -> x
-(MOVWreg x:(MOVBreg _)) -> x
-(MOVWreg x:(MOVBZreg _)) -> x
-(MOVWreg x:(MOVHreg _)) -> x
-(MOVWreg x:(MOVHreg _)) -> x
-(MOVWreg x:(MOVWreg _)) -> x
-(MOVWZreg x:(MOVBZreg _)) -> x
-(MOVWZreg x:(MOVHZreg _)) -> x
-(MOVWZreg x:(MOVWZreg _)) -> x
+(MOVBreg x:(MOVBreg _)) -> (MOVDreg x)
+(MOVBZreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBreg _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVHreg x:(MOVHreg _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVHZreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVWreg _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVHZreg _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVWZreg _)) -> (MOVDreg x)
 
 // fold extensions into constants
 (MOVBreg (MOVDconst [c])) -> (MOVDconst [int64(int8(c))])
@@ -641,10 +648,10 @@
 (MOVWZreg x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZloadidx <v.Type> [off] {sym} ptr idx mem)
 
 // replace load from same location as preceding store with copy
-(MOVBZload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-(MOVHZload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-(MOVWZload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
+(MOVBZload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
+(MOVHZload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
+(MOVWZload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
+(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
 
 // Don't extend before storing
 (MOVWstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
diff --git a/src/cmd/compile/internal/ssa/gen/S390XOps.go b/src/cmd/compile/internal/ssa/gen/S390XOps.go
index 7a25c26784..29383f6c10 100644
--- a/src/cmd/compile/internal/ssa/gen/S390XOps.go
+++ b/src/cmd/compile/internal/ssa/gen/S390XOps.go
@@ -311,6 +311,9 @@ func init() {
 		{name: "MOVHZreg", argLength: 1, reg: gp11sp, asm: "MOVHZ", typ: "UInt64"}, // zero extend arg0 from int16 to int64
 		{name: "MOVWreg", argLength: 1, reg: gp11sp, asm: "MOVW", typ: "Int64"},    // sign extend arg0 from int32 to int64
 		{name: "MOVWZreg", argLength: 1, reg: gp11sp, asm: "MOVWZ", typ: "UInt64"}, // zero extend arg0 from int32 to int64
+		{name: "MOVDreg", argLength: 1, reg: gp11sp, asm: "MOVD"},                  // move from arg0
+
+		{name: "MOVDnop", argLength: 1, reg: gp11, resultInArg0: true}, // nop, return arg0 in same register
 
 		{name: "MOVDconst", reg: gp01, asm: "MOVD", typ: "UInt64", aux: "Int64", rematerializeable: true}, // auxint
 
diff --git a/src/cmd/compile/internal/ssa/opGen.go b/src/cmd/compile/internal/ssa/opGen.go
index a63c5b9953..9d11d03793 100644
--- a/src/cmd/compile/internal/ssa/opGen.go
+++ b/src/cmd/compile/internal/ssa/opGen.go
@@ -1473,6 +1473,8 @@ const (
 	OpS390XMOVHZreg
 	OpS390XMOVWreg
 	OpS390XMOVWZreg
+	OpS390XMOVDreg
+	OpS390XMOVDnop
 	OpS390XMOVDconst
 	OpS390XCFDBRA
 	OpS390XCGDBRA
@@ -18571,6 +18573,32 @@ var opcodeTable = [...]opInfo{
 		},
 	},
 	{
+		name:   "MOVDreg",
+		argLen: 1,
+		asm:    s390x.AMOVD,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 54271}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14 SP
+			},
+			outputs: []outputInfo{
+				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
+			},
+		},
+	},
+	{
+		name:         "MOVDnop",
+		argLen:       1,
+		resultInArg0: true,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
+			},
+			outputs: []outputInfo{
+				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
+			},
+		},
+	},
+	{
 		name:              "MOVDconst",
 		auxType:           auxInt64,
 		argLen:            0,
diff --git a/src/cmd/compile/internal/ssa/rewriteS390X.go b/src/cmd/compile/internal/ssa/rewriteS390X.go
index 5acaf2dbdc..0425ced330 100644
--- a/src/cmd/compile/internal/ssa/rewriteS390X.go
+++ b/src/cmd/compile/internal/ssa/rewriteS390X.go
@@ -524,6 +524,8 @@ func rewriteValueS390X(v *Value, config *Config) bool {
 		return rewriteValueS390X_OpS390XMOVDload(v, config)
