third_party/libdav1d/src/lr_apply_tmpl.c - cobalt - Git at Google

 /*
  * Copyright © 2018, VideoLAN and dav1d authors
  * Copyright © 2018, Two Orioles, LLC
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
  *
  * 1. Redistributions of source code must retain the above copyright notice, this
  *    list of conditions and the following disclaimer.
  *
  * 2. Redistributions in binary form must reproduce the above copyright notice,
  *    this list of conditions and the following disclaimer in the documentation
  *    and/or other materials provided with the distribution.
  *
  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */

 #include "config.h"

 #include <stdio.h>

 #include "common/intops.h"

 #include "src/lr_apply.h"

 enum LrRestorePlanes {
     LR_RESTORE_Y = 1 << 0,
     LR_RESTORE_U = 1 << 1,
     LR_RESTORE_V = 1 << 2,
 };

 // The loop filter buffer stores 12 rows of pixels. A superblock block will
 // contain at most 2 stripes. Each stripe requires 4 rows pixels (2 above
 // and 2 below) the final 4 rows are used to swap the bottom of the last
 // stripe with the top of the next super block row.
 static void backup_lpf(const Dav1dFrameContext *const f,
                        pixel *dst, const ptrdiff_t dst_stride,
                        const pixel *src, const ptrdiff_t src_stride,
                        const int ss_ver, const int sb128,
                        int row, const int row_h, const int src_w,
                        const int h, const int ss_hor)
 {
     const int dst_w = f->frame_hdr->super_res.enabled ?
                       (f->frame_hdr->width[1] + ss_hor) >> ss_hor : src_w;

     // The first stripe of the frame is shorter by 8 luma pixel rows.
     int stripe_h = (64 - 8 * !row) >> ss_ver;

     if (row) {
         const int top = 4 << sb128;
         // Copy the top part of the stored loop filtered pixels from the
         // previous sb row needed above the first stripe of this sb row.
         pixel_copy(&dst[PXSTRIDE(dst_stride) *  0],
                    &dst[PXSTRIDE(dst_stride) *  top],      dst_w);
         pixel_copy(&dst[PXSTRIDE(dst_stride) *  1],
                    &dst[PXSTRIDE(dst_stride) * (top + 1)], dst_w);
         pixel_copy(&dst[PXSTRIDE(dst_stride) *  2],
                    &dst[PXSTRIDE(dst_stride) * (top + 2)], dst_w);
         pixel_copy(&dst[PXSTRIDE(dst_stride) *  3],
                    &dst[PXSTRIDE(dst_stride) * (top + 3)], dst_w);
     }

     dst += 4 * PXSTRIDE(dst_stride);
     src += (stripe_h - 2) * PXSTRIDE(src_stride);

     if (f->frame_hdr->super_res.enabled) {
         while (row + stripe_h <= row_h) {
             const int n_lines = 4 - (row + stripe_h + 1 == h);
             f->dsp->mc.resize(dst, dst_stride, src, src_stride,
                               dst_w, src_w, n_lines, f->resize_step[ss_hor],
                               f->resize_start[ss_hor] HIGHBD_CALL_SUFFIX);
             row += stripe_h; // unmodified stripe_h for the 1st stripe
             stripe_h = 64 >> ss_ver;
             src += stripe_h * PXSTRIDE(src_stride);
             dst += n_lines * PXSTRIDE(dst_stride);
             if (n_lines == 3) {
                 pixel_copy(dst, &dst[-PXSTRIDE(dst_stride)], dst_w);
                 dst += PXSTRIDE(dst_stride);
             }
         }
     } else {
         while (row + stripe_h <= row_h) {
             const int n_lines = 4 - (row + stripe_h + 1 == h);
             for (int i = 0; i < 4; i++) {
                 pixel_copy(dst, i == n_lines ? &dst[-PXSTRIDE(dst_stride)] :
                                                src, src_w);
                 dst += PXSTRIDE(dst_stride);
                 src += PXSTRIDE(src_stride);
             }
             row += stripe_h; // unmodified stripe_h for the 1st stripe
             stripe_h = 64 >> ss_ver;
             src += (stripe_h - 4) * PXSTRIDE(src_stride);
         }
     }
 }

 void bytefn(dav1d_lr_copy_lpf)(Dav1dFrameContext *const f,
                                /*const*/ pixel *const src[3], const int sby)
 {
     const int offset = 8 * !!sby;
     const ptrdiff_t *const src_stride = f->cur.stride;
     const ptrdiff_t lr_stride = ((f->sr_cur.p.p.w + 31) & ~31) * sizeof(pixel);

