libav1/dx/shaders/restoration.h - av1-xbox-one - Git at Google

 /*
  * Copyright 2020 Google LLC
  *
  */

 /*
  * Copyright (c) 2016, Alliance for Open Media. All rights reserved
  *
  * This source code is subject to the terms of the BSD 2 Clause License and
  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
  * was not distributed with this source code in the LICENSE file, you can
  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
  * Media Patent License 1.0 was not distributed with this source code in the
  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
  */

 #ifndef AOM_AV1_COMMON_RESTORATION_H_
 #define AOM_AV1_COMMON_RESTORATION_H_

 #ifdef __cplusplus
 extern "C" {
 #endif

 // Border for Loop restoration buffer
 #define AOM_RESTORATION_FRAME_BORDER 32
 #define CLIP(x, lo, hi) ((x) < (lo) ? (lo) : (x) > (hi) ? (hi) : (x))
 #define RINT(x) ((x) < 0 ? (int)((x)-0.5) : (int)((x) + 0.5))

 #define RESTORATION_PROC_UNIT_SIZE 64

 // Filter tile grid offset upwards compared to the superblock grid
 #define RESTORATION_UNIT_OFFSET 8

 #define SGRPROJ_BORDER_VERT 3  // Vertical border used for Sgr
 #define SGRPROJ_BORDER_HORZ 3  // Horizontal border used for Sgr

 #define WIENER_BORDER_VERT 2  // Vertical border used for Wiener
 #define WIENER_HALFWIN 3
 #define WIENER_BORDER_HORZ (WIENER_HALFWIN)  // Horizontal border for Wiener

 // RESTORATION_BORDER_VERT determines line buffer requirement for LR.
 // Should be set at the max of SGRPROJ_BORDER_VERT and WIENER_BORDER_VERT.
 // Note the line buffer needed is twice the value of this macro.
 #if SGRPROJ_BORDER_VERT >= WIENER_BORDER_VERT
 #define RESTORATION_BORDER_VERT (SGRPROJ_BORDER_VERT)
 #else
 #define RESTORATION_BORDER_VERT (WIENER_BORDER_VERT)
 #endif  // SGRPROJ_BORDER_VERT >= WIENER_BORDER_VERT

 #if SGRPROJ_BORDER_HORZ >= WIENER_BORDER_HORZ
 #define RESTORATION_BORDER_HORZ (SGRPROJ_BORDER_HORZ)
 #else
 #define RESTORATION_BORDER_HORZ (WIENER_BORDER_HORZ)
 #endif  // SGRPROJ_BORDER_VERT >= WIENER_BORDER_VERT

 // How many border pixels do we need for each processing unit?
 #define RESTORATION_BORDER 3

 // How many rows of deblocked pixels do we save above/below each processing
 // stripe?
 #define RESTORATION_CTX_VERT 2

 // Additional pixels to the left and right in above/below buffers
 // It is RESTORATION_BORDER_HORZ rounded up to get nicer buffer alignment
 #define RESTORATION_EXTRA_HORZ 4

 // Pad up to 20 more (may be much less is needed)
 #define RESTORATION_PADDING 20
 #define RESTORATION_PROC_UNIT_PELS                                                    \
   ((RESTORATION_PROC_UNIT_SIZE + RESTORATION_BORDER_HORZ * 2 + RESTORATION_PADDING) * \
    (RESTORATION_PROC_UNIT_SIZE + RESTORATION_BORDER_VERT * 2 + RESTORATION_PADDING))

 #define RESTORATION_UNITSIZE_MAX 256
 #define RESTORATION_UNITPELS_HORZ_MAX (RESTORATION_UNITSIZE_MAX * 3 / 2 + 2 * RESTORATION_BORDER_HORZ + 16)
 #define RESTORATION_UNITPELS_VERT_MAX \
   ((RESTORATION_UNITSIZE_MAX * 3 / 2 + 2 * RESTORATION_BORDER_VERT + RESTORATION_UNIT_OFFSET))
 #define RESTORATION_UNITPELS_MAX (RESTORATION_UNITPELS_HORZ_MAX * RESTORATION_UNITPELS_VERT_MAX)

