FFmpeg
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_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);
60 void checkasm_check_hevc_idct(void);
61 void checkasm_check_hevc_qpel(void);
66 void checkasm_check_hevc_epel(void);
71 void checkasm_check_hevc_sao(void);
72 void checkasm_check_huffyuvdsp(void);
74 void checkasm_check_llviddsp(void);
76 void checkasm_check_nlmeans(void);
77 void checkasm_check_opusdsp(void);
79 void checkasm_check_sbrdsp(void);
81 void checkasm_check_sw_rgb(void);
82 void checkasm_check_sw_scale(void);
83 void checkasm_check_utvideodsp(void);
84 void checkasm_check_v210dec(void);
85 void checkasm_check_v210enc(void);
86 void checkasm_check_vf_eq(void);
87 void checkasm_check_vf_gblur(void);
88 void checkasm_check_vf_hflip(void);
90 void checkasm_check_vp8dsp(void);
91 void checkasm_check_vp9dsp(void);
92 void checkasm_check_videodsp(void);
93 
94 struct CheckasmPerf;
95 
96 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
97 int checkasm_bench_func(void);
98 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2);
100 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
101 
102 /* float compare utilities */
103 int float_near_ulp(float a, float b, unsigned max_ulp);
104 int float_near_abs_eps(float a, float b, float eps);
105 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp);
106 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
107  unsigned len);
108 int float_near_abs_eps_array(const float *a, const float *b, float eps,
109  unsigned len);
110 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
111  unsigned max_ulp, unsigned len);
112 int double_near_abs_eps(double a, double b, double eps);
113 int double_near_abs_eps_array(const double *a, const double *b, double eps,
114  unsigned len);
115 
116 extern AVLFG checkasm_lfg;
117 #define rnd() av_lfg_get(&checkasm_lfg)
118 
119 static av_unused void *func_ref, *func_new;
120 
121 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
122 
123 /* Decide whether or not the specified function needs to be tested */
124 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__))
125 
126 /* Declare the function prototype. The first argument is the return value, the remaining
127  * arguments are the function parameters. Naming parameters is optional. */
128 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
129 #define declare_func_float(ret, ...) declare_new_float(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
130 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
131 
132 /* Indicate that the current test has failed */
133 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
134 
135 /* Print the test outcome */
136 #define report checkasm_report
137 
138 /* Call the reference function */
139 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
140 
141 #if ARCH_X86 && HAVE_X86ASM
142 /* Verifies that clobbered callee-saved registers are properly saved and restored
143  * and that either no MMX registers are touched or emms is issued */
144 void checkasm_checked_call(void *func, ...);
145 /* Verifies that clobbered callee-saved registers are properly saved and restored
146  * and issues emms for asm functions which are not required to do so */
147 void checkasm_checked_call_emms(void *func, ...);
148 /* Verifies that clobbered callee-saved registers are properly saved and restored
149  * but doesn't issue emms. Meant for dsp functions returning float or double */
150 void checkasm_checked_call_float(void *func, ...);
151 
152 #if ARCH_X86_64
153 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
154  * This is done by clobbering the stack with junk around the stack pointer and calling the
155  * assembly function through checked_call() with added dummy arguments which forces all
156  * real arguments to be passed on the stack and not in registers. For 32-bit arguments the
157  * upper half of the 64-bit register locations on the stack will now contain junk which will
158  * cause misbehaving functions to either produce incorrect output or segfault. Note that
159  * even though this works extremely well in practice, it's technically not guaranteed
160  * and false negatives is theoretically possible, but there can never be any false positives.
161  */
162 void checkasm_stack_clobber(uint64_t clobber, ...);
163 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
164  = (void *)checkasm_checked_call;
165 #define declare_new_float(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
166  = (void *)checkasm_checked_call_float;
167 #define declare_new_emms(cpu_flags, ret, ...) \
168  ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \
169  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
170  (void *)checkasm_checked_call;
171 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
172 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
173  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
174  checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__))
175 #elif ARCH_X86_32
176 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
177 #define declare_new_float(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call_float;
178 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \
179  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
180  (void *)checkasm_checked_call;
181 #define call_new(...) checked_call(func_new, __VA_ARGS__)
182 #endif
183 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
184 /* Use a dummy argument, to offset the real parameters by 2, not only 1.
