cmd/compile/internal/ssa: expand runtime.memequal for length {3,5,6,7}

This CL slightly speeds up strings.HasPrefix when testing constant
prefixes of length {3,5,6,7}.

goos: linux
goarch: amd64
cpu: Intel(R) Core(TM) i7-8650U CPU @ 1.90GHz
                │      old     │                 new                 │
                │    sec/op    │   sec/op     vs base                │
StringPrefix3-8   11.125n ± 2%   8.539n ± 1%  -23.25% (p=0.000 n=20)
StringPrefix5-8   11.170n ± 2%   8.700n ± 1%  -22.11% (p=0.000 n=20)
StringPrefix6-8   11.190n ± 2%   8.655n ± 1%  -22.65% (p=0.000 n=20)
StringPrefix7-8   11.095n ± 1%   8.878n ± 1%  -19.98% (p=0.000 n=20)

Change-Id: I510a80d59cf78680b57d68780d35d212d24030e2
Reviewed-on: https://go-review.googlesource.com/c/go/+/700816
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Mark Freeman <markfreeman@google.com>
Auto-Submit: Keith Randall <khr@golang.org>

3 files changed, 519 insertions, 1 deletions

diff --git a/src/cmd/compile/internal/ssa/_gen/generic.rules b/src/cmd/compile/internal/ssa/_gen/generic.rules
index c0806c2243..58872ca85a 100644
--- a/src/cmd/compile/internal/ssa/_gen/generic.rules
+++ b/src/cmd/compile/internal/ssa/_gen/generic.rules
@@ -2084,7 +2084,7 @@
 (NilCheck ptr:(NilCheck _ _) _ ) => ptr
 
 // for late-expanded calls, recognize memequal applied to a single constant byte
-// Support is limited by 1, 2, 4, 8 byte sizes
+// Support is limited by [1-8] byte sizes
 (StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [1]) mem)
   && isSameCall(callAux, "runtime.memequal")
   && symIsRO(scon)
@@ -2131,6 +2131,118 @@
   && canLoadUnaligned(config) && config.PtrSize == 8
   => (MakeResult (Eq64 (Load <typ.Int64> sptr mem) (Const64 <typ.Int64> [int64(read64(scon,0,config.ctxt.Arch.ByteOrder))])) mem)
 
