vp9: split out loopfilter functions in their own source file.
authorRonald S. Bultje <rsbultje@gmail.com>
Mon, 27 Mar 2017 20:59:06 +0000 (16:59 -0400)
committerRonald S. Bultje <rsbultje@gmail.com>
Tue, 28 Mar 2017 22:04:26 +0000 (18:04 -0400)
libavcodec/Makefile
libavcodec/vp9.c
libavcodec/vp9dec.h
libavcodec/vp9lpf.c [new file with mode: 0644]

index cf7ba25..5651526 100644 (file)
@@ -610,7 +610,7 @@ OBJS-$(CONFIG_VP8_DECODER)             += vp8.o vp56rac.o
 OBJS-$(CONFIG_VP8_CUVID_DECODER)       += cuvid.o
 OBJS-$(CONFIG_VP8_MEDIACODEC_DECODER)  += mediacodecdec.o
 OBJS-$(CONFIG_VP8_VAAPI_ENCODER)       += vaapi_encode_vp8.o
-OBJS-$(CONFIG_VP9_DECODER)             += vp9.o vp9data.o vp9dsp.o \
+OBJS-$(CONFIG_VP9_DECODER)             += vp9.o vp9data.o vp9dsp.o vp9lpf.o \
                                           vp9block.o vp9prob.o vp9mvs.o vp56rac.o \
                                           vp9dsp_8bpp.o vp9dsp_10bpp.o vp9dsp_12bpp.o
 OBJS-$(CONFIG_VP9_CUVID_DECODER)       += cuvid.o
index 01b6a1b..4d7310f 100644 (file)
@@ -1056,184 +1056,6 @@ static void decode_sb_mem(AVCodecContext *avctx, int row, int col, VP9Filter *lf
     }
 }
 
-static av_always_inline void filter_plane_cols(VP9Context *s, int col, int ss_h, int ss_v,
-                                               uint8_t *lvl, uint8_t (*mask)[4],
-                                               uint8_t *dst, ptrdiff_t ls)
-{
-    int y, x, bytesperpixel = s->bytesperpixel;
-
-    // filter edges between columns (e.g. block1 | block2)
-    for (y = 0; y < 8; y += 2 << ss_v, dst += 16 * ls, lvl += 16 << ss_v) {
-        uint8_t *ptr = dst, *l = lvl, *hmask1 = mask[y], *hmask2 = mask[y + 1 + ss_v];
-        unsigned hm1 = hmask1[0] | hmask1[1] | hmask1[2], hm13 = hmask1[3];
-        unsigned hm2 = hmask2[1] | hmask2[2], hm23 = hmask2[3];
-        unsigned hm = hm1 | hm2 | hm13 | hm23;
-
-        for (x = 1; hm & ~(x - 1); x <<= 1, ptr += 8 * bytesperpixel >> ss_h) {
-            if (col || x > 1) {
-                if (hm1 & x) {
-                    int L = *l, H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    if (hmask1[0] & x) {
-                        if (hmask2[0] & x) {
-                            av_assert2(l[8 << ss_v] == L);
-                            s->dsp.loop_filter_16[0](ptr, ls, E, I, H);
-                        } else {
-                            s->dsp.loop_filter_8[2][0](ptr, ls, E, I, H);
-                        }
-                    } else if (hm2 & x) {
-                        L = l[8 << ss_v];
-                        H |= (L >> 4) << 8;
-                        E |= s->filter_lut.mblim_lut[L] << 8;
-                        I |= s->filter_lut.lim_lut[L] << 8;
-                        s->dsp.loop_filter_mix2[!!(hmask1[1] & x)]
-                                               [!!(hmask2[1] & x)]
-                                               [0](ptr, ls, E, I, H);
-                    } else {
-                        s->dsp.loop_filter_8[!!(hmask1[1] & x)]
-                                            [0](ptr, ls, E, I, H);
-                    }
-                } else if (hm2 & x) {
-                    int L = l[8 << ss_v], H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    s->dsp.loop_filter_8[!!(hmask2[1] & x)]
-                                        [0](ptr + 8 * ls, ls, E, I, H);
-                }
-            }
-            if (ss_h) {
-                if (x & 0xAA)
-                    l += 2;
-            } else {
-                if (hm13 & x) {
-                    int L = *l, H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    if (hm23 & x) {
-                        L = l[8 << ss_v];
-                        H |= (L >> 4) << 8;
-                        E |= s->filter_lut.mblim_lut[L] << 8;
-                        I |= s->filter_lut.lim_lut[L] << 8;
-                        s->dsp.loop_filter_mix2[0][0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
-                    } else {
-                        s->dsp.loop_filter_8[0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
-                    }
