Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  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. #include <stdarg.h>
  24. #include <stdio.h>
  25. #include <stdlib.h>
  26. #include <string.h>
  27. #include "checkasm.h"
  28. #include "libavutil/common.h"
  29. #include "libavutil/cpu.h"
  30. #include "libavutil/random_seed.h"
  31.  
  32. #if HAVE_IO_H
  33. #include <io.h>
  34. #endif
  35.  
  36. #if HAVE_SETCONSOLETEXTATTRIBUTE
  37. #include <windows.h>
  38. #define COLOR_RED    FOREGROUND_RED
  39. #define COLOR_GREEN  FOREGROUND_GREEN
  40. #define COLOR_YELLOW (FOREGROUND_RED|FOREGROUND_GREEN)
  41. #else
  42. #define COLOR_RED    1
  43. #define COLOR_GREEN  2
  44. #define COLOR_YELLOW 3
  45. #endif
  46.  
  47. #if HAVE_UNISTD_H
  48. #include <unistd.h>
  49. #endif
  50.  
  51. #if !HAVE_ISATTY
  52. #define isatty(fd) 1
  53. #endif
  54.  
  55. /* List of tests to invoke */
  56. static const struct {
  57.     const char *name;
  58.     void (*func)(void);
  59. } tests[] = {
  60. #if CONFIG_BSWAPDSP
  61.     { "bswapdsp", checkasm_check_bswapdsp },
  62. #endif
  63. #if CONFIG_H264PRED
  64.     { "h264pred", checkasm_check_h264pred },
  65. #endif
  66. #if CONFIG_H264QPEL
  67.     { "h264qpel", checkasm_check_h264qpel },
  68. #endif
  69.     { NULL }
  70. };
  71.  
  72. /* List of cpu flags to check */
  73. static const struct {
  74.     const char *name;
  75.     const char *suffix;
  76.     int flag;
  77. } cpus[] = {
  78. #if   ARCH_AARCH64
  79.     { "ARMV8",    "armv8",    AV_CPU_FLAG_ARMV8 },
  80.     { "NEON",     "neon",     AV_CPU_FLAG_NEON },
  81. #elif ARCH_ARM
  82.     { "ARMV5TE",  "armv5te",  AV_CPU_FLAG_ARMV5TE },
  83.     { "ARMV6",    "armv6",    AV_CPU_FLAG_ARMV6 },
  84.     { "ARMV6T2",  "armv6t2",  AV_CPU_FLAG_ARMV6T2 },
  85.     { "VFP",      "vfp",      AV_CPU_FLAG_VFP },
  86.     { "VFPV3",    "vfp3",     AV_CPU_FLAG_VFPV3 },
  87.     { "NEON",     "neon",     AV_CPU_FLAG_NEON },
  88. #elif ARCH_PPC
  89.     { "ALTIVEC",  "altivec",  AV_CPU_FLAG_ALTIVEC },
  90.     { "VSX",      "vsx",      AV_CPU_FLAG_VSX },
  91.     { "POWER8",   "power8",   AV_CPU_FLAG_POWER8 },
  92. #elif ARCH_X86
  93.     { "MMX",      "mmx",      AV_CPU_FLAG_MMX|AV_CPU_FLAG_CMOV },
  94.     { "MMXEXT",   "mmxext",   AV_CPU_FLAG_MMXEXT },
  95.     { "3DNOW",    "3dnow",    AV_CPU_FLAG_3DNOW },
  96.     { "3DNOWEXT", "3dnowext", AV_CPU_FLAG_3DNOWEXT },
  97.     { "SSE",      "sse",      AV_CPU_FLAG_SSE },
  98.     { "SSE2",     "sse2",     AV_CPU_FLAG_SSE2|AV_CPU_FLAG_SSE2SLOW },
  99.     { "SSE3",     "sse3",     AV_CPU_FLAG_SSE3|AV_CPU_FLAG_SSE3SLOW },
  100.     { "SSSE3",    "ssse3",    AV_CPU_FLAG_SSSE3|AV_CPU_FLAG_ATOM },
  101.     { "SSE4.1",   "sse4",     AV_CPU_FLAG_SSE4 },
  102.     { "SSE4.2",   "sse42",    AV_CPU_FLAG_SSE42 },
  103.     { "AVX",      "avx",      AV_CPU_FLAG_AVX },
  104.     { "XOP",      "xop",      AV_CPU_FLAG_XOP },
  105.     { "FMA3",     "fma3",     AV_CPU_FLAG_FMA3 },
  106.     { "FMA4",     "fma4",     AV_CPU_FLAG_FMA4 },
  107.     { "AVX2",     "avx2",     AV_CPU_FLAG_AVX2 },
  108. #endif
  109.     { NULL }
  110. };
  111.  
