ffmpeg中AVFrame解码linesize确定

一?测试环境

机型：Intel(R)?Xeon(R)?Gold?5218

ffmpeg版本4.3-dev

解码视频分辨率?720x1280?,?解码后?AVFrame?linesize?768

解码视频分辨率?480x640,?解码后AVFrame?linesize?512

二?linesize计算过程

avcodec_default_get_buffer2

-->update_frame_pool()

-->video_get_buffer()

video_get_buffer()-->

for?(i = 0;?i < 4 && pool->pools[i];?i++)?{

pic->linesize[i]?= pool->linesize[i];

由上面代码可见，AVFrame的linesize源自于?pool的linesize

三?pool?的linesize计算过程

update_frame_pool()

case AVMEDIA_TYPE_VIDEO:?{

int linesize[4];

int w = frame->width;

int h = frame->height;

int unaligned;

ptrdiff_t linesize1[4];

size_t size[4];

avcodec_align_dimensions2(avctx,?&w,?&h,?pool->stride_align);

do?{

//?NOTE:?do?not?align?linesizes?individually,?this?breaks?e.g.?assumptions

//?that?linesize[0]?==?2*linesize[1]?in?the?MPEG-encoder?for?4:2:2

ret = av_image_fill_linesizes(linesize,?avctx->pix_fmt,?w);

if?(ret < 0)

goto fail;

//?increase?alignment?of?w?for?next?try?(rhs?gives?the?lowest?bit?set?in?w)

w += w & ~(w - 1);

unaligned = 0;

for?(i = 0;?i < 4;?i++)

unaligned |= linesize[i]?% pool->stride_align[i];

????????}?while?(unaligned);

for?(i = 0;?i < 4;?i++)

linesize1[i]?= linesize[i];

ret = av_image_fill_plane_sizes(size,?avctx->pix_fmt,?h,?linesize1);

if?(ret < 0)

goto fail;

for?(i = 0;?i < 4;?i++)?{

pool->linesize[i]?= linesize[i];?//这里赋值给pool的linesize

linesize列表?通过?av_image_fill_linesizes获取，基本就是等于?width。

循环中

w?+=?w?&?~(w?-?1);?//每次将最低有效二进制位+1，这样做的目的，比如当前是16的整数倍，调整为32的整数倍。然后64的整数倍。

直到最终对?stride_align[i]取余为0.

avcodec_align_dimensions2(avctx,?&w,?&h,?pool->stride_align);

函数获取?stride_align

for?(i = 0;?i < 4;?i++)

linesize_align[i]?= STRIDE_ALIGN;

#if?HAVE_SIMD_ALIGN_64

#???define?STRIDE_ALIGN?64?/*?AVX-512?*/

#elif?HAVE_SIMD_ALIGN_32

#???define?STRIDE_ALIGN?32

#elif?HAVE_SIMD_ALIGN_16

#???define?STRIDE_ALIGN?16

#else

#???define?STRIDE_ALIGN?8

#endif

当机器支持avx512,?STRIDE_ALIGN?64.

总结：ffmpeg的AVFrame?linesize对齐方式和机器支持的simd有关，avx512就是64字节，avx256就是32字节