improvements/tally-v2): c51ce99e — Combine extra_lbits/base_length and extra_dbits/base_dist lookup tables1b880ba9 — Make extra length/distance bits computation branchless using bit masking
Hoist s->bi_buf and s->bi_valid into local variables in compress_block() and pass them by pointer to the emit functions. This eliminates redundant load/store pairs between zng_emit_lit and zng_emit_dist calls within the main compression loop.
Comparison of develop (08fa4859) vs HEAD (conditional preload with MIN_HAVE=15).
The patch decodes the next iteration's Huffman symbol before performing the
chunk copy, allowing the table lookup latency to overlap with copy operations.
A can_preload flag skips the preload when the bit accumulator is low (the
UNLIKELY 2+ literal path), keeping INFLATE_FAST_MIN_HAVE at 15 instead of 22.
Extracted by inspecting undefined symbols in functable.c.o for each build — these are the function pointers the functable actually assigns at runtime. Builds use clang -target x86_64-apple-macos with runtime CPU detection enabled (the default).
-march |
SSE2 | SSSE3 | SSE4.1 | SSE4.2 | PCLMUL | AVX2 | AVX-512 | AVX512VNNI | VPCLMUL |
|---|---|---|---|---|---|---|---|---|---|
| x86-64 | - | - | - | - | - | - | - | - | - |
| nehalem | native | native | native | native | - | - | - | - | - |
| /* =========================================================================== | |
| * Symbol buffer write/read macros. | |
| * | |
| * The symbol buffer stores literal and distance/length pairs. The storage | |
| * format differs based on LIT_MEM (separate buffers) vs sym_buf (interleaved), | |
| * and on whether the platform supports fast unaligned 32-bit access | |
| * (OPTIMAL_CMP >= 32), which allows packing a 3-byte symbol into a single | |
| * 32-bit write/read. | |
| * | |
| * SYM_WRITE_LIT and SYM_WRITE_DIST write a symbol and advance sym_next. |
| import re, sys | |
| def parse_results(path): | |
| results = {} | |
| with open(path) as f: | |
| for line in f: | |
| m = re.match(r'\s+(\d+|avg\d?)\s', line) | |
| if not m: | |
| continue | |
| level = m.group(1) |
| name: Benchmark | |
| on: | |
| issue_comment: | |
| types: [created] | |
| workflow_dispatch: | |
| inputs: | |
| pr_number: | |
| description: 'PR number to benchmark (results posted as PR comment)' | |
| required: false | |
| type: number |
| // Set up generic C code fallbacks | |
| #ifndef WITH_ALL_FALLBACKS | |
| // Only use necessary generic functions when no suitable simd versions are available. | |
| // These conditions mirror the native_* defines in arch/*_functions.h headers. | |
| # if (defined(X86_SSE2) && defined(__SSE2__)) || (defined(ARCH_X86) && defined(ARCH_64BIT)) | |
| ft.adler32 = &adler32_c; | |
| ft.adler32_copy = &adler32_copy_c; | |
| ft.crc32 = &crc32_braid; | |
| ft.crc32_copy = &crc32_copy_braid; |
| /* benchmark_tally.cc -- benchmark sym_buf read/write strategies | |
| * Copyright (C) 2024 zlib-ng contributors | |
| * For conditions of distribution and use, see copyright notice in zlib.h | |
| * | |
| * Compares: | |
| * 1. LIT_MEM (separate d_buf/l_buf arrays) | |
| * 2. sym_buf with zng_memread_4/zng_memwrite_4 (batched) | |
| * 3. sym_buf with byte-by-byte access (original) | |
| */ |
| #include <benchmark/benchmark.h> | |
| #include <cstdint> | |
| static inline uint32_t count_matching_bytes_ctzll(uint64_t mask) { | |
| return __builtin_ctzll(mask); | |
| } | |
| static inline uint32_t count_matching_bytes_ctz32(uint64_t mask) { | |
| uint32_t lo = (uint32_t)mask; | |
| if (lo) |