From: Hans Kristian Rosbach Date: Wed, 14 Jan 2026 20:18:56 +0000 (+0100) Subject: Unify compare256/compare256_rle benchmarks and add rolling misalignment X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=b7cd3d8d8589ad061b04fcb519e2c2d0212483f7;p=thirdparty%2Fzlib-ng.git Unify compare256/compare256_rle benchmarks and add rolling misalignment --- diff --git a/test/benchmarks/benchmark_compare256.cc b/test/benchmarks/benchmark_compare256.cc index 6853ca2b1..4b616f5a5 100644 --- a/test/benchmarks/benchmark_compare256.cc +++ b/test/benchmarks/benchmark_compare256.cc @@ -3,19 +3,16 @@ * For conditions of distribution and use, see copyright notice in zlib.h */ -#include - #include extern "C" { # include "zbuild.h" -# include "zutil_p.h" # include "arch_functions.h" # include "../test_cpu_features.h" # include "arch/generic/compare256_p.h" } -#define MAX_COMPARE_SIZE (256) +#define MAX_COMPARE_SIZE (256 + 64) class compare256: public benchmark::Fixture { private: @@ -23,41 +20,53 @@ private: uint8_t *str2; public: - void SetUp(const ::benchmark::State&) { - str1 = (uint8_t *)zng_alloc(MAX_COMPARE_SIZE); - assert(str1 != NULL); - memset(str1, 'a', MAX_COMPARE_SIZE); + void SetUp(::benchmark::State& state) { + str1 = (uint8_t *)malloc(MAX_COMPARE_SIZE); + str2 = (uint8_t *)malloc(MAX_COMPARE_SIZE); + if (str1 == NULL || str2 == NULL) { + state.SkipWithError("malloc failed"); + return; + } - str2 = (uint8_t *)zng_alloc(MAX_COMPARE_SIZE); - assert(str2 != NULL); + memset(str1, 'a', MAX_COMPARE_SIZE); memset(str2, 'a', MAX_COMPARE_SIZE); } + // Benchmark compare256, with rolling buffer misalignment for consistent results void Bench(benchmark::State& state, compare256_func compare256) { + int misalign = 0; int32_t match_len = (int32_t)state.range(0) - 1; uint32_t len = 0; - str2[match_len] = 0; for (auto _ : state) { - len = compare256((const uint8_t *)str1, (const uint8_t *)str2); + str2[match_len + misalign] = 0; // Set new match limit + + len = compare256((const uint8_t *)str1 + misalign, (const uint8_t *)str2 + misalign); + + str2[match_len + misalign] = 'a'; // Reset match limit + + if (misalign >= 63) + misalign = 0; + else + misalign++; } - str2[match_len] = 'a'; + // Prevent the result from being optimized away benchmark::DoNotOptimize(len); } void TearDown(const ::benchmark::State&) { - zng_free(str1); - zng_free(str2); + free(str1); + free(str2); } }; -#define BENCHMARK_COMPARE256(name, fptr, support_flag) \ +#define BENCHMARK_COMPARE256(name, comparefunc, support_flag) \ BENCHMARK_DEFINE_F(compare256, name)(benchmark::State& state) { \ if (!(support_flag)) { \ state.SkipWithError("CPU does not support " #name); \ } \ - Bench(state, fptr); \ + Bench(state, comparefunc); \ } \ BENCHMARK_REGISTER_F(compare256, name)->Arg(1)->Arg(10)->Arg(40)->Arg(80)->Arg(100)->Arg(175)->Arg(256); diff --git a/test/benchmarks/benchmark_compare256_rle.cc b/test/benchmarks/benchmark_compare256_rle.cc index c7da50511..7e6551dbd 100644 --- a/test/benchmarks/benchmark_compare256_rle.cc +++ b/test/benchmarks/benchmark_compare256_rle.cc @@ -3,17 +3,14 @@ * For conditions of distribution and use, see copyright notice in zlib.h */ -#include - #include extern "C" { # include "zbuild.h" -# include "zutil_p.h" # include "compare256_rle.h" } -#define MAX_COMPARE_SIZE (256) +#define MAX_COMPARE_SIZE (256 + 64) class compare256_rle: public benchmark::Fixture { private: @@ -21,43 +18,55 @@ private: uint8_t *str2; public: - void SetUp(const ::benchmark::State&) { - str1 = (uint8_t *)zng_alloc(MAX_COMPARE_SIZE); - assert(str1 != NULL); - memset(str1, 'a', MAX_COMPARE_SIZE); + void SetUp(::benchmark::State& state) { + str1 = (uint8_t *)malloc(MAX_COMPARE_SIZE); + str2 = (uint8_t *)malloc(MAX_COMPARE_SIZE); + if (str1 == NULL || str2 == NULL) { + state.SkipWithError("malloc failed"); + return; + } - str2 = (uint8_t *)zng_alloc(MAX_COMPARE_SIZE); - assert(str2 != NULL); + memset(str1, 'a', MAX_COMPARE_SIZE); memset(str2, 'a', MAX_COMPARE_SIZE); } + // Benchmark compare256_rle, with rolling buffer misalignment for consistent results void Bench(benchmark::State& state, compare256_rle_func compare256_rle) { + int misalign = 0; int32_t match_len = (int32_t)state.range(0) - 1; uint32_t len = 0; - str2[match_len] = 0; for (auto _ : state) { - len = compare256_rle((const uint8_t *)str1, (const uint8_t *)str2); + str2[match_len + misalign] = 0; // Set new match limit + + len = compare256_rle((const uint8_t *)str1 + misalign, (const uint8_t *)str2 + misalign); + + str2[match_len + misalign] = 'a'; // Reset match limit + + if (misalign >= 63) + misalign = 0; + else + misalign++; } - str2[match_len] = 'a'; + // Prevent the result from being optimized away benchmark::DoNotOptimize(len); } void TearDown(const ::benchmark::State&) { - zng_free(str1); - zng_free(str2); + free(str1); + free(str2); } }; -#define BENCHMARK_COMPARE256_RLE(name, fptr, support_flag) \ +#define BENCHMARK_COMPARE256_RLE(name, comparefunc, support_flag) \ BENCHMARK_DEFINE_F(compare256_rle, name)(benchmark::State& state) { \ if (!(support_flag)) { \ state.SkipWithError("CPU does not support " #name); \ } \ - Bench(state, fptr); \ + Bench(state, comparefunc); \ } \ - BENCHMARK_REGISTER_F(compare256_rle, name)->Range(1, MAX_COMPARE_SIZE); + BENCHMARK_REGISTER_F(compare256_rle, name)->Arg(1)->Arg(10)->Arg(40)->Arg(80)->Arg(100)->Arg(175)->Arg(256);; BENCHMARK_COMPARE256_RLE(8, compare256_rle_8, 1); BENCHMARK_COMPARE256_RLE(16, compare256_rle_16, 1);