FFmpeg  4.0
checkasm.h
Go to the documentation of this file.
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);
59 void checkasm_check_hevc_idct(void);
60 void checkasm_check_hevc_sao(void);
61 void checkasm_check_huffyuvdsp(void);
63 void checkasm_check_llviddsp(void);
66 void checkasm_check_sbrdsp(void);
68 void checkasm_check_sw_rgb(void);
69 void checkasm_check_utvideodsp(void);
70 void checkasm_check_v210enc(void);
71 void checkasm_check_vf_hflip(void);
73 void checkasm_check_vp8dsp(void);
74 void checkasm_check_vp9dsp(void);
75 void checkasm_check_videodsp(void);
76 
77 struct CheckasmPerf;
78 
79 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
80 int checkasm_bench_func(void);
81 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2);
83 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
84 
85 /* float compare utilities */
86 int float_near_ulp(float a, float b, unsigned max_ulp);
87 int float_near_abs_eps(float a, float b, float eps);
88 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp);
89 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
90  unsigned len);
91 int float_near_abs_eps_array(const float *a, const float *b, float eps,
92  unsigned len);
93 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
94  unsigned max_ulp, unsigned len);
95 int double_near_abs_eps(double a, double b, double eps);
96 int double_near_abs_eps_array(const double *a, const double *b, double eps,
97  unsigned len);
98 
99 extern AVLFG checkasm_lfg;
100 #define rnd() av_lfg_get(&checkasm_lfg)
101 
102 static av_unused void *func_ref, *func_new;
103 
104 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
105 
106 /* Decide whether or not the specified function needs to be tested */
107 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__))
108 
109 /* Declare the function prototype. The first argument is the return value, the remaining
110  * arguments are the function parameters. Naming parameters is optional. */
111 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
112 #define declare_func_float(ret, ...) declare_new_float(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
113 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
114 
115 /* Indicate that the current test has failed */
116 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
117 
118 /* Print the test outcome */
119 #define report checkasm_report
120 
121 /* Call the reference function */
122 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
123 
124 #if ARCH_X86 && HAVE_X86ASM
125 /* Verifies that clobbered callee-saved registers are properly saved and restored
126  * and that either no MMX registers are touched or emms is issued */
127 void checkasm_checked_call(void *func, ...);
128 /* Verifies that clobbered callee-saved registers are properly saved and restored
129  * and issues emms for asm functions which are not required to do so */
130 void checkasm_checked_call_emms(void *func, ...);
131 /* Verifies that clobbered callee-saved registers are properly saved and restored
132  * but doesn't issue emms. Meant for dsp functions returning float or double */
133 void checkasm_checked_call_float(void *func, ...);
134 
135 #if ARCH_X86_64
136 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
137  * This is done by clobbering the stack with junk around the stack pointer and calling the
138  * assembly function through checked_call() with added dummy arguments which forces all
139  * real arguments to be passed on the stack and not in registers. For 32-bit arguments the
140  * upper half of the 64-bit register locations on the stack will now contain junk which will
141  * cause misbehaving functions to either produce incorrect output or segfault. Note that
142  * even though this works extremely well in practice, it's technically not guaranteed
143  * and false negatives is theoretically possible, but there can never be any false positives.
144  */
145 void checkasm_stack_clobber(uint64_t clobber, ...);
146 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
147  = (void *)checkasm_checked_call;
148 #define declare_new_float(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
149  = (void *)checkasm_checked_call_float;
150 #define declare_new_emms(cpu_flags, ret, ...) \
151  ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \
152  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
153  (void *)checkasm_checked_call;
154 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
155 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
156  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
157  checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__))
158 #elif ARCH_X86_32
159 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
160 #define declare_new_float(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call_float;
161 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \
162  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
163  (void *)checkasm_checked_call;
164 #define call_new(...) checked_call(func_new, __VA_ARGS__)
165 #endif
166 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
167 /* Use a dummy argument, to offset the real parameters by 2, not only 1.
168  * This makes sure that potential 8-byte-alignment of parameters is kept the same
169  * even when the extra parameters have been removed. */
170 void checkasm_checked_call_vfp(void *func, int dummy, ...);
171 void checkasm_checked_call_novfp(void *func, int dummy, ...);
172 extern void (*checkasm_checked_call)(void *func, int dummy, ...);
173 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__) = (void *)checkasm_checked_call;
174 #define call_new(...) checked_call(func_new, 0, __VA_ARGS__)
175 #elif ARCH_AARCH64 && !defined(__APPLE__)
176 void checkasm_stack_clobber(uint64_t clobber, ...);
177 void checkasm_checked_call(void *func, ...);
178 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, int, int, __VA_ARGS__)\
179  = (void *)checkasm_checked_call;
180 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
181 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
182  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
183  checked_call(func_new, 0, 0, 0, 0, 0, 0, 0, __VA_ARGS__))
184 #else
185 #define declare_new(ret, ...)
186 #define declare_new_float(ret, ...)
187 #define declare_new_emms(cpu_flags, ret, ...)
