void I420_R5G5B5_MMX(void * pHandle, uint8_t *p_y, uint8_t *p_u, uint8_t *p_v, uint16_t *p_d)
{
PSOFTCONV pconv = pHandle;
if (!pHandle)
return;
int b_hscale = pconv->b_hscale; /* horizontal scaling type */
unsigned int i_vscale = pconv->i_vscale; /* vertical scaling type */
unsigned int i_x, i_y; /* horizontal and vertical indexes */
CC_RECT * dst_rect = &pconv->dst_dis_rect;
CC_RECT * src_rect = &pconv->src_rect;
int dst_stride = pconv->dst_pitch;
uint16_t *p_pic = (uint16_t *)((uint8_t *)p_d + pconv->dst_offset);
int i_right_margin = pconv->i_right_margin;
int i_rewind;
int i_scale_count; /* scale modulo counter */
int i_chroma_width = WIDTHOF(src_rect) / 2; /* chroma width */
uint16_t * p_pic_start; /* beginning of the current line for copy */
/* Conversion buffer pointer */
uint16_t * p_buffer_start = pconv->pConvBuffer;
uint16_t * p_buffer;
/* Offset array pointer */
int * p_offset_start = pconv->pOffsetBuffer;
int * p_offset;
const int i_source_margin = pconv->i_source_margin;
const int i_source_margin_c = pconv->i_source_margin_c;
/*
* Perform conversion
*/
i_scale_count = ( i_vscale == 1 ) ?
HEIGHTOF(dst_rect) :
HEIGHTOF(src_rect);
if( WIDTHOF(src_rect) & 7 )
{
i_rewind = 8 - ( WIDTHOF(src_rect) & 7 );
}
else
{
i_rewind = 0;
}
for( i_y = 0; i_y < HEIGHTOF(src_rect); i_y++ )
{
p_pic_start = p_pic;
p_buffer = b_hscale ? p_buffer_start : p_pic;
for ( i_x = WIDTHOF(src_rect) / 8; i_x--; )
{
MMX_CALL (
MMX_INIT_16
MMX_YUV_MUL
MMX_YUV_ADD
MMX_UNPACK_15
);
p_y += 8;
p_u += 4;
p_v += 4;
p_buffer += 8;
}
/* Here we do some unaligned reads and duplicate conversions, but
* at least we have all the pixels */
if( i_rewind )
{
p_y -= i_rewind;
p_u -= i_rewind >> 1;
p_v -= i_rewind >> 1;
p_buffer -= i_rewind;
MMX_CALL (
MMX_INIT_16
MMX_YUV_MUL
MMX_YUV_ADD
MMX_UNPACK_15
);
p_y += 8;
p_u += 4;
p_v += 4;
p_buffer += 8;
}
SCALE_WIDTH;
SCALE_HEIGHT( 2 );
p_y += i_source_margin;
if( i_y % 2 )
{
p_u += i_source_margin_c;
p_v += i_source_margin_c;
}
}
/* re-enable FPU registers */
MMX_END;
}
VLC_TARGET
void I420_R5G5B5( filter_t *p_filter, picture_t *p_src, picture_t *p_dest )
{
/* We got this one from the old arguments */
uint16_t *p_pic = (uint16_t*)p_dest->p->p_pixels;
uint8_t *p_y = p_src->Y_PIXELS;
uint8_t *p_u = p_src->U_PIXELS;
uint8_t *p_v = p_src->V_PIXELS;
bool b_hscale; /* horizontal scaling type */
unsigned int i_vscale; /* vertical scaling type */
unsigned int i_x, i_y; /* horizontal and vertical indexes */
int i_right_margin;
int i_rewind;
int i_scale_count; /* scale modulo counter */
int i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
uint16_t * p_pic_start; /* beginning of the current line for copy */
/* Conversion buffer pointer */
uint16_t * p_buffer_start = (uint16_t*)p_filter->p_sys->p_buffer;
uint16_t * p_buffer;
/* Offset array pointer */
int * p_offset_start = p_filter->p_sys->p_offset;
int * p_offset;
const int i_source_margin = p_src->p[0].i_pitch
- p_src->p[0].i_visible_pitch
- p_filter->fmt_in.video.i_x_offset;
const int i_source_margin_c = p_src->p[1].i_pitch
- p_src->p[1].i_visible_pitch
- ( p_filter->fmt_in.video.i_x_offset / 2 );
i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;
/* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
* on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
* then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
SetOffset( (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width),
(p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height),
(p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
(p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
&b_hscale, &i_vscale, p_offset_start );
/*
* Perform conversion
*/
i_scale_count = ( i_vscale == 1 ) ?
