diff --git a/blake3.go b/blake3.go index 46fbc5a..7019ca6 100644 --- a/blake3.go +++ b/blake3.go @@ -45,19 +45,16 @@ func wordsToBytes(words []uint32, bytes []byte) { } } -// The g function, split into two parts so that the compiler will inline it. -func gx(state *[16]uint32, a, b, c, d int, mx uint32) { - state[a] += state[b] + mx - state[d] = bits.RotateLeft32(state[d]^state[a], -16) - state[c] += state[d] - state[b] = bits.RotateLeft32(state[b]^state[c], -12) -} - -func gy(state *[16]uint32, a, b, c, d int, my uint32) { - state[a] += state[b] + my - state[d] = bits.RotateLeft32(state[d]^state[a], -8) - state[c] += state[d] - state[b] = bits.RotateLeft32(state[b]^state[c], -7) +func g(a, b, c, d, mx, my uint32) (uint32, uint32, uint32, uint32) { + a += b + mx + d = bits.RotateLeft32(d^a, -16) + c += d + b = bits.RotateLeft32(b^c, -12) + a += b + my + d = bits.RotateLeft32(d^a, -8) + c += d + b = bits.RotateLeft32(b^c, -7) + return a, b, c, d } // A node represents a chunk or parent in the BLAKE3 Merkle tree. In BLAKE3 @@ -83,172 +80,114 @@ type node struct { // node. When nodes are being merged into parents, only the first 8 words are // used. When the root node is being used to generate output, the full 16 words // are used. -func (n node) compress() (state [16]uint32) { - state = [16]uint32{ - n.cv[0], n.cv[1], n.cv[2], n.cv[3], - n.cv[4], n.cv[5], n.cv[6], n.cv[7], - iv[0], iv[1], iv[2], iv[3], - uint32(n.counter), uint32(n.counter >> 32), n.blockLen, n.flags, - } - - // round1 - - // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[0]) - gy(&state, 0, 4, 8, 12, n.block[1]) - gx(&state, 1, 5, 9, 13, n.block[2]) - gy(&state, 1, 5, 9, 13, n.block[3]) - gx(&state, 2, 6, 10, 14, n.block[4]) - gy(&state, 2, 6, 10, 14, n.block[5]) - gx(&state, 3, 7, 11, 15, n.block[6]) - gy(&state, 3, 7, 11, 15, n.block[7]) +func (n node) compress() (s [16]uint32) { + // round1 rather than init s and mix, do both. + // mix the columns. + s[0], s[4], s[8], s[12] = g(n.cv[0], n.cv[4], iv[0], uint32(n.counter), n.block[0], n.block[1]) + s[1], s[5], s[9], s[13] = g(n.cv[1], n.cv[5], iv[1], uint32(n.counter>>32), n.block[2], n.block[3]) + s[2], s[6], s[10], s[14] = g(n.cv[2], n.cv[6], iv[2], n.blockLen, n.block[4], n.block[5]) + s[3], s[7], s[11], s[15] = g(n.cv[3], n.cv[7], iv[3], n.flags, n.block[6], n.block[7]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[8]) - gy(&state, 0, 5, 10, 15, n.block[9]) - gx(&state, 1, 6, 11, 12, n.block[10]) - gy(&state, 1, 6, 11, 12, n.block[11]) - gx(&state, 2, 7, 8, 13, n.block[12]) - gy(&state, 2, 7, 8, 13, n.block[13]) - gx(&state, 3, 4, 9, 14, n.block[14]) - gy(&state, 3, 4, 9, 14, n.block[15]) - - // round2 + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[8], n.block[9]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[10], n.block[11]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[12], n.block[13]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[14], n.block[15]) + // round 2 // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[2]) - gy(&state, 0, 4, 8, 12, n.block[6]) - gx(&state, 1, 5, 9, 13, n.block[3]) - gy(&state, 1, 5, 9, 13, n.block[10]) - gx(&state, 2, 6, 10, 14, n.block[7]) - gy(&state, 