1 /* 2 * Assembly testing and benchmarking tool 3 * Copyright (c) 2015 Henrik Gramner 4 * Copyright (c) 2008 Loren Merritt 5 * 6 * This file is part of FFmpeg. 7 * 8 * FFmpeg is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation; either version 2 of the License, or 11 * (at your option) any later version. 12 * 13 * FFmpeg is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 * GNU General Public License for more details. 17 * 18 * You should have received a copy of the GNU General Public License along 19 * with FFmpeg; if not, write to the Free Software Foundation, Inc., 20 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. 21 */ 22 23 #ifndef TESTS_CHECKASM_CHECKASM_H 24 #define TESTS_CHECKASM_CHECKASM_H 25 26 #include <stdint.h> 27 #include "config.h" 28 29 #if CONFIG_LINUX_PERF 30 #include <unistd.h> // read(3) 31 #include <sys/ioctl.h> 32 #include <asm/unistd.h> 33 #include <linux/perf_event.h> 34 #endif 35 36 #include "libavutil/avstring.h" 37 #include "libavutil/cpu.h" 38 #include "libavutil/internal.h" 39 #include "libavutil/lfg.h" 40 #include "libavutil/timer.h" 41 42 void checkasm_check_aacpsdsp(void); 43 void checkasm_check_alacdsp(void); 44 void checkasm_check_audiodsp(void); 45 void checkasm_check_blend(void); 46 void checkasm_check_blockdsp(void); 47 void checkasm_check_bswapdsp(void); 48 void checkasm_check_colorspace(void); 49 void checkasm_check_exrdsp(void); 50 void checkasm_check_fixed_dsp(void); 51 void checkasm_check_flacdsp(void); 52 void checkasm_check_float_dsp(void); 53 void checkasm_check_fmtconvert(void); 54 void checkasm_check_g722dsp(void); 55 void checkasm_check_h264dsp(void); 56 void checkasm_check_h264pred(void); 57 void checkasm_check_h264qpel(void); 58 void checkasm_check_hevc_add_res(void); 59 void checkasm_check_hevc_idct(void); 60 void checkasm_check_hevc_sao(void); 61 void checkasm_check_huffyuvdsp(void); 62 void checkasm_check_jpeg2000dsp(void); 63 void checkasm_check_llviddsp(void); 64 void checkasm_check_llviddspenc(void); 65 void checkasm_check_nlmeans(void); 66 void checkasm_check_pixblockdsp(void); 67 void checkasm_check_sbrdsp(void); 68 void checkasm_check_synth_filter(void); 69 void checkasm_check_sw_rgb(void); 70 void checkasm_check_utvideodsp(void); 71 void checkasm_check_v210enc(void); 72 void checkasm_check_vf_hflip(void); 73 void checkasm_check_vf_threshold(void); 74 void checkasm_check_vp8dsp(void); 75 void checkasm_check_vp9dsp(void); 76 void checkasm_check_videodsp(void); 77 78 struct CheckasmPerf; 79 80 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3); 81 int checkasm_bench_func(void); 82 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2); 83 struct CheckasmPerf *checkasm_get_perf_context(void); 84 void checkasm_report(const char *name, ...) av_printf_format(1, 2); 85 86 /* float compare utilities */ 87 int float_near_ulp(float a, float b, unsigned max_ulp); 88 int float_near_abs_eps(float a, float b, float eps); 89 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp); 90 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, 91 unsigned len); 92 int float_near_abs_eps_array(const float *a, const float *b, float eps, 93 unsigned len); 94 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, 95 unsigned max_ulp, unsigned len); 96 int double_near_abs_eps(double a, double b, double eps); 97 int double_near_abs_eps_array(const double *a, const double *b, double eps, 98 unsigned len); 99 100 extern AVLFG checkasm_lfg; 101 #define rnd() av_lfg_get(&checkasm_lfg) 102 103 static av_unused void *func_ref, *func_new; 104 105 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */ 106 107 /* Decide whether or not the specified function needs to be tested */ 108 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__)) 109 110 /* Declare the function prototype. The first argument is the return value, the remaining 111 * arguments are the function parameters. Naming parameters is optional. */ 112 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__) 113 #define declare_func_float(ret, ...) declare_new_float(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__) 114 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__) 115 116 /* Indicate that the current test has failed */ 117 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__) 118 119 /* Print the test outcome */ 120 #define report checkasm_report 121 122 /* Call the reference function */ 123 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__) 124 125 #if ARCH_X86 && HAVE_X86ASM 126 /* Verifies that clobbered callee-saved registers are properly saved and restored 127 * and that either no MMX registers are touched or emms is issued */ 128 void checkasm_checked_call(void *func, ...); 129 /* Verifies that clobbered callee-saved registers are properly saved and restored 130 * and issues emms for asm functions which are not required to do so */ 131 void checkasm_checked_call_emms(void *func, ...); 132 /* Verifies that clobbered callee-saved registers are properly saved and restored 133 * but doesn't