 // Two 32-bit buffers needed for the restored versions from two filters
 // TODO(debargha, rupert): Refactor to not need the large tilesize to be stored
 // on the decoder side.
 #define SGRPROJ_TMPBUF_SIZE (RESTORATION_UNITPELS_MAX * 2 * sizeof(int32_t))

 #define SGRPROJ_EXTBUF_SIZE (0)
 #define SGRPROJ_PARAMS_BITS 4
 #define SGRPROJ_PARAMS (1 << SGRPROJ_PARAMS_BITS)

 // Precision bits for projection
 #define SGRPROJ_PRJ_BITS 7
 // Restoration precision bits generated higher than source before projection
 #define SGRPROJ_RST_BITS 4
 // Internal precision bits for core selfguided_restoration
 #define SGRPROJ_SGR_BITS 8
 #define SGRPROJ_SGR (1 << SGRPROJ_SGR_BITS)

 #define SGRPROJ_PRJ_MIN0 (-(1 << SGRPROJ_PRJ_BITS) * 3 / 4)
 #define SGRPROJ_PRJ_MAX0 (SGRPROJ_PRJ_MIN0 + (1 << SGRPROJ_PRJ_BITS) - 1)
 #define SGRPROJ_PRJ_MIN1 (-(1 << SGRPROJ_PRJ_BITS) / 4)
 #define SGRPROJ_PRJ_MAX1 (SGRPROJ_PRJ_MIN1 + (1 << SGRPROJ_PRJ_BITS) - 1)

 #define SGRPROJ_PRJ_SUBEXP_K 4

 #define SGRPROJ_BITS (SGRPROJ_PRJ_BITS * 2 + SGRPROJ_PARAMS_BITS)

 #define MAX_RADIUS 2  // Only 1, 2, 3 allowed
 #define MAX_NELEM ((2 * MAX_RADIUS + 1) * (2 * MAX_RADIUS + 1))
 #define SGRPROJ_MTABLE_BITS 20
 #define SGRPROJ_RECIP_BITS 12

 #define WIENER_HALFWIN1 (WIENER_HALFWIN + 1)
 #define WIENER_WIN (2 * WIENER_HALFWIN + 1)
 #define WIENER_WIN2 ((WIENER_WIN) * (WIENER_WIN))
 #define WIENER_TMPBUF_SIZE (0)
 #define WIENER_EXTBUF_SIZE (0)

 // If WIENER_WIN_CHROMA == WIENER_WIN - 2, that implies 5x5 filters are used for
 // chroma. To use 7x7 for chroma set WIENER_WIN_CHROMA to WIENER_WIN.
 #define WIENER_WIN_CHROMA (WIENER_WIN - 2)
 #define WIENER_WIN2_CHROMA ((WIENER_WIN_CHROMA) * (WIENER_WIN_CHROMA))

 #define WIENER_FILT_PREC_BITS 7
 #define WIENER_FILT_STEP (1 << WIENER_FILT_PREC_BITS)

 // Central values for the taps
 #define WIENER_FILT_TAP0_MIDV (3)
 #define WIENER_FILT_TAP1_MIDV (-7)
 #define WIENER_FILT_TAP2_MIDV (15)
 #define WIENER_FILT_TAP3_MIDV \
   (WIENER_FILT_STEP - 2 * (WIENER_FILT_TAP0_MIDV + WIENER_FILT_TAP1_MIDV + WIENER_FILT_TAP2_MIDV))

 #define WIENER_FILT_TAP0_BITS 4
 #define WIENER_FILT_TAP1_BITS 5
 #define WIENER_FILT_TAP2_BITS 6

 #define WIENER_FILT_BITS ((WIENER_FILT_TAP0_BITS + WIENER_FILT_TAP1_BITS + WIENER_FILT_TAP2_BITS) * 2)