2, 6, 10, 14, n.block[0]) - gx(&state, 3, 7, 11, 15, n.block[4]) - gy(&state, 3, 7, 11, 15, n.block[13]) + s[0], s[4], s[8], s[12] = g(s[0], s[4], s[8], s[12], n.block[2], n.block[6]) + s[1], s[5], s[9], s[13] = g(s[1], s[5], s[9], s[13], n.block[3], n.block[10]) + s[2], s[6], s[10], s[14] = g(s[2], s[6], s[10], s[14], n.block[7], n.block[0]) + s[3], s[7], s[11], s[15] = g(s[3], s[7], s[11], s[15], n.block[4], n.block[13]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[1]) - gy(&state, 0, 5, 10, 15, n.block[11]) - gx(&state, 1, 6, 11, 12, n.block[12]) - gy(&state, 1, 6, 11, 12, n.block[5]) - gx(&state, 2, 7, 8, 13, n.block[9]) - gy(&state, 2, 7, 8, 13, n.block[14]) - gx(&state, 3, 4, 9, 14, n.block[15]) - gy(&state, 3, 4, 9, 14, n.block[8]) - - // round3 + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[1], n.block[11]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[12], n.block[5]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[9], n.block[14]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[15], n.block[8]) + // round 3 // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[3]) - gy(&state, 0, 4, 8, 12, n.block[4]) - gx(&state, 1, 5, 9, 13, n.block[10]) - gy(&state, 1, 5, 9, 13, n.block[12]) - gx(&state, 2, 6, 10, 14, n.block[13]) - gy(&state, 2, 6, 10, 14, n.block[2]) - gx(&state, 3, 7, 11, 15, n.block[7]) - gy(&state, 3, 7, 11, 15, n.block[14]) + s[0], s[4], s[8], s[12] = g(s[0], s[4], s[8], s[12], n.block[3], n.block[4]) + s[1], s[5], s[9], s[13] = g(s[1], s[5], s[9], s[13], n.block[10], n.block[12]) + s[2], s[6], s[10], s[14] = g(s[2], s[6], s[10], s[14], n.block[13], n.block[2]) + s[3], s[7], s[11], s[15] = g(s[3], s[7], s[11], s[15], n.block[7], n.block[14]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[6]) - gy(&state, 0, 5, 10, 15, n.block[5]) - gx(&state, 1, 6, 11, 12, n.block[9]) - gy(&state, 1, 6, 11, 12, n.block[0]) - gx(&state, 2, 7, 8, 13, n.block[11]) - gy(&state, 2, 7, 8, 13, n.block[15]) - gx(&state, 3, 4, 9, 14, n.block[8]) - gy(&state, 3, 4, 9, 14, n.block[1]) - - // round4 + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[6], n.block[5]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[9], n.block[0]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[11], n.block[15]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[8], n.block[1]) + // round 4 // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[10]) - gy(&state, 0, 4, 8, 12, n.block[7]) - gx(&state, 1, 5, 9, 13, n.block[12]) - gy(&state, 1, 5, 9, 13, n.block[9]) - gx(&state, 2, 6, 10, 14, n.block[14]) - gy(&state, 2, 6, 10, 14, n.block[3]) - gx(&state, 3, 7, 11, 15, n.block[13]) - gy(&state, 3, 7, 11, 15, n.block[15]) + s[0], s[4], s[8], s[12] = g(s[0], s[4], s[8], s[12], n.block[10], n.block[7]) + s[1], s[5], s[9], s[13] = g(s[1], s[5], s[9], s[13], n.block[12], n.block[9]) + s[2], s[6], s[10], s[14] = g(s[2], s[6], s[10], s[14], n.block[14], n.block[3]) + s[3], s[7], s[11], s[15] = g(s[3], s[7], s[11], s[15], n.block[13], n.block[15]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[4]) - gy(&state, 0, 5, 10, 15, n.block[0]) - gx(&state, 1, 6, 11, 12, n.block[11]) - gy(&state, 1, 6, 11, 12, n.block[2]) - gx(&state, 