  112. typedef struct CheckasmFuncVersion {
  113.     struct CheckasmFuncVersion *next;
  114.     void *func;
  115.     int ok;
  116.     int cpu;
  117.     int iterations;
  118.     uint64_t cycles;
  119. } CheckasmFuncVersion;
  120.  
  121. /* Binary search tree node */
  122. typedef struct CheckasmFunc {
  123.     struct CheckasmFunc *child[2];
  124.     CheckasmFuncVersion versions;
  125.     char name[1];
  126. } CheckasmFunc;
  127.  
  128. /* Internal state */
  129. static struct {
  130.     CheckasmFunc *funcs;
  131.     CheckasmFunc *current_func;
  132.     CheckasmFuncVersion *current_func_ver;
  133.     const char *current_test_name;
  134.     const char *bench_pattern;
  135.     int bench_pattern_len;
  136.     int num_checked;
  137.     int num_failed;
  138.     int nop_time;
  139.     int cpu_flag;
  140.     const char *cpu_flag_name;
  141. } state;
  142.  
  143. /* PRNG state */
  144. AVLFG checkasm_lfg;
  145.  
  146. /* Print colored text to stderr if the terminal supports it */
  147. static void color_printf(int color, const char *fmt, ...)
  148. {
  149.     static int use_color = -1;
  150.     va_list arg;
  151.  
  152. #if HAVE_SETCONSOLETEXTATTRIBUTE
  153.     static HANDLE con;
  154.     static WORD org_attributes;
  155.  
  156.     if (use_color < 0) {
  157.         CONSOLE_SCREEN_BUFFER_INFO con_info;
  158.         con = GetStdHandle(STD_ERROR_HANDLE);
  159.         if (con && con != INVALID_HANDLE_VALUE && GetConsoleScreenBufferInfo(con, &con_info)) {
  160.             org_attributes = con_info.wAttributes;
  161.             use_color = 1;
  162.         } else
  163.             use_color = 0;
  164.     }
  165.     if (use_color)
  166.         SetConsoleTextAttribute(con, (org_attributes & 0xfff0) | (color & 0x0f));
  167. #else
  168.     if (use_color < 0) {
  169.         const char *term = getenv("TERM");
  170.         use_color = term && strcmp(term, "dumb") && isatty(2);
  171.     }
  172.     if (use_color)
  173.         fprintf(stderr, "\x1b[%d;3%dm", (color & 0x08) >> 3, color & 0x07);
  174. #endif
  175.  
  176.     va_start(arg, fmt);
  177.     vfprintf(stderr, fmt, arg);
  178.     va_end(arg);
  179.  
  180.     if (use_color) {
  181. #if HAVE_SETCONSOLETEXTATTRIBUTE
  182.         SetConsoleTextAttribute(con, org_attributes);
  183. #else
  184.         fprintf(stderr, "\x1b[0m");
  185. #endif
  186.     }
  187. }
  188.  
  189. /* Deallocate a tree */
  190. static void destroy_func_tree(CheckasmFunc *f)
  191. {
  192.     if (f) {
  193.         CheckasmFuncVersion *v = f->versions.next;
  194.         while (v) {
  195.             CheckasmFuncVersion *next = v->next;
  196.             free(v);
  197.             v = next;
  198.         }
  199.  