 #define WIENER_FILT_TAP0_MINV (WIENER_FILT_TAP0_MIDV - (1 << WIENER_FILT_TAP0_BITS) / 2)
 #define WIENER_FILT_TAP1_MINV (WIENER_FILT_TAP1_MIDV - (1 << WIENER_FILT_TAP1_BITS) / 2)
 #define WIENER_FILT_TAP2_MINV (WIENER_FILT_TAP2_MIDV - (1 << WIENER_FILT_TAP2_BITS) / 2)

 #define WIENER_FILT_TAP0_MAXV (WIENER_FILT_TAP0_MIDV - 1 + (1 << WIENER_FILT_TAP0_BITS) / 2)
 #define WIENER_FILT_TAP1_MAXV (WIENER_FILT_TAP1_MIDV - 1 + (1 << WIENER_FILT_TAP1_BITS) / 2)
 #define WIENER_FILT_TAP2_MAXV (WIENER_FILT_TAP2_MIDV - 1 + (1 << WIENER_FILT_TAP2_BITS) / 2)

 #define WIENER_FILT_TAP0_SUBEXP_K 1
 #define WIENER_FILT_TAP1_SUBEXP_K 2
 #define WIENER_FILT_TAP2_SUBEXP_K 3

 // Max of SGRPROJ_TMPBUF_SIZE, DOMAINTXFMRF_TMPBUF_SIZE, WIENER_TMPBUF_SIZE
 #define RESTORATION_TMPBUF_SIZE (SGRPROJ_TMPBUF_SIZE)

 // Max of SGRPROJ_EXTBUF_SIZE, WIENER_EXTBUF_SIZE
 #define RESTORATION_EXTBUF_SIZE (WIENER_EXTBUF_SIZE)

 #if WIENER_FILT_PREC_BITS != 7
 #error "Wiener filter currently only works if WIENER_FILT_PREC_BITS == 7"
 #endif

 #define LR_TILE_ROW 0
 #define LR_TILE_COL 0
 #define LR_TILE_COLS 1
 #ifdef __cplusplus
 }  // extern "C"
 #endif

 #endif  // AOM_AV1_COMMON_RESTORATION_H_
	/*
	* Copyright 2020 Google LLC
	*
	*/

	/*
	* Copyright (c) 2016, Alliance for Open Media. All rights reserved
	*
	* This source code is subject to the terms of the BSD 2 Clause License and
	* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
	* was not distributed with this source code in the LICENSE file, you can
	* obtain it at www.aomedia.org/license/software. If the Alliance for Open
	* Media Patent License 1.0 was not distributed with this source code in the
	* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
	*/

	#ifndef AOM_AV1_COMMON_RESTORATION_H_
	#define AOM_AV1_COMMON_RESTORATION_H_

	#ifdef __cplusplus
	extern "C" {
	#endif

	// Border for Loop restoration buffer
	#define AOM_RESTORATION_FRAME_BORDER 32
	#define CLIP(x, lo, hi) ((x) < (lo) ? (lo) : (x) > (hi) ? (hi) : (x))
	#define RINT(x) ((x) < 0 ? (int)((x)-0.5) : (int)((x) + 0.5))

	#define RESTORATION_PROC_UNIT_SIZE 64

	// Filter tile grid offset upwards compared to the superblock grid
	#define RESTORATION_UNIT_OFFSET 8

	#define SGRPROJ_BORDER_VERT 3 // Vertical border used for Sgr
	#define SGRPROJ_BORDER_HORZ 3 // Horizontal border used for Sgr

	#define WIENER_BORDER_VERT 2 // Vertical border used for Wiener
	#define WIENER_HALFWIN 3
	#define WIENER_BORDER_HORZ (WIENER_HALFWIN) // Horizontal border for Wiener

	// RESTORATION_BORDER_VERT determines line buffer requirement for LR.
	// Should be set at the max of SGRPROJ_BORDER_VERT and WIENER_BORDER_VERT.
	// Note the line buffer needed is twice the value of this macro.
	#if SGRPROJ_BORDER_VERT >= WIENER_BORDER_VERT
	#define RESTORATION_BORDER_VERT (SGRPROJ_BORDER_VERT)
	#else
	#define RESTORATION_BORDER_VERT (WIENER_BORDER_VERT)
	#endif // SGRPROJ_BORDER_VERT >= WIENER_BORDER_VERT

