int x264_threadpool_init( x264_threadpool_t **p_pool, int threads,
void (*init_func)(void *), void *init_arg )
{
if( threads <= 0 )
return -1;
x264_threadpool_t *pool;
CHECKED_MALLOCZERO( pool, sizeof(x264_threadpool_t) );
*p_pool = pool;
pool->init_func = init_func;
pool->init_arg = init_arg;
pool->threads = threads;
CHECKED_MALLOC( pool->thread_handle, pool->threads * sizeof(x264_pthread_t) );
if( x264_sync_frame_list_init( &pool->uninit, pool->threads ) ||
x264_sync_frame_list_init( &pool->run, pool->threads ) ||
x264_sync_frame_list_init( &pool->done, pool->threads ) )
goto fail;
for( int i = 0; i < pool->threads; i++ )
{
x264_threadpool_job_t *job;
CHECKED_MALLOC( job, sizeof(x264_threadpool_job_t) );
x264_sync_frame_list_push( &pool->uninit, (void*)job );
}
for( int i = 0; i < pool->threads; i++ )
if( x264_pthread_create( pool->thread_handle+i, NULL, (void*)x264_threadpool_thread, pool ) )
goto fail;
return 0;
fail:
return -1;
}
int x264_sei_version_write( x264_t *h, bs_t *s )
{
// random ID number generated according to ISO-11578
static const uint8_t uuid[16] =
{
0xdc, 0x45, 0xe9, 0xbd, 0xe6, 0xd9, 0x48, 0xb7,
0x96, 0x2c, 0xd8, 0x20, 0xd9, 0x23, 0xee, 0xef
};
char *opts = x264_param2string( &h->param, 0 );
char *payload;
int length;
if( !opts )
return -1;
CHECKED_MALLOC( payload, 200 + strlen( opts ) );
memcpy( payload, uuid, 16 );
sprintf( payload+16, "x264 - core %d%s - H.264/MPEG-4 AVC codec - "
"Copy%s 2003-2011 - http://www.videolan.org/x264.html - options: %s",
X264_BUILD, X264_VERSION, HAVE_GPL?"left":"right", opts );
length = strlen(payload)+1;
x264_sei_write( s, (uint8_t *)payload, length, SEI_USER_DATA_UNREGISTERED );
x264_free( opts );
x264_free( payload );
return 0;
fail:
x264_free( opts );
return -1;
}
int x264_visualize_init( x264_t *h )
{
CHECKED_MALLOC( h->visualize, h->mb.i_mb_width * h->mb.i_mb_height * sizeof(visualize_t) );
return 0;
fail:
return -1;
}
/* {{{ [fold] void x264_visualize_init( x264_t *h ) */
int x264_visualize_init( x264_t *h )
{
int mb = h->sps->i_mb_width * h->sps->i_mb_height;
CHECKED_MALLOC( h->visualize, mb * sizeof(visualize_t) );
return 0;
fail:
return -1;
}
/* Try to load the cached compiled program binary, verify the device context is
* still valid before reuse */
static cl_program x264_opencl_cache_load( x264_t *h, const char *dev_name, const char *dev_vendor, const char *driver_version )
{
/* try to load cached program binary */
FILE *fp = x264_fopen( h->param.psz_clbin_file, "rb" );
x264_opencl_function_t *ocl;
cl_program program;
uint8_t *binary;
size_t size;
const uint8_t *ptr;
cl_int status;
if( !fp )
return NULL;
ocl = h->opencl.ocl;
program = NULL;
binary = NULL;
fseek( fp, 0, SEEK_END );
size = ftell( fp );
rewind( fp );
CHECKED_MALLOC( binary, size );
if ( fread( binary, 1, size, fp ) != size )
goto fail;
ptr = (const uint8_t*)binary;
#define CHECK_STRING( STR )\
do {\
size_t len = strlen( STR );\
if( size <= len || strncmp( (char*)ptr, STR, len ) )\
goto fail;\
else {\
size -= (len+1); ptr += (len+1);\
}\
} while( 0 )
CHECK_STRING( dev_name );
CHECK_STRING( dev_vendor );
CHECK_STRING( driver_version );
CHECK_STRING( x264_opencl_source_hash );