  200.         destroy_func_tree(f->child[0]);
  201.         destroy_func_tree(f->child[1]);
  202.         free(f);
  203.     }
  204. }
  205.  
  206. /* Allocate a zero-initialized block, clean up and exit on failure */
  207. static void *checkasm_malloc(size_t size)
  208. {
  209.     void *ptr = calloc(1, size);
  210.     if (!ptr) {
  211.         fprintf(stderr, "checkasm: malloc failed\n");
  212.         destroy_func_tree(state.funcs);
  213.         exit(1);
  214.     }
  215.     return ptr;
  216. }
  217.  
  218. /* Get the suffix of the specified cpu flag */
  219. static const char *cpu_suffix(int cpu)
  220. {
  221.     int i = FF_ARRAY_ELEMS(cpus);
  222.  
  223.     while (--i >= 0)
  224.         if (cpu & cpus[i].flag)
  225.             return cpus[i].suffix;
  226.  
  227.     return "c";
  228. }
  229.  
  230. #ifdef AV_READ_TIME
  231. static int cmp_nop(const void *a, const void *b)
  232. {
  233.     return *(const uint16_t*)a - *(const uint16_t*)b;
  234. }
  235.  
  236. /* Measure the overhead of the timing code (in decicycles) */
  237. static int measure_nop_time(void)
  238. {
  239.     uint16_t nops[10000];
  240.     int i, nop_sum = 0;
  241.  
  242.     for (i = 0; i < 10000; i++) {
  243.         uint64_t t = AV_READ_TIME();
  244.         nops[i] = AV_READ_TIME() - t;
  245.     }
  246.  
  247.     qsort(nops, 10000, sizeof(uint16_t), cmp_nop);
  248.     for (i = 2500; i < 7500; i++)
  249.         nop_sum += nops[i];
  250.  
  251.     return nop_sum / 500;
  252. }
  253.  
  254. /* Print benchmark results */
  255. static void print_benchs(CheckasmFunc *f)
  256. {
  257.     if (f) {
  258.         print_benchs(f->child[0]);
  259.  
  260.         /* Only print functions with at least one assembly version */
  261.         if (f->versions.cpu || f->versions.next) {
  262.             CheckasmFuncVersion *v = &f->versions;
  263.             do {
  264.                 if (v->iterations) {
  265.                     int decicycles = (10*v->cycles/v->iterations - state.nop_time) / 4;
  266.                     printf("%s_%s: %d.%d\n", f->name, cpu_suffix(v->cpu), decicycles/10, decicycles%10);
  267.                 }
  268.             } while ((v = v->next));
  269.         }
  270.  
  271.         print_benchs(f->child[1]);
  272.     }
  273. }
  274. #endif
  275.  
  276. /* ASCIIbetical sort except preserving natural order for numbers */
  277. static int cmp_func_names(const char *a, const char *b)
  278. {
  279.     int ascii_diff, digit_diff;
  280.  
  281.     for (; !(ascii_diff = *a - *b) && *a; a++, b++);
  282.     for (; av_isdigit(*a) && av_isdigit(*b); a++, b++);
  283.  
  284.     return (digit_diff = av_isdigit(*a) - av_isdigit(*b)) ? digit_diff : ascii_diff;
  285. }
  286.  
  287. /* Get a node with the specified name, creating it if it doesn't exist */
  288. static CheckasmFunc *get_func(const char *name, int length)
  289. {
  290.     CheckasmFunc *f, **f_ptr = &state.funcs;
  291.  
  292.     /* Search the tree for a matching node */
  293.     while ((f = *f_ptr)) {
  294.         int cmp = cmp_func_names(name, f->name);
  295.         if (!cmp)
  296.             return f;
  297.  
  298.         f_ptr = &f->child[(cmp > 0)];
  299.     }
  300.  
