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 #elif CONFIG_MACOS_KPERF
35 #include "libavutil/macos_kperf.h"
36 #endif
37 
38 #include "libavutil/avstring.h"
39 #include "libavutil/cpu.h"
40 #include "libavutil/internal.h"
41 #include "libavutil/lfg.h"
42 #include "libavutil/timer.h"
43 
44 void checkasm_check_aacpsdsp(void);
45 void checkasm_check_afir(void);
46 void checkasm_check_alacdsp(void);
47 void checkasm_check_audiodsp(void);
48 void checkasm_check_av_tx(void);
49 void checkasm_check_blend(void);
50 void checkasm_check_blockdsp(void);
51 void checkasm_check_bswapdsp(void);
52 void checkasm_check_colorspace(void);
53 void checkasm_check_exrdsp(void);
54 void checkasm_check_fixed_dsp(void);
55 void checkasm_check_flacdsp(void);
56 void checkasm_check_float_dsp(void);
57 void checkasm_check_fmtconvert(void);
58 void checkasm_check_g722dsp(void);
59 void checkasm_check_h264dsp(void);
60 void checkasm_check_h264pred(void);
61 void checkasm_check_h264qpel(void);
63 void checkasm_check_hevc_idct(void);
64 void checkasm_check_hevc_qpel(void);
69 void checkasm_check_hevc_epel(void);
74 void checkasm_check_hevc_sao(void);
75 void checkasm_check_huffyuvdsp(void);
77 void checkasm_check_llviddsp(void);
79 void checkasm_check_nlmeans(void);
80 void checkasm_check_opusdsp(void);
82 void checkasm_check_sbrdsp(void);
84 void checkasm_check_sw_rgb(void);
85 void checkasm_check_sw_scale(void);
86 void checkasm_check_utvideodsp(void);
87 void checkasm_check_v210dec(void);
88 void checkasm_check_v210enc(void);
89 void checkasm_check_vf_eq(void);
90 void checkasm_check_vf_gblur(void);
91 void checkasm_check_vf_hflip(void);
93 void checkasm_check_vp8dsp(void);
94 void checkasm_check_vp9dsp(void);
95 void checkasm_check_videodsp(void);
96 
97 struct CheckasmPerf;
98 
99 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
100 int checkasm_bench_func(void);
101 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2);
103 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
104 
105 /* float compare utilities */
106 int float_near_ulp(float a, float b, unsigned max_ulp);
107 int float_near_abs_eps(float a, float b, float eps);
108 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp);
109 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
110  unsigned len);
111 int float_near_abs_eps_array(const float *a, const float *b, float eps,
112  unsigned len);
113 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
114  unsigned max_ulp, unsigned len);
115 int double_near_abs_eps(double a, double b, double eps);
116 int double_near_abs_eps_array(const double *a, const double *b, double eps,
117  unsigned len);
118 
119 extern AVLFG checkasm_lfg;
120 #define rnd() av_lfg_get(&checkasm_lfg)
121 
122 static av_unused void *func_ref, *func_new;
123 
124 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
125 
126 /* Decide whether or not the specified function needs to be tested */
127 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__))
128 
129 /* Declare the function prototype. The first argument is the return value, the remaining
130  * arguments are the function parameters. Naming parameters is optional. */
131 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
132 #define declare_func_float(ret, ...) declare_new_float(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
133 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
134 
135 /* Indicate that the current test has failed */
136 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
137 
138 /* Print the test outcome */
139 #define report checkasm_report
140 
141 /* Call the reference function */
142 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
143 
144 #if ARCH_X86 && HAVE_X86ASM
145 /* Verifies that clobbered callee-saved registers are properly saved and restored
146  * and that either no MMX registers are touched or emms is issued */
147 void checkasm_checked_call(void *func, ...);
148 /* Verifies that clobbered callee-saved registers are properly saved and restored
149  * and issues emms for asm functions which are not required to do so */
150 void checkasm_checked_call_emms(void *func, ...);
151 /* Verifies that clobbered callee-saved registers are properly saved and restored
152  * but doesn't issue emms. Meant for dsp functions returning float or double */
153 void checkasm_checked_call_float(void *func, ...);
154 
155 #if ARCH_X86_64
156 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
157  * This is done by clobbering the stack with junk around the stack pointer and calling the
158  * assembly function through checked_call() with added dummy arguments which forces all
159  * real arguments to be passed on the stack and not in registers. For 32-bit arguments the
160  * upper half of the 64-bit register locations on the stack will now contain junk which will
161  * cause misbehaving functions to either produce incorrect output or segfault. Note that
162  * even though this works extremely well in practice, it's technically not guaranteed
163  * and false negatives is theoretically possible, but there can never be any false positives.