#undef CHECK_STRING
program = ocl->clCreateProgramWithBinary( h->opencl.context, 1, &h->opencl.device, &size, &ptr, NULL, &status );
if( status != CL_SUCCESS )
program = NULL;
fail:
fclose( fp );
x264_free( binary );
return program;
}
/* Save the compiled program binary to a file for later reuse. Device context
* is also saved in the cache file so we do not reuse stale binaries */
static void x264_opencl_cache_save( x264_t *h, cl_program program, const char *dev_name, const char *dev_vendor, const char *driver_version )
{
FILE *fp = x264_fopen( h->param.psz_clbin_file, "wb" );
x264_opencl_function_t *ocl;
uint8_t *binary;
size_t size;
cl_int status;
if( !fp )
{
x264_log( h, X264_LOG_INFO, "OpenCL: unable to open clbin file for write\n" );
return;
}
ocl = h->opencl.ocl;
binary = NULL;
size = 0;
status = ocl->clGetProgramInfo( program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t), &size, NULL );
if( status != CL_SUCCESS || !size )
{
x264_log( h, X264_LOG_INFO, "OpenCL: Unable to query program binary size, no cache file generated\n" );
goto fail;
}
CHECKED_MALLOC( binary, size );
status = ocl->clGetProgramInfo( program, CL_PROGRAM_BINARIES, sizeof(uint8_t *), &binary, NULL );
if( status != CL_SUCCESS )
{
x264_log( h, X264_LOG_INFO, "OpenCL: Unable to query program binary, no cache file generated\n" );
goto fail;
}
fputs( dev_name, fp );
fputc( '\n', fp );
fputs( dev_vendor, fp );
fputc( '\n', fp );
fputs( driver_version, fp );
fputc( '\n', fp );
fputs( x264_opencl_source_hash, fp );
fputc( '\n', fp );
fwrite( binary, 1, size, fp );
fail:
fclose( fp );
x264_free( binary );
return;
}
int x264_sei_version_write( x264_t *h, bs_t *s )
{
int i;
// random ID number generated according to ISO-11578
static const uint8_t uuid[16] =
{
0xdc, 0x45, 0xe9, 0xbd, 0xe6, 0xd9, 0x48, 0xb7,
0x96, 0x2c, 0xd8, 0x20, 0xd9, 0x23, 0xee, 0xef
};
char *opts = x264_param2string( &h->param, 0 );
char *version;
int length;
if( !opts )
return -1;
CHECKED_MALLOC( version, 200 + strlen( opts ) );
sprintf( version, "x264 - core %d%s - H.264/MPEG-4 AVC codec - "
"Copyleft 2003-2010 - http://www.videolan.org/x264.html - options: %s",
X264_BUILD, X264_VERSION, opts );
length = strlen(version)+1+16;
bs_realign( s );
bs_write( s, 8, SEI_USER_DATA_UNREGISTERED );
// payload_size
for( i = 0; i <= length-255; i += 255 )
bs_write( s, 8, 255 );
bs_write( s, 8, length-i );
for( int j = 0; j < 16; j++ )
bs_write( s, 8, uuid[j] );
for( int j = 0; j < length-16; j++ )
bs_write( s, 8, version[j] );
bs_rbsp_trailing( s );
bs_flush( s );
x264_free( opts );
x264_free( version );
return 0;
fail:
x264_free( opts );
return -1;
}
int x265_global_init ( x265_t *h )
{
int32_t loop = 0, i_bit = 0 ;
CHECKED_MALLOC( h->global.convert_to_bit, (X265_MAX_CU_SIZE + 1) * sizeof(int8_t) );
memset ( h->global.convert_to_bit, -1, (X265_MAX_CU_SIZE + 1)* sizeof(int8_t) ) ;
i_bit = 0 ;
for ( loop = 4 ; loop < X265_MAX_CU_SIZE ; loop *= 2 )
{
h->global.convert_to_bit[loop] = i_bit ;
++ i_bit ;
}
h->global.convert_to_bit [loop] = i_bit ;
return 0 ;
fail:
x265_global_deinit ( h ) ;
return -1 ;
}
x264_frame_t *x264_frame_new( x264_t *h )
{
x264_frame_t *frame = x264_malloc( sizeof(x264_frame_t) );