-                } else if (hm23 & x) {
-                    int L = l[8 << ss_v], H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    s->dsp.loop_filter_8[0][0](ptr + 8 * ls + 4 * bytesperpixel, ls, E, I, H);
-                }
-                l++;
-            }
-        }
-    }
-}
-
-static av_always_inline void filter_plane_rows(VP9Context *s, int row, int ss_h, int ss_v,
-                                               uint8_t *lvl, uint8_t (*mask)[4],
-                                               uint8_t *dst, ptrdiff_t ls)
-{
-    int y, x, bytesperpixel = s->bytesperpixel;
-
-    //                                 block1
-    // filter edges between rows (e.g. ------)
-    //                                 block2
-    for (y = 0; y < 8; y++, dst += 8 * ls >> ss_v) {
-        uint8_t *ptr = dst, *l = lvl, *vmask = mask[y];
-        unsigned vm = vmask[0] | vmask[1] | vmask[2], vm3 = vmask[3];
-
-        for (x = 1; vm & ~(x - 1); x <<= (2 << ss_h), ptr += 16 * bytesperpixel, l += 2 << ss_h) {
-            if (row || y) {
-                if (vm & x) {
-                    int L = *l, H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    if (vmask[0] & x) {
-                        if (vmask[0] & (x << (1 + ss_h))) {
-                            av_assert2(l[1 + ss_h] == L);
-                            s->dsp.loop_filter_16[1](ptr, ls, E, I, H);
-                        } else {
-                            s->dsp.loop_filter_8[2][1](ptr, ls, E, I, H);
-                        }
-                    } else if (vm & (x << (1 + ss_h))) {
-                        L = l[1 + ss_h];
-                        H |= (L >> 4) << 8;
-                        E |= s->filter_lut.mblim_lut[L] << 8;
-                        I |= s->filter_lut.lim_lut[L] << 8;
-                        s->dsp.loop_filter_mix2[!!(vmask[1] &  x)]
-                                               [!!(vmask[1] & (x << (1 + ss_h)))]
-                                               [1](ptr, ls, E, I, H);
-                    } else {
-                        s->dsp.loop_filter_8[!!(vmask[1] & x)]
-                                            [1](ptr, ls, E, I, H);
-                    }
-                } else if (vm & (x << (1 + ss_h))) {
-                    int L = l[1 + ss_h], H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    s->dsp.loop_filter_8[!!(vmask[1] & (x << (1 + ss_h)))]
-                                        [1](ptr + 8 * bytesperpixel, ls, E, I, H);
-                }
-            }
-            if (!ss_v) {
-                if (vm3 & x) {
-                    int L = *l, H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    if (vm3 & (x << (1 + ss_h))) {
-                        L = l[1 + ss_h];
-                        H |= (L >> 4) << 8;
-                        E |= s->filter_lut.mblim_lut[L] << 8;
-                        I |= s->filter_lut.lim_lut[L] << 8;
-                        s->dsp.loop_filter_mix2[0][0][1](ptr + ls * 4, ls, E, I, H);
-                    } else {
-                        s->dsp.loop_filter_8[0][1](ptr + ls * 4, ls, E, I, H);
-                    }
-                } else if (vm3 & (x << (1 + ss_h))) {
-                    int L = l[1 + ss_h], H = L >> 4;
-                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
-
-                    s->dsp.loop_filter_8[0][1](ptr + ls * 4 + 8 * bytesperpixel, ls, E, I, H);
-                }
-            }
-        }
-        if (ss_v) {
-            if (y & 1)
-                lvl += 16;
-        } else {
-            lvl += 8;
-        }
-    }
-}
-
-static void loopfilter_sb(AVCodecContext *avctx, VP9Filter *lflvl,
-                          int row, int col, ptrdiff_t yoff, ptrdiff_t uvoff)
-{
-    VP9Context *s = avctx->priv_data;
-    AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
-    uint8_t *dst = f->data[0] + yoff;
-    ptrdiff_t ls_y = f->linesize[0], ls_uv = f->linesize[1];
-    uint8_t (*uv_masks)[8][4] = lflvl->mask[s->ss_h | s->ss_v];
-    int p;
-
-    /* FIXME: In how far can we interleave the v/h loopfilter calls? E.g.