     // TODO Also check block level restore type to reduce copying.
     const int restore_planes = f->lf.restore_planes;

     if (restore_planes & LR_RESTORE_Y) {
         const int h = f->cur.p.h;
         const int w = f->bw << 2;
         const int row_h = imin((sby + 1) << (6 + f->seq_hdr->sb128), h - 1);
         const int y_stripe = (sby << (6 + f->seq_hdr->sb128)) - offset;
         backup_lpf(f, f->lf.lr_lpf_line[0], lr_stride,
                    src[0] - offset * PXSTRIDE(src_stride[0]), src_stride[0],
                    0, f->seq_hdr->sb128, y_stripe, row_h, w, h, 0);
     }
     if (restore_planes & (LR_RESTORE_U | LR_RESTORE_V)) {
         const int ss_ver = f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
         const int ss_hor = f->sr_cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444;
         const int h = (f->cur.p.h + ss_ver) >> ss_ver;
         const int w = f->bw << (2 - ss_hor);
         const int row_h = imin((sby + 1) << ((6 - ss_ver) + f->seq_hdr->sb128), h - 1);
         const int offset_uv = offset >> ss_ver;
         const int y_stripe =
             (sby << ((6 - ss_ver) + f->seq_hdr->sb128)) - offset_uv;

         if (restore_planes & LR_RESTORE_U) {
             backup_lpf(f, f->lf.lr_lpf_line[1], lr_stride,
                        src[1] - offset_uv * PXSTRIDE(src_stride[1]), src_stride[1],
                        ss_ver, f->seq_hdr->sb128, y_stripe, row_h, w, h, ss_hor);
         }
         if (restore_planes & LR_RESTORE_V) {
             backup_lpf(f, f->lf.lr_lpf_line[2], lr_stride,
                        src[2] - offset_uv * PXSTRIDE(src_stride[1]), src_stride[1],
                        ss_ver, f->seq_hdr->sb128, y_stripe, row_h, w, h, ss_hor);
         }
     }
 }

 static void lr_stripe(const Dav1dFrameContext *const f, pixel *p,
                       const pixel (*left)[4], int x, int y,
                       const int plane, const int unit_w, const int row_h,
                       const Av1RestorationUnit *const lr, enum LrEdgeFlags edges)
 {
     const Dav1dDSPContext *const dsp = f->dsp;
     const int chroma = !!plane;
     const int ss_ver = chroma & (f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420);
     const int sbrow_has_bottom = (edges & LR_HAVE_BOTTOM);
     const pixel *lpf = f->lf.lr_lpf_line[plane] + x;
     const ptrdiff_t p_stride = f->sr_cur.p.stride[chroma];
     const ptrdiff_t lpf_stride = sizeof(pixel) * ((f->sr_cur.p.p.w + 31) & ~31);

     // The first stripe of the frame is shorter by 8 luma pixel rows.
     int stripe_h = imin((64 - 8 * !y) >> ss_ver, row_h - y);

     // FIXME [8] might be easier for SIMD
     int16_t filterh[7], filterv[7];
     if (lr->type == DAV1D_RESTORATION_WIENER) {
         filterh[0] = filterh[6] = lr->filter_h[0];
         filterh[1] = filterh[5] = lr->filter_h[1];
         filterh[2] = filterh[4] = lr->filter_h[2];
         filterh[3] = -((filterh[0] + filterh[1] + filterh[2]) * 2);

         filterv[0] = filterv[6] = lr->filter_v[0];
         filterv[1] = filterv[5] = lr->filter_v[1];
         filterv[2] = filterv[4] = lr->filter_v[2];
         filterv[3] = -((filterv[0] + filterv[1] + filterv[2]) * 2);
     }

