encoding.base64: speed up encoding and decoding (#11055)

* add new function headers

* new encoding function

* rename to decode_micro and add helper functions

* implement new decoding function

* add test for buffer to buffer decoding

* - add notice to GO code
- v fmt base64.v

* implement new decoding function

* fix base64_memory_test.v by commenting a few lines

* vfmt base64.v

* add some more asserts to base64_memory_test.v

* remove unused decoding function

* add bounds check, when detecting the padding

* use union for storing the decoded data

Co-authored-by: Delyan Angelov <delian66@gmail.com>

vlib/encoding/base64/base64.v

@@ -1,6 +1,8 @@
 // Copyright (c) 2019-2021 Alexander Medvednikov. All rights reserved.
 // Use of this source code is governed by an MIT license
 // that can be found in the LICENSE file.
+// Based off:   https://github.com/golang/go/blob/master/src/encoding/base64/base64.go
+// Last commit: https://github.com/golang/go/commit/9a93baf4d7d13d7d5c67388c93960d78abc8e11e
 module base64
 
 const (
@@ -13,6 +15,18 @@ const (
 	enc_table    = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/'
 )
 
+union B64_64_datablock {
+mut:
+	data      u64
+	data_byte [8]byte
+}
+
+union B64_32_datablock {
+mut:
+	data      u32
+	data_byte [4]byte
+}
+
 // decode decodes the base64 encoded `string` value passed in `data`.
 // Please note: If you need to decode many strings repeatedly, take a look at `decode_in_buffer`.
 // Example: assert base64.decode('ViBpbiBiYXNlIDY0') == 'V in base 64'
@@ -104,63 +118,122 @@ pub fn url_encode_str(data string) string {
 	return encode_str(data).replace_each(['+', '-', '/', '_', '=', ''])
 }
 
+// assemble64 assembles 8 base64 digits into 6 bytes.
+// Each digit comes from the decode map.
+// Please note: Invalid base64 digits are not expected and not handled.
+fn assemble64(n1 byte, n2 byte, n3 byte, n4 byte, n5 byte, n6 byte, n7 byte, n8 byte) u64 {
+	return u64(n1) << 58 | u64(n2) << 52 | u64(n3) << 46 | u64(n4) << 40 | u64(n5) << 34 | u64(n6) << 28 | u64(n7) << 22 | u64(n8) << 16
+}
+
+// assemble32 assembles 4 base64 digits into 3 bytes.
+// Each digit comes from the decode map.
+// Please note: Invalid base64 digits are not expected and not handled.
+fn assemble32(n1 byte, n2 byte, n3 byte, n4 byte) u32 {
+	return u32(n1) << 26 | u32(n2) << 20 | u32(n3) << 14 | u32(n4) << 8
+}
+
 // decode_in_buffer decodes the base64 encoded `string` reference passed in `data` into `buffer`.
 // decode_in_buffer returns the size of the decoded data in the buffer.
 // Please note: The `buffer` should be large enough (i.e. 3/4 of the data.len, or larger)
 // to hold the decoded data.
 // Please note: This function does NOT allocate new memory, and is thus suitable for handling very large strings.
 pub fn decode_in_buffer(data &string, buffer &byte) int {
+	return decode_from_buffer(buffer, data.str, data.len)
+}
+
+// decode_from_buffer decodes the base64 encoded ASCII bytes from `data` into `buffer`.
+// decode_from_buffer returns the size of the decoded data in the buffer.
+// Please note: The `buffer` should be large enough (i.e. 3/4 of the data.len, or larger)
+// to hold the decoded data.
+// Please note: This function does NOT allocate new memory, and is thus suitable for handling very large strings.
+pub fn decode_in_buffer_bytes(data []byte, buffer &byte) int {
+	return decode_from_buffer(buffer, data.data, data.len)
+}
+
+// decode_from_buffer decodes the base64 encoded ASCII bytes from `src` into `dest`.
+// decode_from_buffer returns the size of the decoded data in the buffer.
+// Please note: The `dest` buffer should be large enough (i.e. 3/4 of the `src_len`, or larger)
+// to hold the decoded data.
+// Please note: This function does NOT allocate new memory, and is thus suitable for handling very large strings.
+// Please note: This function is for internal base64 decoding
+fn decode_from_buffer(dest &byte, src &byte, src_len int) int {
+	if src_len < 4 {
+		return 0
+	}
+
 	mut padding := 0
-	if data.ends_with('=') {
+	if unsafe { src[src_len - 1] == `=` } {
-		if data.ends_with('==') {
+		if unsafe { src[src_len - 2] == `=` } {
 			padding = 2
 		} else {
 			padding = 1
 		}
 	}
-	// input_length is the length of meaningful data
-	input_length := data.len - padding
-	output_length := input_length * 3 / 4
 
