FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
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);
61 void checkasm_check_llviddsp(void);
63 void checkasm_check_sbrdsp(void);
65 void checkasm_check_v210enc(void);
66 void checkasm_check_vp8dsp(void);
67 void checkasm_check_vp9dsp(void);
68 void checkasm_check_videodsp(void);
69 
70 struct CheckasmPerf;
71 
72 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
73 int checkasm_bench_func(void);
74 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2);
76 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
77 
78 /* float compare utilities */
79 int float_near_ulp(float a, float b, unsigned max_ulp);
80 int float_near_abs_eps(float a, float b, float eps);
81 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp);
82 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
83  unsigned len);
84 int float_near_abs_eps_array(const float *a, const float *b, float eps,
85  unsigned len);
86 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
87  unsigned max_ulp, unsigned len);
88 int double_near_abs_eps(double a, double b, double eps);
89 int double_near_abs_eps_array(const double *a, const double *b, double eps,
90  unsigned len);
91 
92 extern AVLFG checkasm_lfg;
93 #define rnd() av_lfg_get(&checkasm_lfg)
94 
95 static av_unused void *func_ref, *func_new;
96 
97 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
98 
99 /* Decide whether or not the specified function needs to be tested */
100 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__))
101 
102 /* Declare the function prototype. The first argument is the return value, the remaining
103  * arguments are the function parameters. Naming parameters is optional. */
104 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
105 #define declare_func_float(ret, ...) declare_new_float(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
106 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
107 
108 /* Indicate that the current test has failed */
109 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
110 
111 /* Print the test outcome */
112 #define report checkasm_report
113 
114 /* Call the reference function */
115 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
116 
117 #if ARCH_X86 && HAVE_X86ASM
118 /* Verifies that clobbered callee-saved registers are properly saved and restored
119  * and that either no MMX registers are touched or emms is issued */
120 void checkasm_checked_call(void *func, ...);
121 /* Verifies that clobbered callee-saved registers are properly saved and restored
122  * and issues emms for asm functions which are not required to do so */
123 void checkasm_checked_call_emms(void *func, ...);
124 /* Verifies that clobbered callee-saved registers are properly saved and restored
125  * but doesn't issue emms. Meant for dsp functions returning float or double */
126 void checkasm_checked_call_float(void *func, ...);
127 
128 #if ARCH_X86_64
129 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
130  * This is done by clobbering the stack with junk around the stack pointer and calling the
131  * assembly function through checked_call() with added dummy arguments which forces all
132  * real arguments to be passed on the stack and not in registers. For 32-bit arguments the
133  * upper half of the 64-bit register locations on the stack will now contain junk which will
134  * cause misbehaving functions to either produce incorrect output or segfault. Note that
135  * even though this works extremely well in practice, it's technically not guaranteed
136  * and false negatives is theoretically possible, but there can never be any false positives.
137  */
138 void checkasm_stack_clobber(uint64_t clobber, ...);
139 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
140  = (void *)checkasm_checked_call;
141 #define declare_new_float(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
142  = (void *)checkasm_checked_call_float;
143 #define declare_new_emms(cpu_flags, ret, ...) \
144  ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \
145  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
146  (void *)checkasm_checked_call;
147 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
148 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
149  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
150  checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__))
151 #elif ARCH_X86_32
152 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
153 #define declare_new_float(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call_float;
154 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \
155  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
156  (void *)checkasm_checked_call;
157 #define call_new(...) checked_call(func_new, __VA_ARGS__)
158 #endif
159 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
160 /* Use a dummy argument, to offset the real parameters by 2, not only 1.
