| // Copyright 2026 The Wuffs Authors. |
| // |
| // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
| // https://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
| // <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your |
| // option. This file may not be copied, modified, or distributed |
| // except according to those terms. |
| // |
| // SPDX-License-Identifier: Apache-2.0 OR MIT |
| |
| // -------- |
| |
| // Reconstructing YUV samples involves adding predictions (based on previous |
| // YUV samples, in macroblocks one column left or one column above) to |
| // residuals (decoded from the bitstream, dequantized and IDCT'ed). |
| // |
| // There are 16×16 = 256 luma and 2×8×8 = 128 chroma samples per macroblock. |
| // The decoder.yuv_cache field is a bit larger than that, being an array[26] |
| // array[32] base.u8. Here's the layout. |
| // |
| // 0 1 2 3 4 5 6 7 8 9 A B C D E F 0 1 2 3 4 5 6 7 8 9 A B C D E F |
| // . . . . . . . a b b b b b b b b b b b b b b b b c c c c . . . . 00 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 01 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 02 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 03 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y c c c c . . . . 04 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 05 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 06 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 07 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y c c c c . . . . 08 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 09 |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 0A |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 0B |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y c c c c . . . . 0C |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 0D |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 0E |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 0F |
| // . . . . . . . d Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y . . . . . . . . 10 |
| // . . . . . . . e f f f f f f f f . . . . . . . g h h h h h h h h 11 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 12 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 13 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 14 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 15 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 16 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 17 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 18 |
| // . . . . . . . i U U U U U U U U . . . . . . . j V V V V V V V V 19 |
| // |
| // Y, U and V are this macroblock's luma and chroma values. |
| // |
| // The Y predictions (per 4×4 subblock) depend on the row above's Y values |
| // (either abc or dYc) and the column left's Y values (ad or bY). |
| // |
| // The U predictions (per 8×8 block) depend on the row above's U values (ef) |
| // and the column left's U values (i). |
| // |
| // The V predictions (per 8×8 block) depend on the row above's V values (gh) |
| // and the column left's V values (j). |
| // |
| // For top-row (mby == 0) macroblocks, abcefgh is initialized to 0x7F. For |
| // other macroblocks, these are copied from the bottom-most row of the |
| // macroblock above (before filtering, for I-Frames), from the workbuf. |
| // |
| // The c values are then duplicated from row 0x00 to rows 0x04, 0x08 and 0x0C. |
| // |
| // For left-column (mbx == 0) macroblocks, adeigj are initialized to 0x81 (but |
| // aeg is 0x7F if mby is also 0). For other macroblocks, these are copied from |
| // the right-most column of the macroblock to the left (before filtering, for |
| // I-Frames), from the yuv_cache to itself, shifted by 8 or 16 columns. |
| |
| pri func decoder.reconstruct!( |
| workbuf: slice base.u8, |
| mbx: base.u32[..= 0x3FF], |
| mby: base.u32[..= 0x3FF], |
| luma_mode: base.u32[..= 4], |
| chroma_mode: base.u32[..= 3]) { |
| var mode : base.u32[..= 12] |
| var b : base.u32 |
| var mask : base.u32 |
| |
| this.prepare_yuv_cache!(workbuf: args.workbuf, mbx: args.mbx, mby: args.mby) |
| |
| if args.luma_mode < 4 { |
| mode = this.substitute_dc_top_left(mode: args.luma_mode, mbx: args.mbx, mby: args.mby) |
| this.predict_y16!(mode: mode) |
| } |
| |
| b = 0 |
| while b < 16 { |
| if args.luma_mode >= 4 { |
| this.predict_y4!(b: b) |
| } |
| |
| mask = (1 as base.u32) << (b + 0x00) |
| if (this.mb_ac_nz & mask) <> 0 { |
| this.inverse_dct_full!( |
| cachex: ((b & 0x3) * 4) + 0x08, |
| cachey: ((b & 0xC) * 1) + 0x01, |
| b: (b + 0x00)) |
| } else if (this.mb_dc_nz & mask) <> 0 { |
