diff --git a/_gen/gen.go b/_gen/gen.go
index b17eb0d0..30b71780 100644
--- a/_gen/gen.go
+++ b/_gen/gen.go
@@ -1,5 +1,4 @@
 //go:build generate
-// +build generate
 
 // Copyright 2022+, Klaus Post. See LICENSE for details.
 
@@ -110,7 +109,6 @@ import (
 )
 
 `)
-	w.WriteString(fmt.Sprintf("const pshufb = %v\n\n", pshufb))
 	w.WriteString(fmt.Sprintf(`const (
 avx2CodeGen = true
 maxAvx2Inputs = %d
diff --git a/galois_amd64.go b/galois_amd64.go
index 23adfa42..c7ab3663 100644
--- a/galois_amd64.go
+++ b/galois_amd64.go
@@ -4,6 +4,8 @@
 
 package reedsolomon
 
+const pshufb = true
+
 //go:noescape
 func galMulSSSE3(low, high, in, out []byte)
 
diff --git a/galois_arm64.go b/galois_arm64.go
index 0479c902..8ef402bf 100644
--- a/galois_arm64.go
+++ b/galois_arm64.go
@@ -1,19 +1,18 @@
-//go:build !noasm && !appengine && !gccgo && pshufb
+//go:build !noasm && !appengine && !gccgo && !nopshufb
 
 // Copyright 2015, Klaus Post, see LICENSE for details.
 // Copyright 2017, Minio, Inc.
 
 package reedsolomon
 
+const pshufb = true
+
 //go:noescape
 func galMulNEON(low, high, in, out []byte)
 
 //go:noescape
 func galMulXorNEON(low, high, in, out []byte)
 
-//go:noescape
-func galXorNEON(in, out []byte)
-
 func galMulSlice(c byte, in, out []byte, o *options) {
 	if c == 1 {
 		copy(out, in)
@@ -50,20 +49,6 @@ func galMulSliceXor(c byte, in, out []byte, o *options) {
 	}
 }
 
-// simple slice xor
-func sliceXor(in, out []byte, o *options) {
-
-	galXorNEON(in, out)
-	done := (len(in) >> 5) << 5
-
-	remain := len(in) - done
-	if remain > 0 {
-		for i := done; i < len(in); i++ {
-			out[i] ^= in[i]
-		}
-	}
-}
-
 // 4-way butterfly
 func ifftDIT4(work [][]byte, dist int, log_m01, log_m23, log_m02 ffe, o *options) {
 	ifftDIT4Ref(work, dist, log_m01, log_m23, log_m02, o)
@@ -89,7 +74,7 @@ func fftDIT2(x, y []byte, log_m ffe, o *options) {
 	// Reference version:
 	refMulAdd(x, y, log_m)
 	// 64 byte aligned, always full.
-	galXorNEON(x, y)
+	xorSliceNEON(x, y)
 }
 
 // 2-way butterfly forward
@@ -102,7 +87,7 @@ func fftDIT28(x, y []byte, log_m ffe8, o *options) {
 // 2-way butterfly
 func ifftDIT2(x, y []byte, log_m ffe, o *options) {
 	// 64 byte aligned, always full.
-	galXorNEON(x, y)
+	xorSliceNEON(x, y)
 	// Reference version:
 	refMulAdd(x, y, log_m)
 }
diff --git a/galois_arm64.s b/galois_arm64.s
index 5c042bea..772dfac9 100644
--- a/galois_arm64.s
+++ b/galois_arm64.s
@@ -1,7 +1,7 @@
 //+build !noasm
 //+build !appengine
 //+build !gccgo
-//+build !pshufb
+//+build !nopshufb
 
 // Copyright 2015, Klaus Post, see LICENSE for details.
 // Copyright 2017, Minio, Inc.
@@ -100,29 +100,3 @@ loopXor:
 
 completeXor:
 	RET
-
-// func galXorNEON(in, out []byte)
-TEXT ·galXorNEON(SB), 7, $0
-	MOVD in_base+0(FP), R1
-	MOVD in_len+8(FP), R2    // length of message
-	MOVD out_base+24(FP), R5
-	SUBS $32, R2
-	BMI  completeXor
-
-loopXor:
-	// Main loop
-	VLD1.P 32(R1), [V0.B16, V1.B16]
-	VLD1   (R5), [V20.B16, V21.B16]
-
-	VEOR V20.B16, V0.B16, V4.B16
-	VEOR V21.B16, V1.B16, V5.B16
-
-	// Store result
-	VST1.P [V4.D2, V5.D2], 32(R5)
-
-	SUBS $32, R2
-	BPL  loopXor
-
-completeXor:
-	RET
-
diff --git a/galois_gen_switch_amd64.go b/galois_gen_switch_amd64.go
index d3c1fa66..28c50658 100644
--- a/galois_gen_switch_amd64.go
+++ b/galois_gen_switch_amd64.go
@@ -9,8 +9,6 @@ import (
 	"fmt"
 )
 
-const pshufb = true
-
 const (
 	avx2CodeGen    = true
 	maxAvx2Inputs  = 10
diff --git a/galois_gen_switch_nopshufb_amd64.go b/galois_gen_switch_nopshufb_amd64.go
index a900c2b9..888df307 100644
--- a/galois_gen_switch_nopshufb_amd64.go
+++ b/galois_gen_switch_nopshufb_amd64.go
@@ -9,8 +9,6 @@ import (
 	"fmt"
 )
 
