git.sesse.net Git - x264/blob - common/win32thread.c

   1 /*****************************************************************************
   2  * win32thread.c: windows threading
   3  *****************************************************************************
   4  * Copyright (C) 2010-2012 x264 project
   5  *
   6  * Authors: Steven Walters <kemuri9@gmail.com>
   7  *          Pegasys Inc. <http://www.pegasys-inc.com>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  22  *
  23  * This program is also available under a commercial proprietary license.
  24  * For more information, contact us at licensing@x264.com.
  25  *****************************************************************************/
  26
  27 /* Microsoft's way of supporting systems with >64 logical cpus can be found at
  28  * http://www.microsoft.com/whdc/system/Sysinternals/MoreThan64proc.mspx */
  29
  30 /* Based on the agreed standing that x264 does not need to utilize >64 logical cpus,
  31  * this API does not detect nor utilize more than 64 cpus for systems that have them. */
  32
  33 #include "common.h"
  34 #include <process.h>
  35
  36 /* number of times to spin a thread about to block on a locked mutex before retrying and sleeping if still locked */
  37 #define X264_SPIN_COUNT 0
  38
  39 /* GROUP_AFFINITY struct */
  40 typedef struct
  41 {
  42     ULONG_PTR mask; // KAFFINITY = ULONG_PTR
  43     USHORT group;
  44     USHORT reserved[3];
  45 } x264_group_affinity_t;
  46
  47 typedef struct
  48 {
  49     /* global mutex for replacing MUTEX_INITIALIZER instances */
  50     x264_pthread_mutex_t static_mutex;
  51
  52     /* function pointers to conditional variable API on windows 6.0+ kernels */
  53     void (WINAPI *cond_broadcast)( x264_pthread_cond_t *cond );
  54     void (WINAPI *cond_init)( x264_pthread_cond_t *cond );
  55     void (WINAPI *cond_signal)( x264_pthread_cond_t *cond );
  56     BOOL (WINAPI *cond_wait)( x264_pthread_cond_t *cond, x264_pthread_mutex_t *mutex, DWORD milliseconds );
  57 } x264_win32thread_control_t;
  58
  59 static x264_win32thread_control_t thread_control;
  60
  61 /* _beginthreadex requires that the start routine is __stdcall */
  62 static unsigned __stdcall x264_win32thread_worker( void *arg )
  63 {
  64     x264_pthread_t *h = arg;
  65     h->ret = h->func( h->arg );
  66     return 0;
  67 }
  68
  69 int x264_pthread_create( x264_pthread_t *thread, const x264_pthread_attr_t *attr,
  70                          void *(*start_routine)( void* ), void *arg )
  71 {
  72     thread->func   = start_routine;
  73     thread->arg    = arg;
  74     thread->handle = (void*)_beginthreadex( NULL, 0, x264_win32thread_worker, thread, 0, NULL );
  75     return !thread->handle;
  76 }
  77
  78 int x264_pthread_join( x264_pthread_t thread, void **value_ptr )
  79 {
  80     DWORD ret = WaitForSingleObject( thread.handle, INFINITE );
  81     if( ret != WAIT_OBJECT_0 )
  82         return -1;
  83     if( value_ptr )
  84         *value_ptr = thread.ret;
  85     CloseHandle( thread.handle );
  86     return 0;
  87 }
  88
  89 int x264_pthread_mutex_init( x264_pthread_mutex_t *mutex, const x264_pthread_mutexattr_t *attr )
  90 {
  91     return !InitializeCriticalSectionAndSpinCount( mutex, X264_SPIN_COUNT );
  92 }
  93
  94 int x264_pthread_mutex_destroy( x264_pthread_mutex_t *mutex )
  95 {
  96     DeleteCriticalSection( mutex );
  97     return 0;
  98 }
  99
 100 int x264_pthread_mutex_lock( x264_pthread_mutex_t *mutex )
 101 {
 102     static x264_pthread_mutex_t init = X264_PTHREAD_MUTEX_INITIALIZER;
 103     if( !memcmp( mutex, &init, sizeof(x264_pthread_mutex_t) ) )
 104         *mutex = thread_control.static_mutex;
 105     EnterCriticalSection( mutex );
 106     return 0;
 107 }
 108