188 /* Call the function */
189 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__)
190 #endif
191 
192 #ifndef declare_new_emms
193 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
194 #endif
195 #ifndef declare_new_float
196 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__)
197 #endif
198 
199 typedef struct CheckasmPerf {
200  int sysfd;
201  uint64_t cycles;
203 } CheckasmPerf;
204 
205 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF
206 
207 #if CONFIG_LINUX_PERF
208 #define PERF_START(t) do { \
209  ioctl(sysfd, PERF_EVENT_IOC_RESET, 0); \
210  ioctl(sysfd, PERF_EVENT_IOC_ENABLE, 0); \
211 } while (0)
212 #define PERF_STOP(t) do { \
213  ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \
214  read(sysfd, &t, sizeof(t)); \
215 } while (0)
216 #else
217 #define PERF_START(t) t = AV_READ_TIME()
218 #define PERF_STOP(t) t = AV_READ_TIME() - t
219 #endif
220 
221 /* Benchmark the function */
222 #define bench_new(...)\
223  do {\
224  if (checkasm_bench_func()) {\
225  struct CheckasmPerf *perf = checkasm_get_perf_context();\
226  av_unused const int sysfd = perf->sysfd;\
227  func_type *tfunc = func_new;\
228  uint64_t tsum = 0;\
229  int ti, tcount = 0;\
230  uint64_t t = 0; \
231  for (ti = 0; ti < BENCH_RUNS; ti++) {\
232  PERF_START(t);\
233  tfunc(__VA_ARGS__);\
234  tfunc(__VA_ARGS__);\
235  tfunc(__VA_ARGS__);\
236  tfunc(__VA_ARGS__);\
237  PERF_STOP(t);\
238  if (t*tcount <= tsum*4 && ti > 0) {\
239  tsum += t;\
240  tcount++;\
241  }\
242  }\
243  emms_c();\
244  perf->cycles += t;\
245  perf->iterations++;\
246  }\
247  } while (0)
248 #else
249 #define bench_new(...) while(0)
250 #define PERF_START(t) while(0)
251 #define PERF_STOP(t) while(0)
252 #endif
253 
254 #endif /* TESTS_CHECKASM_CHECKASM_H */
const char * name
Definition: avisynth_c.h:775
Definition: lfg.h:27
void int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:264
void * checkasm_check_func(void *func, const char *name,...) av_printf_format(2
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
Definition: checkasm.c:317
void checkasm_check_v210enc(void)
Definition: v210enc.c:81
void struct CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:774
uint64_t cycles
Definition: checkasm.h:201
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:625
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:59
void checkasm_check_vf_threshold(void)
Definition: vf_threshold.c:78
const char * b
Definition: vf_curves.c:113
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:56
void checkasm_check_h264pred(void)
Definition: h264pred.c:232
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:54
void checkasm_check_colorspace(void)
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:507
void checkasm_check_sbrdsp(void)
Definition: sbrdsp.c:250
void checkasm_check_h264qpel(void)
Definition: h264qpel.c:50
high precision timer, useful to profile code
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:91
void checkasm_check_llviddspenc(void)
Definition: llviddspenc.c:104
void int checkasm_bench_func(void)
Definition: checkasm.c:749
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:315
void checkasm_fail_func(const char *msg,...) av_printf_format(1
void checkasm_check_hevc_add_res(void)
Definition: hevc_add_res.c:74
#define av_printf_format(fmtpos, attrpos)
Definition: attributes.h:155
void checkasm_check_videodsp(void)
Definition: videodsp.c:80
common internal API header
void checkasm_check_fixed_dsp(void)
Definition: fixed_dsp.c:132
AVLFG checkasm_lfg
Definition: checkasm.c:256
void checkasm_check_utvideodsp(void)
Definition: utvideodsp.c:90
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:341
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:51
void checkasm_check_sw_rgb(void)
Definition: sw_rgb.c:67
void checkasm_check_exrdsp(void)
Definition: exrdsp.c:76
static av_unused void * func_ref
Definition: checkasm.h:102
int dummy
Definition: motion.c:64
int iterations
Definition: checkasm.h:202
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:86
typedef void(RENAME(mix_any_func_type))
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:334
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:44
int(* func)(AVBPrint *dst, const char *in, const char *arg)
Definition: jacosubdec.c:67
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:215
void checkasm_report(const char *name,...) av_printf_format(1
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:195
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:115
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:255
void checkasm_check_huffyuvdsp(void)
Definition: huffyuvdsp.c:67
static av_unused void * func_new
Definition: checkasm.h:102
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:294
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:282
int len
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:81
void checkasm_check_hevc_sao(void)
Definition: hevc_sao.c:131
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:44
void checkasm_check_g722dsp(void)
Definition: g722dsp.c:53
void checkasm_check_vf_hflip(void)
Definition: vf_hflip.c:69
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
Definition: checkasm.c:322
void checkasm_check_blend(void)
Definition: vf_blend.c:88
#define av_unused
Definition: attributes.h:125
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:305