  301.     /* Allocate and insert a new node into the tree */
  302.     f = *f_ptr = checkasm_malloc(sizeof(CheckasmFunc) + length);
  303.     memcpy(f->name, name, length+1);
  304.  
  305.     return f;
  306. }
  307.  
  308. /* Perform tests and benchmarks for the specified cpu flag if supported by the host */
  309. static void check_cpu_flag(const char *name, int flag)
  310. {
  311.     int old_cpu_flag = state.cpu_flag;
  312.  
  313.     flag |= old_cpu_flag;
  314.     av_set_cpu_flags_mask(flag);
  315.     state.cpu_flag = av_get_cpu_flags();
  316.  
  317.     if (!flag || state.cpu_flag != old_cpu_flag) {
  318.         int i;
  319.  
  320.         state.cpu_flag_name = name;
  321.         for (i = 0; tests[i].func; i++) {
  322.             state.current_test_name = tests[i].name;
  323.             tests[i].func();
  324.         }
  325.     }
  326. }
  327.  
  328. /* Print the name of the current CPU flag, but only do it once */
  329. static void print_cpu_name(void)
  330. {
  331.     if (state.cpu_flag_name) {
  332.         color_printf(COLOR_YELLOW, "%s:\n", state.cpu_flag_name);
  333.         state.cpu_flag_name = NULL;
  334.     }
  335. }
  336.  
  337. int main(int argc, char *argv[])
  338. {
  339.     int i, seed, ret = 0;
  340.  
  341.     if (!tests[0].func || !cpus[0].flag) {
  342.         fprintf(stderr, "checkasm: no tests to perform\n");
  343.         return 0;
  344.     }
  345.  
  346.     if (argc > 1 && !strncmp(argv[1], "--bench", 7)) {
  347. #ifndef AV_READ_TIME
  348.         fprintf(stderr, "checkasm: --bench is not supported on your system\n");
  349.         return 1;
  350. #endif
  351.         if (argv[1][7] == '=') {
  352.             state.bench_pattern = argv[1] + 8;
  353.             state.bench_pattern_len = strlen(state.bench_pattern);
  354.         } else
  355.             state.bench_pattern = "";
  356.  
  357.         argc--;
  358.         argv++;
  359.     }
  360.  
  361.     seed = (argc > 1) ? atoi(argv[1]) : av_get_random_seed();
  362.     fprintf(stderr, "checkasm: using random seed %u\n", seed);
  363.     av_lfg_init(&checkasm_lfg, seed);
  364.  
  365.     check_cpu_flag(NULL, 0);
  366.     for (i = 0; cpus[i].flag; i++)
  367.         check_cpu_flag(cpus[i].name, cpus[i].flag);
  368.  
  369.     if (state.num_failed) {
  370.         fprintf(stderr, "checkasm: %d of %d tests have failed\n", state.num_failed, state.num_checked);
  371.         ret = 1;
  372.     } else {
  373.         fprintf(stderr, "checkasm: all %d tests passed\n", state.num_checked);
  374. #ifdef AV_READ_TIME
  375.         if (state.bench_pattern) {
  376.             state.nop_time = measure_nop_time();
  377.             printf("nop: %d.%d\n", state.nop_time/10, state.nop_time%10);
  378.             print_benchs(state.funcs);
  379.         }
  380. #endif
  381.     }
  382.  
  383.     destroy_func_tree(state.funcs);
  384.     return ret;
  385. }
  386.  
  387. /* Decide whether or not the specified function needs to be tested and
  388.  * allocate/initialize data structures if needed. Returns a pointer to a
  389.  * reference function if the function should be tested, otherwise NULL */
  390. void *checkasm_check_func(void *func, const char *name, ...)
  391. {
  392.     char name_buf[256];
  393.     void *ref = func;
  394.     CheckasmFuncVersion *v;
  395.     int name_length;
  396.     va_list arg;
  397.  