-     * if you think of them as acting on a 8x8 block max, we can interleave
-     * each v/h within the single x loop, but that only works if we work on
-     * 8 pixel blocks, and we won't always do that (we want at least 16px
-     * to use SSE2 optimizations, perhaps 32 for AVX2) */
-
-    filter_plane_cols(s, col, 0, 0, lflvl->level, lflvl->mask[0][0], dst, ls_y);
-    filter_plane_rows(s, row, 0, 0, lflvl->level, lflvl->mask[0][1], dst, ls_y);
-
-    for (p = 0; p < 2; p++) {
-        dst = f->data[1 + p] + uvoff;
-        filter_plane_cols(s, col, s->ss_h, s->ss_v, lflvl->level, uv_masks[0], dst, ls_uv);
-        filter_plane_rows(s, row, s->ss_h, s->ss_v, lflvl->level, uv_masks[1], dst, ls_uv);
-    }
-}
-
 static void set_tile_offset(int *start, int *end, int idx, int log2_n, int n)
 {
     int sb_start = ( idx      * n) >> log2_n;
@@ -1522,7 +1344,8 @@ FF_ENABLE_DEPRECATION_WARNINGS
                     for (col = 0; col < s->cols;
                          col += 8, yoff2 += 64 * bytesperpixel,
                          uvoff2 += 64 * bytesperpixel >> s->ss_h, lflvl_ptr++) {
-                        loopfilter_sb(avctx, lflvl_ptr, row, col, yoff2, uvoff2);
+                        ff_vp9_loopfilter_sb(avctx, lflvl_ptr, row, col,
+                                             yoff2, uvoff2);
                     }
                 }
 
index ba821f3..1f3348a 100644 (file)
@@ -203,4 +203,7 @@ void ff_vp9_decode_block(AVCodecContext *ctx, int row, int col,
                          VP9Filter *lflvl, ptrdiff_t yoff, ptrdiff_t uvoff,
                          enum BlockLevel bl, enum BlockPartition bp);
 
+void ff_vp9_loopfilter_sb(AVCodecContext *avctx, VP9Filter *lflvl,
+                          int row, int col, ptrdiff_t yoff, ptrdiff_t uvoff);
+
 #endif /* AVCODEC_VP9DEC_H */
diff --git a/libavcodec/vp9lpf.c b/libavcodec/vp9lpf.c
new file mode 100644 (file)
index 0000000..414cede
--- /dev/null
@@ -0,0 +1,202 @@
+/*
+ * VP9 compatible video decoder
+ *
+ * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
+ * Copyright (C) 2013 Clément Bœsch <u pkh me>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "vp9dec.h"
+
+static av_always_inline void filter_plane_cols(VP9Context *s, int col, int ss_h, int ss_v,
+                                               uint8_t *lvl, uint8_t (*mask)[4],
+                                               uint8_t *dst, ptrdiff_t ls)
+{
+    int y, x, bytesperpixel = s->bytesperpixel;
+
+    // filter edges between columns (e.g. block1 | block2)
+    for (y = 0; y < 8; y += 2 << ss_v, dst += 16 * ls, lvl += 16 << ss_v) {
+        uint8_t *ptr = dst, *l = lvl, *hmask1 = mask[y], *hmask2 = mask[y + 1 + ss_v];
+        unsigned hm1 = hmask1[0] | hmask1[1] | hmask1[2], hm13 = hmask1[3];
+        unsigned hm2 = hmask2[1] | hmask2[2], hm23 = hmask2[3];
+        unsigned hm = hm1 | hm2 | hm13 | hm23;
+
+        for (x = 1; hm & ~(x - 1); x <<= 1, ptr += 8 * bytesperpixel >> ss_h) {
+            if (col || x > 1) {
+                if (hm1 & x) {
+                    int L = *l, H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    if (hmask1[0] & x) {
+                        if (hmask2[0] & x) {
+                            av_assert2(l[8 << ss_v] == L);
+                            s->dsp.loop_filter_16[0](ptr, ls, E, I, H);
