1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
use crate::{
    engine::{general_purpose::INVALID_VALUE, DecodeEstimate, DecodePaddingMode},
    DecodeError, PAD_BYTE,
};

// decode logic operates on chunks of 8 input bytes without padding
const INPUT_CHUNK_LEN: usize = 8;
const DECODED_CHUNK_LEN: usize = 6;

// we read a u64 and write a u64, but a u64 of input only yields 6 bytes of output, so the last
// 2 bytes of any output u64 should not be counted as written to (but must be available in a
// slice).
const DECODED_CHUNK_SUFFIX: usize = 2;

// how many u64's of input to handle at a time
const CHUNKS_PER_FAST_LOOP_BLOCK: usize = 4;

const INPUT_BLOCK_LEN: usize = CHUNKS_PER_FAST_LOOP_BLOCK * INPUT_CHUNK_LEN;

// includes the trailing 2 bytes for the final u64 write
const DECODED_BLOCK_LEN: usize =
    CHUNKS_PER_FAST_LOOP_BLOCK * DECODED_CHUNK_LEN + DECODED_CHUNK_SUFFIX;

#[doc(hidden)]
pub struct GeneralPurposeEstimate {
    /// Total number of decode chunks, including a possibly partial last chunk
    num_chunks: usize,
    decoded_len_estimate: usize,
}

impl GeneralPurposeEstimate {
    pub(crate) fn new(encoded_len: usize) -> Self {
        Self {
            num_chunks: encoded_len
                .checked_add(INPUT_CHUNK_LEN - 1)
                .expect("Overflow when calculating number of chunks in input")
                / INPUT_CHUNK_LEN,
            decoded_len_estimate: encoded_len
                .checked_add(3)
                .expect("Overflow when calculating decoded len estimate")
                / 4
                * 3,
        }
    }
}

impl DecodeEstimate for GeneralPurposeEstimate {
    fn decoded_len_estimate(&self) -> usize {
        self.decoded_len_estimate
    }
}

/// Helper to avoid duplicating num_chunks calculation, which is costly on short inputs.
/// Returns the number of bytes written, or an error.
// We're on the fragile edge of compiler heuristics here. If this is not inlined, slow. If this is
// inlined(always), a different slow. plain ol' inline makes the benchmarks happiest at the moment,
// but this is fragile and the best setting changes with only minor code modifications.
#[inline]
pub(crate) fn decode_helper(
    input: &[u8],
    estimate: GeneralPurposeEstimate,
    output: &mut [u8],
    decode_table: &[u8; 256],
    decode_allow_trailing_bits: bool,
    padding_mode: DecodePaddingMode,
) -> Result<usize, DecodeError> {
    let remainder_len = input.len() % INPUT_CHUNK_LEN;

    // Because the fast decode loop writes in groups of 8 bytes (unrolled to
    // CHUNKS_PER_FAST_LOOP_BLOCK times 8 bytes, where possible) and outputs 8 bytes at a time (of
    // which only 6 are valid data), we need to be sure that we stop using the fast decode loop
    // soon enough that there will always be 2 more bytes of valid data written after that loop.
    let trailing_bytes_to_skip = match remainder_len {
        // if input is a multiple of the chunk size, ignore the last chunk as it may have padding,
        // and the fast decode logic cannot handle padding
        0 => INPUT_CHUNK_LEN,
        // 1 and 5 trailing bytes are illegal: can't decode 6 bits of input into a byte
        1 | 5 => {
            // trailing whitespace is so common that it's worth it to check the last byte to
            // possibly return a better error message
            if let Some(b) = input.last() {
                if *b != PAD_BYTE && decode_table[*b as usize] == INVALID_VALUE {
                    return Err(DecodeError::InvalidByte(input.len() - 1, *b));
                }
            }

            return Err(DecodeError::InvalidLength);
        }
        // This will decode to one output byte, which isn't enough to overwrite the 2 extra bytes
        // written by the fast decode loop. So, we have to ignore both these 2 bytes and the
        // previous chunk.
        2 => INPUT_CHUNK_LEN + 2,
        // If this is 3 un-padded chars, then it would actually decode to 2 bytes. However, if this
        // is an erroneous 2 chars + 1 pad char that would decode to 1 byte, then it should fail
        // with an error, not panic from going past the bounds of the output slice, so we let it
        // use stage 3 + 4.
        3 => INPUT_CHUNK_LEN + 3,
        // This can also decode to one output byte because it may be 2 input chars + 2 padding
        // chars, which would decode to 1 byte.
        4 => INPUT_CHUNK_LEN + 4,
        // Everything else is a legal decode len (given that we don't require padding), and will
        // decode to at least 2 bytes of output.
        _ => remainder_len,
    };

    // rounded up to include partial chunks
    let mut remaining_chunks = estimate.num_chunks;

    let mut input_index = 0;
    let mut output_index = 0;

    {
        let length_of_fast_decode_chunks = input.len().saturating_sub(trailing_bytes_to_skip);

        // Fast loop, stage 1
        // manual unroll to CHUNKS_PER_FAST_LOOP_BLOCK of u64s to amortize slice bounds checks
        if let Some(max_start_index) = length_of_fast_decode_chunks.checked_sub(INPUT_BLOCK_LEN) {
            while input_index <= max_start_index {
                let input_slice = &input[input_index..(input_index + INPUT_BLOCK_LEN)];
                let output_slice = &mut output[output_index..(output_index + DECODED_BLOCK_LEN)];

                decode_chunk(
                    &input_slice[0..],
                    input_index,
                    decode_table,
                    &mut output_slice[0..],
                )?;
                decode_chunk(
                    &input_slice[8..],
                    input_index + 8,
                    decode_table,
                    &mut output_slice[6..],
                )?;
                decode_chunk(
                    &input_slice[16..],
                    input_index + 16,
                    decode_table,
                    &mut output_slice[12..],
                )?;
                decode_chunk(
                    &input_slice[24..],
                    input_index + 24,
                    decode_table,
                    &mut output_slice[18..],
                )?;

                input_index += INPUT_BLOCK_LEN;
                output_index += DECODED_BLOCK_LEN - DECODED_CHUNK_SUFFIX;
                remaining_chunks -= CHUNKS_PER_FAST_LOOP_BLOCK;
            }
        }

        // Fast loop, stage 2 (aka still pretty fast loop)
        // 8 bytes at a time for whatever we didn't do in stage 1.
        if let Some(max_start_index) = length_of_fast_decode_chunks.checked_sub(INPUT_CHUNK_LEN) {
            while input_index < max_start_index {
                decode_chunk(
                    &input[input_index..(input_index + INPUT_CHUNK_LEN)],
                    input_index,
                    decode_table,
                    &mut output
                        [output_index..(output_index + DECODED_CHUNK_LEN + DECODED_CHUNK_SUFFIX)],
                )?;

                output_index += DECODED_CHUNK_LEN;
                input_index += INPUT_CHUNK_LEN;
                remaining_chunks -= 1;
            }
        }
    }

    // Stage 3
    // If input length was such that a chunk had to be deferred until after the fast loop
    // because decoding it would have produced 2 trailing bytes that wouldn't then be
    // overwritten, we decode that chunk here. This way is slower but doesn't write the 2
    // trailing bytes.
    // However, we still need to avoid the last chunk (partial or complete) because it could
    // have padding, so we always do 1 fewer to avoid the last chunk.
    for _ in 1..remaining_chunks {
        decode_chunk_precise(
            &input[input_index..],
            input_index,
            decode_table,
            &mut output[output_index..(output_index + DECODED_CHUNK_LEN)],
        )?;

        input_index += INPUT_CHUNK_LEN;
        output_index += DECODED_CHUNK_LEN;
    }

    // always have one more (possibly partial) block of 8 input
    debug_assert!(input.len() - input_index > 1 || input.is_empty());
    debug_assert!(input.len() - input_index <= 8);

    super::decode_suffix::decode_suffix(
        input,
        input_index,
        output,
        output_index,
        decode_table,
        decode_allow_trailing_bits,
        padding_mode,
    )
}

/// Decode 8 bytes of input into 6 bytes of output. 8 bytes of output will be written, but only the
/// first 6 of those contain meaningful data.
///
/// `input` is the bytes to decode, of which the first 8 bytes will be processed.
/// `index_at_start_of_input` is the offset in the overall input (used for reporting errors
/// accurately)
/// `decode_table` is the lookup table for the particular base64 alphabet.
/// `output` will have its first 8 bytes overwritten, of which only the first 6 are valid decoded
/// data.
// yes, really inline (worth 30-50% speedup)
#[inline(always)]
fn decode_chunk(
    input: &[u8],
    index_at_start_of_input: usize,
    decode_table: &[u8; 256],
    output: &mut [u8],
) -> Result<(), DecodeError> {
    let morsel = decode_table[input[0] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(index_at_start_of_input, input[0]));
    }
    let mut accum = (morsel as u64) << 58;

    let morsel = decode_table[input[1] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(
            index_at_start_of_input + 1,
            input[1],
        ));
    }
    accum |= (morsel as u64) << 52;

    let morsel = decode_table[input[2] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(
            index_at_start_of_input + 2,
            input[2],
        ));
    }
    accum |= (morsel as u64) << 46;

    let morsel = decode_table[input[3] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(
            index_at_start_of_input + 3,
            input[3],
        ));
    }
    accum |= (morsel as u64) << 40;

    let morsel = decode_table[input[4] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(
            index_at_start_of_input + 4,
            input[4],
        ));
    }
    accum |= (morsel as u64) << 34;

    let morsel = decode_table[input[5] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(
            index_at_start_of_input + 5,
            input[5],
        ));
    }
    accum |= (morsel as u64) << 28;

    let morsel = decode_table[input[6] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(
            index_at_start_of_input + 6,
            input[6],
        ));
    }
    accum |= (morsel as u64) << 22;

    let morsel = decode_table[input[7] as usize];
    if morsel == INVALID_VALUE {
        return Err(DecodeError::InvalidByte(
            index_at_start_of_input + 7,
            input[7],
        ));
    }
    accum |= (morsel as u64) << 16;

    write_u64(output, accum);

    Ok(())
}

/// Decode an 8-byte chunk, but only write the 6 bytes actually decoded instead of including 2
/// trailing garbage bytes.
#[inline]
fn decode_chunk_precise(
    input: &[u8],
    index_at_start_of_input: usize,
    decode_table: &[u8; 256],
    output: &mut [u8],
) -> Result<(), DecodeError> {
    let mut tmp_buf = [0_u8; 8];

    decode_chunk(
        input,
        index_at_start_of_input,
        decode_table,
        &mut tmp_buf[..],
    )?;

    output[0..6].copy_from_slice(&tmp_buf[0..6]);

    Ok(())
}

#[inline]
fn write_u64(output: &mut [u8], value: u64) {
    output[..8].copy_from_slice(&value.to_be_bytes());
}

#[cfg(test)]
mod tests {
    use super::*;

    use crate::engine::general_purpose::STANDARD;

    #[test]
    fn decode_chunk_precise_writes_only_6_bytes() {
        let input = b"Zm9vYmFy"; // "foobar"
        let mut output = [0_u8, 1, 2, 3, 4, 5, 6, 7];

        decode_chunk_precise(&input[..], 0, &STANDARD.decode_table, &mut output).unwrap();
        assert_eq!(&vec![b'f', b'o', b'o', b'b', b'a', b'r', 6, 7], &output);
    }

    #[test]
    fn decode_chunk_writes_8_bytes() {
        let input = b"Zm9vYmFy"; // "foobar"
        let mut output = [0_u8, 1, 2, 3, 4, 5, 6, 7];

        decode_chunk(&input[..], 0, &STANDARD.decode_table, &mut output).unwrap();
        assert_eq!(&vec![b'f', b'o', b'o', b'b', b'a', b'r', 0, 0], &output);
    }
}