2, 7, 8, 13, n.block[5]) - gy(&state, 2, 7, 8, 13, n.block[8]) - gx(&state, 3, 4, 9, 14, n.block[1]) - gy(&state, 3, 4, 9, 14, n.block[6]) - - // round5 + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[4], n.block[0]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[11], n.block[2]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[5], n.block[8]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[1], n.block[6]) + // round 5 // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[12]) - gy(&state, 0, 4, 8, 12, n.block[13]) - gx(&state, 1, 5, 9, 13, n.block[9]) - gy(&state, 1, 5, 9, 13, n.block[11]) - gx(&state, 2, 6, 10, 14, n.block[15]) - gy(&state, 2, 6, 10, 14, n.block[10]) - gx(&state, 3, 7, 11, 15, n.block[14]) - gy(&state, 3, 7, 11, 15, n.block[8]) + s[0], s[4], s[8], s[12] = g(s[0], s[4], s[8], s[12], n.block[12], n.block[13]) + s[1], s[5], s[9], s[13] = g(s[1], s[5], s[9], s[13], n.block[9], n.block[11]) + s[2], s[6], s[10], s[14] = g(s[2], s[6], s[10], s[14], n.block[15], n.block[10]) + s[3], s[7], s[11], s[15] = g(s[3], s[7], s[11], s[15], n.block[14], n.block[8]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[7]) - gy(&state, 0, 5, 10, 15, n.block[2]) - gx(&state, 1, 6, 11, 12, n.block[5]) - gy(&state, 1, 6, 11, 12, n.block[3]) - gx(&state, 2, 7, 8, 13, n.block[0]) - gy(&state, 2, 7, 8, 13, n.block[1]) - gx(&state, 3, 4, 9, 14, n.block[6]) - gy(&state, 3, 4, 9, 14, n.block[4]) - - // round6 + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[7], n.block[2]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[5], n.block[3]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[0], n.block[1]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[6], n.block[4]) + // round 6 // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[9]) - gy(&state, 0, 4, 8, 12, n.block[14]) - gx(&state, 1, 5, 9, 13, n.block[11]) - gy(&state, 1, 5, 9, 13, n.block[5]) - gx(&state, 2, 6, 10, 14, n.block[8]) - gy(&state, 2, 6, 10, 14, n.block[12]) - gx(&state, 3, 7, 11, 15, n.block[15]) - gy(&state, 3, 7, 11, 15, n.block[1]) + s[0], s[4], s[8], s[12] = g(s[0], s[4], s[8], s[12], n.block[9], n.block[14]) + s[1], s[5], s[9], s[13] = g(s[1], s[5], s[9], s[13], n.block[11], n.block[5]) + s[2], s[6], s[10], s[14] = g(s[2], s[6], s[10], s[14], n.block[8], n.block[12]) + s[3], s[7], s[11], s[15] = g(s[3], s[7], s[11], s[15], n.block[15], n.block[1]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[13]) - gy(&state, 0, 5, 10, 15, n.block[3]) - gx(&state, 1, 6, 11, 12, n.block[0]) - gy(&state, 1, 6, 11, 12, n.block[10]) - gx(&state, 2, 7, 8, 13, n.block[2]) - gy(&state, 2, 7, 8, 13, n.block[6]) - gx(&state, 3, 4, 9, 14, n.block[4]) - gy(&state, 3, 4, 9, 14, n.block[7]) - - // round7 + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[13], n.block[3]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[0], n.block[10]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[2], n.block[6]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[4], n.block[7]) + // round 7 // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[11]) - gy(&state, 0, 4, 8, 12, n.block[15]) - gx(&state, 1, 5, 9, 13, n.block[5]) - gy(&state, 1, 5, 9, 