+                        } else {
+                            s->dsp.loop_filter_8[2][0](ptr, ls, E, I, H);
+                        }
+                    } else if (hm2 & x) {
+                        L = l[8 << ss_v];
+                        H |= (L >> 4) << 8;
+                        E |= s->filter_lut.mblim_lut[L] << 8;
+                        I |= s->filter_lut.lim_lut[L] << 8;
+                        s->dsp.loop_filter_mix2[!!(hmask1[1] & x)]
+                                               [!!(hmask2[1] & x)]
+                                               [0](ptr, ls, E, I, H);
+                    } else {
+                        s->dsp.loop_filter_8[!!(hmask1[1] & x)]
+                                            [0](ptr, ls, E, I, H);
+                    }
+                } else if (hm2 & x) {
+                    int L = l[8 << ss_v], H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    s->dsp.loop_filter_8[!!(hmask2[1] & x)]
+                                        [0](ptr + 8 * ls, ls, E, I, H);
+                }
+            }
+            if (ss_h) {
+                if (x & 0xAA)
+                    l += 2;
+            } else {
+                if (hm13 & x) {
+                    int L = *l, H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    if (hm23 & x) {
+                        L = l[8 << ss_v];
+                        H |= (L >> 4) << 8;
+                        E |= s->filter_lut.mblim_lut[L] << 8;
+                        I |= s->filter_lut.lim_lut[L] << 8;
+                        s->dsp.loop_filter_mix2[0][0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
+                    } else {
+                        s->dsp.loop_filter_8[0][0](ptr + 4 * bytesperpixel, ls, E, I, H);
+                    }
+                } else if (hm23 & x) {
+                    int L = l[8 << ss_v], H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    s->dsp.loop_filter_8[0][0](ptr + 8 * ls + 4 * bytesperpixel, ls, E, I, H);
+                }
+                l++;
+            }
+        }
+    }
+}
+
+static av_always_inline void filter_plane_rows(VP9Context *s, int row, int ss_h, int ss_v,
+                                               uint8_t *lvl, uint8_t (*mask)[4],
+                                               uint8_t *dst, ptrdiff_t ls)
+{
+    int y, x, bytesperpixel = s->bytesperpixel;
+
+    //                                 block1
+    // filter edges between rows (e.g. ------)
+    //                                 block2
+    for (y = 0; y < 8; y++, dst += 8 * ls >> ss_v) {
+        uint8_t *ptr = dst, *l = lvl, *vmask = mask[y];
+        unsigned vm = vmask[0] | vmask[1] | vmask[2], vm3 = vmask[3];
+
+        for (x = 1; vm & ~(x - 1); x <<= (2 << ss_h), ptr += 16 * bytesperpixel, l += 2 << ss_h) {
+            if (row || y) {
+                if (vm & x) {
+                    int L = *l, H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    if (vmask[0] & x) {
+                        if (vmask[0] & (x << (1 + ss_h))) {
+                            av_assert2(l[1 + ss_h] == L);
+                            s->dsp.loop_filter_16[1](ptr, ls, E, I, H);
+                        } else {
+                            s->dsp.loop_filter_8[2][1](ptr, ls, E, I, H);
+                        }