 109 int x264_pthread_mutex_unlock( x264_pthread_mutex_t *mutex )
 110 {
 111     LeaveCriticalSection( mutex );
 112     return 0;
 113 }
 114
 115 /* for pre-Windows 6.0 platforms we need to define and use our own condition variable and api */
 116 typedef struct
 117 {
 118     x264_pthread_mutex_t mtx_broadcast;
 119     x264_pthread_mutex_t mtx_waiter_count;
 120     volatile int waiter_count;
 121     HANDLE semaphore;
 122     HANDLE waiters_done;
 123     volatile int is_broadcast;
 124 } x264_win32_cond_t;
 125
 126 int x264_pthread_cond_init( x264_pthread_cond_t *cond, const x264_pthread_condattr_t *attr )
 127 {
 128     if( thread_control.cond_init )
 129     {
 130         thread_control.cond_init( cond );
 131         return 0;
 132     }
 133
 134     /* non native condition variables */
 135     x264_win32_cond_t *win32_cond = calloc( 1, sizeof(x264_win32_cond_t) );
 136     if( !win32_cond )
 137         return -1;
 138     cond->ptr = win32_cond;
 139     win32_cond->semaphore = CreateSemaphore( NULL, 0, 0x7fffffff, NULL );
 140     if( !win32_cond->semaphore )
 141         return -1;
 142
 143     if( x264_pthread_mutex_init( &win32_cond->mtx_waiter_count, NULL ) )
 144         return -1;
 145     if( x264_pthread_mutex_init( &win32_cond->mtx_broadcast, NULL ) )
 146         return -1;
 147
 148     win32_cond->waiters_done = CreateEvent( NULL, FALSE, FALSE, NULL );
 149     if( !win32_cond->waiters_done )
 150         return -1;
 151
 152     return 0;
 153 }
 154
 155 int x264_pthread_cond_destroy( x264_pthread_cond_t *cond )
 156 {
 157     /* native condition variables do not destroy */
 158     if( thread_control.cond_init )
 159         return 0;
 160
 161     /* non native condition variables */
 162     x264_win32_cond_t *win32_cond = cond->ptr;
 163     CloseHandle( win32_cond->semaphore );
 164     CloseHandle( win32_cond->waiters_done );
 165     x264_pthread_mutex_destroy( &win32_cond->mtx_broadcast );
 166     x264_pthread_mutex_destroy( &win32_cond->mtx_waiter_count );
 167     free( win32_cond );
 168
 169     return 0;
 170 }
 171
 172 int x264_pthread_cond_broadcast( x264_pthread_cond_t *cond )
 173 {
 174     if( thread_control.cond_broadcast )
 175     {
 176         thread_control.cond_broadcast( cond );
 177         return 0;
 178     }
 179
 180     /* non native condition variables */
 181     x264_win32_cond_t *win32_cond = cond->ptr;
 182     x264_pthread_mutex_lock( &win32_cond->mtx_broadcast );
 183     x264_pthread_mutex_lock( &win32_cond->mtx_waiter_count );
 184     int have_waiter = 0;
 185
 186     if( win32_cond->waiter_count )
 187     {
 188         win32_cond->is_broadcast = 1;
 189         have_waiter = 1;
 190     }
 191
 192     if( have_waiter )
 193     {
 194         ReleaseSemaphore( win32_cond->semaphore, win32_cond->waiter_count, NULL );
 195         x264_pthread_mutex_unlock( &win32_cond->mtx_waiter_count );
 196         WaitForSingleObject( win32_cond->waiters_done, INFINITE );
 197         win32_cond->is_broadcast = 0;
 198     }
 199     else
 200         x264_pthread_mutex_unlock( &win32_cond->mtx_waiter_count );
 201     return x264_pthread_mutex_unlock( &win32_cond->mtx_broadcast );
 202 }
 203
 204 int x264_pthread_cond_signal( x264_pthread_cond_t *cond )
 205 {
 206     if( thread_control.cond_signal )
 207     {
 208         thread_control.cond_signal( cond );
 209         return 0;
 210     }
 211
 212     /* non-native condition variables */
 213     x264_win32_cond_t *win32_cond = cond->ptr;
 214
 215     x264_pthread_mutex_lock( &win32_cond->mtx_broadcast );
 216     x264_pthread_mutex_lock( &win32_cond->mtx_waiter_count );
 217     int have_waiter = win32_cond->waiter_count;
 218     x264_pthread_mutex_unlock( &win32_cond->mtx_waiter_count );
 219
 220     if( have_waiter )
 221     {
