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_afir(void); 44 void checkasm_check_alacdsp(void); 45 void checkasm_check_audiodsp(void); 46 void checkasm_check_blend(void); 47 void checkasm_check_blockdsp(void); 48 void checkasm_check_bswapdsp(void); 49 void checkasm_check_colorspace(void); 50 void checkasm_check_exrdsp(void); 51 void checkasm_check_fixed_dsp(void); 52 void checkasm_check_flacdsp(void); 53 void checkasm_check_float_dsp(void); 54 void checkasm_check_fmtconvert(void); 55 void checkasm_check_g722dsp(void); 56 void checkasm_check_h264dsp(void); 57 void checkasm_check_h264pred(void); 58 void checkasm_check_h264qpel(void); 59 void checkasm_check_hevc_add_res(void); 60 void checkasm_check_hevc_idct(void); 61 void checkasm_check_hevc_sao(void); 62 void checkasm_check_huffyuvdsp(void); 63 void checkasm_check_jpeg2000dsp(void); 64 void checkasm_check_llviddsp(void); 65 void checkasm_check_llviddspenc(void); 66 void checkasm_check_nlmeans(void); 67 void checkasm_check_opusdsp(void); 68 void checkasm_check_pixblockdsp(void); 69 void checkasm_check_sbrdsp(void); 70 void checkasm_check_synth_filter(void); 71 void checkasm_check_sw_rgb(void); 72 void checkasm_check_sw_scale(void); 73 void checkasm_check_utvideodsp(void); 74 void checkasm_check_v210dec(void); 75 void checkasm_check_v210enc(void); 76 void checkasm_check_vf_eq(void); 77 void checkasm_check_vf_gblur(void); 78 void checkasm_check_vf_hflip(void); 79 void checkasm_check_vf_threshold(void); 80 void checkasm_check_vp8dsp(void); 81 void checkasm_check_vp9dsp(void); 82 void checkasm_check_videodsp(void); 83 84 struct CheckasmPerf; 85 86 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3); 87 int checkasm_bench_func(void); 88 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2); 89 struct CheckasmPerf *checkasm_get_perf_context(void); 90 void checkasm_report(const char *name, ...) av_printf_format(1, 2); 91 92 /* float compare utilities */ 93 int float_near_ulp(float a, float b, unsigned max_ulp); 94 int float_near_abs_eps(float a, float b, float eps); 95 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp); 96 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, 97 unsigned len); 98 int float_near_abs_eps_array(const float *a, const float *b, float eps, 99 unsigned len); 100 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, 101 unsigned max_ulp, unsigned len); 102 int double_near_abs_eps(double a, double b, double eps); 103 int double_near_abs_eps_array(const double *a, const double *b, double eps, 104 unsigned len); 105 106 extern AVLFG checkasm_lfg; 107 #define rnd() av_lfg_get(&checkasm_lfg) 108 109 static av_unused void *func_ref, *func_new; 110 111 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */ 112 113 /* Decide whether or not the specified function needs to be tested */ 114 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__)) 115 116 /* Declare the function prototype. The first argument is the return value, the remaining 117 * arguments are the function parameters. Naming parameters is optional. */ 118 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__) 119 #define declare_func_float(ret, ...) declare_new_float(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__) 120 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__) 121 122 /* Indicate that the current test has failed */ 123 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__) 124 125 /* Print the test outcome */ 126 #define report checkasm_report 127 128 /* Call the reference function */ 129 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__) 130 131 #if ARCH_X86 && HAVE_X86ASM 132 /* Verifies that clobbered callee-saved registers are properly saved and restored 133 * and that either no MMX registers are touched or emms is issued */ 134 void checkasm_checked_call(void *func, ...); 135 /* Verifies that clobbered callee-saved registers are properly saved and restored 136 * and issues emms for asm functions which are not required to do so */ 137 void checkasm_checked_call_emms(void *func, ...); 138 /* Verifies that clobbered callee-saved registers are properly saved and restored 139 * but doesn't issue emms. Meant for dsp functions returning float or double */ 140 void checkasm_checked_call_float(void *func, ...); 141 142 #if ARCH_X86_64 143 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit. 144 * This is done by clobbering the stack with junk around the stack pointer and calling the 145 * assembly function through checked_call() with added dummy arguments which forces all 146 * real arguments to be passed on the stack and not in registers. For 32-bit arguments the 147 * upper half of the 64-bit register locations on the stack will now contain junk which will 148 * cause misbehaving functions to either produce incorrect output or segfault. Note that 149 * even though this works extremely well in practice, it's technically not guaranteed 150 * and false negatives is theoretically possible, but there can never be any false positives. 