     while (y + stripe_h <= row_h) {
         // Change HAVE_BOTTOM bit in edges to (y + stripe_h != row_h)
         edges ^= (-(y + stripe_h != row_h) ^ edges) & LR_HAVE_BOTTOM;
         if (lr->type == DAV1D_RESTORATION_WIENER) {
             dsp->lr.wiener(p, p_stride, left, lpf, lpf_stride, unit_w, stripe_h,
                            filterh, filterv, edges HIGHBD_CALL_SUFFIX);
         } else {
             assert(lr->type == DAV1D_RESTORATION_SGRPROJ);
             dsp->lr.selfguided(p, p_stride, left, lpf, lpf_stride, unit_w, stripe_h,
                                lr->sgr_idx, lr->sgr_weights, edges HIGHBD_CALL_SUFFIX);
         }

         left += stripe_h;
         y += stripe_h;
         if (y + stripe_h > row_h && sbrow_has_bottom) break;
         p += stripe_h * PXSTRIDE(p_stride);
         edges |= LR_HAVE_TOP;
         stripe_h = imin(64 >> ss_ver, row_h - y);
         if (stripe_h == 0) break;
         lpf += 4 * PXSTRIDE(lpf_stride);
     }
 }

 static void backup4xU(pixel (*dst)[4], const pixel *src, const ptrdiff_t src_stride,
                       int u)
 {
     for (; u > 0; u--, dst++, src += PXSTRIDE(src_stride))
         pixel_copy(dst, src, 4);
 }

 static void lr_sbrow(const Dav1dFrameContext *const f, pixel *p, const int y,
                      const int w, const int h, const int row_h, const int plane)
 {
     const int chroma = !!plane;
     const int ss_ver = chroma & (f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420);
     const int ss_hor = chroma & (f->sr_cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444);
     const ptrdiff_t p_stride = f->sr_cur.p.stride[chroma];

     const int unit_size_log2 = f->frame_hdr->restoration.unit_size[!!plane];
     const int unit_size = 1 << unit_size_log2;
     const int half_unit_size = unit_size >> 1;
     const int max_unit_size = unit_size + half_unit_size;

     // Y coordinate of the sbrow (y is 8 luma pixel rows above row_y)
     const int row_y = y + ((8 >> ss_ver) * !!y);

     // FIXME This is an ugly hack to lookup the proper AV1Filter unit for
     // chroma planes. Question: For Multithreaded decoding, is it better
     // to store the chroma LR information with collocated Luma information?
     // In other words. For a chroma restoration unit locate at 128,128 and
     // with a 4:2:0 chroma subsampling, do we store the filter information at
     // the AV1Filter unit located at (128,128) or (256,256)
     // TODO Support chroma subsampling.
     const int shift_hor = 7 - ss_hor;

     pixel pre_lr_border[2][128 + 8 /* maximum sbrow height is 128 + 8 rows offset */][4];
     const Av1RestorationUnit *lr[2];

     enum LrEdgeFlags edges = (y > 0 ? LR_HAVE_TOP : 0) | LR_HAVE_RIGHT |
                              (row_h < h ? LR_HAVE_BOTTOM : 0);

     int aligned_unit_pos = row_y & ~(unit_size - 1);
     if (aligned_unit_pos && aligned_unit_pos + half_unit_size > h)
         aligned_unit_pos -= unit_size;
     aligned_unit_pos <<= ss_ver;
     const int sb_idx = (aligned_unit_pos >> 7) * f->sr_sb128w;
     const int unit_idx = ((aligned_unit_pos >> 6) & 1) << 1;
     lr[0] = &f->lf.lr_mask[sb_idx].lr[plane][unit_idx];
     int restore = lr[0]->type != DAV1D_RESTORATION_NONE;
     int x = 0, bit = 0;
     for (; x + max_unit_size <= w; p += unit_size, edges |= LR_HAVE_LEFT, bit ^= 1) {
         const int next_x = x + unit_size;
         const int next_u_idx = unit_idx + ((next_x >> (shift_hor - 1)) & 1);
         lr[!bit] =
             &f->lf.lr_mask[sb_idx + (next_x >> shift_hor)].lr[plane][next_u_idx];
         const int restore_next = lr[!bit]->type != DAV1D_RESTORATION_NONE;
         if (restore_next)
             backup4xU(pre_lr_border[bit], p + unit_size - 4, p_stride, row_h - y);
         if (restore)
             lr_stripe(f, p, pre_lr_border[!bit], x, y, plane, unit_size, row_h,
                       lr[bit], edges);
         x = next_x;
         restore = restore_next;
     }
     if (restore) {
         edges &= ~LR_HAVE_RIGHT;
         const int unit_w = w - x;
         lr_stripe(f, p, pre_lr_border[!bit], x, y, plane, unit_w, row_h, lr[bit], edges);
     }
 }