	#if SGRPROJ_BORDER_HORZ >= WIENER_BORDER_HORZ
	#define RESTORATION_BORDER_HORZ (SGRPROJ_BORDER_HORZ)
	#else
	#define RESTORATION_BORDER_HORZ (WIENER_BORDER_HORZ)
	#endif // SGRPROJ_BORDER_VERT >= WIENER_BORDER_VERT

	// How many border pixels do we need for each processing unit?
	#define RESTORATION_BORDER 3

	// How many rows of deblocked pixels do we save above/below each processing
	// stripe?
	#define RESTORATION_CTX_VERT 2

	// Additional pixels to the left and right in above/below buffers
	// It is RESTORATION_BORDER_HORZ rounded up to get nicer buffer alignment
	#define RESTORATION_EXTRA_HORZ 4

	// Pad up to 20 more (may be much less is needed)
	#define RESTORATION_PADDING 20
	#define RESTORATION_PROC_UNIT_PELS \
	((RESTORATION_PROC_UNIT_SIZE + RESTORATION_BORDER_HORZ * 2 + RESTORATION_PADDING) * \
	(RESTORATION_PROC_UNIT_SIZE + RESTORATION_BORDER_VERT * 2 + RESTORATION_PADDING))

	#define RESTORATION_UNITSIZE_MAX 256
	#define RESTORATION_UNITPELS_HORZ_MAX (RESTORATION_UNITSIZE_MAX * 3 / 2 + 2 * RESTORATION_BORDER_HORZ + 16)
	#define RESTORATION_UNITPELS_VERT_MAX \
	((RESTORATION_UNITSIZE_MAX * 3 / 2 + 2 * RESTORATION_BORDER_VERT + RESTORATION_UNIT_OFFSET))
	#define RESTORATION_UNITPELS_MAX (RESTORATION_UNITPELS_HORZ_MAX * RESTORATION_UNITPELS_VERT_MAX)

	// Two 32-bit buffers needed for the restored versions from two filters
	// TODO(debargha, rupert): Refactor to not need the large tilesize to be stored
	// on the decoder side.
	#define SGRPROJ_TMPBUF_SIZE (RESTORATION_UNITPELS_MAX * 2 * sizeof(int32_t))

	#define SGRPROJ_EXTBUF_SIZE (0)
	#define SGRPROJ_PARAMS_BITS 4
	#define SGRPROJ_PARAMS (1 << SGRPROJ_PARAMS_BITS)

	// Precision bits for projection
	#define SGRPROJ_PRJ_BITS 7
	// Restoration precision bits generated higher than source before projection
	#define SGRPROJ_RST_BITS 4
	// Internal precision bits for core selfguided_restoration
	#define SGRPROJ_SGR_BITS 8
	#define SGRPROJ_SGR (1 << SGRPROJ_SGR_BITS)

	#define SGRPROJ_PRJ_MIN0 (-(1 << SGRPROJ_PRJ_BITS) * 3 / 4)
	#define SGRPROJ_PRJ_MAX0 (SGRPROJ_PRJ_MIN0 + (1 << SGRPROJ_PRJ_BITS) - 1)
	#define SGRPROJ_PRJ_MIN1 (-(1 << SGRPROJ_PRJ_BITS) / 4)
	#define SGRPROJ_PRJ_MAX1 (SGRPROJ_PRJ_MIN1 + (1 << SGRPROJ_PRJ_BITS) - 1)

	#define SGRPROJ_PRJ_SUBEXP_K 4

	#define SGRPROJ_BITS (SGRPROJ_PRJ_BITS * 2 + SGRPROJ_PARAMS_BITS)

	#define MAX_RADIUS 2 // Only 1, 2, 3 allowed
	#define MAX_NELEM ((2 * MAX_RADIUS + 1) * (2 * MAX_RADIUS + 1))
	#define SGRPROJ_MTABLE_BITS 20
	#define SGRPROJ_RECIP_BITS 12

