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_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);
59 void checkasm_check_hevc_add_res(void);
60 void checkasm_check_hevc_idct(void);
61 void checkasm_check_hevc_sao(void);
62 void checkasm_check_huffyuvdsp(void);
63 void checkasm_check_jpeg2000dsp(void);
64 void checkasm_check_llviddsp(void);
65 void checkasm_check_llviddspenc(void);
66 void checkasm_check_nlmeans(void);
67 void checkasm_check_pixblockdsp(void);
68 void checkasm_check_sbrdsp(void);
69 void checkasm_check_synth_filter(void);
70 void checkasm_check_sw_rgb(void);
71 void checkasm_check_utvideodsp(void);
72 void checkasm_check_v210enc(void);
73 void checkasm_check_vf_hflip(void);
74 void checkasm_check_vf_threshold(void);
75 void checkasm_check_vp8dsp(void);
76 void checkasm_check_vp9dsp(void);
77 void checkasm_check_videodsp(void);
81 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
82 int checkasm_bench_func(void);
83 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2);
84 struct CheckasmPerf *checkasm_get_perf_context(void);
85 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
87 /* float compare utilities */
88 int float_near_ulp(float a, float b, unsigned max_ulp);
89 int float_near_abs_eps(float a, float b, float eps);
90 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp);
91 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
93 int float_near_abs_eps_array(const float *a, const float *b, float eps,
95 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
96 unsigned max_ulp, unsigned len);
97 int double_near_abs_eps(double a, double b, double eps);
98 int double_near_abs_eps_array(const double *a, const double *b, double eps,
101 extern AVLFG checkasm_lfg;
102 #define rnd() av_lfg_get(&checkasm_lfg)
104 static av_unused void *func_ref, *func_new;
106 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
108 /* Decide whether or not the specified function needs to be tested */
109 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__))
111 /* Declare the function prototype. The first argument is the return value, the remaining
112 * arguments are the function parameters. Naming parameters is optional. */
113 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
114 #define declare_func_float(ret, ...) declare_new_float(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
115 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
117 /* Indicate that the current test has failed */
118 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
120 /* Print the test outcome */
121 #define report checkasm_report
123 /* Call the reference function */
124 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
126 #if ARCH_X86 && HAVE_X86ASM
127 /* Verifies that clobbered callee-saved registers are properly saved and restored
128 * and that either no MMX registers are touched or emms is issued */
129 void checkasm_checked_call(void *func, ...);
130 /* Verifies that clobbered callee-saved registers are properly saved and restored
131 * and issues emms for asm functions which are not required to do so */
132 void checkasm_checked_call_emms(void *func, ...);
133 /* Verifies that clobbered callee-saved registers are properly saved and restored
134 * but doesn't issue emms. Meant for dsp functions returning float or double */
135 void checkasm_checked_call_float(void *func, ...);
138 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
139 * This is done by clobbering the stack with junk around the stack pointer and calling the
140 * assembly function through checked_call() with added dummy arguments which forces all
141 * real arguments to be passed on the stack and not in registers. For 32-bit arguments the
142 * upper half of the 64-bit register locations on the stack will now contain junk which will
143 * cause misbehaving functions to either produce incorrect output or segfault. Note that
144 * even though this works extremely well in practice, it's technically not guaranteed
145 * and false negatives is theoretically possible, but there can never be any false positives.
147 void checkasm_stack_clobber(uint64_t clobber, ...);
148 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
149 = (void *)checkasm_checked_call;
150 #define declare_new_float(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
151 = (void *)checkasm_checked_call_float;
152 #define declare_new_emms(cpu_flags, ret, ...) \
153 ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \
154 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
155 (void *)checkasm_checked_call;
156 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
157 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
158 CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
159 checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__))
161 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
162 #define declare_new_float(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call_float;
163 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \
164 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
165 (void *)checkasm_checked_call;
166 #define call_new(...) checked_call(func_new, __VA_ARGS__)
168 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
169 /* Use a dummy argument, to offset the real parameters by 2, not only 1.
170 * This makes sure that potential 8-byte-alignment of parameters is kept the same
171 * even when the extra parameters have been removed. */
172 void checkasm_checked_call_vfp(void *func, int dummy, ...);
173 void checkasm_checked_call_novfp(void *func, int dummy, ...);
174 extern void (*checkasm_checked_call)(void *func, int dummy, ...);
175 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__) = (void *)checkasm_checked_call;
176 #define call_new(...) checked_call(func_new, 0, __VA_ARGS__)
177 #elif ARCH_AARCH64 && !defined(__APPLE__)
178 void checkasm_stack_clobber(uint64_t clobber, ...);
179 void checkasm_checked_call(void *func, ...);
180 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, int, int, __VA_ARGS__)\
181 = (void *)checkasm_checked_call;
182 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
183 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
184 CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
185 checked_call(func_new, 0, 0, 0, 0, 0, 0, 0, __VA_ARGS__))
187 #define declare_new(ret, ...)
188 #define declare_new_float(ret, ...)
189 #define declare_new_emms(cpu_flags, ret, ...)
190 /* Call the function */
191 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__)
194 #ifndef declare_new_emms
195 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
197 #ifndef declare_new_float
198 #define declare_new_float(ret, ...) declare_new(ret, __VA_ARGS__)
201 typedef struct CheckasmPerf {
207 #if defined(AV_READ_TIME) || CONFIG_LINUX_PERF
209 #if CONFIG_LINUX_PERF
210 #define PERF_START(t) do { \
211 ioctl(sysfd, PERF_EVENT_IOC_RESET, 0); \
212 ioctl(sysfd, PERF_EVENT_IOC_ENABLE, 0); \
214 #define PERF_STOP(t) do { \
215 ioctl(sysfd, PERF_EVENT_IOC_DISABLE, 0); \
216 read(sysfd, &t, sizeof(t)); \
219 #define PERF_START(t) t = AV_READ_TIME()
220 #define PERF_STOP(t) t = AV_READ_TIME() - t
223 /* Benchmark the function */
224 #define bench_new(...)\
226 if (checkasm_bench_func()) {\
227 struct CheckasmPerf *perf = checkasm_get_perf_context();\
228 av_unused const int sysfd = perf->sysfd;\
229 func_type *tfunc = func_new;\
233 for (ti = 0; ti < BENCH_RUNS; ti++) {\
240 if (t*tcount <= tsum*4 && ti > 0) {\
251 #define bench_new(...) while(0)
252 #define PERF_START(t) while(0)
253 #define PERF_STOP(t) while(0)
256 #endif /* TESTS_CHECKASM_CHECKASM_H */