int i, j;
int i_mb_count = h->mb.i_mb_count;
int i_stride;
int i_lines;
if( !frame ) return NULL;
memset( frame, 0, sizeof(x264_frame_t) );
/* allocate frame data (+64 for extra data for me) */
i_stride = ( ( h->param.i_width + 15 )&0xfffff0 )+ 64;
i_lines = ( ( h->param.i_height + 15 )&0xfffff0 );
frame->i_plane = 3;
for( i = 0; i < 3; i++ )
{
int i_divh = 1;
int i_divw = 1;
if( i > 0 )
{
if( h->param.i_csp == X264_CSP_I420 )
i_divh = i_divw = 2;
else if( h->param.i_csp == X264_CSP_I422 )
i_divw = 2;
}
frame->i_stride[i] = i_stride / i_divw;
frame->i_lines[i] = i_lines / i_divh;
CHECKED_MALLOC( frame->buffer[i],
frame->i_stride[i] * ( frame->i_lines[i] + 64 / i_divh ) );
// frame->buffer[i]=x264_malloc(frame->i_stride[i] * ( frame->i_lines[i] + 64 / i_divh ) );
frame->plane[i] = ((uint8_t*)frame->buffer[i]) +
frame->i_stride[i] * 32 / i_divh + 32 / i_divw;
}
frame->i_stride[3] = 0;
frame->i_lines[3] = 0;
frame->buffer[3] = NULL;
frame->plane[3] = NULL;
frame->filtered[0] = frame->plane[0];
for( i = 0; i < 3; i++ )
{
CHECKED_MALLOC( frame->buffer[4+i],
frame->i_stride[0] * ( frame->i_lines[0] + 64 ) );
frame->filtered[i+1] = ((uint8_t*)frame->buffer[4+i]) +
frame->i_stride[0] * 32 + 32;
}
if( h->frames.b_have_lowres )
{
frame->i_stride_lowres = frame->i_stride[0]/2 + 32;
frame->i_lines_lowres = frame->i_lines[0]/2;
for( i = 0; i < 4; i++ )
{
CHECKED_MALLOC( frame->buffer[7+i],
frame->i_stride_lowres * ( frame->i_lines[0]/2 + 64 ) );
frame->lowres[i] = ((uint8_t*)frame->buffer[7+i]) +
frame->i_stride_lowres * 32 + 32;
}
}
if( h->param.analyse.i_me_method == X264_ME_ESA )
{
CHECKED_MALLOC( frame->buffer[11],
frame->i_stride[0] * (frame->i_lines[0] + 64) * sizeof(uint16_t) );
frame->integral = (uint16_t*)frame->buffer[11] + frame->i_stride[0] * 32 + 32;
}
frame->i_poc = -1;
frame->i_type = X264_TYPE_AUTO;
frame->i_qpplus1 = 0;
frame->i_pts = -1;
frame->i_frame = -1;
frame->i_frame_num = -1;
CHECKED_MALLOC( frame->mb_type, i_mb_count * sizeof(int8_t));
CHECKED_MALLOC( frame->mv[0], 2*16 * i_mb_count * sizeof(int16_t) );
CHECKED_MALLOC( frame->ref[0], 4 * i_mb_count * sizeof(int8_t) );
if( h->param.i_bframe )
{
CHECKED_MALLOC( frame->mv[1], 2*16 * i_mb_count * sizeof(int16_t) );
CHECKED_MALLOC( frame->ref[1], 4 * i_mb_count * sizeof(int8_t) );
}
else
{
frame->mv[1] = NULL;
frame->ref[1] = NULL;
}
CHECKED_MALLOC( frame->i_row_bits, i_lines/16 * sizeof(int) );
CHECKED_MALLOC( frame->i_row_qp, i_lines/16 * sizeof(int) );
for( i = 0; i < h->param.i_bframe + 2; i++ )
for( j = 0; j < h->param.i_bframe + 2; j++ )
CHECKED_MALLOC( frame->i_row_satds[i][j], i_lines/16 * sizeof(int) );
return frame;
fail:
x264_frame_delete( frame );
return NULL;
}
/* The OpenCL source under common/opencl will be merged into common/oclobj.h by
* the Makefile. It defines a x264_opencl_source byte array which we will pass
* to clCreateProgramWithSource(). We also attempt to use a cache file for the
* compiled binary, stored in the current working folder. */
static cl_program x264_opencl_compile( x264_t *h )
{
x264_opencl_function_t *ocl = h->opencl.ocl;
cl_program program;
cl_int status;
char devname[64];
char devvendor[64];
char driverversion[64];