 	case OpS390XMOVDloadidx:
 		return rewriteValueS390X_OpS390XMOVDloadidx(v, config)
+	case OpS390XMOVDreg:
+		return rewriteValueS390X_OpS390XMOVDreg(v, config)
 	case OpS390XMOVDstore:
 		return rewriteValueS390X_OpS390XMOVDstore(v, config)
 	case OpS390XMOVDstoreconst:
@@ -3236,13 +3238,28 @@ func rewriteValueS390X_OpLoad(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (Load <t> ptr mem)
-	// cond: is32BitInt(t)
+	// cond: is32BitInt(t) && isSigned(t)
+	// result: (MOVWload ptr mem)
+	for {
+		t := v.Type
+		ptr := v.Args[0]
+		mem := v.Args[1]
+		if !(is32BitInt(t) && isSigned(t)) {
+			break
+		}
+		v.reset(OpS390XMOVWload)
+		v.AddArg(ptr)
+		v.AddArg(mem)
+		return true
+	}
+	// match: (Load <t> ptr mem)
+	// cond: is32BitInt(t) && !isSigned(t)
 	// result: (MOVWZload ptr mem)
 	for {
 		t := v.Type
 		ptr := v.Args[0]
 		mem := v.Args[1]
-		if !(is32BitInt(t)) {
+		if !(is32BitInt(t) && !isSigned(t)) {
 			break
 		}
 		v.reset(OpS390XMOVWZload)
@@ -3251,13 +3268,28 @@ func rewriteValueS390X_OpLoad(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (Load <t> ptr mem)
-	// cond: is16BitInt(t)
+	// cond: is16BitInt(t) && isSigned(t)
+	// result: (MOVHload ptr mem)
+	for {
+		t := v.Type
+		ptr := v.Args[0]
+		mem := v.Args[1]
+		if !(is16BitInt(t) && isSigned(t)) {
+			break
+		}
+		v.reset(OpS390XMOVHload)
+		v.AddArg(ptr)
+		v.AddArg(mem)
+		return true
+	}
+	// match: (Load <t> ptr mem)
+	// cond: is16BitInt(t) && !isSigned(t)
 	// result: (MOVHZload ptr mem)
 	for {
 		t := v.Type
 		ptr := v.Args[0]
 		mem := v.Args[1]
-		if !(is16BitInt(t)) {
+		if !(is16BitInt(t) && !isSigned(t)) {
 			break
 		}
 		v.reset(OpS390XMOVHZload)
@@ -3266,13 +3298,28 @@ func rewriteValueS390X_OpLoad(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (Load <t> ptr mem)
-	// cond: (t.IsBoolean() || is8BitInt(t))
+	// cond: is8BitInt(t) && isSigned(t)
+	// result: (MOVBload ptr mem)
+	for {
+		t := v.Type
+		ptr := v.Args[0]
+		mem := v.Args[1]
+		if !(is8BitInt(t) && isSigned(t)) {
+			break
+		}
+		v.reset(OpS390XMOVBload)
+		v.AddArg(ptr)
+		v.AddArg(mem)
+		return true
+	}
+	// match: (Load <t> ptr mem)
+	// cond: (t.IsBoolean() || (is8BitInt(t) && !isSigned(t)))
 	// result: (MOVBZload ptr mem)
 	for {
 		t := v.Type
 		ptr := v.Args[0]
 		mem := v.Args[1]
-		if !(t.IsBoolean() || is8BitInt(t)) {
+		if !(t.IsBoolean() || (is8BitInt(t) && !isSigned(t))) {
 			break
 		}
 		v.reset(OpS390XMOVBZload)
@@ -7802,7 +7849,7 @@ func rewriteValueS390X_OpS390XMOVBZload(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVBZload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _))
 	// cond: sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		off := v.AuxInt
 		sym := v.Aux
@@ -7818,8 +7865,7 @@ func rewriteValueS390X_OpS390XMOVBZload(v *Value, config *Config) bool {
 		if !(sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -7976,7 +8022,7 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVBZreg x:(MOVDLT (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDLT {
@@ -7995,14 +8041,13 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVDLE (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDLE {
@@ -8021,14 +8066,13 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVDGT (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDGT {
@@ -8047,14 +8091,13 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVDGE (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDGE {
@@ -8073,14 +8116,13 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVDEQ (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDEQ {
@@ -8099,14 +8141,13 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVDNE (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDNE {
@@ -8125,14 +8166,13 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVDGTnoinv (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDGTnoinv {
@@ -8151,14 +8191,13 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVDGEnoinv (MOVDconst [c]) (MOVDconst [d]) _))