161  * This makes sure that potential 8-byte-alignment of parameters is kept the same
162  * even when the extra parameters have been removed. */
163 void checkasm_checked_call_vfp(void *func, int dummy, ...);
164 void checkasm_checked_call_novfp(void *func, int dummy, ...);
165 extern void (*checkasm_checked_call)(void *func, int dummy, ...);
166 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__) = (void *)checkasm_checked_call;
167 #define call_new(...) checked_call(func_new, 0, __VA_ARGS__)
168 #elif ARCH_AARCH64 && !defined(__APPLE__)
169 void checkasm_stack_clobber(uint64_t clobber, ...);
170 void checkasm_checked_call(void *func, ...);
171 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, int, int, __VA_ARGS__)\
172  = (void *)checkasm_checked_call;
173 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
174 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
175  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
176  checked_call(func_new, 0, 0, 0, 0, 0, 0, 0, __VA_ARGS__))
177 #else
178 #define declare_new(ret, ...)
179 #define declare_new_float(ret, ...)
180 #define declare_new_emms(cpu_flags, ret, ...)
181 /* Call the function */
182 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__)
183 #endif
184 
185 #ifndef declare_new_emms
186 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
187 #endif
188 #ifndef declare_new_float
189 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__)
190 #endif
191 
192 typedef struct CheckasmPerf {
193  int sysfd;
194  uint64_t cycles;
196 } CheckasmPerf;
197 
198 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF
199 
200 #if CONFIG_LINUX_PERF
201 #define PERF_START(t) do { \
202  ioctl(sysfd, PERF_EVENT_IOC_RESET, 0); \
203  ioctl(sysfd, PERF_EVENT_IOC_ENABLE, 0); \
204 } while (0)
205 #define PERF_STOP(t) do { \
206  ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \
207  read(sysfd, &t, sizeof(t)); \
208 } while (0)
209 #else
210 #define PERF_START(t) t = AV_READ_TIME()
211 #define PERF_STOP(t) t = AV_READ_TIME() - t
212 #endif
213 
214 /* Benchmark the function */
215 #define bench_new(...)\
216  do {\
217  if (checkasm_bench_func()) {\
218  struct CheckasmPerf *perf = checkasm_get_perf_context();\
219  av_unused const int sysfd = perf->sysfd;\
220  func_type *tfunc = func_new;\
221  uint64_t tsum = 0;\
222  int ti, tcount = 0;\
223  uint64_t t = 0; \
224  for (ti = 0; ti < BENCH_RUNS; ti++) {\
225  PERF_START(t);\
226  tfunc(__VA_ARGS__);\
227  tfunc(__VA_ARGS__);\
228  tfunc(__VA_ARGS__);\
229  tfunc(__VA_ARGS__);\
230  PERF_STOP(t);\
231  if (t*tcount <= tsum*4 && ti > 0) {\
232  tsum += t;\
233  tcount++;\
234  }\
235  }\
236  emms_c();\
237  perf->cycles += t;\
238  perf->iterations++;\
239  }\
240  } while (0)
241 #else
242 #define bench_new(...) while(0)
243 #define PERF_START(t) while(0)
244 #define PERF_STOP(t) while(0)
245 #endif
246 
247 #endif /* TESTS_CHECKASM_CHECKASM_H */
Definition: lfg.h:27
void int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:244
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:293
void checkasm_check_v210enc(void)
Definition: v210enc.c:81
void struct CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:750
uint64_t cycles
Definition: checkasm.h:194
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:625
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:59
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:58
void int checkasm_bench_func(void)
Definition: checkasm.c:725
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:229
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:236
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:317
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:51
typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)
void checkasm_check_exrdsp(void)
Definition: exrdsp.c:76
static av_unused void * func_ref
Definition: checkasm.h:95
int dummy
Definition: motion.c:64
int iterations
Definition: checkasm.h:195
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:86
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:310
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:44
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:215
int(* func)(AVBPrint *dst, const char *in, const char *arg)
Definition: jacosubdec.c:67
void checkasm_report(const char *name,...) av_printf_format(1
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:66
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:115
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:254
static av_unused void * func_new
Definition: checkasm.h:95
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:274
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:262
int len
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:81
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:44
void checkasm_check_g722dsp(void)
Definition: g722dsp.c:53
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
Definition: checkasm.c:298
void checkasm_check_blend(void)
Definition: vf_blend.c:87
#define av_unused
Definition: attributes.h:125
const char * name
Definition: opengl_enc.c:103
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:281