From ccfb03ecc4f64eee75b934671264e9b6a90ae7bd Mon Sep 17 00:00:00 2001 From: "Ronald S. Bultje" Date: Sat, 16 May 2015 08:36:19 -0400 Subject: vp9: fix scaled motion vector clipping for sub8x8 blocks. To match the obscure clipping bug behaviour in libvpx. --- libavcodec/vp9_mc_template.c | 104 ++++++++++++++++++++++--------------------- 1 file changed, 54 insertions(+), 50 deletions(-) (limited to 'libavcodec/vp9_mc_template.c') diff --git a/libavcodec/vp9_mc_template.c b/libavcodec/vp9_mc_template.c index 5e6ee870b8..f4eb4e56ac 100644 --- a/libavcodec/vp9_mc_template.c +++ b/libavcodec/vp9_mc_template.c @@ -57,11 +57,11 @@ static void FN(inter_pred)(AVCodecContext *ctx) if (b->bs == BS_8x4) { mc_luma_dir(s, mc[3][b->filter][0], s->dst[0], ls_y, ref1->data[0], ref1->linesize[0], tref1, - row << 3, col << 3, &b->mv[0][0], 8, 4, w1, h1, 0); + row << 3, col << 3, &b->mv[0][0],,,,, 8, 4, w1, h1, 0); mc_luma_dir(s, mc[3][b->filter][0], s->dst[0] + 4 * ls_y, ls_y, ref1->data[0], ref1->linesize[0], tref1, - (row << 3) + 4, col << 3, &b->mv[2][0], 8, 4, w1, h1, 0); + (row << 3) + 4, col << 3, &b->mv[2][0],,,,, 8, 4, w1, h1, 0); w1 = (w1 + s->ss_h) >> s->ss_h; if (s->ss_v) { h1 = (h1 + 1) >> 1; @@ -71,14 +71,14 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 2, col << (3 - s->ss_h), - &uvmv, 8 >> s->ss_h, 4, w1, h1, 0); + &uvmv,,,,, 8 >> s->ss_h, 4, w1, h1, 0); } else { mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][0], s->dst[1], s->dst[2], ls_uv, ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 3, col << (3 - s->ss_h), - &b->mv[0][0], 8 >> s->ss_h, 4, w1, h1, 0); + &b->mv[0][0],,,,, 8 >> s->ss_h, 4, w1, h1, 0); // BUG for 4:2:2 bs=8x4, libvpx uses the wrong block index // to get the motion vector for the bottom 4x4 block // https://code.google.com/p/webm/issues/detail?id=993 @@ -92,17 +92,17 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, (row << 3) + 4, col << (3 - s->ss_h), - &uvmv, 8 >> s->ss_h, 4, w1, h1, 0); + &uvmv,,,,, 8 >> s->ss_h, 4, w1, h1, 0); } if (b->comp) { mc_luma_dir(s, mc[3][b->filter][1], s->dst[0], ls_y, ref2->data[0], ref2->linesize[0], tref2, - row << 3, col << 3, &b->mv[0][1], 8, 4, w2, h2, 1); + row << 3, col << 3, &b->mv[0][1],,,,, 8, 4, w2, h2, 1); mc_luma_dir(s, mc[3][b->filter][1], s->dst[0] + 4 * ls_y, ls_y, ref2->data[0], ref2->linesize[0], tref2, - (row << 3) + 4, col << 3, &b->mv[2][1], 8, 4, w2, h2, 1); + (row << 3) + 4, col << 3, &b->mv[2][1],,,,, 8, 4, w2, h2, 1); w2 = (w2 + s->ss_h) >> s->ss_h; if (s->ss_v) { h2 = (h2 + 1) >> 1; @@ -112,14 +112,14 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 2, col << (3 - s->ss_h), - &uvmv, 8 >> s->ss_h, 4, w2, h2, 1); + &uvmv,,,,, 8 >> s->ss_h, 4, w2, h2, 1); } else { mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][1], s->dst[1], s->dst[2], ls_uv, ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 3, col << (3 - s->ss_h), - &b->mv[0][1], 8 >> s->ss_h, 4, w2, h2, 1); + &b->mv[0][1],,,,, 8 >> s->ss_h, 4, w2, h2, 1); // BUG for 4:2:2 bs=8x4, libvpx uses the wrong block index // to get the motion vector for the bottom 4x4 block // https://code.google.com/p/webm/issues/detail?id=993 @@ -133,16 +133,16 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, (row << 3) + 4, col << (3 - s->ss_h), - &uvmv, 8 >> s->ss_h, 4, w2, h2, 1); + &uvmv,,,,, 8 >> s->ss_h, 4, w2, h2, 1); } } } else if (b->bs == BS_4x8) { mc_luma_dir(s, mc[4][b->filter][0], s->dst[0], ls_y, ref1->data[0], ref1->linesize[0], tref1, - row << 3, col << 3, &b->mv[0][0], 4, 8, w1, h1, 0); + row << 3, col << 3, &b->mv[0][0],,,,, 4, 8, w1, h1, 0); mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4 * bytesperpixel, ls_y, ref1->data[0], ref1->linesize[0], tref1, - row << 3, (col << 3) + 4, &b->mv[1][0], 4, 8, w1, h1, 0); + row << 3, (col << 3) + 4, &b->mv[1][0],,,,, 4, 8, w1, h1, 0); h1 = (h1 + s->ss_v) >> s->ss_v; if (s->ss_h) { w1 = (w1 + 1) >> 1; @@ -152,30 +152,30 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << (3 - s->ss_v), col << 2, - &uvmv, 4, 8 >> s->ss_v, w1, h1, 0); + &uvmv,,,,, 4, 8 >> s->ss_v, w1, h1, 0); } else { mc_chroma_dir(s, mc[4][b->filter][0], s->dst[1], s->dst[2], ls_uv, ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << (3 - s->ss_v), col << 3, - &b->mv[0][0], 4, 8 >> s->ss_v, w1, h1, 0); + &b->mv[0][0],,,,, 4, 8 >> s->ss_v, w1, h1, 0); mc_chroma_dir(s, mc[4][b->filter][0], s->dst[1] + 4 * bytesperpixel, s->dst[2] + 4 * bytesperpixel, ls_uv, ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << (3 - s->ss_v), (col << 3) + 4, - &b->mv[1][0], 4, 8 >> s->ss_v, w1, h1, 0); + &b->mv[1][0],,,,, 4, 8 >> s->ss_v, w1, h1, 0); } if (b->comp) { mc_luma_dir(s, mc[4][b->filter][1], s->dst[0], ls_y, ref2->data[0], ref2->linesize[0], tref2, - row << 3, col << 3, &b->mv[0][1], 4, 8, w2, h2, 1); + row << 3, col << 3, &b->mv[0][1],,,,, 4, 8, w2, h2, 1); mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4 * bytesperpixel, ls_y, ref2->data[0], ref2->linesize[0], tref2, - row << 3, (col << 3) + 4, &b->mv[1][1], 4, 8, w2, h2, 1); + row << 3, (col << 3) + 4, &b->mv[1][1],,,,, 4, 8, w2, h2, 1); h2 = (h2 + s->ss_v) >> s->ss_v; if (s->ss_h) { w2 = (w2 + 1) >> 1; @@ -185,21 +185,21 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << (3 - s->ss_v), col << 2, - &uvmv, 4, 8 >> s->ss_v, w2, h2, 1); + &uvmv,,,,, 4, 8 >> s->ss_v, w2, h2, 1); } else { mc_chroma_dir(s, mc[4][b->filter][1], s->dst[1], s->dst[2], ls_uv, ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << (3 - s->ss_v), col << 3, - &b->mv[0][1], 4, 8 >> s->ss_v, w2, h2, 1); + &b->mv[0][1],,,,, 4, 8 >> s->ss_v, w2, h2, 1); mc_chroma_dir(s, mc[4][b->filter][1], s->dst[1] + 4 * bytesperpixel, s->dst[2] + 4 * bytesperpixel, ls_uv, ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << (3 - s->ss_v), (col << 3) + 4, - &b->mv[1][1], 4, 8 >> s->ss_v, w2, h2, 1); + &b->mv[1][1],,,,, 4, 8 >> s->ss_v, w2, h2, 1); } } } else @@ -211,18 +211,22 @@ static void FN(inter_pred)(AVCodecContext *ctx) // do a w8 instead of a w4 call mc_luma_dir(s, mc[4][b->filter][0], s->dst[0], ls_y, ref1->data[0], ref1->linesize[0], tref1, - row << 3, col << 3, &b->mv[0][0], 4, 4, w1, h1, 0); + row << 3, col << 3, &b->mv[0][0], + 0, 0, 8, 8, 4, 4, w1, h1, 0); mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4 * bytesperpixel, ls_y, ref1->data[0], ref1->linesize[0], tref1, - row << 3, (col << 3) + 4, &b->mv[1][0], 4, 4, w1, h1, 0); + row << 3, (col << 3) + 4, &b->mv[1][0], + 4, 0, 8, 8, 4, 4, w1, h1, 0); mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4 * ls_y, ls_y, ref1->data[0], ref1->linesize[0], tref1, - (row << 3) + 4, col << 3, &b->mv[2][0], 4, 4, w1, h1, 0); + (row << 3) + 4, col << 3, &b->mv[2][0], + 0, 4, 8, 8, 4, 4, w1, h1, 0); mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4 * ls_y + 4 * bytesperpixel, ls_y, ref1->data[0], ref1->linesize[0], tref1, - (row << 3) + 4, (col << 3) + 4, &b->mv[3][0], 4, 4, w1, h1, 0); + (row << 3) + 4, (col << 3) + 4, &b->mv[3][0], + 4, 4, 8, 8, 4, 4, w1, h1, 0); if (s->ss_v) { h1 = (h1 + 1) >> 1; if (s->ss_h) { @@ -234,7 +238,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 2, col << 2, - &uvmv, 4, 4, w1, h1, 0); + &uvmv, 0, 0, 4, 4, 4, 4, w1, h1, 0); } else { uvmv = ROUNDED_DIV_MVx2(b->mv[0][0], b->mv[2][0]); mc_chroma_dir(s, mc[4][b->filter][0], @@ -242,7 +246,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 2, col << 3, - &uvmv, 4, 4, w1, h1, 0); + &uvmv, 0, 0, 8, 4, 4, 4, w1, h1, 0); uvmv = ROUNDED_DIV_MVx2(b->mv[1][0], b->mv[3][0]); mc_chroma_dir(s, mc[4][b->filter][0], s->dst[1] + 4 * bytesperpixel, @@ -250,7 +254,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 2, (col << 3) + 4, - &uvmv, 4, 4, w1, h1, 0); + &uvmv, 4, 0, 8, 4, 4, 4, w1, h1, 0); } } else { if (s->ss_h) { @@ -261,7 +265,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 3, col << 2, - &uvmv, 4, 4, w1, h1, 0); + &uvmv, 0, 0, 4, 8, 4, 4, w1, h1, 0); // BUG libvpx uses wrong block index for 4:2:2 bs=4x4 // bottom block // https://code.google.com/p/webm/issues/detail?id=993 @@ -271,52 +275,52 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, (row << 3) + 4, col << 2, - &uvmv, 4, 4, w1, h1, 0); + &uvmv, 0, 4, 4, 8, 4, 4, w1, h1, 0); } else { mc_chroma_dir(s, mc[4][b->filter][0], s->dst[1], s->dst[2], ls_uv, ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 