 	// cond: int64(uint8(c)) == c && int64(uint8(d)) == d
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVDGEnoinv {
@@ -8177,27 +8216,25 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(int64(uint8(c)) == c && int64(uint8(d)) == d) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVBZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(Arg <t>))
 	// cond: is8BitInt(t) && !isSigned(t)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpArg {
@@ -8207,21 +8244,19 @@ func rewriteValueS390X_OpS390XMOVBZreg(v *Value, config *Config) bool {
 		if !(is8BitInt(t) && !isSigned(t)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBZreg x:(MOVBZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -8349,20 +8384,19 @@ func rewriteValueS390X_OpS390XMOVBreg(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVBreg x:(MOVBload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBreg x:(Arg <t>))
 	// cond: is8BitInt(t) && isSigned(t)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpArg {
@@ -8372,21 +8406,19 @@ func rewriteValueS390X_OpS390XMOVBreg(v *Value, config *Config) bool {
 		if !(is8BitInt(t) && isSigned(t)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVBreg x:(MOVBreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -9995,7 +10027,7 @@ func rewriteValueS390X_OpS390XMOVDload(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _))
 	// cond: sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		off := v.AuxInt
 		sym := v.Aux
@@ -10011,8 +10043,7 @@ func rewriteValueS390X_OpS390XMOVDload(v *Value, config *Config) bool {
 		if !(sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -10164,6 +10195,23 @@ func rewriteValueS390X_OpS390XMOVDloadidx(v *Value, config *Config) bool {
 	}
 	return false
 }
+func rewriteValueS390X_OpS390XMOVDreg(v *Value, config *Config) bool {
+	b := v.Block
+	_ = b
+	// match: (MOVDreg x)
+	// cond: x.Uses == 1
+	// result: (MOVDnop x)
+	for {
+		x := v.Args[0]
+		if !(x.Uses == 1) {
+			break
+		}
+		v.reset(OpS390XMOVDnop)
+		v.AddArg(x)
+		return true
+	}
+	return false
+}
 func rewriteValueS390X_OpS390XMOVDstore(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
@@ -10912,7 +10960,7 @@ func rewriteValueS390X_OpS390XMOVHZload(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVHZload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _))
 	// cond: sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		off := v.AuxInt
 		sym := v.Aux
@@ -10928,8 +10976,7 @@ func rewriteValueS390X_OpS390XMOVHZload(v *Value, config *Config) bool {
 		if !(sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -11086,33 +11133,31 @@ func rewriteValueS390X_OpS390XMOVHZreg(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVHZreg x:(MOVBZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHZreg x:(MOVHZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHZreg x:(Arg <t>))
 	// cond: (is8BitInt(t) || is16BitInt(t)) && !isSigned(t)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpArg {
@@ -11122,34 +11167,31 @@ func rewriteValueS390X_OpS390XMOVHZreg(v *Value, config *Config) bool {
 		if !((is8BitInt(t) || is16BitInt(t)) && !isSigned(t)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHZreg x:(MOVBZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHZreg x:(MOVHZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -11277,46 +11319,43 @@ func rewriteValueS390X_OpS390XMOVHreg(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVHreg x:(MOVBload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHreg x:(MOVBZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHreg x:(MOVHload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHreg x:(Arg <t>))
 	// cond: (is8BitInt(t) || is16BitInt(t)) && isSigned(t)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpArg {
@@ -11326,47 +11365,43 @@ func rewriteValueS390X_OpS390XMOVHreg(v *Value, config *Config) bool {