| this.inverse_dct_dc_only!( |
| cachex: ((b & 0x3) * 4) + 0x08, |
| cachey: ((b & 0xC) * 1) + 0x01, |
| b: (b + 0x00)) |
| } |
| b += 1 |
| } |
| |
| mode = this.substitute_dc_top_left(mode: args.chroma_mode, mbx: args.mbx, mby: args.mby) |
| |
| this.predict_uv8!(b: 0, mode: mode) |
| |
| if (this.mb_ac_nz & 0x0F_0000) <> 0 { |
| b = 0 |
| while b < 4 { |
| this.inverse_dct_full!( |
| cachex: ((b & 0x1) * 4) + 0x08, |
| cachey: ((b & 0x2) * 2) + 0x12, |
| b: (b + 0x10)) |
| b += 1 |
| } |
| } else if (this.mb_dc_nz & 0x0F_0000) <> 0 { |
| b = 0 |
| while b < 4 { |
| this.inverse_dct_dc_only!( |
| cachex: ((b & 0x1) * 4) + 0x08, |
| cachey: ((b & 0x2) * 2) + 0x12, |
| b: (b + 0x10)) |
| b += 1 |
| } |
| } |
| |
| this.predict_uv8!(b: 1, mode: mode) |
| |
| if (this.mb_ac_nz & 0xF0_0000) <> 0 { |
| b = 0 |
| while b < 4 { |
| this.inverse_dct_full!( |
| cachex: ((b & 0x1) * 4) + 0x18, |
| cachey: ((b & 0x2) * 2) + 0x12, |
| b: (b + 0x14)) |
| b += 1 |
| } |
| } else if (this.mb_dc_nz & 0xF0_0000) <> 0 { |
| b = 0 |
| while b < 4 { |
| this.inverse_dct_dc_only!( |
| cachex: ((b & 0x1) * 4) + 0x18, |
| cachey: ((b & 0x2) * 2) + 0x12, |
| b: (b + 0x14)) |
| b += 1 |
| } |
| } |
| } |
| |
| pri func decoder.substitute_dc_top_left( |
| mode: base.u32[..= 3], |
| mbx: base.u32[..= 0x3FF], |
| mby: base.u32[..= 0x3FF]) base.u32[..= 12] { |
| if args.mode == 0 { // DC. |
| if args.mbx == 0 { |
| if args.mby == 0 { |
| return 12 // DCTopLeft. |
| } else { |
| return 11 // DCLeft. |
| } |
| } else if args.mby == 0 { |
| return 10 // DCTop. |
| } |
| } |
| return args.mode |
| } |
| |
| pri func decoder.prepare_yuv_cache!( |
| workbuf: roslice base.u8, |
| mbx: base.u32[..= 0x3FF], |
| mby: base.u32[..= 0x3FF]) { |
| var i : base.u32 |
| var j : base.u32 |
| var offset : base.u64[..= 0x17FE_FFF8] |
| var o : base.u64 |
| |
| // For abcdefghij comments, see the layout at the top of this file. |
| |
| if args.mbx <= 0 { |
| // Initialize adeigj to 0x81. |
| j = 0x00 |
| while j < 0x11 { |
| this.yuv_cache[j][0x07] = 0x81 |
| j += 1 |
| } |
| while j < 0x1A { |
| this.yuv_cache[j][0x07] = 0x81 |
| this.yuv_cache[j][0x17] = 0x81 |
| j += 1 |
| } |
| |
| } else { |
| // Copy from the previous (left) macroblock, in yuv_cache, to adeigj. |
| j = 0x00 |
| while j < 0x11 { |
| this.yuv_cache[j][0x07] = this.yuv_cache[j][0x17] |
| j += 1 |
| } |
| while j < 0x1A { |
| this.yuv_cache[j][0x07] = this.yuv_cache[j][0x0F] |
| this.yuv_cache[j][0x17] = this.yuv_cache[j][0x1F] |
| j += 1 |
| } |
| } |
| |
| if args.mby <= 0 { |
| // Initialize abcefgh to 0x7F. |
| i = 0x00 |
| while i < 0x20 { |
| this.yuv_cache[0x00][i] = 0x7F |
| i += 1 |
| } |
| i = 0x00 |
| while i < 0x20 { |
| this.yuv_cache[0x11][i] = 0x7F |
| i += 1 |
| } |
| |
| } else { |
| // Copy from the previous (above) macroblock, in args.workbuf, to bcfh. |
| |
| offset = 0 + |
| (((((args.mby * 16) - 1) * this.workbuf_yuv_y_stride) + (args.mbx * 16)) as base.u64) |
| i = 0x00 |
| while i < 0x10, |
| inv args.mby > 0, |
| { |
| o = offset + (i as base.u64) |
| if o < args.workbuf.length() { |
| this.yuv_cache[0x00][0x08 + i] = args.workbuf[o] |
| } |
| i += 1 |
| } |
| |
| if (args.mbx + 1) < this.mbw { |
| while i < 0x14, |
| inv args.mby > 0, |
| { |
| o = offset + (i as base.u64) |
| if o < args.workbuf.length() { |
| this.yuv_cache[0x00][0x08 + i] = args.workbuf[o] |
| } |
| i += 1 |
| } |
| } else { |
| while i < 0x14, |
| inv args.mby > 0, |
| { |
| o = offset + (0x0F as base.u64) |
| if o < args.workbuf.length() { |
| this.yuv_cache[0x00][0x08 + i] = args.workbuf[o] |
| } |
| i += 1 |
| } |
| } |
| |
| offset = this.workbuf_yuv_y_end + |
| (((((args.mby * 8) - 1) * this.workbuf_yuv_uv_stride) + (args.mbx * 8)) as base.u64) |
| i = 0x00 |
| while i < 0x08, |
| inv args.mby > 0, |
| { |
| o = offset + (i as base.u64) |
| if o < args.workbuf.length() { |
| this.yuv_cache[0x11][0x08 + i] = args.workbuf[o] |
| } |
| i += 1 |
| } |
| |
| offset = this.workbuf_yuv_u_end + |
| (((((args.mby * 8) - 1) * this.workbuf_yuv_uv_stride) + (args.mbx * 8)) as base.u64) |
| i = 0x00 |
| while i < 0x08, |
| inv args.mby > 0, |
| { |
| o = offset + (i as base.u64) |
| if o < args.workbuf.length() { |
| this.yuv_cache[0x11][0x18 + i] = args.workbuf[o] |
| } |
| i += 1 |
| } |
| } |
| |
| // Duplicate the c values. |
| j = 0x04 |
| while j < 0x10 { |
| this.yuv_cache[j][0x18] = this.yuv_cache[0][0x18] |
| this.yuv_cache[j][0x19] = this.yuv_cache[0][0x19] |
| this.yuv_cache[j][0x1A] = this.yuv_cache[0][0x1A] |
| this.yuv_cache[j][0x1B] = this.yuv_cache[0][0x1B] |
| j += 4 |
| } |
| } |