issue emms. Meant for dsp functions returning float or double */ 134 void checkasm_checked_call_float(void *func, ...); 135 136 #if ARCH_X86_64 137 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit. 138 * This is done by clobbering the stack with junk around the stack pointer and calling the 139 * assembly function through checked_call() with added dummy arguments which forces all 140 * real arguments to be passed on the stack and not in registers. For 32-bit arguments the 141 * upper half of the 64-bit register locations on the stack will now contain junk which will 142 * cause misbehaving functions to either produce incorrect output or segfault. Note that 143 * even though this works extremely well in practice, it's technically not guaranteed 144 * and false negatives is theoretically possible, but there can never be any false positives. 145 */ 146 void checkasm_stack_clobber(uint64_t clobber, ...); 147 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\ 148 = (void *)checkasm_checked_call; 149 #define declare_new_float(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\ 150 = (void *)checkasm_checked_call_float; 151 #define declare_new_emms(cpu_flags, ret, ...) \ 152 ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \ 153 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \ 154 (void *)checkasm_checked_call; 155 #define CLOB (UINT64_C(0xdeadbeefdeadbeef)) 156 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\ 157 CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\ 158 checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__)) 159 #elif ARCH_X86_32 160 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call; 161 #define declare_new_float(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call_float; 162 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \ 163 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \ 164 (void *)checkasm_checked_call; 165 #define call_new(...) checked_call(func_new, __VA_ARGS__) 166 #endif 167 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL 168 /* Use a dummy argument, to offset the real parameters by 2, not only 1. 169 * This makes sure that potential 8-byte-alignment of parameters is kept the same 170 * even when the extra parameters have been removed. */ 171 void checkasm_checked_call_vfp(void *func, int dummy, ...); 172 void checkasm_checked_call_novfp(void *func, int dummy, ...); 173 extern void (*checkasm_checked_call)(void *func, int dummy, ...); 174 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__) = (void *)checkasm_checked_call; 175 #define call_new(...) checked_call(func_new, 0, __VA_ARGS__) 176 #elif ARCH_AARCH64 && !defined(__APPLE__) 177 void checkasm_stack_clobber(uint64_t clobber, ...); 178 void checkasm_checked_call(void *func, ...); 179 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, int, int, __VA_ARGS__)\ 180 = (void *)checkasm_checked_call; 181 #define CLOB (UINT64_C(0xdeadbeefdeadbeef)) 182 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\ 183 CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\ 184 checked_call(func_new, 0, 0, 0, 0, 0, 0, 0, __VA_ARGS__)) 185 #else 186 #define declare_new(ret, ...) 187 #define declare_new_float(ret, ...) 188 #define declare_new_emms(cpu_flags, ret, ...) 189 /* Call the function */ 190 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__) 191 #endif 192 193 #ifndef declare_new_emms 194 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__) 195 #endif 196 #ifndef declare_new_float 197 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__) 198 #endif 199 200 typedef struct CheckasmPerf { 201 int sysfd; 202 uint64_t cycles; 203 int iterations; 204 } CheckasmPerf; 205 206 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF 207 208 #if CONFIG_LINUX_PERF 209 #define PERF_START(t) do { \ 210 ioctl(sysfd, PERF_EVENT_IOC_RESET, 0); \ 211 ioctl(sysfd, PERF_EVENT_IOC_ENABLE, 0); \ 212 } while (0) 213 #define PERF_STOP(t) do { \ 214 ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \ 215 read(sysfd, &t, sizeof(t)); \ 216 } while (0) 217 #else 218 #define PERF_START(t) t = AV_READ_TIME() 219 #define PERF_STOP(t) t = AV_READ_TIME() - t 220 #endif 221 222 /* Benchmark the function */ 223 #define bench_new(...)\ 224 do {\ 225 if (checkasm_bench_func()) {\ 226 struct CheckasmPerf *perf = checkasm_get_perf_context();\ 227 av_unused const int sysfd = perf->sysfd;\ 228 func_type *tfunc = func_new;\ 229 uint64_t tsum = 0;\ 230 int ti, tcount = 0;\ 231 uint64_t t = 0; \ 232 for (ti = 0; ti < BENCH_RUNS; ti++) {\ 233 PERF_START(t);\ 234 tfunc(__VA_ARGS__);\ 235 tfunc(__VA_ARGS__);\ 236 tfunc(__VA_ARGS__);\ 237 tfunc(__VA_ARGS__);\ 238 PERF_STOP(t);\ 239 if (t*tcount <= tsum*4 && ti > 0) {\ 240 tsum += t;\ 241 tcount++;\ 242 }\ 243 }\ 244 emms_c();\ 245 perf->cycles += t;\ 246 perf->iterations++;\ 247 }\ 248 } while (0) 249 #else 250 #define bench_new(...) while(0) 251 #define PERF_START(t) while(0) 252 #define PERF_STOP(t) while(0) 253 #endif 254 255 #endif /* TESTS_CHECKASM_CHECKASM_H */ 256