164  */
165 void checkasm_stack_clobber(uint64_t clobber, ...);
166 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
167  = (void *)checkasm_checked_call;
168 #define declare_new_float(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
169  = (void *)checkasm_checked_call_float;
170 #define declare_new_emms(cpu_flags, ret, ...) \
171  ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \
172  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
173  (void *)checkasm_checked_call;
174 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
175 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
176  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
177  checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__))
178 #elif ARCH_X86_32
179 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
180 #define declare_new_float(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call_float;
181 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \
182  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
183  (void *)checkasm_checked_call;
184 #define call_new(...) checked_call(func_new, __VA_ARGS__)
185 #endif
186 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
187 /* Use a dummy argument, to offset the real parameters by 2, not only 1.
188  * This makes sure that potential 8-byte-alignment of parameters is kept the same
189  * even when the extra parameters have been removed. */
190 void checkasm_checked_call_vfp(void *func, int dummy, ...);
191 void checkasm_checked_call_novfp(void *func, int dummy, ...);
192 extern void (*checkasm_checked_call)(void *func, int dummy, ...);
193 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__, \
194  int, int, int, int, int, int, int, int, \
195  int, int, int, int, int, int, int) = (void *)checkasm_checked_call;
196 #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)
197 #elif ARCH_AARCH64 && !defined(__APPLE__)
198 void checkasm_stack_clobber(uint64_t clobber, ...);
199 void checkasm_checked_call(void *func, ...);
200 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, int, int, __VA_ARGS__,\
201  int, int, int, int, int, int, int, int,\
202  int, int, int, int, int, int, int)\
203  = (void *)checkasm_checked_call;
204 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
205 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
206  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
207  checked_call(func_new, 0, 0, 0, 0, 0, 0, 0, __VA_ARGS__,\
208  7, 6, 5, 4, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0))
209 #else
210 #define declare_new(ret, ...)
211 #define declare_new_float(ret, ...)
212 #define declare_new_emms(cpu_flags, ret, ...)
213 /* Call the function */
214 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__)
215 #endif
216 
217 #ifndef declare_new_emms
218 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
219 #endif
220 #ifndef declare_new_float
221 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__)
222 #endif
223 
224 typedef struct CheckasmPerf {
225  int sysfd;
226  uint64_t cycles;
228 } CheckasmPerf;
229 
230 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF || CONFIG_MACOS_KPERF
231 
232 #if CONFIG_LINUX_PERF
233 #define PERF_START(t) do { \
234  ioctl(sysfd, PERF_EVENT_IOC_RESET, 0); \
235  ioctl(sysfd, PERF_EVENT_IOC_ENABLE, 0); \
236 } while (0)
237 #define PERF_STOP(t) do { \
238  ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \
239  read(sysfd, &t, sizeof(t)); \
240 } while (0)
241 #elif CONFIG_MACOS_KPERF
242 #define PERF_START(t) t = ff_kperf_cycles()
243 #define PERF_STOP(t) t = ff_kperf_cycles() - t
244 #else
245 #define PERF_START(t) t = AV_READ_TIME()
246 #define PERF_STOP(t) t = AV_READ_TIME() - t
247 #endif
248 
249 /* Benchmark the function */
250 #define bench_new(...)\
251  do {\
252  if (checkasm_bench_func()) {\
253  struct CheckasmPerf *perf = checkasm_get_perf_context();\
254  av_unused const int sysfd = perf->sysfd;\
255  func_type *tfunc = func_new;\
256  uint64_t tsum = 0;\
257  int ti, tcount = 0;\
258  uint64_t t = 0; \
259  for (ti = 0; ti < BENCH_RUNS; ti++) {\
260  PERF_START(t);\
261  tfunc(__VA_ARGS__);\
262  tfunc(__VA_ARGS__);\
263  tfunc(__VA_ARGS__);\
264  tfunc(__VA_ARGS__);\
265  PERF_STOP(t);\
266  if (t*tcount <= tsum*4 && ti > 0) {\
267  tsum += t;\
268  tcount++;\
269  }\
270  }\
271  emms_c();\
272  perf->cycles += t;\
273  perf->iterations++;\
274  }\
275  } while (0)
276 #else