-	mut i := 0
+	mut d := unsafe { src }
-	mut j := 0
+	mut b := unsafe { dest }
-	mut b := &byte(0)
+
-	mut d := &byte(0)
 	unsafe {
-		d = &byte(data.str)
+		mut n_decoded_bytes := 0 // padding bytes are also counted towards this.
-		b = &byte(buffer)
+		mut si := 0
+
+		mut datablock_64 := B64_64_datablock{
+			data: 0
 		}
-	for i < input_length {
+		mut datablock_32 := B64_32_datablock{
-		mut char_a := 0
+			data: 0
-		mut char_b := 0
-		mut char_c := 0
-		mut char_d := 0
-		if i < input_length {
-			char_a = base64.index[unsafe { d[i] }]
-			i++
-		}
-		if i < input_length {
-			char_b = base64.index[unsafe { d[i] }]
-			i++
-		}
-		if i < input_length {
-			char_c = base64.index[unsafe { d[i] }]
-			i++
-		}
-		if i < input_length {
-			char_d = base64.index[unsafe { d[i] }]
-			i++
 		}
 
-		decoded_bytes := (char_a << 18) | (char_b << 12) | (char_c << 6) | (char_d << 0)
+		for src_len - si >= 8 {
-		unsafe {
+			// Converting 8 bytes of input into 6 bytes of output. Storing these in the upper bytes of an u64.
-			b[j] = byte(decoded_bytes >> 16)
+			datablock_64.data = assemble64(byte(base64.index[d[si + 0]]), byte(base64.index[d[si + 1]]),
-			b[j + 1] = byte((decoded_bytes >> 8) & 0xff)
+				byte(base64.index[d[si + 2]]), byte(base64.index[d[si + 3]]), byte(base64.index[d[
-			b[j + 2] = byte((decoded_bytes >> 0) & 0xff)
+				si + 4]]), byte(base64.index[d[si + 5]]), byte(base64.index[d[si + 6]]),
+				byte(base64.index[d[si + 7]]))
+
+			// Reading out the individual bytes from the u64. Watch out with endianess.
+			$if little_endian {
+				b[n_decoded_bytes + 0] = datablock_64.data_byte[7]
+				b[n_decoded_bytes + 1] = datablock_64.data_byte[6]
+				b[n_decoded_bytes + 2] = datablock_64.data_byte[5]
+				b[n_decoded_bytes + 3] = datablock_64.data_byte[4]
+				b[n_decoded_bytes + 4] = datablock_64.data_byte[3]
+				b[n_decoded_bytes + 5] = datablock_64.data_byte[2]
+			} $else {
+				b[n_decoded_bytes + 0] = datablock_64.data_byte[0]
+				b[n_decoded_bytes + 1] = datablock_64.data_byte[1]
+				b[n_decoded_bytes + 2] = datablock_64.data_byte[2]
+				b[n_decoded_bytes + 3] = datablock_64.data_byte[3]
+				b[n_decoded_bytes + 4] = datablock_64.data_byte[4]
+				b[n_decoded_bytes + 5] = datablock_64.data_byte[5]
 			}
-		j += 3
+
+			n_decoded_bytes += 6
+			si += 8
+		}
+
+		for src_len - si >= 4 {
+			datablock_32.data = assemble32(byte(base64.index[d[si + 0]]), byte(base64.index[d[si + 1]]),
+				byte(base64.index[d[si + 2]]), byte(base64.index[d[si + 3]]))
+
+			$if little_endian {
+				b[n_decoded_bytes + 0] = datablock_32.data_byte[3]
+				b[n_decoded_bytes + 1] = datablock_32.data_byte[2]
+				b[n_decoded_bytes + 2] = datablock_32.data_byte[1]
+				b[n_decoded_bytes + 3] = datablock_32.data_byte[0]
+			} $else {
+				b[n_decoded_bytes + 0] = datablock_32.data_byte[0]
+				b[n_decoded_bytes + 1] = datablock_32.data_byte[1]
+				b[n_decoded_bytes + 2] = datablock_32.data_byte[2]
+				b[n_decoded_bytes + 3] = datablock_32.data_byte[3]
+			}
+
+			n_decoded_bytes += 3
+			si += 4
+		}
+
+		return n_decoded_bytes - padding
 	}
-	return output_length
 }
 
 // encode_in_buffer base64 encodes the `[]byte` passed in `data` into `buffer`.
@@ -176,48 +249,59 @@ pub fn encode_in_buffer(data []byte, buffer &byte) int {
 // Please note: The `dest` buffer should be large enough (i.e. 4/3 of the src_len, or larger) to hold the encoded data.
 // Please note: This function is for internal base64 encoding
 fn encode_from_buffer(dest &byte, src &byte, src_len int) int {
-	input_length := src_len
+	if src_len == 0 {
-	output_length := 4 * ((input_length + 2) / 3)
+		return 0
-
+	}
-	mut i := 0
+	output_length := 4 * ((src_len + 2) / 3)
-	mut j := 0
 
 	mut d := unsafe { src }
 	mut b := unsafe { dest }
-	mut etable := base64.enc_table.str
+	etable := base64.enc_table.str
-	for i < input_length {
-		mut octet_a := 0
-		mut octet_b := 0
-		mut octet_c := 0
-
-		if i < input_length {
-			octet_a = int(unsafe { d[i] })
-			i++
-		}
-		if i < input_length {
-			octet_b = int(unsafe { d[i] })
-			i++
-		}
-		if i < input_length {
-			octet_c = int(unsafe { d[i] })
-			i++
-		}
-
-		triple := ((octet_a << 0x10) + (octet_b << 0x08) + octet_c)
 