 void bytefn(dav1d_lr_sbrow)(Dav1dFrameContext *const f, pixel *const dst[3],
                             const int sby)
 {
     const int offset_y = 8 * !!sby;
     const ptrdiff_t *const dst_stride = f->sr_cur.p.stride;
     const int restore_planes = f->lf.restore_planes;

     if (restore_planes & LR_RESTORE_Y) {
         const int h = f->sr_cur.p.p.h;
         const int w = f->sr_cur.p.p.w;
         const int row_h = imin((sby + 1) << (6 + f->seq_hdr->sb128), h);
         const int y_stripe = (sby << (6 + f->seq_hdr->sb128)) - offset_y;
         lr_sbrow(f, dst[0] - offset_y * PXSTRIDE(dst_stride[0]), y_stripe, w,
                  h, row_h, 0);
     }
     if (restore_planes & (LR_RESTORE_U | LR_RESTORE_V)) {
         const int ss_ver = f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
         const int ss_hor = f->sr_cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444;
         const int h = (f->sr_cur.p.p.h + ss_ver) >> ss_ver;
         const int w = (f->sr_cur.p.p.w + ss_hor) >> ss_hor;
         const int row_h = imin((sby + 1) << ((6 - ss_ver) + f->seq_hdr->sb128), h);
         const int offset_uv = offset_y >> ss_ver;
         const int y_stripe =
             (sby << ((6 - ss_ver) + f->seq_hdr->sb128)) - offset_uv;
         if (restore_planes & LR_RESTORE_U)
             lr_sbrow(f, dst[1] - offset_uv * PXSTRIDE(dst_stride[1]), y_stripe,
                      w, h, row_h, 1);

         if (restore_planes & LR_RESTORE_V)
             lr_sbrow(f, dst[2] - offset_uv * PXSTRIDE(dst_stride[1]), y_stripe,
                      w, h, row_h, 2);
     }
 }
	/*
	* Copyright © 2018, VideoLAN and dav1d authors
	* Copyright © 2018, Two Orioles, LLC
	* All rights reserved.
	*
	* Redistribution and use in source and binary forms, with or without
	* modification, are permitted provided that the following conditions are met:
	*
	* 1. Redistributions of source code must retain the above copyright notice, this
	* list of conditions and the following disclaimer.
	*
	* 2. Redistributions in binary form must reproduce the above copyright notice,
	* this list of conditions and the following disclaimer in the documentation
	* and/or other materials provided with the distribution.
	*
	* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
	* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
	* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
	* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
	* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
	* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
	* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
	* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
	* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
	*/

	#include "config.h"

	#include <stdio.h>

	#include "common/intops.h"

	#include "src/lr_apply.h"

	enum LrRestorePlanes {
	LR_RESTORE_Y = 1 << 0,
	LR_RESTORE_U = 1 << 1,
	LR_RESTORE_V = 1 << 2,
	};

	// The loop filter buffer stores 12 rows of pixels. A superblock block will
	// contain at most 2 stripes. Each stripe requires 4 rows pixels (2 above
	// and 2 below) the final 4 rows are used to swap the bottom of the last
	// stripe with the top of the next super block row.
	static void backup_lpf(const Dav1dFrameContext *const f,
	pixel *dst, const ptrdiff_t dst_stride,
	const pixel *src, const ptrdiff_t src_stride,
	const int ss_ver, const int sb128,
	int row, const int row_h, const int src_w,
	const int h, const int ss_hor)
	{
	const int dst_w = f->frame_hdr->super_res.enabled ?
	(f->frame_hdr->width[1] + ss_hor) >> ss_hor : src_w;