3, col << 3, - &b->mv[0][0], 4, 4, w1, h1, 0); + &b->mv[0][0], 0, 0, 8, 8, 4, 4, w1, h1, 0); mc_chroma_dir(s, mc[4][b->filter][0], s->dst[1] + 4 * bytesperpixel, s->dst[2] + 4 * bytesperpixel, ls_uv, ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << 3, (col << 3) + 4, - &b->mv[1][0], 4, 4, w1, h1, 0); + &b->mv[1][0], 4, 0, 8, 8, 4, 4, w1, h1, 0); mc_chroma_dir(s, mc[4][b->filter][0], s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv, ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, (row << 3) + 4, col << 3, - &b->mv[2][0], 4, 4, w1, h1, 0); + &b->mv[2][0], 0, 4, 8, 8, 4, 4, w1, h1, 0); mc_chroma_dir(s, mc[4][b->filter][0], s->dst[1] + 4 * ls_uv + 4 * bytesperpixel, s->dst[2] + 4 * ls_uv + 4 * bytesperpixel, ls_uv, ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, (row << 3) + 4, (col << 3) + 4, - &b->mv[3][0], 4, 4, w1, h1, 0); + &b->mv[3][0], 4, 4, 8, 8, 4, 4, w1, h1, 0); } } if (b->comp) { mc_luma_dir(s, mc[4][b->filter][1], s->dst[0], ls_y, ref2->data[0], ref2->linesize[0], tref2, - row << 3, col << 3, &b->mv[0][1], 4, 4, w2, h2, 1); + row << 3, col << 3, &b->mv[0][1], 0, 0, 8, 8, 4, 4, w2, h2, 1); mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4 * bytesperpixel, ls_y, ref2->data[0], ref2->linesize[0], tref2, - row << 3, (col << 3) + 4, &b->mv[1][1], 4, 4, w2, h2, 1); + row << 3, (col << 3) + 4, &b->mv[1][1], 4, 0, 8, 8, 4, 4, w2, h2, 1); mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4 * ls_y, ls_y, ref2->data[0], ref2->linesize[0], tref2, - (row << 3) + 4, col << 3, &b->mv[2][1], 4, 4, w2, h2, 1); + (row << 3) + 4, col << 3, &b->mv[2][1], 0, 4, 8, 8, 4, 4, w2, h2, 1); mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4 * ls_y + 4 * bytesperpixel, ls_y, ref2->data[0], ref2->linesize[0], tref2, - (row << 3) + 4, (col << 3) + 4, &b->mv[3][1], 4, 4, w2, h2, 1); + (row << 3) + 4, (col << 3) + 4, &b->mv[3][1], 4, 4, 8, 8, 4, 4, w2, h2, 1); if (s->ss_v) { h2 = (h2 + 1) >> 1; if (s->ss_h) { @@ -328,7 +332,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 2, col << 2, - &uvmv, 4, 4, w2, h2, 1); + &uvmv, 0, 0, 4, 4, 4, 4, w2, h2, 1); } else { uvmv = ROUNDED_DIV_MVx2(b->mv[0][1], b->mv[2][1]); mc_chroma_dir(s, mc[4][b->filter][1], @@ -336,7 +340,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 2, col << 3, - &uvmv, 4, 4, w2, h2, 1); + &uvmv, 0, 0, 8, 4, 4, 4, w2, h2, 1); uvmv = ROUNDED_DIV_MVx2(b->mv[1][1], b->mv[3][1]); mc_chroma_dir(s, mc[4][b->filter][1], s->dst[1] + 4 * bytesperpixel, @@ -344,7 +348,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 2, (col << 3) + 4, - &uvmv, 4, 4, w2, h2, 1); + &uvmv, 4, 0, 8, 4, 4, 4, w2, h2, 1); } } else { if (s->ss_h) { @@ -355,7 +359,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 3, col << 