151 */ 152 void checkasm_stack_clobber(uint64_t clobber, ...); 153 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\ 154 = (void *)checkasm_checked_call; 155 #define declare_new_float(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\ 156 = (void *)checkasm_checked_call_float; 157 #define declare_new_emms(cpu_flags, ret, ...) \ 158 ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \ 159 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \ 160 (void *)checkasm_checked_call; 161 #define CLOB (UINT64_C(0xdeadbeefdeadbeef)) 162 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\ 163 CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\ 164 checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__)) 165 #elif ARCH_X86_32 166 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call; 167 #define declare_new_float(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call_float; 168 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \ 169 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \ 170 (void *)checkasm_checked_call; 171 #define call_new(...) checked_call(func_new, __VA_ARGS__) 172 #endif 173 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL 174 /* Use a dummy argument, to offset the real parameters by 2, not only 1. 175 * This makes sure that potential 8-byte-alignment of parameters is kept the same 176 * even when the extra parameters have been removed. */ 177 void checkasm_checked_call_vfp(void *func, int dummy, ...); 178 void checkasm_checked_call_novfp(void *func, int dummy, ...); 179 extern void (*checkasm_checked_call)(void *func, int dummy, ...); 180 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__, \ 181 int, int, int, int, int, int, int, int, \ 182 int, int, int, int, int, int, int) = (void *)checkasm_checked_call; 183 #define call_new(...) checked_call(func_new, 0, __VA_ARGS__, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0, 0, 0, 0) 184 #elif ARCH_AARCH64 && !defined(__APPLE__) 185 void checkasm_stack_clobber(uint64_t clobber, ...); 186 void checkasm_checked_call(void *func, ...); 187 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, int, int, __VA_ARGS__,\ 188 int, int, int, int, int, int, int, int,\ 189 int, int, int, int, int, int, int)\ 190 = (void *)checkasm_checked_call; 191 #define CLOB (UINT64_C(0xdeadbeefdeadbeef)) 192 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\ 193 CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\ 194 checked_call(func_new, 0, 0, 0, 0, 0, 0, 0, __VA_ARGS__,\ 195 7, 6, 5, 4, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0)) 196 #else 197 #define declare_new(ret, ...) 198 #define declare_new_float(ret, ...) 199 #define declare_new_emms(cpu_flags, ret, ...) 200 /* Call the function */ 201 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__) 202 #endif 203 204 #ifndef declare_new_emms 205 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__) 206 #endif 207 #ifndef declare_new_float 208 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__) 209 #endif 210 211 typedef struct CheckasmPerf { 212 int sysfd; 213 uint64_t cycles; 214 int iterations; 215 } CheckasmPerf; 216 217 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF 218 219 #if CONFIG_LINUX_PERF 220 #define PERF_START(t) do { \ 221 ioctl(sysfd, PERF_EVENT_IOC_RESET, 0); \ 222 ioctl(sysfd, PERF_EVENT_IOC_ENABLE, 0); \ 223 } while (0) 224 #define PERF_STOP(t) do { \ 225 ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \ 226 read(sysfd, &t, sizeof(t)); \ 227 } while (0) 228 #else 229 #define PERF_START(t) t = AV_READ_TIME() 230 #define PERF_STOP(t) t = AV_READ_TIME() - t 231 #endif 232 233 /* Benchmark the function */ 234 #define bench_new(...)\ 235 do {\ 236 if (checkasm_bench_func()) {\ 237 struct CheckasmPerf *perf = checkasm_get_perf_context();\ 238 av_unused const int sysfd = perf->sysfd;\ 239 func_type *tfunc = func_new;\ 240 uint64_t tsum = 0;\ 241 int ti, tcount = 0;\ 242 uint64_t t = 0; \ 243 for (ti = 0; ti < BENCH_RUNS; ti++) {\ 244 PERF_START(t);\ 245 tfunc(__VA_ARGS__);\ 246 tfunc(__VA_ARGS__);\ 247 tfunc(__VA_ARGS__);\ 248 tfunc(__VA_ARGS__);\ 249 PERF_STOP(t);\ 250 if (t*tcount <= tsum*4 && ti > 0) {\ 251 tsum += t;\ 252 tcount++;\ 253 }\ 254 }\ 255 emms_c();\ 256 perf->cycles += t;\ 257 perf->iterations++;\ 258 }\ 259 } while (0) 260 #else 261 #define bench_new(...) while(0) 262 #define PERF_START(t) while(0) 263 #define PERF_STOP(t) while(0) 264 #endif 265 266 #define DECL_CHECKASM_CHECK_FUNC(type) \ 267 int checkasm_check_##type(const char *const file, const int line, \ 268 const type *const buf1, const ptrdiff_t stride1, \ 269 const type *const buf2, const ptrdiff_t stride2, \ 270 const int w, const int h, const char *const name) 271 272 DECL_CHECKASM_CHECK_FUNC(uint8_t); 273 DECL_CHECKASM_CHECK_FUNC(uint16_t); 274 DECL_CHECKASM_CHECK_FUNC(int16_t); 275 DECL_CHECKASM_CHECK_FUNC(int32_t); 276 277 #define PASTE(a,b) a ## b 278 #define CONCAT(a,b) PASTE(a,b) 279 280 #define checkasm_check(prefix, ...) CONCAT(checkasm_check_, prefix)(__FILE__, __LINE__, __VA_ARGS__) 281 282 #endif /* TESTS_CHECKASM_CHECKASM_H */ 283