+(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [3]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) =>
+  (MakeResult
+    (Eq32
+      (Or32 <typ.Int32>
+        (ZeroExt16to32 <typ.Int32> (Load <typ.Int16> sptr mem))
+        (Lsh32x32 <typ.Int32>
+          (ZeroExt8to32 <typ.Int32> (Load <typ.Int8> (OffPtr <typ.BytePtr> [2] sptr) mem))
+          (Const32 <typ.Int32> [16])))
+      (Const32 <typ.Int32> [int32(uint32(read16(scon,0,config.ctxt.Arch.ByteOrder))|(uint32(read8(scon,2))<<16))]))
+    mem)
+
+(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [3]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) =>
+  (MakeResult
+    (Eq32
+      (Or32 <typ.Int32>
+        (ZeroExt16to32 <typ.Int32> (Load <typ.Int16> sptr mem))
+        (Lsh32x32 <typ.Int32>
+          (ZeroExt8to32 <typ.Int32> (Load <typ.Int8> (OffPtr <typ.BytePtr> [2] sptr) mem))
+          (Const32 <typ.Int32> [16])))
+      (Const32 <typ.Int32> [int32(uint32(read16(scon,0,config.ctxt.Arch.ByteOrder))|(uint32(read8(scon,2))<<16))]))
+    mem)
+
+(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [5]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) && config.PtrSize == 8 =>
+  (MakeResult
+    (Eq64
+      (Or64 <typ.Int64>
+        (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem))
+        (Lsh64x64 <typ.Int64>
+          (ZeroExt8to64 <typ.Int64> (Load <typ.Int8> (OffPtr <typ.BytePtr> [4] sptr) mem))
+          (Const64 <typ.Int64> [32])))
+      (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read8(scon,4))<<32))]))
+    mem)
+
+(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [5]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) && config.PtrSize == 8 =>
+  (MakeResult
+    (Eq64
+      (Or64 <typ.Int64>
+        (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem))
+        (Lsh64x64 <typ.Int64>
+          (ZeroExt8to64 <typ.Int64> (Load <typ.Int8> (OffPtr <typ.BytePtr> [4] sptr) mem))
+          (Const64 <typ.Int64> [32])))
+      (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read8(scon,4))<<32))]))
+    mem)
+
+(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [6]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) && config.PtrSize == 8 =>
+  (MakeResult
+    (Eq64
+      (Or64 <typ.Int64>
+        (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem))
+        (Lsh64x64 <typ.Int64>
+          (ZeroExt16to64 <typ.Int64> (Load <typ.Int16> (OffPtr <typ.BytePtr> [4] sptr) mem))
+          (Const64 <typ.Int64> [32])))
+      (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read16(scon,4,config.ctxt.Arch.ByteOrder))<<32))]))
+    mem)
+
+(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [6]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) && config.PtrSize == 8 =>
+  (MakeResult
+    (Eq64
+      (Or64 <typ.Int64>
+        (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem))
+        (Lsh64x64 <typ.Int64>
+          (ZeroExt16to64 <typ.Int64> (Load <typ.Int16> (OffPtr <typ.BytePtr> [4] sptr) mem))
+          (Const64 <typ.Int64> [32])))
+      (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read16(scon,4,config.ctxt.Arch.ByteOrder))<<32))]))
+    mem)
+
+(StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [7]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) && config.PtrSize == 8 =>
+  (MakeResult
+    (Eq64
+      (Or64 <typ.Int64>
+        (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem))
+        (Lsh64x64 <typ.Int64>
+          (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> (OffPtr <typ.BytePtr> [3] sptr) mem))
+          (Const64 <typ.Int64> [32])))
+      (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read32(scon,3,config.ctxt.Arch.ByteOrder))<<32))]))
+    mem)
+
+(StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [7]) mem)
+  && isSameCall(callAux, "runtime.memequal")
+  && symIsRO(scon)
+  && canLoadUnaligned(config) && config.PtrSize == 8 =>
+  (MakeResult
+    (Eq64
+      (Or64 <typ.Int64>
+        (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem))
+        (Lsh64x64 <typ.Int64>
+          (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> (OffPtr <typ.BytePtr> [3] sptr) mem))
+          (Const64 <typ.Int64> [32])))
+      (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read32(scon,3,config.ctxt.Arch.ByteOrder))<<32))]))
+    mem)
+
 (StaticLECall {callAux} _ _ (Const64 [0]) mem)
   && isSameCall(callAux, "runtime.memequal")
   => (MakeResult (ConstBool <typ.Bool> [true]) mem)
diff --git a/src/cmd/compile/internal/ssa/rewritegeneric.go b/src/cmd/compile/internal/ssa/rewritegeneric.go
index 28d14e7076..7e23194e6a 100644
--- a/src/cmd/compile/internal/ssa/rewritegeneric.go
+++ b/src/cmd/compile/internal/ssa/rewritegeneric.go
@@ -30884,6 +30884,390 @@ func rewriteValuegeneric_OpStaticLECall(v *Value) bool {
 		v.AddArg2(v0, mem)
 		return true
 	}
+	// match: (StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [3]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config)
+	// result: (MakeResult (Eq32 (Or32 <typ.Int32> (ZeroExt16to32 <typ.Int32> (Load <typ.Int16> sptr mem)) (Lsh32x32 <typ.Int32> (ZeroExt8to32 <typ.Int32> (Load <typ.Int8> (OffPtr <typ.BytePtr> [2] sptr) mem)) (Const32 <typ.Int32> [16]))) (Const32 <typ.Int32> [int32(uint32(read16(scon,0,config.ctxt.Arch.ByteOrder))|(uint32(read8(scon,2))<<16))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		sptr := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_1.Aux)
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpSB {
+			break
+		}
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 3 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config)) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq32, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr32, typ.Int32)
+		v2 := b.NewValue0(v.Pos, OpZeroExt16to32, typ.Int32)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int16)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh32x32, typ.Int32)