2, - &uvmv, 4, 4, w2, h2, 1); + &uvmv, 0, 0, 4, 8, 4, 4, w2, h2, 1); // BUG libvpx uses wrong block index for 4:2:2 bs=4x4 // bottom block // https://code.google.com/p/webm/issues/detail?id=993 @@ -365,34 +369,34 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, (row << 3) + 4, col << 2, - &uvmv, 4, 4, w2, h2, 1); + &uvmv, 0, 4, 4, 8, 4, 4, w2, h2, 1); } else { mc_chroma_dir(s, mc[4][b->filter][1], s->dst[1], s->dst[2], ls_uv, ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 3, col << 3, - &b->mv[0][1], 4, 4, w2, h2, 1); + &b->mv[0][1], 0, 0, 8, 8, 4, 4, w2, h2, 1); mc_chroma_dir(s, mc[4][b->filter][1], s->dst[1] + 4 * bytesperpixel, s->dst[2] + 4 * bytesperpixel, ls_uv, ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << 3, (col << 3) + 4, - &b->mv[1][1], 4, 4, w2, h2, 1); + &b->mv[1][1], 4, 0, 8, 8, 4, 4, w2, h2, 1); mc_chroma_dir(s, mc[4][b->filter][1], s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv, ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, (row << 3) + 4, col << 3, - &b->mv[2][1], 4, 4, w2, h2, 1); + &b->mv[2][1], 0, 4, 8, 8, 4, 4, w2, h2, 1); mc_chroma_dir(s, mc[4][b->filter][1], s->dst[1] + 4 * ls_uv + 4 * bytesperpixel, s->dst[2] + 4 * ls_uv + 4 * bytesperpixel, ls_uv, ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, (row << 3) + 4, (col << 3) + 4, - &b->mv[3][1], 4, 4, w2, h2, 1); + &b->mv[3][1], 4, 4, 8, 8, 4, 4, w2, h2, 1); } } } @@ -404,7 +408,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) mc_luma_dir(s, mc[bwl][b->filter][0], s->dst[0], ls_y, ref1->data[0], ref1->linesize[0], tref1, - row << 3, col << 3, &b->mv[0][0], bw, bh, w1, h1, 0); + row << 3, col << 3, &b->mv[0][0], 0, 0, bw, bh, bw, bh, w1, h1, 0); w1 = (w1 + s->ss_h) >> s->ss_h; h1 = (h1 + s->ss_v) >> s->ss_v; mc_chroma_dir(s, mc[bwl + s->ss_h][b->filter][0], @@ -412,12 +416,12 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref1->data[1], ref1->linesize[1], ref1->data[2], ref1->linesize[2], tref1, row << (3 - s->ss_v), col << (3 - s->ss_h), - &b->mv[0][0], uvbw, uvbh, w1, h1, 0); + &b->mv[0][0], 0, 0, uvbw, uvbh, uvbw, uvbh, w1, h1, 0); if (b->comp) { mc_luma_dir(s, mc[bwl][b->filter][1], s->dst[0], ls_y, ref2->data[0], ref2->linesize[0], tref2, - row << 3, col << 3, &b->mv[0][1], bw, bh, w2, h2, 1); + row << 3, col << 3, &b->mv[0][1], 0, 0, bw, bh, bw, bh, w2, h2, 1); w2 = (w2 + s->ss_h) >> s->ss_h; h2 = (h2 + s->ss_v) >> s->ss_v; mc_chroma_dir(s, mc[bwl + s->ss_h][b->filter][1], @@ -425,7 +429,7 @@ static void FN(inter_pred)(AVCodecContext *ctx) ref2->data[1], ref2->linesize[1], ref2->data[2], ref2->linesize[2], tref2, row << (3 - s->ss_v), col << (3 - s->ss_h), - &b->mv[0][1], uvbw, uvbh, w2, h2, 1); + &b->mv[0][1], 0, 0, uvbw, uvbh, uvbw, uvbh, w2, h2, 1); } } } -- cgit v1.2.3