+	mut di := 0
+	mut si := 0
+	n := (src_len / 3) * 3
+	for si < n {
+		// Convert 3x 8bit source bytes into 4 bytes
 		unsafe {
-			b[j] = etable[(triple >> 3 * 6) & 63] // 63 is 0x3F
+			val := u32(d[si + 0]) << 16 | u32(d[si + 1]) << 8 | u32(d[si + 2])
-			b[j + 1] = etable[(triple >> 2 * 6) & 63]
+
-			b[j + 2] = etable[(triple >> 1 * 6) & 63]
+			b[di + 0] = etable[val >> 18 & 0x3F]
-			b[j + 3] = etable[(triple >> 0 * 6) & 63]
+			b[di + 1] = etable[val >> 12 & 0x3F]
+			b[di + 2] = etable[val >> 6 & 0x3F]
+			b[di + 3] = etable[val & 0x3F]
 		}
-		j += 4
+		si += 3
+		di += 4
 	}
 
-	padding_length := base64.ending_table[input_length % 3]
+	remain := src_len - si
-	for i = 0; i < padding_length; i++ {
+	if remain == 0 {
+		return output_length
+	}
+
+	// Add the remaining small block and padding
 	unsafe {
-			b[output_length - 1 - i] = `=`
+		mut val := u32(d[si + 0]) << 16
+		if remain == 2 {
+			val |= u32(d[si + 1]) << 8
+		}
+
+		b[di + 0] = etable[val >> 18 & 0x3F]
+		b[di + 1] = etable[val >> 12 & 0x3F]
+
+		match remain {
+			2 {
+				b[di + 2] = etable[val >> 6 & 0x3F]
+				b[di + 3] = byte(`=`)
+			}
+			1 {
+				b[di + 2] = byte(`=`)
+				b[di + 3] = byte(`=`)
+			}
+			else {
+				panic('base64: This case should never occur.')
+			}
 		}
 	}
 	return output_length

vlib/encoding/base64/base64_memory_test.v

@@ -11,16 +11,37 @@ fn test_long_encoding() {
 	assert s_encoded.len > s_original.len
 	assert s_original == s_decoded
 
-	mut s := 0
-
 	ebuffer := unsafe { malloc(s_encoded.len) }
+	dbuffer := unsafe { malloc(s_decoded.len) }
+	//
+	encoded_size := base64.encode_in_buffer(s_original, ebuffer)
+	mut encoded_in_buf := []byte{len: encoded_size}
+	unsafe { C.memcpy(encoded_in_buf.data, ebuffer, encoded_size) }
+	assert input_size * 4 / 3 == encoded_size
+	assert encoded_in_buf[0] == `Y`
+	assert encoded_in_buf[1] == `W`
+	assert encoded_in_buf[2] == `F`
+	assert encoded_in_buf[3] == `h`
+
+	assert encoded_in_buf[encoded_size - 4] == `Y`
+	assert encoded_in_buf[encoded_size - 3] == `W`
+	assert encoded_in_buf[encoded_size - 2] == `F`
+	assert encoded_in_buf[encoded_size - 1] == `h`
+	assert encoded_in_buf == s_encoded.bytes()
+
+	decoded_size := base64.decode_in_buffer(s_encoded, dbuffer)
+	assert decoded_size == input_size
+	mut decoded_in_buf := []byte{len: decoded_size}
+	unsafe { C.memcpy(decoded_in_buf.data, dbuffer, decoded_size) }
+	assert decoded_in_buf == s_original
+
+	mut s := 0
 	for _ in 0 .. repeats {
 		resultsize := base64.encode_in_buffer(s_original, ebuffer)
 		s += resultsize
 		assert resultsize == s_encoded.len
 	}
 
-	dbuffer := unsafe { malloc(s_decoded.len) }
 	for _ in 0 .. repeats {
 		resultsize := base64.decode_in_buffer(s_encoded, dbuffer)
 		s += resultsize

vlib/encoding/base64/base64_test.v

@@ -130,3 +130,21 @@ fn test_decode_null_byte_str() {
 	s := [byte(`A`), 0, `C`].bytestr()
 	assert base64.decode_str('QQBD') == s
 }
+
+fn test_decode_in_buffer_bytes() {
+	rfc4648_pairs := [
+		TestPair{'foob', 'Zm9vYg=='},
+		TestPair{'fooba', 'Zm9vYmE='},
+		TestPair{'foobar', 'Zm9vYmFy'},
+	]
+	mut src_dec_buf := []byte{len: 8}
+	mut src_enc_buf := []byte{len: 8}
+	mut out_buf := []byte{len: 8}
+
+	for p in rfc4648_pairs {
+		src_dec_buf = p.decoded.bytes()
+		src_enc_buf = p.encoded.bytes()
+		n := base64.decode_in_buffer_bytes(src_enc_buf, out_buf.data)
+		assert src_dec_buf == out_buf[..n]
+	}
+}