 		if !((is8BitInt(t) || is16BitInt(t)) && isSigned(t)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHreg x:(MOVBreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHreg x:(MOVBZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVHreg x:(MOVHreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -12310,7 +12345,7 @@ func rewriteValueS390X_OpS390XMOVWZload(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVWZload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _))
 	// cond: sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		off := v.AuxInt
 		sym := v.Aux
@@ -12326,8 +12361,7 @@ func rewriteValueS390X_OpS390XMOVWZload(v *Value, config *Config) bool {
 		if !(sym == sym2 && off == off2 && isSamePtr(ptr, ptr2)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -12484,46 +12518,43 @@ func rewriteValueS390X_OpS390XMOVWZreg(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVWZreg x:(MOVBZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWZreg x:(MOVHZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWZreg x:(MOVWZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVWZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWZreg x:(Arg <t>))
 	// cond: (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && !isSigned(t)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpArg {
@@ -12533,47 +12564,43 @@ func rewriteValueS390X_OpS390XMOVWZreg(v *Value, config *Config) bool {
 		if !((is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && !isSigned(t)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWZreg x:(MOVBZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWZreg x:(MOVHZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWZreg x:(MOVWZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVWZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
@@ -12701,72 +12728,67 @@ func rewriteValueS390X_OpS390XMOVWreg(v *Value, config *Config) bool {
 	_ = b
 	// match: (MOVWreg x:(MOVBload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVBZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVHload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVHZload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHZload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVWload _ _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVWload {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(Arg <t>))
 	// cond: (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && isSigned(t)
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpArg {
@@ -12776,73 +12798,67 @@ func rewriteValueS390X_OpS390XMOVWreg(v *Value, config *Config) bool {
 		if !((is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && isSigned(t)) {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVBreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVBZreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVBZreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVHreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVHreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVHreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
 	// match: (MOVWreg x:(MOVWreg _))
 	// cond:
-	// result: x
+	// result: (MOVDreg x)
 	for {
 		x := v.Args[0]
 		if x.Op != OpS390XMOVWreg {
 			break
 		}
-		v.reset(OpCopy)
-		v.Type = x.Type
+		v.reset(OpS390XMOVDreg)
 		v.AddArg(x)
 		return true
 	}
diff --git a/test/fixedbugs/issue18906.go b/test/fixedbugs/issue18906.go
new file mode 100644
index 0000000000..544400be25
--- /dev/null
+++ b/test/fixedbugs/issue18906.go
@@ -0,0 +1,36 @@
+// run
+
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+//go:noinline
+func f(x int) {
+}
+
+//go:noinline
+func val() int8 {
+	return -1
+}
+
+var (
+	array = [257]int{}
+	slice = array[1:]
+)
+
+func init() {
+	for i := range array {
+		array[i] = i - 1
+	}
+}
+
+func main() {
+	x := val()
+	y := int(uint8(x))
+	f(y) // try and force y to be calculated and spilled
+	if slice[y] != 255 {
+		panic("incorrect value")
+	}
+}