277 #define bench_new(...) while(0)
278 #define PERF_START(t) while(0)
279 #define PERF_STOP(t) while(0)
280 #endif
281 
282 #define DECL_CHECKASM_CHECK_FUNC(type) \
283 int checkasm_check_##type(const char *const file, const int line, \
284  const type *const buf1, const ptrdiff_t stride1, \
285  const type *const buf2, const ptrdiff_t stride2, \
286  const int w, const int h, const char *const name)
287 
288 DECL_CHECKASM_CHECK_FUNC(uint8_t);
289 DECL_CHECKASM_CHECK_FUNC(uint16_t);
290 DECL_CHECKASM_CHECK_FUNC(int16_t);
292 
293 #define PASTE(a,b) a ## b
294 #define CONCAT(a,b) PASTE(a,b)
295 
296 #define checkasm_check(prefix, ...) CONCAT(checkasm_check_, prefix)(__FILE__, __LINE__, __VA_ARGS__)
297 
298 #endif /* TESTS_CHECKASM_CHECKASM_H */
checkasm_check_nlmeans
void checkasm_check_nlmeans(void)
Definition: vf_nlmeans.c:31
func
int(* func)(AVBPrint *dst, const char *in, const char *arg)
Definition: jacosubdec.c:68
checkasm_lfg
AVLFG checkasm_lfg
Definition: checkasm.c:290
checkasm_check_vp8dsp
void checkasm_check_vp8dsp(void)
Definition: vp8dsp.c:508
name
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default minimum maximum flags name is the option name
Definition: writing_filters.txt:88
checkasm_check_hevc_epel
void checkasm_check_hevc_epel(void)
Definition: hevc_pel.c:297
checkasm_check_videodsp
void checkasm_check_videodsp(void)
Definition: videodsp.c:81
checkasm_check_vf_eq
void checkasm_check_vf_eq(void)
Definition: vf_eq.c:76
checkasm_check_v210dec
void checkasm_check_v210dec(void)
Definition: v210dec.c:47
av_unused
#define av_unused
Definition: attributes.h:131
checkasm_check_hevc_idct
void checkasm_check_hevc_idct(void)
Definition: hevc_idct.c:87
checkasm_check_sw_scale
void checkasm_check_sw_scale(void)
Definition: sw_scale.c:229
checkasm_bench_func
void int checkasm_bench_func(void)
Definition: checkasm.c:793
b
#define b
Definition: input.c:40
func_ref
static av_unused void * func_ref
Definition: checkasm.h:122
checkasm_check_aacpsdsp
void checkasm_check_aacpsdsp(void)
Definition: aacpsdsp.c:234
checkasm_check_func
void * checkasm_check_func(void *func, const char *name,...) av_printf_format(2
CheckasmPerf::sysfd
int sysfd
Definition: checkasm.h:225
checkasm_check_alacdsp
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:116
checkasm_check_llviddspenc
void checkasm_check_llviddspenc(void)
Definition: llviddspenc.c:104
double_near_abs_eps
int double_near_abs_eps(double a, double b, double eps)
Definition: checkasm.c:368
checkasm_check_h264dsp
void checkasm_check_h264dsp(void)
Definition: h264dsp.c:442
checkasm_check_hevc_qpel
void checkasm_check_hevc_qpel(void)
Definition: hevc_pel.c:68
func_new
static av_unused void * func_new
Definition: checkasm.h:122
float_near_abs_eps_ulp
int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp)
Definition: checkasm.c:351
CheckasmPerf
Definition: checkasm.h:224
checkasm_check_vp9dsp
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:626
checkasm_check_vf_threshold
void checkasm_check_vf_threshold(void)
Definition: vf_threshold.c:79
checkasm_check_vf_gblur
void checkasm_check_vf_gblur(void)
Definition: vf_gblur.c:65
checkasm_check_hevc_qpel_uni_w
void checkasm_check_hevc_qpel_uni_w(void)
Definition: hevc_pel.c:153
checkasm_report
void checkasm_report(const char *name,...) av_printf_format(1
checkasm_check_float_dsp
void checkasm_check_float_dsp(void)
Definition: float_dsp.c:282
double_near_abs_eps_array
int double_near_abs_eps_array(const double *a, const double *b, double eps, unsigned len)
Definition: checkasm.c:375
checkasm_check_pixblockdsp
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:82
checkasm_check_bswapdsp
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:60
lfg.h
checkasm_check_sw_rgb
void checkasm_check_sw_rgb(void)
Definition: sw_rgb.c:182
checkasm_check_hevc_sao
void checkasm_check_hevc_sao(void)
Definition: hevc_sao.c:132
CheckasmPerf::iterations
int iterations
Definition: checkasm.h:227
checkasm_check_fmtconvert
void checkasm_check_fmtconvert(void)
Definition: fmtconvert.c:46
macos_kperf.h
float_near_abs_eps_array
int float_near_abs_eps_array(const float *a, const float *b, float eps, unsigned len)
Definition: checkasm.c:339
av_printf_format
#define av_printf_format(fmtpos, attrpos)
Definition: attributes.h:161
float_near_abs_eps_array_ulp
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
Definition: checkasm.c:356
timer.h
AVLFG
Context structure for the Lagged Fibonacci PRNG.