  398.     va_start(arg, name);
  399.     name_length = vsnprintf(name_buf, sizeof(name_buf), name, arg);
  400.     va_end(arg);
  401.  
  402.     if (!func || name_length <= 0 || name_length >= sizeof(name_buf))
  403.         return NULL;
  404.  
  405.     state.current_func = get_func(name_buf, name_length);
  406.     v = &state.current_func->versions;
  407.  
  408.     if (v->func) {
  409.         CheckasmFuncVersion *prev;
  410.         do {
  411.             /* Only test functions that haven't already been tested */
  412.             if (v->func == func)
  413.                 return NULL;
  414.  
  415.             if (v->ok)
  416.                 ref = v->func;
  417.  
  418.             prev = v;
  419.         } while ((v = v->next));
  420.  
  421.         v = prev->next = checkasm_malloc(sizeof(CheckasmFuncVersion));
  422.     }
  423.  
  424.     v->func = func;
  425.     v->ok = 1;
  426.     v->cpu = state.cpu_flag;
  427.     state.current_func_ver = v;
  428.  
  429.     if (state.cpu_flag)
  430.         state.num_checked++;
  431.  
  432.     return ref;
  433. }
  434.  
  435. /* Decide whether or not the current function needs to be benchmarked */
  436. int checkasm_bench_func(void)
  437. {
  438.     return !state.num_failed && state.bench_pattern &&
  439.            !strncmp(state.current_func->name, state.bench_pattern, state.bench_pattern_len);
  440. }
  441.  
  442. /* Indicate that the current test has failed */
  443. void checkasm_fail_func(const char *msg, ...)
  444. {
  445.     if (state.current_func_ver->cpu && state.current_func_ver->ok) {
  446.         va_list arg;
  447.  
  448.         print_cpu_name();
  449.         fprintf(stderr, "   %s_%s (", state.current_func->name, cpu_suffix(state.current_func_ver->cpu));
  450.         va_start(arg, msg);
  451.         vfprintf(stderr, msg, arg);
  452.         va_end(arg);
  453.         fprintf(stderr, ")\n");
  454.  
  455.         state.current_func_ver->ok = 0;
  456.         state.num_failed++;
  457.     }
  458. }
  459.  
  460. /* Update benchmark results of the current function */
  461. void checkasm_update_bench(int iterations, uint64_t cycles)
  462. {
  463.     state.current_func_ver->iterations += iterations;
  464.     state.current_func_ver->cycles += cycles;
  465. }
  466.  
  467. /* Print the outcome of all tests performed since the last time this function was called */
  468. void checkasm_report(const char *name, ...)
  469. {
  470.     static int prev_checked, prev_failed, max_length;
  471.  
  472.     if (state.num_checked > prev_checked) {
  473.         int pad_length = max_length + 4;
  474.         va_list arg;
  475.  
  476.         print_cpu_name();
  477.         pad_length -= fprintf(stderr, " - %s.", state.current_test_name);
  478.         va_start(arg, name);
  479.         pad_length -= vfprintf(stderr, name, arg);
  480.         va_end(arg);
  481.         fprintf(stderr, "%*c", FFMAX(pad_length, 0) + 2, '[');
  482.  
  483.         if (state.num_failed == prev_failed)
  484.             color_printf(COLOR_GREEN, "OK");
  485.         else
  486.             color_printf(COLOR_RED, "FAILED");
  487.         fprintf(stderr, "]\n");
  488.  
  489.         prev_checked = state.num_checked;
  490.         prev_failed  = state.num_failed;
  491.     } else if (!state.cpu_flag) {
  492.         /* Calculate the amount of padding required to make the output vertically aligned */
  493.         int length = strlen(state.current_test_name);
  494.         va_list arg;
  495.  
  496.         va_start(arg, name);
  497.         length += vsnprintf(NULL, 0, name, arg);
  498.         va_end(arg);
  499.  
  500.         if (length > max_length)
  501.             max_length = length;
  502.     }
  503. }
  504.