13, n.block[0]) - gx(&state, 2, 6, 10, 14, n.block[1]) - gy(&state, 2, 6, 10, 14, n.block[9]) - gx(&state, 3, 7, 11, 15, n.block[8]) - gy(&state, 3, 7, 11, 15, n.block[6]) + s[0], s[4], s[8], s[12] = g(s[0], s[4], s[8], s[12], n.block[11], n.block[15]) + s[1], s[5], s[9], s[13] = g(s[1], s[5], s[9], s[13], n.block[5], n.block[0]) + s[2], s[6], s[10], s[14] = g(s[2], s[6], s[10], s[14], n.block[1], n.block[9]) + s[3], s[7], s[11], s[15] = g(s[3], s[7], s[11], s[15], n.block[8], n.block[6]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[14]) - gy(&state, 0, 5, 10, 15, n.block[10]) - gx(&state, 1, 6, 11, 12, n.block[2]) - gy(&state, 1, 6, 11, 12, n.block[12]) - gx(&state, 2, 7, 8, 13, n.block[3]) - gy(&state, 2, 7, 8, 13, n.block[4]) - gx(&state, 3, 4, 9, 14, n.block[7]) - gy(&state, 3, 4, 9, 14, n.block[13]) + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[14], n.block[10]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[2], n.block[12]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[3], n.block[4]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[7], n.block[13]) - for i := range n.cv { - state[i] ^= state[i+8] - state[i+8] ^= n.cv[i] - } + s[0] ^= s[0+8] + s[1] ^= s[1+8] + s[2] ^= s[2+8] + s[3] ^= s[3+8] + s[4] ^= s[4+8] + s[5] ^= s[5+8] + s[6] ^= s[6+8] + s[7] ^= s[7+8] + s[0+8] ^= n.cv[0] + s[1+8] ^= n.cv[1] + s[2+8] ^= n.cv[2] + s[3+8] ^= n.cv[3] + s[4+8] ^= n.cv[4] + s[5+8] ^= n.cv[5] + s[6+8] ^= n.cv[6] + s[7+8] ^= n.cv[7] return } diff --git a/gen/gen.go b/gen/gen.go index 5c3179d..7999b37 100644 --- a/gen/gen.go +++ b/gen/gen.go @@ -24,26 +24,18 @@ func main() { fmt.Printf(`// round%d // Mix the columns. - gx(&state, 0, 4, 8, 12, n.block[%d]) - gy(&state, 0, 4, 8, 12, n.block[%d]) - gx(&state, 1, 5, 9, 13, n.block[%d]) - gy(&state, 1, 5, 9, 13, n.block[%d]) - gx(&state, 2, 6, 10, 14, n.block[%d]) - gy(&state, 2, 6, 10, 14, n.block[%d]) - gx(&state, 3, 7, 11, 15, n.block[%d]) - gy(&state, 3, 7, 11, 15, n.block[%d]) + s[0], s[4], s[8], s[12] = g(s[0], s[4], s[8], s[12], n.block[%d], n.block[%d]) + s[1], s[5], s[9], s[13] = g(s[1], s[5], s[9], s[13], n.block[%d], n.block[%d]) + s[2], s[6], s[10], s[14] = g(s[2], s[6], s[10], s[14], n.block[%d], n.block[%d]) + s[3], s[7], s[11], s[15] = g(s[3], s[7], s[11], s[15], n.block[%d], n.block[%d]) // Mix the diagonals. - gx(&state, 0, 5, 10, 15, n.block[%d]) - gy(&state, 0, 5, 10, 15, n.block[%d]) - gx(&state, 1, 6, 11, 12, n.block[%d]) - gy(&state, 1, 6, 11, 12, n.block[%d]) - gx(&state, 2, 7, 8, 13, n.block[%d]) - gy(&state, 2, 7, 8, 13, n.block[%d]) - gx(&state, 3, 4, 9, 14, n.block[%d]) - gy(&state, 3, 4, 9, 14, n.block[%d]) - -`, x, m[0], m[1], m[2], m[3], m[4], m[5], m[6], m[7], m[8], m[9], m[10], m[11], m[12], m[13], m[14], m[15]) + s[0], s[5], s[10], s[15] = g(s[0], s[5], s[10], s[15], n.block[%d], n.block[%d]) + s[1], s[6], s[11], s[12] = g(s[1], s[6], s[11], s[12], n.block[%d], n.block[%d]) + s[2], s[7], s[8], s[13] = g(s[2], s[7], s[8], s[13], n.block[%d], n.block[%d]) + s[3], s[4], s[9], s[14] = g(s[3], s[4], s[9], s[14], n.block[%d], n.block[%d]) +`, x, + m[0], m[1], m[2], m[3], m[4], m[5], m[6], m[7], m[8], m[9], m[10], m[11], m[12], m[13], m[14], m[15]) permute(&m) } }