Increase chrow row alignment to 16 bytes.

This is done by expanding luma row to 32-byte alignment, since
 there is currently a bunch of code that assumes that
 uv_stride == y_stride/2 (see, for example, vp8/common/postproc.c,
 common/reconinter.c, common/arm/neon/recon16x16mb_neon.asm,
 encoder/temporal_filter.c, and possibly others; I haven't done a
 full audit).
It also uses replaces the hardcoded border of 16 in a number of
 encoder buffers with VP8BORDERINPIXELS (currently 32), as the
 chroma rows start at an offset of border/2.
Together, these two changes have the nice advantage that simply
 dumping the frame memory as a contiguous blob produces a valid,
 if padded, image.

Change-Id: Iaf5ea722ae5c82d5daa50f6e2dade9de753f1003
This commit is contained in:
Timothy B. Terriberry 2011-07-20 10:20:31 -07:00
Родитель b2d9700f53
Коммит 7d1b37cdac
4 изменённых файлов: 54 добавлений и 27 удалений

Просмотреть файл

@ -86,7 +86,8 @@ vp8_lookahead_init(unsigned int width,
if(!ctx->buf)
goto bail;
for(i=0; i<depth; i++)
if (vp8_yv12_alloc_frame_buffer(&ctx->buf[i].img, width, height, 16))
if (vp8_yv12_alloc_frame_buffer(&ctx->buf[i].img,
width, height, VP8BORDERINPIXELS))
goto bail;
}
return ctx;

Просмотреть файл

@ -1240,7 +1240,7 @@ static void alloc_raw_frame_buffers(VP8_COMP *cpi)
#if VP8_TEMPORAL_ALT_REF
if (vp8_yv12_alloc_frame_buffer(&cpi->alt_ref_buffer,
width, height, 16))
width, height, VP8BORDERINPIXELS))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate altref buffer");
@ -1290,7 +1290,8 @@ void vp8_alloc_compressor_data(VP8_COMP *cpi)
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate last frame buffer");
if (vp8_yv12_alloc_frame_buffer(&cpi->scaled_source, width, height, 16))
if (vp8_yv12_alloc_frame_buffer(&cpi->scaled_source,
width, height, VP8BORDERINPIXELS))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate scaled source buffer");

Просмотреть файл

@ -301,6 +301,36 @@ update_error_state(vpx_codec_alg_priv_t *ctx,
return res;
}
static void yuvconfig2image(vpx_image_t *img,
const YV12_BUFFER_CONFIG *yv12,
void *user_priv)
{
/** vpx_img_wrap() doesn't allow specifying independent strides for
* the Y, U, and V planes, nor other alignment adjustments that
* might be representable by a YV12_BUFFER_CONFIG, so we just
* initialize all the fields.*/
img->fmt = yv12->clrtype == REG_YUV ?
VPX_IMG_FMT_I420 : VPX_IMG_FMT_VPXI420;
img->w = yv12->y_stride;
img->h = (yv12->y_height + 2 * VP8BORDERINPIXELS + 15) & ~15;
img->d_w = yv12->y_width;
img->d_h = yv12->y_height;
img->x_chroma_shift = 1;
img->y_chroma_shift = 1;
img->planes[VPX_PLANE_Y] = yv12->y_buffer;
img->planes[VPX_PLANE_U] = yv12->u_buffer;
img->planes[VPX_PLANE_V] = yv12->v_buffer;
img->planes[VPX_PLANE_ALPHA] = NULL;
img->stride[VPX_PLANE_Y] = yv12->y_stride;
img->stride[VPX_PLANE_U] = yv12->uv_stride;
img->stride[VPX_PLANE_V] = yv12->uv_stride;
img->stride[VPX_PLANE_ALPHA] = yv12->y_stride;
img->bps = 12;
img->user_priv = user_priv;
img->img_data = yv12->buffer_alloc;
img->img_data_owner = 0;
img->self_allocd = 0;
}
static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t *ctx,
const uint8_t *data,
@ -429,21 +459,8 @@ static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t *ctx,
if (!res && 0 == vp8dx_get_raw_frame(ctx->pbi, &sd, &time_stamp, &time_end_stamp, &flags))
{
/* Align width/height */
unsigned int a_w = (sd.y_width + 15) & ~15;
unsigned int a_h = (sd.y_height + 15) & ~15;
vpx_img_wrap(&ctx->img, VPX_IMG_FMT_I420,
a_w + 2 * VP8BORDERINPIXELS,
a_h + 2 * VP8BORDERINPIXELS,
1,
sd.buffer_alloc);
vpx_img_set_rect(&ctx->img,
VP8BORDERINPIXELS, VP8BORDERINPIXELS,
sd.y_width, sd.y_height);
ctx->img.user_priv = user_priv;
yuvconfig2image(&ctx->img, &sd, user_priv);
ctx->img_avail = 1;
}
}

Просмотреть файл

@ -49,25 +49,33 @@ vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, int
if (ybf)
{
int y_stride = ((width + 2 * border) + 31) & ~31;
int yplane_size = (height + 2 * border) * y_stride;
int uv_width = width >> 1;
int uv_height = height >> 1;
int yplane_size = (height + 2 * border) * (width + 2 * border);
int uvplane_size = (uv_height + border) * (uv_width + border);
/** There is currently a bunch of code which assumes
* uv_stride == y_stride/2, so enforce this here. */
int uv_stride = y_stride >> 1;
int uvplane_size = (uv_height + border) * uv_stride;
vp8_yv12_de_alloc_frame_buffer(ybf);
/* only support allocating buffers that have
a height and width that are multiples of 16 */
if ((width & 0xf) | (height & 0xf))
/** Only support allocating buffers that have a height and width that
* are multiples of 16, and a border that's a multiple of 32.
* The border restriction is required to get 16-byte alignment of the
* start of the chroma rows without intoducing an arbitrary gap
* between planes, which would break the semantics of things like
* vpx_img_set_rect(). */
if ((width & 0xf) | (height & 0xf) | (border & 0x1f))
return -3;
ybf->y_width = width;
ybf->y_height = height;
ybf->y_stride = width + 2 * border;
ybf->y_stride = y_stride;
ybf->uv_width = uv_width;
ybf->uv_height = uv_height;
ybf->uv_stride = uv_width + border;
ybf->uv_stride = uv_stride;
ybf->border = border;
ybf->frame_size = yplane_size + 2 * uvplane_size;
@ -77,9 +85,9 @@ vp8_yv12_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, int
if (ybf->buffer_alloc == NULL)
return -1;
ybf->y_buffer = ybf->buffer_alloc + (border * ybf->y_stride) + border;
ybf->u_buffer = ybf->buffer_alloc + yplane_size + (border / 2 * ybf->uv_stride) + border / 2;
ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size + (border / 2 * ybf->uv_stride) + border / 2;
ybf->y_buffer = ybf->buffer_alloc + (border * y_stride) + border;
ybf->u_buffer = ybf->buffer_alloc + yplane_size + (border / 2 * uv_stride) + border / 2;
ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size + (border / 2 * uv_stride) + border / 2;
ybf->corrupted = 0; /* assume not currupted by errors */
}