 222         ReleaseSemaphore( win32_cond->semaphore, 1, NULL );
 223         WaitForSingleObject( win32_cond->waiters_done, INFINITE );
 224     }
 225
 226     return x264_pthread_mutex_unlock( &win32_cond->mtx_broadcast );
 227 }
 228
 229 int x264_pthread_cond_wait( x264_pthread_cond_t *cond, x264_pthread_mutex_t *mutex )
 230 {
 231     if( thread_control.cond_wait )
 232         return !thread_control.cond_wait( cond, mutex, INFINITE );
 233
 234     /* non native condition variables */
 235     x264_win32_cond_t *win32_cond = cond->ptr;
 236
 237     x264_pthread_mutex_lock( &win32_cond->mtx_broadcast );
 238     x264_pthread_mutex_lock( &win32_cond->mtx_waiter_count );
 239     win32_cond->waiter_count++;
 240     x264_pthread_mutex_unlock( &win32_cond->mtx_waiter_count );
 241     x264_pthread_mutex_unlock( &win32_cond->mtx_broadcast );
 242
 243     // unlock the external mutex
 244     x264_pthread_mutex_unlock( mutex );
 245     WaitForSingleObject( win32_cond->semaphore, INFINITE );
 246
 247     x264_pthread_mutex_lock( &win32_cond->mtx_waiter_count );
 248     win32_cond->waiter_count--;
 249     int last_waiter = !win32_cond->waiter_count || !win32_cond->is_broadcast;
 250     x264_pthread_mutex_unlock( &win32_cond->mtx_waiter_count );
 251
 252     if( last_waiter )
 253         SetEvent( win32_cond->waiters_done );
 254
 255     // lock the external mutex
 256     return x264_pthread_mutex_lock( mutex );
 257 }
 258
 259 int x264_win32_threading_init( void )
 260 {
 261     /* find function pointers to API functions, if they exist */
 262     HANDLE kernel_dll = GetModuleHandle( TEXT( "kernel32.dll" ) );
 263     thread_control.cond_init = (void*)GetProcAddress( kernel_dll, "InitializeConditionVariable" );
 264     if( thread_control.cond_init )
 265     {
 266         /* we're on a windows 6.0+ kernel, acquire the rest of the functions */
 267         thread_control.cond_broadcast = (void*)GetProcAddress( kernel_dll, "WakeAllConditionVariable" );
 268         thread_control.cond_signal = (void*)GetProcAddress( kernel_dll, "WakeConditionVariable" );
 269         thread_control.cond_wait = (void*)GetProcAddress( kernel_dll, "SleepConditionVariableCS" );
 270     }
 271     return x264_pthread_mutex_init( &thread_control.static_mutex, NULL );
 272 }
 273
 274 void x264_win32_threading_destroy( void )
 275 {
 276     x264_pthread_mutex_destroy( &thread_control.static_mutex );
 277     memset( &thread_control, 0, sizeof(x264_win32thread_control_t) );
 278 }
 279
 280 int x264_pthread_num_processors_np()
 281 {
 282     DWORD_PTR system_cpus, process_cpus = 0;
 283     int cpus = 0;
 284
 285     /* GetProcessAffinityMask returns affinities of 0 when the process has threads in multiple processor groups.
 286      * On platforms that support processor grouping, use GetThreadGroupAffinity to get the current thread's affinity instead. */
 287 #if ARCH_X86_64
 288     /* find function pointers to API functions specific to x86_64 platforms, if they exist */
 289     HANDLE kernel_dll = GetModuleHandle( TEXT( "kernel32.dll" ) );
 290     BOOL (*get_thread_affinity)( HANDLE thread, x264_group_affinity_t *group_affinity ) = (void*)GetProcAddress( kernel_dll, "GetThreadGroupAffinity" );
 291     if( get_thread_affinity )
 292     {
 293         /* running on a platform that supports >64 logical cpus */
 294         x264_group_affinity_t thread_affinity;
 295         if( get_thread_affinity( GetCurrentThread(), &thread_affinity ) )
 296             process_cpus = thread_affinity.mask;
 297     }
 298 #endif
 299     if( !process_cpus )
 300         GetProcessAffinityMask( GetCurrentProcess(), &process_cpus, &system_cpus );
 301     for( DWORD_PTR bit = 1; bit; bit <<= 1 )
 302         cpus += !!(process_cpus & bit);
 303
 304     return cpus ? cpus : 1;
 305 }