Do fast by one multiplication (#130)

When multiplying by one we can use faster math.

galois_amd64.go

@@ -53,6 +53,10 @@ func galMulSSSE3Xor(low, high, in, out []byte) {
 const bigSwitchover = 128
 
 func galMulSlice(c byte, in, out []byte, o *options) {
+	if c == 1 {
+		copy(out, in)
+		return
+	}
 	if o.useAVX2 {
 		if len(in) >= bigSwitchover {
 			galMulAVX2_64(mulTableLow[c][:], mulTableHigh[c][:], in, out)
@@ -80,6 +84,11 @@ func galMulSlice(c byte, in, out []byte, o *options) {
 }
 
 func galMulSliceXor(c byte, in, out []byte, o *options) {
+	if c == 1 {
+		sliceXor(in, out, o)
+		return
+	}
+
 	if o.useAVX2 {
 		if len(in) >= bigSwitchover {
 			galMulAVX2Xor_64(mulTableLow[c][:], mulTableHigh[c][:], in, out)
@@ -107,8 +116,8 @@ func galMulSliceXor(c byte, in, out []byte, o *options) {
 }
 
 // slice galois add
-func sliceXor(in, out []byte, sse2 bool) {
+func sliceXor(in, out []byte, o *options) {
-	if sse2 {
+	if o.useSSE2 {
 		if len(in) >= bigSwitchover {
 			sSE2XorSlice_64(in, out)
 			done := (len(in) >> 6) << 6

galois_arm64.go

@@ -14,6 +14,10 @@ func galMulNEON(c uint64, in, out []byte)
 func galMulXorNEON(c uint64, in, out []byte)
 
 func galMulSlice(c byte, in, out []byte, o *options) {
+	if c == 1 {
+		copy(out, in)
+		return
+	}
 	var done int
 	galMulNEON(uint64(c), in, out)
 	done = (len(in) >> 5) << 5
@@ -28,6 +32,10 @@ func galMulSlice(c byte, in, out []byte, o *options) {
 }
 
 func galMulSliceXor(c byte, in, out []byte, o *options) {
+	if c == 1 {
+		sliceXor(in, out, o)
+		return
+	}
 	var done int
 	galMulXorNEON(uint64(c), in, out)
 	done = (len(in) >> 5) << 5
@@ -42,7 +50,7 @@ func galMulSliceXor(c byte, in, out []byte, o *options) {
 }
 
 // slice galois add
-func sliceXor(in, out []byte, sse2 bool) {
+func sliceXor(in, out []byte, o *options) {
 	for n, input := range in {
 		out[n] ^= input
 	}

galois_noasm.go

@@ -7,23 +7,33 @@
 package reedsolomon
 
 func galMulSlice(c byte, in, out []byte, o *options) {
-	mt := mulTable[c][:256]
 	out = out[:len(in)]
+	if c == 1 {
+		copy(out, in)
+		return
+	}
+	mt := mulTable[c][:256]
 	for n, input := range in {
 		out[n] = mt[input]
 	}
 }
 
 func galMulSliceXor(c byte, in, out []byte, o *options) {
-	mt := mulTable[c][:256]
 	out = out[:len(in)]
+	if c == 1 {
+		for n, input := range in {
+			out[n] ^= input
+		}
+		return
+	}
+	mt := mulTable[c][:256]
 	for n, input := range in {
 		out[n] ^= mt[input]
 	}
 }
 
 // slice galois add
-func sliceXor(in, out []byte, sse2 bool) {
+func sliceXor(in, out []byte, o *options) {
 	for n, input := range in {
 		out[n] ^= input
 	}

galois_ppc64le.go

@@ -32,6 +32,10 @@ func galMulPpcXor(low, high, in, out []byte) {
 */
 
 func galMulSlice(c byte, in, out []byte, o *options) {
+	if c == 1 {
+		copy(out, in)
+		return
+	}
 	done := (len(in) >> 4) << 4
 	if done > 0 {
 		galMulPpc(mulTableLow[c][:], mulTableHigh[c][:], in[:done], out)
@@ -46,6 +50,10 @@ func galMulSlice(c byte, in, out []byte, o *options) {
 }
 
 func galMulSliceXor(c byte, in, out []byte, o *options) {
+	if c == 1 {
+		sliceXor(in, out, o)
+		return
+	}
 	done := (len(in) >> 4) << 4
 	if done > 0 {
 		galMulPpcXor(mulTableLow[c][:], mulTableHigh[c][:], in[:done], out)
@@ -60,7 +68,7 @@ func galMulSliceXor(c byte, in, out []byte, o *options) {
 }
 
 // slice galois add
-func sliceXor(in, out []byte, sse2 bool) {
+func sliceXor(in, out []byte, o *options) {
 	for n, input := range in {
 		out[n] ^= input
 	}

galois_test.go

@@ -190,7 +190,9 @@ func TestSliceGalAdd(t *testing.T) {
 		for i := range expect {
 			expect[i] = in[i] ^ out[i]
 		}
-		sliceXor(in, out, false)
+		noSSE2 := defaultOptions
+		noSSE2.useSSE2 = false
+		sliceXor(in, out, &noSSE2)
 		if 0 != bytes.Compare(out, expect) {
 			t.Errorf("got %#v, expected %#v", out, expect)
 		}
@@ -198,7 +200,7 @@ func TestSliceGalAdd(t *testing.T) {
 		for i := range expect {
 			expect[i] = in[i] ^ out[i]
 		}
-		sliceXor(in, out, true)
+		sliceXor(in, out, &defaultOptions)
 		if 0 != bytes.Compare(out, expect) {
 			t.Errorf("got %#v, expected %#v", out, expect)
 		}

reedsolomon.go

@@ -392,7 +392,7 @@ func (r reedSolomon) updateParityShards(matrixRows, oldinputs, newinputs, output
 		}
 		oldin := oldinputs[c]
 		// oldinputs data will be change
-		sliceXor(in, oldin, r.o.useSSE2)
+		sliceXor(in, oldin, &r.o)
 		for iRow := 0; iRow < outputCount; iRow++ {
 			galMulSliceXor(matrixRows[iRow][c], oldin, outputs[iRow], &r.o)
 		}
@@ -419,7 +419,7 @@ func (r reedSolomon) updateParityShardsP(matrixRows, oldinputs, newinputs, outpu
 				}
 				oldin := oldinputs[c]
 				// oldinputs data will be change
-				sliceXor(in[start:stop], oldin[start:stop], r.o.useSSE2)
+				sliceXor(in[start:stop], oldin[start:stop], &r.o)
 				for iRow := 0; iRow < outputCount; iRow++ {
 					galMulSliceXor(matrixRows[iRow][c], oldin[start:stop], outputs[iRow][start:stop], &r.o)
 				}

reedsolomon_test.go

@@ -1171,7 +1171,10 @@ func TestCodeSomeShards(t *testing.T) {
 }
 
 func TestStandardMatrices(t *testing.T) {
-	t.Skip("Skipping slow matrix check (~2 min)")
+	if testing.Short() || runtime.GOMAXPROCS(0) < 4 {
+		// Runtime ~15s.
+		t.Skip("Skipping slow matrix check")
+	}
 	var wg sync.WaitGroup
 	wg.Add(256 - 1)
 	for i := 1; i < 256; i++ {