	// The first stripe of the frame is shorter by 8 luma pixel rows.
	int stripe_h = (64 - 8 * !row) >> ss_ver;

	if (row) {
	const int top = 4 << sb128;
	// Copy the top part of the stored loop filtered pixels from the
	// previous sb row needed above the first stripe of this sb row.
	pixel_copy(&dst[PXSTRIDE(dst_stride) * 0],
	&dst[PXSTRIDE(dst_stride) * top], dst_w);
	pixel_copy(&dst[PXSTRIDE(dst_stride) * 1],
	&dst[PXSTRIDE(dst_stride) * (top + 1)], dst_w);
	pixel_copy(&dst[PXSTRIDE(dst_stride) * 2],
	&dst[PXSTRIDE(dst_stride) * (top + 2)], dst_w);
	pixel_copy(&dst[PXSTRIDE(dst_stride) * 3],
	&dst[PXSTRIDE(dst_stride) * (top + 3)], dst_w);
	}

	dst += 4 * PXSTRIDE(dst_stride);
	src += (stripe_h - 2) * PXSTRIDE(src_stride);

	if (f->frame_hdr->super_res.enabled) {
	while (row + stripe_h <= row_h) {
	const int n_lines = 4 - (row + stripe_h + 1 == h);
	f->dsp->mc.resize(dst, dst_stride, src, src_stride,
	dst_w, src_w, n_lines, f->resize_step[ss_hor],
	f->resize_start[ss_hor] HIGHBD_CALL_SUFFIX);
	row += stripe_h; // unmodified stripe_h for the 1st stripe
	stripe_h = 64 >> ss_ver;
	src += stripe_h * PXSTRIDE(src_stride);
	dst += n_lines * PXSTRIDE(dst_stride);
	if (n_lines == 3) {
	pixel_copy(dst, &dst[-PXSTRIDE(dst_stride)], dst_w);
	dst += PXSTRIDE(dst_stride);
	}
	}
	} else {
	while (row + stripe_h <= row_h) {
	const int n_lines = 4 - (row + stripe_h + 1 == h);
	for (int i = 0; i < 4; i++) {
	pixel_copy(dst, i == n_lines ? &dst[-PXSTRIDE(dst_stride)] :
	src, src_w);
	dst += PXSTRIDE(dst_stride);
	src += PXSTRIDE(src_stride);
	}
	row += stripe_h; // unmodified stripe_h for the 1st stripe
	stripe_h = 64 >> ss_ver;
	src += (stripe_h - 4) * PXSTRIDE(src_stride);
	}
	}
	}

	void bytefn(dav1d_lr_copy_lpf)(Dav1dFrameContext *const f,
	/const/ pixel *const src[3], const int sby)
	{
	const int offset = 8 * !!sby;
	const ptrdiff_t *const src_stride = f->cur.stride;
	const ptrdiff_t lr_stride = ((f->sr_cur.p.p.w + 31) & ~31) * sizeof(pixel);

	// TODO Also check block level restore type to reduce copying.
	const int restore_planes = f->lf.restore_planes;

	if (restore_planes & LR_RESTORE_Y) {
	const int h = f->cur.p.h;
	const int w = f->bw << 2;
	const int row_h = imin((sby + 1) << (6 + f->seq_hdr->sb128), h - 1);
	const int y_stripe = (sby << (6 + f->seq_hdr->sb128)) - offset;
	backup_lpf(f, f->lf.lr_lpf_line[0], lr_stride,
	src[0] - offset * PXSTRIDE(src_stride[0]), src_stride[0],
	0, f->seq_hdr->sb128, y_stripe, row_h, w, h, 0);
	}
	if (restore_planes & (LR_RESTORE_U \| LR_RESTORE_V)) {
	const int ss_ver = f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
	const int ss_hor = f->sr_cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444;
	const int h = (f->cur.p.h + ss_ver) >> ss_ver;
	const int w = f->bw << (2 - ss_hor);
	const int row_h = imin((sby + 1) << ((6 - ss_ver) + f->seq_hdr->sb128), h - 1);
	const int offset_uv = offset >> ss_ver;
	const int y_stripe =
	(sby << ((6 - ss_ver) + f->seq_hdr->sb128)) - offset_uv;

	if (restore_planes & LR_RESTORE_U) {
	backup_lpf(f, f->lf.lr_lpf_line[1], lr_stride,
	src[1] - offset_uv * PXSTRIDE(src_stride[1]), src_stride[1],
	ss_ver, f->seq_hdr->sb128, y_stripe, row_h, w, h, ss_hor);
	}
	if (restore_planes & LR_RESTORE_V) {
	backup_lpf(f, f->lf.lr_lpf_line[2], lr_stride,
	src[2] - offset_uv * PXSTRIDE(src_stride[1]), src_stride[1],
	ss_ver, f->seq_hdr->sb128, y_stripe, row_h, w, h, ss_hor);
	}
	}
	}