Definition: lfg.h:33
cpu.h
checkasm_check_llviddsp
void checkasm_check_llviddsp(void)
Definition: llviddsp.c:195
float_near_ulp
void int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:298
checkasm_check_hevc_qpel_uni
void checkasm_check_hevc_qpel_uni(void)
Definition: hevc_pel.c:112
checkasm_check_h264pred
void checkasm_check_h264pred(void)
Definition: h264pred.c:233
a
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:41
checkasm_fail_func
void checkasm_fail_func(const char *msg,...) av_printf_format(1
checkasm_check_huffyuvdsp
void checkasm_check_huffyuvdsp(void)
Definition: huffyuvdsp.c:67
CheckasmPerf::cycles
uint64_t cycles
Definition: checkasm.h:226
float_near_abs_eps
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:328
checkasm_check_fixed_dsp
void checkasm_check_fixed_dsp(void)
Definition: fixed_dsp.c:133
checkasm_check_colorspace
void checkasm_check_colorspace(void)
Definition: vf_colorspace.c:309
checkasm_check_utvideodsp
void checkasm_check_utvideodsp(void)
Definition: utvideodsp.c:91
checkasm_check_hevc_epel_bi_w
void checkasm_check_hevc_epel_bi_w(void)
Definition: hevc_pel.c:474
checkasm_check_hevc_qpel_bi_w
void checkasm_check_hevc_qpel_bi_w(void)
Definition: hevc_pel.c:245
checkasm_check_hevc_epel_uni_w
void checkasm_check_hevc_epel_uni_w(void)
Definition: hevc_pel.c:382
checkasm_check_flacdsp
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:57
internal.h
checkasm_check_synth_filter
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:46
checkasm_check_av_tx
void checkasm_check_av_tx(void)
Definition: av_tx.c:86
checkasm_check_blockdsp
void checkasm_check_blockdsp(void)
Definition: blockdsp.c:55
len
int len
Definition: vorbis_enc_data.h:426
checkasm_check_g722dsp
void checkasm_check_g722dsp(void)
Definition: g722dsp.c:53
checkasm_check_vf_hflip
void checkasm_check_vf_hflip(void)
Definition: vf_hflip.c:71
checkasm_check_blend
void checkasm_check_blend(void)
Definition: vf_blend.c:88
checkasm_check_afir
void checkasm_check_afir(void)
Definition: af_afir.c:80
checkasm_check_h264qpel
void checkasm_check_h264qpel(void)
Definition: h264qpel.c:51
dummy
int dummy
Definition: motion.c:65
checkasm_get_perf_context
void struct CheckasmPerf * checkasm_get_perf_context(void)
Definition: checkasm.c:818
checkasm_check_v210enc
void checkasm_check_v210enc(void)
Definition: v210enc.c:82
checkasm_check_hevc_qpel_bi
void checkasm_check_hevc_qpel_bi(void)
Definition: hevc_pel.c:201
checkasm_check_sbrdsp
void checkasm_check_sbrdsp(void)
Definition: sbrdsp.c:254
checkasm_check_hevc_epel_uni
void checkasm_check_hevc_epel_uni(void)
Definition: hevc_pel.c:341
DECL_CHECKASM_CHECK_FUNC
#define DECL_CHECKASM_CHECK_FUNC(type)
Definition: checkasm.h:282
checkasm_check_jpeg2000dsp
void checkasm_check_jpeg2000dsp(void)
Definition: jpeg2000dsp.c:92
checkasm_check_hevc_add_res
void checkasm_check_hevc_add_res(void)
Definition: hevc_add_res.c:86
int32_t
int32_t
Definition: audioconvert.c:56
avstring.h
checkasm_check_audiodsp
void checkasm_check_audiodsp(void)
Definition: audiodsp.c:52
checkasm_check_hevc_epel_bi
void checkasm_check_hevc_epel_bi(void)
Definition: hevc_pel.c:430
checkasm_check_exrdsp
void checkasm_check_exrdsp(void)
Definition: exrdsp.c:77
float_near_ulp_array
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:316
checkasm_check_opusdsp
void checkasm_check_opusdsp(void)
Definition: opusdsp.c:85