185  * This makes sure that potential 8-byte-alignment of parameters is kept the same
186  * even when the extra parameters have been removed. */
187 void checkasm_checked_call_vfp(void *func, int dummy, ...);
188 void checkasm_checked_call_novfp(void *func, int dummy, ...);
189 extern void (*checkasm_checked_call)(void *func, int dummy, ...);
190 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__, \
191  int, int, int, int, int, int, int, int, \
192  int, int, int, int, int, int, int) = (void *)checkasm_checked_call;
193 #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)
194 #elif ARCH_AARCH64 && !defined(__APPLE__)
195 void checkasm_stack_clobber(uint64_t clobber, ...);
196 void checkasm_checked_call(void *func, ...);
197 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, int, int, __VA_ARGS__,\
198  int, int, int, int, int, int, int, int,\
199  int, int, int, int, int, int, int)\
200  = (void *)checkasm_checked_call;
201 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
202 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
203  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
204  checked_call(func_new, 0, 0, 0, 0, 0, 0, 0, __VA_ARGS__,\
205  7, 6, 5, 4, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0))
206 #else
207 #define declare_new(ret, ...)
208 #define declare_new_float(ret, ...)
209 #define declare_new_emms(cpu_flags, ret, ...)
210 /* Call the function */
211 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__)
212 #endif
213 
214 #ifndef declare_new_emms
215 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
216 #endif
217 #ifndef declare_new_float
218 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__)
219 #endif
220 
221 typedef struct CheckasmPerf {
222  int sysfd;
223  uint64_t cycles;
225 } CheckasmPerf;
226 
227 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF
228 
229 #if CONFIG_LINUX_PERF
230 #define PERF_START(t) do { \
231  ioctl(sysfd, PERF_EVENT_IOC_RESET, 0); \
232  ioctl(sysfd, PERF_EVENT_IOC_ENABLE, 0); \
233 } while (0)
234 #define PERF_STOP(t) do { \
235  ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \
236  read(sysfd, &t, sizeof(t)); \
237 } while (0)
238 #else
239 #define PERF_START(t) t = AV_READ_TIME()
240 #define PERF_STOP(t) t = AV_READ_TIME() - t
241 #endif
242 
243 /* Benchmark the function */
244 #define bench_new(...)\
245  do {\
246  if (checkasm_bench_func()) {\
247  struct CheckasmPerf *perf = checkasm_get_perf_context();\
248  av_unused const int sysfd = perf->sysfd;\
249  func_type *tfunc = func_new;\
250  uint64_t tsum = 0;\
251  int ti, tcount = 0;\
252  uint64_t t = 0; \
253  for (ti = 0; ti < BENCH_RUNS; ti++) {\
254  PERF_START(t);\
255  tfunc(__VA_ARGS__);\
256  tfunc(__VA_ARGS__);\
257  tfunc(__VA_ARGS__);\
258  tfunc(__VA_ARGS__);\
259  PERF_STOP(t);\
260  if (t*tcount <= tsum*4 && ti > 0) {\
261  tsum += t;\
262  tcount++;\
263  }\
264  }\
265  emms_c();\
266  perf->cycles += t;\
267  perf->iterations++;\
268  }\
269  } while (0)
270 #else
271 #define bench_new(...) while(0)
272 #define PERF_START(t) while(0)
273 #define PERF_STOP(t) while(0)
274 #endif
275 
276 #define DECL_CHECKASM_CHECK_FUNC(type) \
277 int checkasm_check_##type(const char *const file, const int line, \
278  const type *const buf1, const ptrdiff_t stride1, \
279  const type *const buf2, const ptrdiff_t stride2, \
280  const int w, const int h, const char *const name)
281 
283 DECL_CHECKASM_CHECK_FUNC(uint16_t);
284 DECL_CHECKASM_CHECK_FUNC(int16_t);
286 
287 #define PASTE(a,b) a ## b
288 #define CONCAT(a,b) PASTE(a,b)
289 
290 #define checkasm_check(prefix, ...) CONCAT(checkasm_check_, prefix)(__FILE__, __LINE__, __VA_ARGS__)
291 
292 #endif /* TESTS_CHECKASM_CHECKASM_H */
void checkasm_check_hevc_qpel(void)
Definition: hevc_pel.c:68
Context structure for the Lagged Fibonacci PRNG.
