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 #include "libavutil/avstring.h"
29 #include "libavutil/cpu.h"
30 #include "libavutil/lfg.h"
31 #include "libavutil/timer.h"
32 
33 void checkasm_check_alacdsp(void);
34 void checkasm_check_blend(void);
35 void checkasm_check_bswapdsp(void);
36 void checkasm_check_colorspace(void);
37 void checkasm_check_flacdsp(void);
38 void checkasm_check_fmtconvert(void);
39 void checkasm_check_h264pred(void);
40 void checkasm_check_h264qpel(void);
44 void checkasm_check_v210enc(void);
45 void checkasm_check_vp9dsp(void);
46 void checkasm_check_videodsp(void);
47 
48 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
49 int checkasm_bench_func(void);
50 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2);
51 void checkasm_update_bench(int iterations, uint64_t cycles);
52 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
53 
54 /* float compare utilities */
55 int float_near_ulp(float a, float b, unsigned max_ulp);
56 int float_near_abs_eps(float a, float b, float eps);
57 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp);
58 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
59  unsigned len);
60 int float_near_abs_eps_array(const float *a, const float *b, float eps,
61  unsigned len);
62 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
63  unsigned max_ulp, unsigned len);
64 
65 extern AVLFG checkasm_lfg;
66 #define rnd() av_lfg_get(&checkasm_lfg)
67 
68 static av_unused void *func_ref, *func_new;
69 
70 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
71 
72 /* Decide whether or not the specified function needs to be tested */
73 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__))
74 
75 /* Declare the function prototype. The first argument is the return value, the remaining
76  * arguments are the function parameters. Naming parameters is optional. */
77 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
78 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
79 
80 /* Indicate that the current test has failed */
81 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
82 
83 /* Print the test outcome */
84 #define report checkasm_report
85 
86 /* Call the reference function */
87 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
88 
89 #if ARCH_X86 && HAVE_YASM
90 /* Verifies that clobbered callee-saved registers are properly saved and restored
91  * and that either no MMX registers are touched or emms is issued */
92 void checkasm_checked_call(void *func, ...);
93 /* Verifies that clobbered callee-saved registers are properly saved and restored
94  * and issues emms for asm functions which are not required to do so */
95 void checkasm_checked_call_emms(void *func, ...);
96 
97 #if ARCH_X86_64
98 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
99  * This is done by clobbering the stack with junk around the stack pointer and calling the
100  * assembly function through checked_call() with added dummy arguments which forces all
101  * real arguments to be passed on the stack and not in registers. For 32-bit arguments the
102  * upper half of the 64-bit register locations on the stack will now contain junk which will
103  * cause misbehaving functions to either produce incorrect output or segfault. Note that
104  * even though this works extremely well in practice, it's technically not guaranteed
105  * and false negatives is theoretically possible, but there can never be any false positives.
106  */
107 void checkasm_stack_clobber(uint64_t clobber, ...);
108 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
109  = (void *)checkasm_checked_call;
110 #define declare_new_emms(cpu_flags, ret, ...) \
111  ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \
112  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
113  (void *)checkasm_checked_call;
114 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
115 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
116  CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
117  checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__))
118 #elif ARCH_X86_32
119 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
120 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \
121  ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
122  (void *)checkasm_checked_call;
123 #define call_new(...) checked_call(func_new, __VA_ARGS__)
124 #endif
125 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
126 /* Use a dummy argument, to offset the real parameters by 2, not only 1.
127  * This makes sure that potential 8-byte-alignment of parameters is kept the same
128  * even when the extra parameters have been removed. */
129 void checkasm_checked_call_vfp(void *func, int dummy, ...);
130 void checkasm_checked_call_novfp(void *func, int dummy, ...);
131 extern void (*checkasm_checked_call)(void *func, int dummy, ...);
132 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__) = (void *)checkasm_checked_call;
133 #define call_new(...) checked_call(func_new, 0, __VA_ARGS__)
134 #elif ARCH_AARCH64 && !defined(__APPLE__)
135 void checkasm_checked_call(void *func, ...);
136 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
137 #define call_new(...) checked_call(func_new, __VA_ARGS__)
138 #else
139 #define declare_new(ret, ...)
140 #define declare_new_emms(cpu_flags, ret, ...)
141 /* Call the function */
142 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__)
143 #endif
144 
145 #ifndef declare_new_emms
146 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
147 #endif
148 
149 /* Benchmark the function */
150 #ifdef AV_READ_TIME
151 #define bench_new(...)\
152  do {\
153  if (checkasm_bench_func()) {\
154  func_type *tfunc = func_new;\
155  uint64_t tsum = 0;\
156  int ti, tcount = 0;\
157  for (ti = 0; ti < BENCH_RUNS; ti++) {\
158  uint64_t t = AV_READ_TIME();\
159  tfunc(__VA_ARGS__);\
160  tfunc(__VA_ARGS__);\
161  tfunc(__VA_ARGS__);\
162  tfunc(__VA_ARGS__);\
163  t = AV_READ_TIME() - t;\
164  if (t*tcount <= tsum*4 && ti > 0) {\
165  tsum += t;\
166  tcount++;\
167  }\
168  }\
169  checkasm_update_bench(tcount, tsum);\
170  }\
171  } while (0)
172 #else
173 #define bench_new(...) while(0)
174 #endif
175 
176 #endif /* TESTS_CHECKASM_CHECKASM_H */
Definition: lfg.h:25
void int float_near_ulp(float a, float b, unsigned max_ulp)
Definition: checkasm.c:199
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:248
void checkasm_check_v210enc(void)
Definition: v210enc.c:81
void checkasm_check_vp9dsp(void)
Definition: vp9dsp.c:617
void checkasm_check_bswapdsp(void)
Definition: bswapdsp.c:59
const char * b
Definition: vf_curves.c:109
void checkasm_check_flacdsp(void)
Definition: flacdsp.c:56
void checkasm_check_h264pred(void)
Definition: h264pred.c:232
void checkasm_check_colorspace(void)
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:599
void void checkasm_update_bench(int iterations, uint64_t cycles)
Definition: checkasm.c:624
void checkasm_fail_func(const char *msg,...) av_printf_format(1
#define av_printf_format(fmtpos, attrpos)
Definition: attributes.h:156
void checkasm_check_videodsp(void)
Definition: videodsp.c:80
AVLFG checkasm_lfg
Definition: checkasm.c:191
typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)
static av_unused void * func_ref
Definition: checkasm.h:68
int dummy
Definition: motion.c:64
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_report(const char *name,...) av_printf_format(1
void checkasm_check_alacdsp(void)
Definition: alacdsp.c:115
static av_unused void * func_new
Definition: checkasm.h:68
int float_near_abs_eps(float a, float b, float eps)
Definition: checkasm.c:229
int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp, unsigned len)
Definition: checkasm.c:217
int len
void checkasm_check_pixblockdsp(void)
Definition: pixblockdsp.c:81
void checkasm_check_synth_filter(void)
Definition: synth_filter.c:44
int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps, unsigned max_ulp, unsigned len)
Definition: checkasm.c:253
void checkasm_check_blend(void)
Definition: vf_blend.c:87
#define av_unused
Definition: attributes.h:126
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:236