diff --git a/release/c/wuffs-unsupported-snapshot.c b/release/c/wuffs-unsupported-snapshot.c index 8e5da2ec..ad1b5b0f 100644 --- a/release/c/wuffs-unsupported-snapshot.c +++ b/release/c/wuffs-unsupported-snapshot.c @@ -81873,7 +81873,7 @@ wuffs_vp8__decoder__initialize_mb_coeffs( WUFFS_BASE__GENERATED_C_CODE static uint32_t -wuffs_vp8__decoder__decode_block_coefficients( +wuffs_vp8__decoder__decode_subblock_coefficients( wuffs_vp8__decoder* self, wuffs_base__slice_u8 a_workbuf, uint32_t a_mby, @@ -82376,7 +82376,7 @@ wuffs_vp8__decoder__decode_coefficients( self->private_data.f_mb_states_left &= 4286578687u; v_tnz = (1u & (self->private_data.f_mb_states_top[a_mbx] >> 23u)); self->private_data.f_mb_states_top[a_mbx] &= 4286578687u; - v_lnz = wuffs_vp8__decoder__decode_block_coefficients(self, + v_lnz = wuffs_vp8__decoder__decode_subblock_coefficients(self, a_workbuf, a_mby, a_seg, @@ -82397,7 +82397,7 @@ wuffs_vp8__decoder__decode_coefficients( v_tnz = (1u & (self->private_data.f_mb_states_top[a_mbx] >> (24u + v_bx))); self->private_data.f_mb_states_top[a_mbx] &= (4294967295u ^ (((uint32_t)(1u)) << (24u + v_bx))); v_b = ((4u * v_by) + v_bx); - v_lnz = wuffs_vp8__decoder__decode_block_coefficients(self, + v_lnz = wuffs_vp8__decoder__decode_subblock_coefficients(self, a_workbuf, a_mby, a_seg, @@ -82430,7 +82430,7 @@ wuffs_vp8__decoder__decode_coefficients( (2u * v_by) + v_bx + 16u); - v_lnz = wuffs_vp8__decoder__decode_block_coefficients(self, + v_lnz = wuffs_vp8__decoder__decode_subblock_coefficients(self, a_workbuf, a_mby, a_seg, @@ -82478,11 +82478,11 @@ wuffs_vp8__decoder__initialize_mb_coeffs( return wuffs_base__make_empty_struct(); } -// -------- func vp8.decoder.decode_block_coefficients +// -------- func vp8.decoder.decode_subblock_coefficients WUFFS_BASE__GENERATED_C_CODE static uint32_t -wuffs_vp8__decoder__decode_block_coefficients( +wuffs_vp8__decoder__decode_subblock_coefficients( wuffs_vp8__decoder* self, wuffs_base__slice_u8 a_workbuf, uint32_t a_mby, diff --git a/std/vp8/decode_bitstream.wuffs b/std/vp8/decode_bitstream.wuffs index 9b5da55c..3262a34a 100644 --- a/std/vp8/decode_bitstream.wuffs +++ b/std/vp8/decode_bitstream.wuffs @@ -8,6 +8,10 @@ // // SPDX-License-Identifier: Apache-2.0 OR MIT +// -------- + +// RFC 6386 Section 7. Boolean Entropy Decoder. + pri func decoder.read_bit!(workbuf: roslice base.u8, part: base.u32[..= 8], prob: base.u32[..= 0xFF]) base.u32[..= 1] { var p : base.u32[..= 8] var c8 : base.u8 diff --git a/std/vp8/decode_coefficients.wuffs b/std/vp8/decode_coefficients.wuffs index 8ad9d99b..94c8c067 100644 --- a/std/vp8/decode_coefficients.wuffs +++ b/std/vp8/decode_coefficients.wuffs @@ -8,6 +8,14 @@ // // SPDX-License-Identifier: Apache-2.0 OR MIT +// -------- + +// RFC 6386 Section 13. DCT Coefficient Decoding. +// +// and +// +// RFC 6386 Section 13.2. Coding of Individual Coefficient Values. + pri func decoder.decode_coefficients!( workbuf: slice base.u8, mbx: base.u32[..= 0x3FF], @@ -36,7 +44,7 @@ pri func decoder.decode_coefficients!( tnz = 1 & (this.mb_states_top[args.mbx] >> 23) this.mb_states_top[args.mbx] &= 0xFF7F_FFFF - lnz = this.decode_block_coefficients!( + lnz = this.decode_subblock_coefficients!( workbuf: args.workbuf, mby: args.mby, seg: args.seg, @@ -67,7 +75,7 @@ pri func decoder.decode_coefficients!( b = (4 * by) + bx - lnz = this.decode_block_coefficients!( + lnz = this.decode_subblock_coefficients!( workbuf: args.workbuf, mby: args.mby, seg: args.seg, @@ -110,7 +118,7 @@ pri func decoder.decode_coefficients!