Definition: lfg.h:33
void int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:297
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:350
void checkasm_check_v210enc(void)
Definition: v210enc.c:82
void struct CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:811
uint64_t cycles
Definition: checkasm.h:223
void checkasm_check_hevc_qpel_uni_w(void)
Definition: hevc_pel.c:153
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:626
void checkasm_check_hevc_qpel_bi_w(void)
Definition: hevc_pel.c:245
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:60
void checkasm_check_vf_threshold(void)
Definition: vf_threshold.c:79
void checkasm_check_hevc_epel_bi(void)
Definition: hevc_pel.c:430
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:57
The reader does not expect b to be semantically here and if the code is changed by maybe adding a a division or other the signedness will almost certainly be mistaken To avoid this confusion a new type was SUINT is the C unsigned type but it holds a signed int to use the same example SUINT a
Definition: undefined.txt:36
void checkasm_check_h264pred(void)
Definition: h264pred.c:233
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:55
void checkasm_check_hevc_qpel_uni(void)
Definition: hevc_pel.c:112
void checkasm_check_afir(void)
Definition: af_afir.c:80
void checkasm_check_colorspace(void)
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:508
void checkasm_check_hevc_epel(void)
Definition: hevc_pel.c:297
uint8_t
void checkasm_check_sbrdsp(void)
Definition: sbrdsp.c:254
void checkasm_check_hevc_epel_uni(void)
Definition: hevc_pel.c:341
void checkasm_check_h264qpel(void)
Definition: h264qpel.c:51
high precision timer, useful to profile code
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:92
void checkasm_check_llviddspenc(void)
Definition: llviddspenc.c:105
void int checkasm_bench_func(void)
Definition: checkasm.c:786
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:442
void checkasm_fail_func(const char *msg,...) av_printf_format(1
void checkasm_check_hevc_add_res(void)
Definition: hevc_add_res.c:86
void checkasm_check_hevc_qpel_bi(void)
Definition: hevc_pel.c:201
void checkasm_check_hevc_epel_bi_w(void)
Definition: hevc_pel.c:474
#define av_printf_format(fmtpos, attrpos)
Definition: attributes.h:161
void checkasm_check_videodsp(void)
Definition: videodsp.c:81
void checkasm_check_hevc_epel_uni_w(void)
Definition: hevc_pel.c:382
common internal API header
void checkasm_check_fixed_dsp(void)
Definition: fixed_dsp.c:133
AVLFG checkasm_lfg
Definition: checkasm.c:289
#define b
Definition: input.c:41
void checkasm_check_utvideodsp(void)
Definition: utvideodsp.c:91
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:374
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:52
void checkasm_check_nlmeans(void)
Definition: vf_nlmeans.c:31
typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)
void checkasm_check_sw_rgb(void)
Definition: sw_rgb.c:183
int32_t
void checkasm_check_exrdsp(void)
Definition: exrdsp.c:77
static av_unused void * func_ref
Definition: checkasm.h:119
int dummy
Definition: motion.c:64
int iterations
Definition: checkasm.h:224
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:87
void checkasm_check_vf_eq(void)
Definition: vf_eq.c:76
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:367
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:46
int(* func)(AVBPrint *dst, const char *in, const char *arg)
Definition: jacosubdec.c:67
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:234
void checkasm_report(const char *name,...) av_printf_format(1
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:195
#define DECL_CHECKASM_CHECK_FUNC(type)
Definition: checkasm.h:276
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:116
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:282
void checkasm_check_huffyuvdsp(void)
Definition: huffyuvdsp.c:67
void checkasm_check_vf_gblur(void)
Definition: vf_gblur.c:65
void checkasm_check_opusdsp(void)
Definition: opusdsp.c:85
static av_unused void * func_new
Definition: checkasm.h:119
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:327
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:315
void checkasm_check_v210dec(void)
Definition: v210dec.c:47
int len
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:82
void checkasm_check_hevc_sao(void)
Definition: hevc_sao.c:132
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:46
void checkasm_check_g722dsp(void)
Definition: g722dsp.c:53
void checkasm_check_sw_scale(void)
Definition: sw_scale.c:230
void checkasm_check_vf_hflip(void)
Definition: vf_hflip.c:71
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
Definition: checkasm.c:355
void checkasm_check_blend(void)
Definition: vf_blend.c:88
#define av_unused
Definition: attributes.h:131
const char * name
Definition: opengl_enc.c:102
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:338