	static void lr_stripe(const Dav1dFrameContext const f, pixel p,
	const pixel (*left)[4], int x, int y,
	const int plane, const int unit_w, const int row_h,
	const Av1RestorationUnit *const lr, enum LrEdgeFlags edges)
	{
	const Dav1dDSPContext *const dsp = f->dsp;
	const int chroma = !!plane;
	const int ss_ver = chroma & (f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420);
	const int sbrow_has_bottom = (edges & LR_HAVE_BOTTOM);
	const pixel *lpf = f->lf.lr_lpf_line[plane] + x;
	const ptrdiff_t p_stride = f->sr_cur.p.stride[chroma];
	const ptrdiff_t lpf_stride = sizeof(pixel) * ((f->sr_cur.p.p.w + 31) & ~31);

	// The first stripe of the frame is shorter by 8 luma pixel rows.
	int stripe_h = imin((64 - 8 * !y) >> ss_ver, row_h - y);

	// FIXME [8] might be easier for SIMD
	int16_t filterh[7], filterv[7];
	if (lr->type == DAV1D_RESTORATION_WIENER) {
	filterh[0] = filterh[6] = lr->filter_h[0];
	filterh[1] = filterh[5] = lr->filter_h[1];
	filterh[2] = filterh[4] = lr->filter_h[2];
	filterh[3] = -((filterh[0] + filterh[1] + filterh[2]) * 2);

	filterv[0] = filterv[6] = lr->filter_v[0];
	filterv[1] = filterv[5] = lr->filter_v[1];
	filterv[2] = filterv[4] = lr->filter_v[2];
	filterv[3] = -((filterv[0] + filterv[1] + filterv[2]) * 2);
	}

	while (y + stripe_h <= row_h) {
	// Change HAVE_BOTTOM bit in edges to (y + stripe_h != row_h)
	edges ^= (-(y + stripe_h != row_h) ^ edges) & LR_HAVE_BOTTOM;
	if (lr->type == DAV1D_RESTORATION_WIENER) {
	dsp->lr.wiener(p, p_stride, left, lpf, lpf_stride, unit_w, stripe_h,
	filterh, filterv, edges HIGHBD_CALL_SUFFIX);
	} else {
	assert(lr->type == DAV1D_RESTORATION_SGRPROJ);
	dsp->lr.selfguided(p, p_stride, left, lpf, lpf_stride, unit_w, stripe_h,
	lr->sgr_idx, lr->sgr_weights, edges HIGHBD_CALL_SUFFIX);
	}

	left += stripe_h;
	y += stripe_h;
	if (y + stripe_h > row_h && sbrow_has_bottom) break;
	p += stripe_h * PXSTRIDE(p_stride);
	edges \|= LR_HAVE_TOP;
	stripe_h = imin(64 >> ss_ver, row_h - y);
	if (stripe_h == 0) break;
	lpf += 4 * PXSTRIDE(lpf_stride);
	}
	}

	static void backup4xU(pixel (dst)[4], const pixel src, const ptrdiff_t src_stride,
	int u)
	{
	for (; u > 0; u--, dst++, src += PXSTRIDE(src_stride))
	pixel_copy(dst, src, 4);
	}

	static void lr_sbrow(const Dav1dFrameContext const f, pixel p, const int y,
	const int w, const int h, const int row_h, const int plane)
	{
	const int chroma = !!plane;
	const int ss_ver = chroma & (f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420);
	const int ss_hor = chroma & (f->sr_cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444);
	const ptrdiff_t p_stride = f->sr_cur.p.stride[chroma];

	const int unit_size_log2 = f->frame_hdr->restoration.unit_size[!!plane];
	const int unit_size = 1 << unit_size_log2;
	const int half_unit_size = unit_size >> 1;
	const int max_unit_size = unit_size + half_unit_size;

	// Y coordinate of the sbrow (y is 8 luma pixel rows above row_y)
	const int row_y = y + ((8 >> ss_ver) * !!y);