status = ocl->clGetDeviceInfo( h->opencl.device, CL_DEVICE_NAME, sizeof(devname), devname, NULL );
status |= ocl->clGetDeviceInfo( h->opencl.device, CL_DEVICE_VENDOR, sizeof(devvendor), devvendor, NULL );
status |= ocl->clGetDeviceInfo( h->opencl.device, CL_DRIVER_VERSION, sizeof(driverversion), driverversion, NULL );
if( status != CL_SUCCESS )
return NULL;
// Most AMD GPUs have vector registers
int vectorize = !strcmp( devvendor, "Advanced Micro Devices, Inc." );
h->opencl.b_device_AMD_SI = 0;
if( vectorize )
{
/* Disable OpenCL on Intel/AMD switchable graphics devices */
if( x264_detect_switchable_graphics() )
{
x264_log( h, X264_LOG_INFO, "OpenCL acceleration disabled, switchable graphics detected\n" );
return NULL;
}
/* Detect AMD SouthernIsland or newer device (single-width registers) */
cl_uint simdwidth = 4;
status = ocl->clGetDeviceInfo( h->opencl.device, CL_DEVICE_SIMD_INSTRUCTION_WIDTH_AMD, sizeof(cl_uint), &simdwidth, NULL );
if( status == CL_SUCCESS && simdwidth == 1 )
{
vectorize = 0;
h->opencl.b_device_AMD_SI = 1;
}
}
x264_log( h, X264_LOG_INFO, "OpenCL acceleration enabled with %s %s %s\n", devvendor, devname, h->opencl.b_device_AMD_SI ? "(SI)" : "" );
program = x264_opencl_cache_load( h, devname, devvendor, driverversion );
if( !program )
{
/* clCreateProgramWithSource() requires a pointer variable, you cannot just use &x264_opencl_source */
x264_log( h, X264_LOG_INFO, "Compiling OpenCL kernels...\n" );
const char *strptr = (const char*)x264_opencl_source;
size_t size = sizeof(x264_opencl_source);
program = ocl->clCreateProgramWithSource( h->opencl.context, 1, &strptr, &size, &status );
if( status != CL_SUCCESS || !program )
{
x264_log( h, X264_LOG_WARNING, "OpenCL: unable to create program\n" );
return NULL;
}
}
/* Build the program binary for the OpenCL device */
const char *buildopts = vectorize ? "-DVECTORIZE=1" : "";
status = ocl->clBuildProgram( program, 1, &h->opencl.device, buildopts, NULL, NULL );
if( status == CL_SUCCESS )
{
x264_opencl_cache_save( h, program, devname, devvendor, driverversion );
return program;
}
/* Compile failure, should not happen with production code. */
size_t build_log_len = 0;
status = ocl->clGetProgramBuildInfo( program, h->opencl.device, CL_PROGRAM_BUILD_LOG, build_log_len, NULL, &build_log_len );
if( status != CL_SUCCESS )
{
x264_log( h, X264_LOG_WARNING, "OpenCL: Compilation failed, unable to query build log\n" );
return NULL;
}
char *build_log;
CHECKED_MALLOC( build_log, build_log_len );
if( !build_log )
{
x264_log( h, X264_LOG_WARNING, "OpenCL: Compilation failed, unable to alloc build log\n" );
return NULL;
}
status = ocl->clGetProgramBuildInfo( program, h->opencl.device, CL_PROGRAM_BUILD_LOG, build_log_len, build_log, NULL );
if( status != CL_SUCCESS )
{
x264_log( h, X264_LOG_WARNING, "OpenCL: Compilation failed, unable to get build log\n" );
x264_free( build_log );
return NULL;
}
FILE *lg = fopen( "x264_kernel_build_log.txt", "w" );
if( lg )
{
fwrite( build_log, 1, build_log_len, lg );
fclose( lg );
x264_log( h, X264_LOG_WARNING, "OpenCL: kernel build errors written to x264_kernel_build_log.txt\n" );
}
x264_free( build_log );
fail:
return NULL;
}