	#define WIENER_HALFWIN1 (WIENER_HALFWIN + 1)
	#define WIENER_WIN (2 * WIENER_HALFWIN + 1)
	#define WIENER_WIN2 ((WIENER_WIN) * (WIENER_WIN))
	#define WIENER_TMPBUF_SIZE (0)
	#define WIENER_EXTBUF_SIZE (0)

	// If WIENER_WIN_CHROMA == WIENER_WIN - 2, that implies 5x5 filters are used for
	// chroma. To use 7x7 for chroma set WIENER_WIN_CHROMA to WIENER_WIN.
	#define WIENER_WIN_CHROMA (WIENER_WIN - 2)
	#define WIENER_WIN2_CHROMA ((WIENER_WIN_CHROMA) * (WIENER_WIN_CHROMA))

	#define WIENER_FILT_PREC_BITS 7
	#define WIENER_FILT_STEP (1 << WIENER_FILT_PREC_BITS)

	// Central values for the taps
	#define WIENER_FILT_TAP0_MIDV (3)
	#define WIENER_FILT_TAP1_MIDV (-7)
	#define WIENER_FILT_TAP2_MIDV (15)
	#define WIENER_FILT_TAP3_MIDV \
	(WIENER_FILT_STEP - 2 * (WIENER_FILT_TAP0_MIDV + WIENER_FILT_TAP1_MIDV + WIENER_FILT_TAP2_MIDV))

	#define WIENER_FILT_TAP0_BITS 4
	#define WIENER_FILT_TAP1_BITS 5
	#define WIENER_FILT_TAP2_BITS 6

	#define WIENER_FILT_BITS ((WIENER_FILT_TAP0_BITS + WIENER_FILT_TAP1_BITS + WIENER_FILT_TAP2_BITS) * 2)

	#define WIENER_FILT_TAP0_MINV (WIENER_FILT_TAP0_MIDV - (1 << WIENER_FILT_TAP0_BITS) / 2)
	#define WIENER_FILT_TAP1_MINV (WIENER_FILT_TAP1_MIDV - (1 << WIENER_FILT_TAP1_BITS) / 2)
	#define WIENER_FILT_TAP2_MINV (WIENER_FILT_TAP2_MIDV - (1 << WIENER_FILT_TAP2_BITS) / 2)

	#define WIENER_FILT_TAP0_MAXV (WIENER_FILT_TAP0_MIDV - 1 + (1 << WIENER_FILT_TAP0_BITS) / 2)
	#define WIENER_FILT_TAP1_MAXV (WIENER_FILT_TAP1_MIDV - 1 + (1 << WIENER_FILT_TAP1_BITS) / 2)
	#define WIENER_FILT_TAP2_MAXV (WIENER_FILT_TAP2_MIDV - 1 + (1 << WIENER_FILT_TAP2_BITS) / 2)

	#define WIENER_FILT_TAP0_SUBEXP_K 1
	#define WIENER_FILT_TAP1_SUBEXP_K 2
	#define WIENER_FILT_TAP2_SUBEXP_K 3

	// Max of SGRPROJ_TMPBUF_SIZE, DOMAINTXFMRF_TMPBUF_SIZE, WIENER_TMPBUF_SIZE
	#define RESTORATION_TMPBUF_SIZE (SGRPROJ_TMPBUF_SIZE)

	// Max of SGRPROJ_EXTBUF_SIZE, WIENER_EXTBUF_SIZE
	#define RESTORATION_EXTBUF_SIZE (WIENER_EXTBUF_SIZE)

	#if WIENER_FILT_PREC_BITS != 7
	#error "Wiener filter currently only works if WIENER_FILT_PREC_BITS == 7"
	#endif

	#define LR_TILE_ROW 0
	#define LR_TILE_COL 0
	#define LR_TILE_COLS 1
	#ifdef __cplusplus
	} // extern "C"
	#endif

	#endif // AOM_AV1_COMMON_RESTORATION_H_