+		v5 := b.NewValue0(v.Pos, OpZeroExt8to32, typ.Int32)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int8)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(2)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst32, typ.Int32)
+		v8.AuxInt = int32ToAuxInt(16)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst32, typ.Int32)
+		v9.AuxInt = int32ToAuxInt(int32(uint32(read16(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint32(read8(scon, 2)) << 16)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
+	// match: (StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [3]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config)
+	// result: (MakeResult (Eq32 (Or32 <typ.Int32> (ZeroExt16to32 <typ.Int32> (Load <typ.Int16> sptr mem)) (Lsh32x32 <typ.Int32> (ZeroExt8to32 <typ.Int32> (Load <typ.Int8> (OffPtr <typ.BytePtr> [2] sptr) mem)) (Const32 <typ.Int32> [16]))) (Const32 <typ.Int32> [int32(uint32(read16(scon,0,config.ctxt.Arch.ByteOrder))|(uint32(read8(scon,2))<<16))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		v_0 := v.Args[0]
+		if v_0.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_0.Aux)
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpSB {
+			break
+		}
+		sptr := v.Args[1]
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 3 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config)) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq32, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr32, typ.Int32)
+		v2 := b.NewValue0(v.Pos, OpZeroExt16to32, typ.Int32)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int16)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh32x32, typ.Int32)
+		v5 := b.NewValue0(v.Pos, OpZeroExt8to32, typ.Int32)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int8)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(2)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst32, typ.Int32)
+		v8.AuxInt = int32ToAuxInt(16)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst32, typ.Int32)
+		v9.AuxInt = int32ToAuxInt(int32(uint32(read16(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint32(read8(scon, 2)) << 16)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
+	// match: (StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [5]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8
+	// result: (MakeResult (Eq64 (Or64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem)) (Lsh64x64 <typ.Int64> (ZeroExt8to64 <typ.Int64> (Load <typ.Int8> (OffPtr <typ.BytePtr> [4] sptr) mem)) (Const64 <typ.Int64> [32]))) (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read8(scon,4))<<32))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		sptr := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_1.Aux)
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpSB {
+			break
+		}
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 5 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq64, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr64, typ.Int64)
+		v2 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh64x64, typ.Int64)
+		v5 := b.NewValue0(v.Pos, OpZeroExt8to64, typ.Int64)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int8)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(4)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v8.AuxInt = int64ToAuxInt(32)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v9.AuxInt = int64ToAuxInt(int64(uint64(read32(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint64(read8(scon, 4)) << 32)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
+	// match: (StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [5]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8
+	// result: (MakeResult (Eq64 (Or64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem)) (Lsh64x64 <typ.Int64> (ZeroExt8to64 <typ.Int64> (Load <typ.Int8> (OffPtr <typ.BytePtr> [4] sptr) mem)) (Const64 <typ.Int64> [32]))) (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read8(scon,4))<<32))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		v_0 := v.Args[0]
+		if v_0.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_0.Aux)
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpSB {
+			break
+		}
+		sptr := v.Args[1]
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 5 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq64, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr64, typ.Int64)
+		v2 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh64x64, typ.Int64)
+		v5 := b.NewValue0(v.Pos, OpZeroExt8to64, typ.Int64)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int8)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(4)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v8.AuxInt = int64ToAuxInt(32)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v9.AuxInt = int64ToAuxInt(int64(uint64(read32(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint64(read8(scon, 4)) << 32)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
+	// match: (StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [6]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8
+	// result: (MakeResult (Eq64 (Or64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem)) (Lsh64x64 <typ.Int64> (ZeroExt16to64 <typ.Int64> (Load <typ.Int16> (OffPtr <typ.BytePtr> [4] sptr) mem)) (Const64 <typ.Int64> [32]))) (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read16(scon,4,config.ctxt.Arch.ByteOrder))<<32))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		sptr := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_1.Aux)
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpSB {
+			break
+		}
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 6 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq64, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr64, typ.Int64)
+		v2 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh64x64, typ.Int64)
