perf(base58): add AVX2 (#479)

* perf(base58): add AVX2

* fix(base58): count leading zeros without TZCNT in AVX2 kernels

Author: sonicfromnewyoke

base58/avx2_amd64.go

@@ -0,0 +1,93 @@
+//go:build amd64 && !purego
+
+package base58
+
+// Kernels in avx2_amd64.s. Encoders return the length n and clobber dst[n:].
+// Decoders take 32..44 / 64..88 chars and return 0 (ok), 1 (ErrInvalidChar)
+// or 2 (ErrValueTooLarge); leading-'1' validation is left to the caller.
+
+//go:noescape
+func encode32AVX2(src *[32]byte, dst *[48]byte) int
+
+//go:noescape
+func encode64AVX2(src *[64]byte, dst *[96]byte) int
+
+//go:noescape
+func decode32AVX2(s *byte, n int, dst *[32]byte) int
+
+//go:noescape
+func decode64AVX2(s *byte, n int, dst *[64]byte) int
+
+// Tables widened to qwords (rows padded to 4-lane multiples) for VPMULUDQ.
+var (
+	encTable32q [binarySz32][8]uint64
+	encTable64q [binarySz64][20]uint64
+	decTable32q [intermediateSz32][8]uint64
+	decTable64q [intermediateSz64][16]uint64
+)
+
+func init() {
+	for i, row := range encTable32 {
+		for k, v := range row {
+			encTable32q[i][k] = uint64(v)
+		}
+	}
+	for i, row := range encTable64 {
+		for k, v := range row {
+			encTable64q[i][k] = uint64(v)
+		}
+	}
+	for i, row := range decTable32 {
+		for k, v := range row {
+			decTable32q[i][k] = uint64(v)
+		}
+	}
+	for i, row := range decTable64 {
+		for k, v := range row {
+			decTable64q[i][k] = uint64(v)
+		}
+	}
+}
+
+// rep tiles b into a 32-byte vector constant.
+//
+//nolint:unused // referenced from avx2_amd64.s
+func rep(b ...byte) (out [32]byte) {
+	for i := range out {
+		out[i] = b[i%len(b)]
+	}
+	return
+}
+
+//nolint:unused // referenced from avx2_amd64.s
+var (
+	avxBswap  = rep(3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12)
+	avxIota   = rep(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
+	avxSpread = rep(0, 1, 1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1)
+	avxP      = rep(0, 1, 2, 3, 5, 6, 7, 8, 10, 11, 12, 13, 0x80, 0x80, 0x80, 0x80)
+	avxQ      = rep(4, 0x80, 0x80, 0x80, 9, 0x80, 0x80, 0x80, 14, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80)
+	avxTail   = [32]byte{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
+		1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0x80}
+	avxW1 = rep(58, 1)
+	avxW2 = rep(0x24, 0x0D, 1, 0) // int16 {3364, 1}
+	// valid base58 char iff lo[c&15]&hi[c>>4] != 0
+	avxLoNib = rep(20, 31, 31, 31, 31, 31, 31, 31, 31, 29, 30, 10, 2, 10, 10, 8)
+	avxHiNib = rep(0, 0, 0, 1, 2, 4, 8, 16, 0, 0, 0, 0, 0, 0, 0, 0)
+	avxB8    = rep(8)
+	avxB16   = rep(16)
+	avxB21   = rep(21)
+	avxB32   = rep(32)
+	avxB43   = rep(43)
+	avxB64   = rep(64)
+	avxB73   = rep(73)
+	avxB79   = rep(79)
+	avxB96   = rep(96)
+	avxB108  = rep(108)
+	avxBm6   = rep(0xFA) // -6
+	avxBm7   = rep(0xF9) // -7
+	avxBm49  = rep(0xCF) // -'1'
+	avxB15   = rep(0x0F)
+	avxD58   = rep(58, 0, 0, 0)
+	avxDivA  = rep(0x09, 0xCB, 0x3D, 0x8D, 0, 0, 0, 0) // 2369637129 = 2^37/58
+	avxDivB  = rep(0x93, 0x20, 0xED, 0x4D, 0, 0, 0, 0) // 1307386003 = 2^42/58^2
+)