	// FIXME This is an ugly hack to lookup the proper AV1Filter unit for
	// chroma planes. Question: For Multithreaded decoding, is it better
	// to store the chroma LR information with collocated Luma information?
	// In other words. For a chroma restoration unit locate at 128,128 and
	// with a 4:2:0 chroma subsampling, do we store the filter information at
	// the AV1Filter unit located at (128,128) or (256,256)
	// TODO Support chroma subsampling.
	const int shift_hor = 7 - ss_hor;

	pixel pre_lr_border[2][128 + 8 /* maximum sbrow height is 128 + 8 rows offset */][4];
	const Av1RestorationUnit *lr[2];

	enum LrEdgeFlags edges = (y > 0 ? LR_HAVE_TOP : 0) \| LR_HAVE_RIGHT \|
	(row_h < h ? LR_HAVE_BOTTOM : 0);

	int aligned_unit_pos = row_y & ~(unit_size - 1);
	if (aligned_unit_pos && aligned_unit_pos + half_unit_size > h)
	aligned_unit_pos -= unit_size;
	aligned_unit_pos <<= ss_ver;
	const int sb_idx = (aligned_unit_pos >> 7) * f->sr_sb128w;
	const int unit_idx = ((aligned_unit_pos >> 6) & 1) << 1;
	lr[0] = &f->lf.lr_mask[sb_idx].lr[plane][unit_idx];
	int restore = lr[0]->type != DAV1D_RESTORATION_NONE;
	int x = 0, bit = 0;
	for (; x + max_unit_size <= w; p += unit_size, edges \|= LR_HAVE_LEFT, bit ^= 1) {
	const int next_x = x + unit_size;
	const int next_u_idx = unit_idx + ((next_x >> (shift_hor - 1)) & 1);
	lr[!bit] =
	&f->lf.lr_mask[sb_idx + (next_x >> shift_hor)].lr[plane][next_u_idx];
	const int restore_next = lr[!bit]->type != DAV1D_RESTORATION_NONE;
	if (restore_next)
	backup4xU(pre_lr_border[bit], p + unit_size - 4, p_stride, row_h - y);
	if (restore)
	lr_stripe(f, p, pre_lr_border[!bit], x, y, plane, unit_size, row_h,
	lr[bit], edges);
	x = next_x;
	restore = restore_next;
	}
	if (restore) {
	edges &= ~LR_HAVE_RIGHT;
	const int unit_w = w - x;
	lr_stripe(f, p, pre_lr_border[!bit], x, y, plane, unit_w, row_h, lr[bit], edges);
	}
	}

	void bytefn(dav1d_lr_sbrow)(Dav1dFrameContext const f, pixel const dst[3],
	const int sby)
	{
	const int offset_y = 8 * !!sby;
	const ptrdiff_t *const dst_stride = f->sr_cur.p.stride;
	const int restore_planes = f->lf.restore_planes;

	if (restore_planes & LR_RESTORE_Y) {
	const int h = f->sr_cur.p.p.h;
	const int w = f->sr_cur.p.p.w;
	const int row_h = imin((sby + 1) << (6 + f->seq_hdr->sb128), h);
	const int y_stripe = (sby << (6 + f->seq_hdr->sb128)) - offset_y;
	lr_sbrow(f, dst[0] - offset_y * PXSTRIDE(dst_stride[0]), y_stripe, w,
	h, row_h, 0);
	}
	if (restore_planes & (LR_RESTORE_U \| LR_RESTORE_V)) {
	const int ss_ver = f->sr_cur.p.p.layout == DAV1D_PIXEL_LAYOUT_I420;
	const int ss_hor = f->sr_cur.p.p.layout != DAV1D_PIXEL_LAYOUT_I444;
	const int h = (f->sr_cur.p.p.h + ss_ver) >> ss_ver;
	const int w = (f->sr_cur.p.p.w + ss_hor) >> ss_hor;
	const int row_h = imin((sby + 1) << ((6 - ss_ver) + f->seq_hdr->sb128), h);
	const int offset_uv = offset_y >> ss_ver;
	const int y_stripe =
	(sby << ((6 - ss_ver) + f->seq_hdr->sb128)) - offset_uv;
	if (restore_planes & LR_RESTORE_U)
	lr_sbrow(f, dst[1] - offset_uv * PXSTRIDE(dst_stride[1]), y_stripe,
	w, h, row_h, 1);

	if (restore_planes & LR_RESTORE_V)
	lr_sbrow(f, dst[2] - offset_uv * PXSTRIDE(dst_stride[1]), y_stripe,
	w, h, row_h, 2);
	}
	}