( this.mb_states_top[args.mbx] &= 0xFFFF_FFFF ^ ((1 as base.u32) << (28 + bc + bx)) b = (2 * bc) + (2 * by) + bx + 16 - lnz = this.decode_block_coefficients!( + lnz = this.decode_subblock_coefficients!( workbuf: args.workbuf, mby: args.mby, seg: args.seg, @@ -158,7 +166,7 @@ pri func decoder.initialize_mb_coeffs!() { } } -pri func decoder.decode_block_coefficients!( +pri func decoder.decode_subblock_coefficients!( workbuf: slice base.u8, mby: base.u32[..= 0x3FF], seg: base.u32[..= 3], diff --git a/std/vp8/decode_idct_default.wuffs b/std/vp8/decode_idct_default.wuffs index 65c31e20..d70a6e88 100644 --- a/std/vp8/decode_idct_default.wuffs +++ b/std/vp8/decode_idct_default.wuffs @@ -8,6 +8,10 @@ // // SPDX-License-Identifier: Apache-2.0 OR MIT +// -------- + +// A simplified and optimized version of decoder.inverse_dct_full when the AC +// coefficients are all zero. pri func decoder.inverse_dct_dc_only!( cachex: base.u32[..= 28], cachey: base.u32[..= 22], @@ -25,6 +29,9 @@ pri func decoder.inverse_dct_dc_only!( } } +// RFC 6386 Section 14.4. Implementation of the DCT Inversion. +// +// It reads from mb_coeffs and yuv_cache and writes to yuv_cache. pri func decoder.inverse_dct_full!( cachex: base.u32[..= 28], cachey: base.u32[..= 22], @@ -119,8 +126,10 @@ pri func decoder.inverse_dct_full!( var q32 : base.u32 var q33 : base.u32 - // 85627 is approximately 65536 * cos(pi/8) * sqrt(2). // 35468 is approximately 65536 * sin(pi/8) * sqrt(2). + // + // 85627 is approximately 65536 * cos(pi/8) * sqrt(2). Compared to the RFC + // implementation 85627 is also (20091 + 65536). // -------- @@ -299,6 +308,9 @@ pri func decoder.inverse_dct_full!( this.yuv_cache[args.cachey + 3][args.cachex + 3] as base.u32) ~mod+ q33)] } +// RFC 6386 Section 14.3. Implementation of the WHT Inversion. +// +// It reads from and writes to mb_coeffs. pri func decoder.inverse_wht!() { var a00 : base.u32 var a01 : base.u32 diff --git a/std/vp8/decode_macroblocks.wuffs b/std/vp8/decode_macroblocks.wuffs index 8bf1460f..3b959b39 100644 --- a/std/vp8/decode_macroblocks.wuffs +++ b/std/vp8/decode_macroblocks.wuffs @@ -28,6 +28,8 @@ pri func decoder.decode_macroblocks!(dst: ptr base.pixel_buffer, workbuf: slice this.mb_states_left = 0 + // Decode this row, using this.mb_coeffs and this.yuv_cache before + // ultimately writing to args.workbuf. mbx = 0 while mbx < this.mbw, inv mby < 0x400, @@ -39,6 +41,7 @@ pri func decoder.decode_macroblocks!(dst: ptr base.pixel_buffer, workbuf: slice mbx += 1 } + // Filter the previous row, already in args.workbuf. if mby > 0 { if this.filt_level <= 0 { // No-op. @@ -52,6 +55,7 @@ pri func decoder.decode_macroblocks!(dst: ptr base.pixel_buffer, workbuf: slice mby += 1 } + // Filter the final row, already in args.workbuf, and swizzle to args.dst. if this.mbh > 0 { if this.filt_level <= 0 { // No-op. @@ -79,6 +83,7 @@ pri func decoder.decode_one_macroblock!(workbuf: slice base.u8, mbx: base.u32[.. var luma_mode : base.u32[..= 4] var chroma_mode : base.u32[..= 3] + // RFC 6386 Section 10. Segment-Based Feature Adjustments. seg = 0 if this.seg_update_map { v1 = this.read_bit!