Implement XL layer - preliminary work.

vendor/github.com/klauspost/reedsolomon/galois_amd64.go

@@ -0,0 +1,77 @@
+//+build !noasm
+//+build !appengine
+
+// Copyright 2015, Klaus Post, see LICENSE for details.
+
+package reedsolomon
+
+import (
+	"github.com/klauspost/cpuid"
+)
+
+//go:noescape
+func galMulSSSE3(low, high, in, out []byte)
+
+//go:noescape
+func galMulSSSE3Xor(low, high, in, out []byte)
+
+//go:noescape
+func galMulAVX2Xor(low, high, in, out []byte)
+
+//go:noescape
+func galMulAVX2(low, high, in, out []byte)
+
+// This is what the assembler rountes does in blocks of 16 bytes:
+/*
+func galMulSSSE3(low, high, in, out []byte) {
+	for n, input := range in {
+		l := input & 0xf
+		h := input >> 4
+		out[n] = low[l] ^ high[h]
+	}
+}
+
+func galMulSSSE3Xor(low, high, in, out []byte) {
+	for n, input := range in {
+		l := input & 0xf
+		h := input >> 4
+		out[n] ^= low[l] ^ high[h]
+	}
+}
+*/
+
+func galMulSlice(c byte, in, out []byte) {
+	var done int
+	if cpuid.CPU.AVX2() {
+		galMulAVX2(mulTableLow[c][:], mulTableHigh[c][:], in, out)
+		done = (len(in) >> 5) << 5
+	} else if cpuid.CPU.SSSE3() {
+		galMulSSSE3(mulTableLow[c][:], mulTableHigh[c][:], in, out)
+		done = (len(in) >> 4) << 4
+	}
+	remain := len(in) - done
+	if remain > 0 {
+		mt := mulTable[c]
+		for i := done; i < len(in); i++ {
+			out[i] = mt[in[i]]
+		}
+	}
+}
+
+func galMulSliceXor(c byte, in, out []byte) {
+	var done int
+	if cpuid.CPU.AVX2() {
+		galMulAVX2Xor(mulTableLow[c][:], mulTableHigh[c][:], in, out)
+		done = (len(in) >> 5) << 5
+	} else if cpuid.CPU.SSSE3() {
+		galMulSSSE3Xor(mulTableLow[c][:], mulTableHigh[c][:], in, out)
+		done = (len(in) >> 4) << 4
+	}
+	remain := len(in) - done
+	if remain > 0 {
+		mt := mulTable[c]
+		for i := done; i < len(in); i++ {
+			out[i] ^= mt[in[i]]
+		}
+	}
+}