-const pshufb = false
-
 const (
 	avx2CodeGen    = true
 	maxAvx2Inputs  = 10
diff --git a/galois_noasm.go b/galois_noasm.go
index 04d184a4..fb5a3b65 100644
--- a/galois_noasm.go
+++ b/galois_noasm.go
@@ -4,6 +4,8 @@
 
 package reedsolomon
 
+const pshufb = false
+
 func galMulSlice(c byte, in, out []byte, o *options) {
 	out = out[:len(in)]
 	if c == 1 {
@@ -28,11 +30,6 @@ func galMulSliceXor(c byte, in, out []byte, o *options) {
 	}
 }
 
-// simple slice xor
-func sliceXor(in, out []byte, o *options) {
-	sliceXorGo(in, out, o)
-}
-
 func init() {
 	defaultOptions.useAVX512 = false
 }
diff --git a/galois_nopshufb_amd64.go b/galois_nopshufb_amd64.go
index 06b32d88..707eb96f 100644
--- a/galois_nopshufb_amd64.go
+++ b/galois_nopshufb_amd64.go
@@ -7,6 +7,8 @@ package reedsolomon
 // bigSwitchover is the size where 64 bytes are processed per loop.
 const bigSwitchover = 128
 
+const pshufb = false
+
 // simple slice xor
 func sliceXor(in, out []byte, o *options) {
 	if o.useSSE2 {
diff --git a/galois_ppc64le.go b/galois_ppc64le.go
index 01458b5b..31b5da9d 100644
--- a/galois_ppc64le.go
+++ b/galois_ppc64le.go
@@ -1,10 +1,12 @@
-//go:build !noasm && !appengine && !gccgo && !pshufb
+//go:build !noasm && !appengine && !gccgo && !nopshufb
 
 // Copyright 2015, Klaus Post, see LICENSE for details.
 // Copyright 2018, Minio, Inc.
 
 package reedsolomon
 
+const pshufb = true
+
 //go:noescape
 func galMulPpc(low, high, in, out []byte)
 
diff --git a/reedsolomon.go b/reedsolomon.go
index 3fad26f9..75ffc4ff 100644
--- a/reedsolomon.go
+++ b/reedsolomon.go
@@ -652,7 +652,7 @@ func (r *reedSolomon) EncodeIdx(dataShard []byte, idx int, parity [][]byte) erro
 		return ErrShardSize
 	}
 
-	if avx2CodeGen && len(dataShard) >= r.o.perRound && len(parity) >= avx2CodeGenMinShards && (r.o.useAVX2 || r.o.useGFNI) {
+	if avx2CodeGen && len(dataShard) >= r.o.perRound && len(parity) >= avx2CodeGenMinShards && ((pshufb && r.o.useAVX2) || r.o.useGFNI) {
 		m := make([][]byte, r.parityShards)
 		for iRow := range m {
 			m[iRow] = r.parity[iRow][idx : idx+1]
@@ -803,7 +803,7 @@ func (r *reedSolomon) Verify(shards [][]byte) (bool, error) {
 }
 
 func (r *reedSolomon) canAVX2C(byteCount int, inputs, outputs int) bool {
-	return avx2CodeGen && r.o.useAVX2 &&
+	return avx2CodeGen && pshufb && r.o.useAVX2 &&
 		byteCount >= avx2CodeGenMinSize && inputs+outputs >= avx2CodeGenMinShards &&
 		inputs <= maxAvx2Inputs && outputs <= maxAvx2Outputs
 }
diff --git a/xor_arm64.go b/xor_arm64.go
new file mode 100644
index 00000000..6f0522f8
--- /dev/null
+++ b/xor_arm64.go
@@ -0,0 +1,19 @@
+//go:build !noasm && !appengine && !gccgo
+
+package reedsolomon
+
+//go:noescape
+func xorSliceNEON(in, out []byte)
+
+// simple slice xor
+func sliceXor(in, out []byte, o *options) {
+	xorSliceNEON(in, out)
+	done := (len(in) >> 5) << 5
+
+	remain := len(in) - done
+	if remain > 0 {
+		for i := done; i < len(in); i++ {
+			out[i] ^= in[i]
+		}
+	}
+}
diff --git a/xor_arm64.s b/xor_arm64.s
new file mode 100644
index 00000000..56298731
--- /dev/null
+++ b/xor_arm64.s
@@ -0,0 +1,29 @@
+//+build !noasm
+//+build !appengine
+//+build !gccgo
+
+// func xorSliceNEON(in, out []byte)
+TEXT ·xorSliceNEON(SB), 7, $0
+	MOVD in_base+0(FP), R1
+	MOVD in_len+8(FP), R2    // length of message
+	MOVD out_base+24(FP), R5
+	SUBS $32, R2
+	BMI  completeXor
+
+loopXor:
+	// Main loop
+	VLD1.P 32(R1), [V0.B16, V1.B16]
+	VLD1   (R5), [V20.B16, V21.B16]
+
+	VEOR V20.B16, V0.B16, V4.B16
+	VEOR V21.B16, V1.B16, V5.B16
+
+	// Store result
+	VST1.P [V4.D2, V5.D2], 32(R5)
+
+	SUBS $32, R2
+	BPL  loopXor
+
+completeXor:
+	RET
+
diff --git a/xor_noasm.go b/xor_noasm.go
new file mode 100644
index 00000000..f1336678
--- /dev/null
+++ b/xor_noasm.go
@@ -0,0 +1,7 @@
+//go:build noasm || gccgo || appengine || (ppc64le && nopshufb)
+
+package reedsolomon
+
+func sliceXor(in, out []byte, o *options) {
+	sliceXorGo(in, out, o)
+}