(workbuf: args.workbuf, part: 0, prob: this.seg_probs[0] as base.u32) diff --git a/std/vp8/decode_predict_default.wuffs b/std/vp8/decode_predict_default.wuffs index cacf5cac..31870b8f 100644 --- a/std/vp8/decode_predict_default.wuffs +++ b/std/vp8/decode_predict_default.wuffs @@ -10,7 +10,13 @@ // -------- -// Predicts a 4×4 luma block, the Xs below, based on samples top and left. +// RFC 6386 Section 12. Intraframe Prediction. +// +// These methods read from and write to yuv_cache. + +// -------- + +// Predicts a 4×4 luma subblock, the Xs below, based on samples top and left. // // a b c d e f g h i // p X X X X diff --git a/std/vp8/decode_reconstruct.wuffs b/std/vp8/decode_reconstruct.wuffs index e64a2d9d..c25a8d02 100644 --- a/std/vp8/decode_reconstruct.wuffs +++ b/std/vp8/decode_reconstruct.wuffs @@ -8,6 +8,66 @@ // // 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], @@ -120,7 +180,10 @@ pri func decoder.prepare_yuv_cache!( 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 @@ -133,6 +196,7 @@ pri func decoder.prepare_yuv_cache!( } } 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] @@ -146,6 +210,7 @@ pri func decoder.prepare_yuv_cache!( } if args.mby <= 0 { + // Initialize abcefgh to 0x7F. i = 0x00 while i < 0x20 { this.yuv_cache[0x00][i] = 0x7F @@ -158,6 +223,8 @@ pri func decoder.prepare_yuv_cache!( } } 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 @@ -220,6 +287,7 @@ pri func decoder.prepare_yuv_cache!( } } + // Duplicate the c values. j = 0x04 while j < 0x10 { this.yuv_cache[j][0x18] = this.yuv_cache[0][0x18] diff --git a/std/vp8/decode_vp8.wuffs b/std/vp8/decode_vp8.wuffs index 593a901d..026010a6 100644 --- a/std/vp8/decode_vp8.wuffs +++ b/std/vp8/decode_vp8.wuffs @@ -33,6 +33,8 @@ pub struct decoder? implements base.image_decoder( frame_config_io_position : base.u64, + // This uses the same representation as an wuffs_base__optional_u63: + // ((value_as_u63 << 1) | has_value_as_u1). quirk_source_length : base.u64, // Just under 144 MiB, for 9 partitions each up to 0xFF_FFFF bytes. @@ -119,6 +121,7 @@ pub struct decoder? implements base.image_decoder( // hold in the 0x100 bit) and prob_skip_false (in the low 8 bits). prob_skip : base.u32[..= 0x1FF], + // 4×4 luma subblock prediction modes. mb_subblock_modes : array[16] base.u8[..= 9], // Two u32 values whose bits are packed like this: @@ -132,6 +135,13 @@ pub struct decoder? implements base.image_decoder( mb_dc_nz : base.u32, mb_ac_nz : base.u32, + // The workspace for the reconstruction-in-progress macroblock, + // containing the predicted and then fully decoded (prediction plus + // post-IDCT residuals) YUV samples to copy out to the workbuf. Its + // edges also hold samples from previous (left and top) macroblocks + // that are used in calculating this macroblock's predictions. + // + // See the comments near decoder.reconstruct for the layout. yuv_cache : array[26] array[32] base.u8, swizzler : base.pixel_swizzler, @@ -142,6 +152,20 @@ pub struct decoder? implements base.image_decoder( // 1056 is (4 planes, 8 bands, 3 contexts, 11 tokens) flattened. coeff_probs : array[1056] base.u8, + // Macroblock coefficients. Per this.decode_one_macroblock: + // + // this.decode_coefficients decodes these macroblock coefficients from + // the bitstream and dequantizes them, writing them to this.mb_coeffs. + // At this point, these are pre-IDCT residuals. + // + // this.reconstruct calculates the predictions (writing them to + // this.yuv_cache, not this.mb_coeffs) and then applies the inverse WHT + // (if applicable) and the inverse DCT to the pre-IDCT residuals, by + // calling this.inverse_dct_etc. + // + // this.inverse_dct_etc reads the pre-IDCT residuals from + // this.mb_coeffs and writes fully decoded samples (predictions plus + // post-IDCT residuals) to this.yuv_cache. mb_coeffs : array[25] array[16] base.u16, // Macroblock states are a u32 whose bits are packed like this: