mirror of
https://github.com/ddidderr/wuffs.git
synced 2026-11-17 16:32:11 +01:00
This gives a rough upper bound on how fast Puffs' generated C code can be, with a sufficiently smart Puffs compiler. A comparison of the 'old', safe, generated C code and the 'new', unsafe, hand-written C code: name old speed new speed delta puffs_flate_decode_1k/clang 127MB/s ± 2% 181MB/s ± 2% +42.27% (p=0.000 n=83+5) puffs_flate_decode_10k/clang 151MB/s ± 1% 235MB/s ± 2% +56.04% (p=0.000 n=82+5) puffs_flate_decode_100k/clang 165MB/s ± 1% 275MB/s ± 3% +66.69% (p=0.000 n=89+5) puffs_flate_decode_1k/gcc 155MB/s ± 2% 181MB/s ± 1% +16.96% (p=0.000 n=91+5) puffs_flate_decode_10k/gcc 146MB/s ± 2% 241MB/s ± 2% +64.78% (p=0.000 n=86+5) puffs_flate_decode_100k/gcc 166MB/s ± 1% 277MB/s ± 3% +66.78% (p=0.000 n=88+5) While we're discussing benchmark numbers, below is an overall snapshot. std/flate puffs_flate_decode_1k/clang 127MB/s ± 2% puffs_flate_decode_10k/clang 151MB/s ± 1% puffs_flate_decode_100k/clang 165MB/s ± 1% puffs_flate_decode_1k/gcc 155MB/s ± 2% puffs_flate_decode_10k/gcc 146MB/s ± 2% puffs_flate_decode_100k/gcc 166MB/s ± 1% mimic_adler32_10k 2.98GB/s ± 1% mimic_adler32_100k 2.93GB/s ± 1% mimic_crc32_10k 1.38GB/s ± 1% mimic_crc32_100k 1.37GB/s ± 1% mimic_flate_decode_1k 220MB/s ± 2% mimic_flate_decode_10k 262MB/s ± 1% mimic_flate_decode_100k 281MB/s ± 1% mimic_gzip_decode_10k 221MB/s ± 1% mimic_gzip_decode_100k 259MB/s ± 1% mimic_zlib_decode_10k 242MB/s ± 2% mimic_zlib_decode_100k 289MB/s ± 1% std/gif puffs_gif_lzw_decode_20k/clang 125MB/s ± 1% puffs_gif_lzw_decode_100k/clang 173MB/s ± 1% puffs_gif_decode_1k/clang 376MB/s ± 3% puffs_gif_decode_10k/clang 128MB/s ± 2% puffs_gif_decode_100k/clang 120MB/s ± 2% puffs_gif_decode_1000k/clang 121MB/s ± 3% puffs_gif_lzw_decode_20k/gcc 153MB/s ± 1% puffs_gif_lzw_decode_100k/gcc 208MB/s ± 1% puffs_gif_decode_1k/gcc 429MB/s ± 1% puffs_gif_decode_10k/gcc 161MB/s ± 1% puffs_gif_decode_100k/gcc 140MB/s ± 1% puffs_gif_decode_1000k/gcc 143MB/s ± 1% mimic_gif_decode_1k 158MB/s ± 1% mimic_gif_decode_10k 93.9MB/s ± 1% mimic_gif_decode_100k 100MB/s ± 1% mimic_gif_decode_1000k 102MB/s ± 1%
235 lines
7.1 KiB
C
235 lines
7.1 KiB
C
// Use of this source code is governed by a BSD-style license that can be found
|
|
// in the LICENSE file.
|
|
|
|
// This file contains a hand-written C implementation of gen/c/std/flate.c's
|
|
// generated puffs_flate_decoder_decode_huffman function.
|
|
//
|
|
// It is not intended to be used in production settings, on untrusted data. Its
|
|
// purpose is to give a rough upper bound on how fast Puffs' generated C code
|
|
// can be, with a sufficiently smart Puffs compiler.
|
|
//
|
|
// To repeat, substituting in this C implementation is NOT SAFE, and may result
|
|
// in buffer overflows. This code exists only to aid optimization of the (safe)
|
|
// std/flate/*.puffs code and the Puffs compiler's code generation.
|
|
//
|
|
// ----------------
|
|
//
|
|
// Having said that, to generate the benchmark numbers with this hand-written C
|
|
// implementation, edit gen/c/std/flate.c and find the line that says
|
|
//
|
|
// C HEADER ENDS HERE.
|
|
//
|
|
// After that, add this line:
|
|
//
|
|
// #include "../../../script/puffs-flate-decoder-decode-huffman.c"
|
|
//
|
|
// Then find the call to puffs_flate_decoder_decode_huffman. It should be
|
|
// inside the puffs_flate_decoder_decode function body, and the line of code
|
|
// should look something like
|
|
//
|
|
// status = puffs_flate_decoder_decode_huffman(self, a_dst, a_src);
|
|
//
|
|
// Change the "puffs" to "c_puffs", i.e. add a "c_" prefix. The net result
|
|
// should look something like:
|
|
//
|
|
// ----------------
|
|
//
|
|
// $ git diff
|
|
// diff --git a/gen/c/std/flate.c b/gen/c/std/flate.c
|
|
// index 0fac466..87166fc 100644
|
|
// --- a/gen/c/std/flate.c
|
|
// +++ b/gen/c/std/flate.c
|
|
// @@ -252,6 +252,8 @@ puffs_flate_status
|
|
// puffs_flate_decoder_decode(puffs_flate_decoder* self,
|
|
//
|
|
// // C HEADER ENDS HERE.
|
|
//
|
|
// +#include "../../../script/puffs-flate-decoder-decode-huffman.c"
|
|
// +
|
|
// #ifndef PUFFS_BASE_IMPL_H
|
|
// #define PUFFS_BASE_IMPL_H
|
|
//
|
|
// @@ -580,7 +582,7 @@ puffs_flate_status
|
|
// puffs_flate_decoder_decode(puffs_flate_decoder* self,
|
|
// }
|
|
// }
|
|
// }
|
|
// - status = puffs_flate_decoder_decode_huffman(self, a_dst, a_src);
|
|
// + status = c_puffs_flate_decoder_decode_huffman(self, a_dst, a_src);
|
|
// if (a_src.buf) {
|
|
// b_rptr_src = a_src.buf->ptr + a_src.buf->ri;
|
|
// size_t len = a_src.buf->wi - a_src.buf->ri;
|
|
//
|
|
// ----------------
|
|
//
|
|
// That concludes the two edits to gen/c/std/flate.c. Run the C test and
|
|
// benchmark programs directly, instead of running "puffs test" or "puffs
|
|
// bench", since the "puffs" tool will re-generate the C code and override your
|
|
// gen/c/std/flate.c edit.
|
|
//
|
|
// cd test/c/std
|
|
// gcc -O3 flate.c && ./a.out && ./a.out -bench; rm -f ./a.out
|
|
|
|
#include <stddef.h>
|
|
#include <stdio.h> // For manual printf debugging.
|
|
|
|
puffs_flate_status c_puffs_flate_decoder_decode_huffman(
|
|
puffs_flate_decoder* self,
|
|
puffs_base_writer1 a_dst,
|
|
puffs_base_reader1 a_src) {
|
|
if (!a_dst.buf || !a_src.buf) {
|
|
return PUFFS_FLATE_ERROR_BAD_ARGUMENT;
|
|
}
|
|
puffs_flate_status status = PUFFS_FLATE_STATUS_OK;
|
|
|
|
// Load contextual state.
|
|
uint8_t* pdst = a_dst.buf->ptr + a_dst.buf->wi;
|
|
uint8_t* psrc = a_src.buf->ptr + a_src.buf->ri;
|
|
uint32_t bits = self->private_impl.f_bits;
|
|
uint32_t n_bits = self->private_impl.f_n_bits;
|
|
|
|
// Initialize other local variables.
|
|
uint8_t* pdst0 = pdst;
|
|
uint32_t lmask =
|
|
((((uint32_t)(1)) << self->private_impl.f_n_huffs_bits[0]) - 1);
|
|
uint32_t dmask =
|
|
((((uint32_t)(1)) << self->private_impl.f_n_huffs_bits[1]) - 1);
|
|
|
|
// NOTE: a reminder that this loop is NOT SAFE, as it does not check that
|
|
// pdst and psrc remain within a_dst's and a_src's bounds.
|
|
|
|
outer_loop:
|
|
while (true) {
|
|
// Ensure that we have at least 15 bits of input.
|
|
if (n_bits < 15) {
|
|
bits |= (((uint32_t)(*psrc++)) << n_bits);
|
|
n_bits += 8;
|
|
bits |= (((uint32_t)(*psrc++)) << n_bits);
|
|
n_bits += 8;
|
|
}
|
|
|
|
// Decode an lcode symbol from H-L.
|
|
uint32_t table_entry = self->private_impl.f_huffs[0][bits & lmask];
|
|
while (true) {
|
|
uint32_t n = table_entry & 0x0F;
|
|
bits >>= n;
|
|
n_bits -= n;
|
|
if ((table_entry >> 30) == 1) {
|
|
*pdst++ = (uint8_t)(table_entry >> 8);
|
|
goto outer_loop;
|
|
}
|
|
if ((table_entry >> 29) == 1) {
|
|
// Back-reference.
|
|
break;
|
|
}
|
|
if ((table_entry >> 28) == 1) {
|
|
// End of block.
|
|
goto end;
|
|
}
|
|
if ((table_entry >> 24) != 0x80) {
|
|
status =
|
|
PUFFS_FLATE_ERROR_INTERNAL_ERROR_INCONSISTENT_HUFFMAN_DECODER_STATE;
|
|
goto end;
|
|
}
|
|
uint32_t top = (table_entry >> 8) & 0xFFFF;
|
|
uint32_t mask = ((((uint32_t)(1)) << ((table_entry >> 4) & 0x0F)) - 1);
|
|
table_entry = self->private_impl.f_huffs[0][top + (bits & mask)];
|
|
}
|
|
|
|
// length = base number + extra bits.
|
|
uint32_t length = (table_entry >> 8) & 0xFFFF;
|
|
{
|
|
uint32_t n = (table_entry >> 4) & 0x0F;
|
|
if (n) {
|
|
if (n_bits < n) {
|
|
bits |= (((uint32_t)(*psrc++)) << n_bits);
|
|
n_bits += 8;
|
|
}
|
|
length += bits & ((((uint32_t)(1)) << n) - 1);
|
|
bits >>= n;
|
|
n_bits -= n;
|
|
}
|
|
}
|
|
|
|
// Ensure that we have at least 15 bits of input.
|
|
if (n_bits < 15) {
|
|
bits |= (((uint32_t)(*psrc++)) << n_bits);
|
|
n_bits += 8;
|
|
bits |= (((uint32_t)(*psrc++)) << n_bits);
|
|
n_bits += 8;
|
|
}
|
|
|
|
// Decode a dcode symbol from H-D.
|
|
table_entry = self->private_impl.f_huffs[1][bits & dmask];
|
|
while (true) {
|
|
uint32_t n = table_entry & 0x0F;
|
|
bits >>= n;
|
|
n_bits -= n;
|
|
if ((table_entry >> 29) == 1) {
|
|
// Back-reference.
|
|
break;
|
|
}
|
|
if ((table_entry >> 24) != 0x80) {
|
|
status =
|
|
PUFFS_FLATE_ERROR_INTERNAL_ERROR_INCONSISTENT_HUFFMAN_DECODER_STATE;
|
|
goto end;
|
|
}
|
|
uint32_t top = (table_entry >> 8) & 0xFFFF;
|
|
uint32_t mask = ((((uint32_t)(1)) << ((table_entry >> 4) & 0x0F)) - 1);
|
|
table_entry = self->private_impl.f_huffs[1][top + (bits & mask)];
|
|
}
|
|
|
|
// distance = base number + extra bits.
|
|
uint32_t distance = (table_entry >> 8) & 0xFFFF;
|
|
{
|
|
uint32_t n = (table_entry >> 4) & 0x0F;
|
|
if (n) {
|
|
if (n_bits < 15) {
|
|
bits |= (((uint32_t)(*psrc++)) << n_bits);
|
|
n_bits += 8;
|
|
bits |= (((uint32_t)(*psrc++)) << n_bits);
|
|
n_bits += 8;
|
|
}
|
|
distance += bits & ((((uint32_t)(1)) << n) - 1);
|
|
bits >>= n;
|
|
n_bits -= n;
|
|
}
|
|
}
|
|
|
|
// TODO: look at a sliding window, not just output written so far to dst.
|
|
if ((ptrdiff_t)(distance) > (pdst - pdst0)) {
|
|
status = PUFFS_FLATE_ERROR_BAD_ARGUMENT;
|
|
goto end;
|
|
}
|
|
|
|
uint8_t* pback = pdst - distance;
|
|
for (; length >= 8; length -= 8) {
|
|
*pdst++ = *pback++;
|
|
*pdst++ = *pback++;
|
|
*pdst++ = *pback++;
|
|
*pdst++ = *pback++;
|
|
*pdst++ = *pback++;
|
|
*pdst++ = *pback++;
|
|
*pdst++ = *pback++;
|
|
*pdst++ = *pback++;
|
|
}
|
|
for (; length; length--) {
|
|
*pdst++ = *pback++;
|
|
}
|
|
}
|
|
|
|
end:
|
|
// Return unused input bytes.
|
|
for (; n_bits >= 8; n_bits -= 8) {
|
|
psrc--;
|
|
}
|
|
|
|
// Save contextual state.
|
|
a_dst.buf->wi = pdst - a_dst.buf->ptr;
|
|
a_src.buf->ri = psrc - a_src.buf->ptr;
|
|
self->private_impl.f_bits = bits;
|
|
self->private_impl.f_n_bits = n_bits;
|
|
|
|
return status;
|
|
}
|