+		v5 := b.NewValue0(v.Pos, OpZeroExt16to64, typ.Int64)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int16)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(4)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v8.AuxInt = int64ToAuxInt(32)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v9.AuxInt = int64ToAuxInt(int64(uint64(read32(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint64(read16(scon, 4, config.ctxt.Arch.ByteOrder)) << 32)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
+	// match: (StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [6]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8
+	// result: (MakeResult (Eq64 (Or64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem)) (Lsh64x64 <typ.Int64> (ZeroExt16to64 <typ.Int64> (Load <typ.Int16> (OffPtr <typ.BytePtr> [4] sptr) mem)) (Const64 <typ.Int64> [32]))) (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read16(scon,4,config.ctxt.Arch.ByteOrder))<<32))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		v_0 := v.Args[0]
+		if v_0.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_0.Aux)
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpSB {
+			break
+		}
+		sptr := v.Args[1]
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 6 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq64, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr64, typ.Int64)
+		v2 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh64x64, typ.Int64)
+		v5 := b.NewValue0(v.Pos, OpZeroExt16to64, typ.Int64)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int16)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(4)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v8.AuxInt = int64ToAuxInt(32)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v9.AuxInt = int64ToAuxInt(int64(uint64(read32(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint64(read16(scon, 4, config.ctxt.Arch.ByteOrder)) << 32)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
+	// match: (StaticLECall {callAux} sptr (Addr {scon} (SB)) (Const64 [7]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8
+	// result: (MakeResult (Eq64 (Or64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem)) (Lsh64x64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> (OffPtr <typ.BytePtr> [3] sptr) mem)) (Const64 <typ.Int64> [32]))) (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read32(scon,3,config.ctxt.Arch.ByteOrder))<<32))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		sptr := v.Args[0]
+		v_1 := v.Args[1]
+		if v_1.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_1.Aux)
+		v_1_0 := v_1.Args[0]
+		if v_1_0.Op != OpSB {
+			break
+		}
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 7 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq64, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr64, typ.Int64)
+		v2 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh64x64, typ.Int64)
+		v5 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(3)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v8.AuxInt = int64ToAuxInt(32)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v9.AuxInt = int64ToAuxInt(int64(uint64(read32(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint64(read32(scon, 3, config.ctxt.Arch.ByteOrder)) << 32)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
+	// match: (StaticLECall {callAux} (Addr {scon} (SB)) sptr (Const64 [7]) mem)
+	// cond: isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8
+	// result: (MakeResult (Eq64 (Or64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> sptr mem)) (Lsh64x64 <typ.Int64> (ZeroExt32to64 <typ.Int64> (Load <typ.Int32> (OffPtr <typ.BytePtr> [3] sptr) mem)) (Const64 <typ.Int64> [32]))) (Const64 <typ.Int64> [int64(uint64(read32(scon,0,config.ctxt.Arch.ByteOrder))|(uint64(read32(scon,3,config.ctxt.Arch.ByteOrder))<<32))])) mem)
+	for {
+		if len(v.Args) != 4 {
+			break
+		}
+		callAux := auxToCall(v.Aux)
+		mem := v.Args[3]
+		v_0 := v.Args[0]
+		if v_0.Op != OpAddr {
+			break
+		}
+		scon := auxToSym(v_0.Aux)
+		v_0_0 := v_0.Args[0]
+		if v_0_0.Op != OpSB {
+			break
+		}
+		sptr := v.Args[1]
+		v_2 := v.Args[2]
+		if v_2.Op != OpConst64 || auxIntToInt64(v_2.AuxInt) != 7 || !(isSameCall(callAux, "runtime.memequal") && symIsRO(scon) && canLoadUnaligned(config) && config.PtrSize == 8) {
+			break
+		}
+		v.reset(OpMakeResult)
+		v0 := b.NewValue0(v.Pos, OpEq64, typ.Bool)
+		v1 := b.NewValue0(v.Pos, OpOr64, typ.Int64)
+		v2 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v3 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v3.AddArg2(sptr, mem)
+		v2.AddArg(v3)
+		v4 := b.NewValue0(v.Pos, OpLsh64x64, typ.Int64)
+		v5 := b.NewValue0(v.Pos, OpZeroExt32to64, typ.Int64)
+		v6 := b.NewValue0(v.Pos, OpLoad, typ.Int32)
+		v7 := b.NewValue0(v.Pos, OpOffPtr, typ.BytePtr)
+		v7.AuxInt = int64ToAuxInt(3)
+		v7.AddArg(sptr)
+		v6.AddArg2(v7, mem)
+		v5.AddArg(v6)
+		v8 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v8.AuxInt = int64ToAuxInt(32)
+		v4.AddArg2(v5, v8)
+		v1.AddArg2(v2, v4)
+		v9 := b.NewValue0(v.Pos, OpConst64, typ.Int64)
+		v9.AuxInt = int64ToAuxInt(int64(uint64(read32(scon, 0, config.ctxt.Arch.ByteOrder)) | (uint64(read32(scon, 3, config.ctxt.Arch.ByteOrder)) << 32)))
+		v0.AddArg2(v1, v9)
+		v.AddArg2(v0, mem)
+		return true
+	}
 	// match: (StaticLECall {callAux} _ _ (Const64 [0]) mem)
 	// cond: isSameCall(callAux, "runtime.memequal")
 	// result: (MakeResult (ConstBool <typ.Bool> [true]) mem)
diff --git a/test/codegen/strings.go b/test/codegen/strings.go
index fbcef57cd1..498c3d398f 100644
--- a/test/codegen/strings.go
+++ b/test/codegen/strings.go
@@ -6,6 +6,8 @@
 
 package codegen
 
+import "strings"
+
 // This file contains code generation tests related to the handling of
 // string types.
 
@@ -89,3 +91,23 @@ func NotEqualSelf(s string) bool {
 }
 
 var bsink []byte
+
+func HasPrefix3(s string) bool {
+	// amd64:-`.*memequal.*`
+	return strings.HasPrefix(s, "str")
+}
+
+func HasPrefix5(s string) bool {
+	// amd64:-`.*memequal.*`
+	return strings.HasPrefix(s, "strin")
+}
+
+func HasPrefix6(s string) bool {
+	// amd64:-`.*memequal.*`
+	return strings.HasPrefix(s, "string")
+}
+
+func HasPrefix7(s string) bool {
+	// amd64:-`.*memequal.*`
+	return strings.HasPrefix(s, "strings")
+}