On Wed, May 1, 2024 at 6:59 PM Andreas Rheinhardt <
andreas.rheinha...@outlook.com> wrote:
> Stone Chen:
> > To prepare for adding AVX2 functions for different block widths, change
> VVCInterDSPContext to contain (*sad[6]) instead of (*sad). This also
> default initializes the pointer array with the scalar function and the
> calling sites to jump to the correct function based on block width. There's
> no change in functionality.
> > ---
> > libavcodec/vvc/dsp.h | 2 +-
> > libavcodec/vvc/inter.c | 4 ++--
> > libavcodec/vvc/inter_template.c | 5 ++++-
> > 3 files changed, 7 insertions(+), 4 deletions(-)
> >
> > diff --git a/libavcodec/vvc/dsp.h b/libavcodec/vvc/dsp.h
> > index 9810ac314c..b06a3ef10e 100644
> > --- a/libavcodec/vvc/dsp.h
> > +++ b/libavcodec/vvc/dsp.h
> > @@ -86,7 +86,7 @@ typedef struct VVCInterDSPContext {
> >
> > void (*apply_bdof)(uint8_t *dst, ptrdiff_t dst_stride, int16_t
> *src0, int16_t *src1, int block_w, int block_h);
> >
> > - int (*sad)(const int16_t *src0, const int16_t *src1, int dx, int
> dy, int block_w, int block_h);
> > + int (*sad[6])(const int16_t *src0, const int16_t *src1, int dx, int
> dy, int block_w, int block_h);
> > void (*dmvr[2][2])(int16_t *dst, const uint8_t *src, ptrdiff_t
> src_stride, int height,
> > intptr_t mx, intptr_t my, int width);
> > } VVCInterDSPContext;
> > diff --git a/libavcodec/vvc/inter.c b/libavcodec/vvc/inter.c
> > index 4a8d1d866a..a68f4f9452 100644
> > --- a/libavcodec/vvc/inter.c
> > +++ b/libavcodec/vvc/inter.c
> > @@ -742,7 +742,7 @@ static void dmvr_mv_refine(VVCLocalContext *lc,
> MvField *mvf, MvField *orig_mv,
> > fc->vvcdsp.inter.dmvr[!!my][!!mx](tmp[i], src, src_stride,
> pred_h, mx, my, pred_w);
> > }
> >
> > - min_sad = fc->vvcdsp.inter.sad(tmp[L0], tmp[L1], dx, dy, block_w,
> block_h);
> > + min_sad = fc->vvcdsp.inter.sad[av_log2(block_w) - 2](tmp[L0],
> tmp[L1], dx, dy, block_w, block_h);
> > min_sad -= min_sad >> 2;
> > sad[dy][dx] = min_sad;
> >
> > @@ -752,7 +752,7 @@ static void dmvr_mv_refine(VVCLocalContext *lc,
> MvField *mvf, MvField *orig_mv,
> > for (dy = 0; dy < SAD_ARRAY_SIZE; dy++) {
> > for (dx = 0; dx < SAD_ARRAY_SIZE; dx++) {
> > if (dx != sr_range || dy != sr_range) {
> > - sad[dy][dx] = fc->vvcdsp.inter.sad(lc->tmp,
> lc->tmp1, dx, dy, block_w, block_h);
> > + sad[dy][dx] = fc->vvcdsp.inter.sad[av_log2(block_w)
> - 2](lc->tmp, lc->tmp1, dx, dy, block_w, block_h);
> > if (sad[dy][dx] < min_sad) {
> > min_sad = sad[dy][dx];
> > min_dx = dx;
> > diff --git a/libavcodec/vvc/inter_template.c
> b/libavcodec/vvc/inter_template.c
> > index e2fbfd4fc0..545e8dd184 100644
> > --- a/libavcodec/vvc/inter_template.c
> > +++ b/libavcodec/vvc/inter_template.c
> > @@ -458,7 +458,10 @@ static void
> FUNC(ff_vvc_inter_dsp_init)(VVCInterDSPContext *const inter)
> > inter->apply_prof_uni_w = FUNC(apply_prof_uni_w);
> > inter->apply_bdof = FUNC(apply_bdof);
> > inter->prof_grad_filter = FUNC(prof_grad_filter);
> > - inter->sad = vvc_sad;
> > +
> > + for (int i = 0; i < FF_ARRAY_ELEMS(inter->sad); i++) {
> > + inter->sad[i] = vvc_sad;
> > + }
> > }
> >
> > #undef FUNCS
>
> Why is the jump depending upon block width not performed inside your
> avx2 implementation?
>
> - Andreas
>
Hi Andreas,
Sorry missed your email,
In hindsight, there's no particular reason, besides that it was the easiest
way (for me) to get jumps to different functions.
I guess I could just use compares to block width and jumps? Or
alternatively figure out how to write a jump table in asm.
Would those methods be better or did you have something different in mind?
Thanks for the feedback!
Stone
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
