2 * Assembly testing and benchmarking tool
3 * Copyright (c) 2015 Henrik Gramner
4 * Copyright (c) 2008 Loren Merritt
6 * This file is part of FFmpeg.
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.
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.
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.
23 #ifndef TESTS_CHECKASM_CHECKASM_H
24 #define TESTS_CHECKASM_CHECKASM_H
30 #include <unistd.h> // read(3)
31 #include <sys/ioctl.h>
32 #include <asm/unistd.h>
33 #include <linux/perf_event.h>
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"
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);
58 void checkasm_check_hevc_add_res(void);
59 void checkasm_check_hevc_idct(void);
60 void checkasm_check_jpeg2000dsp(void);
61 void checkasm_check_llviddsp(void);
62 void checkasm_check_pixblockdsp(void);
63 void checkasm_check_sbrdsp(void);
64 void checkasm_check_synth_filter(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);
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);
75 struct CheckasmPerf *checkasm_get_perf_context(void);
76 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
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,
84 int float_near_abs_eps_array(const float *a, const float *b, float eps,
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,
92 extern AVLFG checkasm_lfg;
93 #define rnd() av_lfg_get(&checkasm_lfg)
95 static av_unused void *func_ref, *func_new;
97 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
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__))
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__)
108 /* Indicate that the current test has failed */
109 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
111 /* Print the test outcome */
112 #define report checkasm_report
114 /* Call the reference function */
115 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
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, ...);
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.
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__))
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__)
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__))
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__)
185 #ifndef declare_new_emms
186 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
188 #ifndef declare_new_float
189 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__)
192 typedef struct CheckasmPerf {
198 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF
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); \
205 #define PERF_STOP(t) do { \
206 ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \
207 read(sysfd, &t, sizeof(t)); \
210 #define PERF_START(t) t = AV_READ_TIME()
211 #define PERF_STOP(t) t = AV_READ_TIME() - t
214 /* Benchmark the function */
215 #define bench_new(...)\
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;\
224 for (ti = 0; ti < BENCH_RUNS; ti++) {\
231 if (t*tcount <= tsum*4 && ti > 0) {\
242 #define bench_new(...) while(0)
243 #define PERF_START(t) while(0)
244 #define PERF_STOP(t) while(0)
247 #endif /* TESTS_CHECKASM_CHECKASM_H */