+                    } else if (vm & (x << (1 + ss_h))) {
+                        L = l[1 + ss_h];
+                        H |= (L >> 4) << 8;
+                        E |= s->filter_lut.mblim_lut[L] << 8;
+                        I |= s->filter_lut.lim_lut[L] << 8;
+                        s->dsp.loop_filter_mix2[!!(vmask[1] &  x)]
+                                               [!!(vmask[1] & (x << (1 + ss_h)))]
+                                               [1](ptr, ls, E, I, H);
+                    } else {
+                        s->dsp.loop_filter_8[!!(vmask[1] & x)]
+                                            [1](ptr, ls, E, I, H);
+                    }
+                } else if (vm & (x << (1 + ss_h))) {
+                    int L = l[1 + ss_h], H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    s->dsp.loop_filter_8[!!(vmask[1] & (x << (1 + ss_h)))]
+                                        [1](ptr + 8 * bytesperpixel, ls, E, I, H);
+                }
+            }
+            if (!ss_v) {
+                if (vm3 & x) {
+                    int L = *l, H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    if (vm3 & (x << (1 + ss_h))) {
+                        L = l[1 + ss_h];
+                        H |= (L >> 4) << 8;
+                        E |= s->filter_lut.mblim_lut[L] << 8;
+                        I |= s->filter_lut.lim_lut[L] << 8;
+                        s->dsp.loop_filter_mix2[0][0][1](ptr + ls * 4, ls, E, I, H);
+                    } else {
+                        s->dsp.loop_filter_8[0][1](ptr + ls * 4, ls, E, I, H);
+                    }
+                } else if (vm3 & (x << (1 + ss_h))) {
+                    int L = l[1 + ss_h], H = L >> 4;
+                    int E = s->filter_lut.mblim_lut[L], I = s->filter_lut.lim_lut[L];
+
+                    s->dsp.loop_filter_8[0][1](ptr + ls * 4 + 8 * bytesperpixel, ls, E, I, H);
+                }
+            }
+        }
+        if (ss_v) {
+            if (y & 1)
+                lvl += 16;
+        } else {
+            lvl += 8;
+        }
+    }
+}
+
+void ff_vp9_loopfilter_sb(AVCodecContext *avctx, VP9Filter *lflvl,
+                          int row, int col, ptrdiff_t yoff, ptrdiff_t uvoff)
+{
+    VP9Context *s = avctx->priv_data;
+    AVFrame *f = s->s.frames[CUR_FRAME].tf.f;
+    uint8_t *dst = f->data[0] + yoff;
+    ptrdiff_t ls_y = f->linesize[0], ls_uv = f->linesize[1];
+    uint8_t (*uv_masks)[8][4] = lflvl->mask[s->ss_h | s->ss_v];
+    int p;
+
+    /* FIXME: In how far can we interleave the v/h loopfilter calls? E.g.
+     * if you think of them as acting on a 8x8 block max, we can interleave
+     * each v/h within the single x loop, but that only works if we work on
+     * 8 pixel blocks, and we won't always do that (we want at least 16px
+     * to use SSE2 optimizations, perhaps 32 for AVX2) */
+
+    filter_plane_cols(s, col, 0, 0, lflvl->level, lflvl->mask[0][0], dst, ls_y);
+    filter_plane_rows(s, row, 0, 0, lflvl->level, lflvl->mask[0][1], dst, ls_y);
+
+    for (p = 0; p < 2; p++) {
+        dst = f->data[1 + p] + uvoff;
+        filter_plane_cols(s, col, s->ss_h, s->ss_v, lflvl->level, uv_masks[0], dst, ls_uv);
+        filter_plane_rows(s, row, s->ss_h, s->ss_v, lflvl->level, uv_masks[1], dst, ls_uv);
+    }
+}