(p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
(p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);
#ifdef SSE2
i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 15;
/*
** SSE2 128 bits fetch/store instructions are faster
** if memory access is 16 bytes aligned
*/
p_buffer = b_hscale ? p_buffer_start : p_pic;
if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
p_dest->p->i_pitch|
((intptr_t)p_y)|
((intptr_t)p_buffer))) )
{
/* use faster SSE2 aligned fetch and store */
for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
{
p_pic_start = p_pic;
for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)/16; i_x--; )
{
SSE2_CALL (
SSE2_INIT_16_ALIGNED
SSE2_YUV_MUL
SSE2_YUV_ADD
SSE2_UNPACK_15_ALIGNED
);
p_y += 16;
p_u += 8;
p_v += 8;
p_buffer += 16;
}
/* Here we do some unaligned reads and duplicate conversions, but
* at least we have all the pixels */
if( i_rewind )
{
p_y -= i_rewind;
p_u -= i_rewind >> 1;
p_v -= i_rewind >> 1;
p_buffer -= i_rewind;
SSE2_CALL (
SSE2_INIT_16_UNALIGNED
SSE2_YUV_MUL
SSE2_YUV_ADD
SSE2_UNPACK_15_UNALIGNED
);
p_y += 16;
p_u += 8;
p_v += 8;
}
SCALE_WIDTH;
SCALE_HEIGHT( 420, 2 );
p_y += i_source_margin;
if( i_y % 2 )
{
p_u += i_source_margin_c;
p_v += i_source_margin_c;
}
p_buffer = b_hscale ? p_buffer_start : p_pic;
}
}
else
{
/* use slower SSE2 unaligned fetch and store */
for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
void I420_RGB32( filter_t *p_filter, picture_t *p_src, picture_t *p_dest )
{
/* We got this one from the old arguments */
uint32_t *p_pic = (uint32_t*)p_dest->p->p_pixels;
uint8_t *p_y = p_src->Y_PIXELS;
uint8_t *p_u = p_src->U_PIXELS;
uint8_t *p_v = p_src->V_PIXELS;
bool b_hscale; /* horizontal scaling type */
unsigned int i_vscale; /* vertical scaling type */
unsigned int i_x, i_y; /* horizontal and vertical indexes */
int i_right_margin;
int i_rewind;
int i_scale_count; /* scale modulo counter */
int i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
uint32_t * p_pic_start; /* beginning of the current line for copy */
int i_uval, i_vval; /* U and V samples */
int i_red, i_green, i_blue; /* U and V modified samples */
uint32_t * p_yuv = p_filter->p_sys->p_rgb32;
uint32_t * p_ybase; /* Y dependant conversion table */
/* Conversion buffer pointer */
uint32_t * p_buffer_start = (uint32_t*)p_filter->p_sys->p_buffer;
uint32_t * p_buffer;
/* Offset array pointer */
int * p_offset_start = p_filter->p_sys->p_offset;
int * p_offset;
const int i_source_margin = p_src->p[0].i_pitch
- p_src->p[0].i_visible_pitch
- p_filter->fmt_in.video.i_x_offset;
const int i_source_margin_c = p_src->p[1].i_pitch
- p_src->p[1].i_visible_pitch
- ( p_filter->fmt_in.video.i_x_offset / 2 );
i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;
i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 7;
/* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
* on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
* then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
SetOffset( (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width),
(p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height),
(p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
(p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
&b_hscale, &i_vscale, p_offset_start );
/*
* Perform conversion
*/
i_scale_count = ( i_vscale == 1 ) ?
(p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
(p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);
for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
{
p_pic_start = p_pic;
p_buffer = b_hscale ? p_buffer_start : p_pic;
for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 8; i_x--; )
{
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
}
/* Here we do some unaligned reads and duplicate conversions, but
* at least we have all the pixels */
if( i_rewind )
{
p_y -= i_rewind;
p_u -= i_rewind >> 1;
p_v -= i_rewind >> 1;
p_buffer -= i_rewind;
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
CONVERT_YUV_PIXEL(4); CONVERT_Y_PIXEL(4);
}
SCALE_WIDTH;
SCALE_HEIGHT( 420, 4 );
p_y += i_source_margin;
if( i_y % 2 )
{
p_u += i_source_margin_c;
p_v += i_source_margin_c;
}
}
}
