// Copyright 2024 The Wuffs Authors. // // Licensed under the Apache License, Version 2.0 or the MIT license // , 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 use "std/vp8" pub status "#bad ALPH+VP8L combination" pub status "#bad Huffman code (over-subscribed)" pub status "#bad Huffman code (under-subscribed)" pub status "#bad Huffman code" pub status "#bad VP8X chunk" pub status "#bad back-reference" pub status "#bad chunk order" pub status "#bad color cache" pub status "#bad header" pub status "#bad transform" pub status "#short chunk" pub status "#truncated input" pub status "#unsupported VP8X/VP8 chunk length" pub status "#unsupported WebP file" pub status "#unsupported number of Huffman groups" pri status "#internal error: inconsistent Huffman code" pri status "#internal error: inconsistent dst buffer" pri status "#internal error: inconsistent n_bits" // TODO: reference vp8.DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE, although the // worst case for VP8L is bigger than for VP8. pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0x4C00_0000 // The default value of the largest supported VP8-inside-VP8X chunk length. // This limitation is a consequence of combining Wuffs' "decode an image" API // with WebP/VP8X/VP8's file format design. // // Wuffs' API only assumes sequential I/O access, not random I/O access, to the // encoded image's bytes. It also is incapable (in the security capability // sense) of dynamically allocating memory, so any O(width × height) pixel // buffers, or any other buffers that aren't O(1) in terms of width and height, // must be allocated by the caller, not the callee (Wuffs code). // // "How much dynamic memory does the caller need to allocate?" is answered by // the decoder.workbuf_len method, which is valid to call as soon as // decoder.decode_image_config returns (and decode_image_config, unlike // decode_frame, does not take a variable-length slice-of-bytes workbuf // argument). workbuf_len must return a worst case estimate, as it has only one // shot at answering "how much dynamic memory?" // // On the other hand, an VP8 image file can have multiple "partitions", which // means that a VP8 decoder must support random I/O access. Wuffs supports this // by, in the worst case, copying the VP8's chunk payload (the encoded image // data) into the workbuf. (Copying into a contiguous buffer is the same as // what the libwebp C library's WebPIAppend "incremental API" does.) But doing // so requires knowing how large the VP8 chunk is, so that the caller can // allocate a sufficiently large workbuf. // // We cannot answer that question precisely until we parse the VP8 chunk, but // in the WebP/VP8X file format, the VP8 chunk must occur after any ALPH chunk. // For Wuffs to parse an ALPH chunk, it must have already "allocated" enough // memory, more than O(1), by answering "how much dynamic memory?" // // As we cannot answer that question precisely (whilst still satisfying Wuffs' // "decode an image" API, which works across numerous image file formats, not // just WebP), we use a made-up upper bound, which is this number. // // Its value is arbitrary, but there is a trade-off. The larger it is, the more // workbuf memory Wuffs will say it needs (and callers will allocate), even if // it later discovers that it over-budgeted. The smaller it is, Wuffs will // reject (as unsupported) more large-but-valid WebP/VP8X/VP8 files. // // This default value is 8 MiB. pub const VP8X_MAX_INCL_VP8_CHUNK_LENGTH_DEFAULT : base.u64 = 8_388608 // TODO: in the future, regarding VP8X_MAX_INCL_VP8_CHUNK_LENGTH_DEFAULT, we // could make this set by a quirk instead of hard-coded. // // Or, we could change the semantics of the wuffs_base__image_decoder API // (which applies to multiple image formats, not just WebP). The API is that // you can call workbuf_len as soon as decode_image_config returns. workbuf_len // returns a range and any value within that range should be a valid length for // the "dynamically allocated, not O(1) in terms of width and height" workbuf // memory passed to decode_frame. // // We could relax this so that decode_frame could return a "$short workbuf" // suspension, if the workbuf length is unknown at the time of the first // decode_frame call, and the caller is expected to essentially realloc the // workbuf, keeping any existing contents when growing it. pub struct decoder? implements base.image_decoder( width : base.u32[..= 0x4000], height : base.u32[..= 0x4000], // The call sequence state machine is discussed in // (/doc/std/image-decoders-call-sequence.md). call_sequence : base.u8, code_length_code_lengths : array[19] base.u8[..= 7], sub_chunk_has_padding : base.bool, // 0x01 = VP8 lossy // 0x02 = VP8L lossless // 0x04 = VP8X and seen VP8X chunks // 0x08 = VP8X and seen VP8X, ICCP chunks // 0x10 = VP8X and seen VP8X, ICCP†, ANIM chunks // 0x20 = VP8X and seen VP8X, ICCP†, ANIM†, ANMF chunks // 0x40 = VP8X and seen VP8X, ICCP†, ANIM†, ANMF†, ALPH chunks // 0x81 = VP8X and seen VP8X, ICCP†, ANIM†, ANMF†, ALPH†, VP8 chunks // 0x82 = VP8X and seen VP8X, ICCP†, ANIM†, ANMF†, ALPH†, VP8L chunks // // Chunks marked with a † are optional. variant : base.u8, // The "Rsv|I|L|E|X|A|R" bits from // https://developers.google.com/speed/webp/docs/riff_container#extended_file_format // // We also re-purpose the otherwise-unused 0x01 bit to mean VP8X-ness, // so that (vp8x_header_flags <> 0) is an easy check for VP8X versus // not-VP8X. Similarly, ((vp8x_header_flags & 0x11) == 0x01) is an easy // check for "VP8X without an ALPH chunk". // // See also libwebp's src/webp/mux_types.h: // // ANIMATION_FLAG = 0x00000002, // XMP_FLAG = 0x00000004, // EXIF_FLAG = 0x00000008, // ALPHA_FLAG = 0x00000010, // ICCP_FLAG = 0x00000020, vp8x_header_flags : base.u8, frame_config_io_position : base.u64, riff_chunk_length : base.u32, sub_chunk_length : base.u32, bits : base.u32, n_bits : base.u32[..= 31], seen_transform : array[4] base.bool, transform_type : array[4] base.u8[..= 3], transform_tile_size_log2 : array[4] base.u8[..= 9], n_transforms : base.u32[..= 4], color_cache_bits : base.u32[..= 11], overall_color_cache_bits : base.u32[..= 11], overall_tile_size_log2 : base.u32[..= 9], overall_n_huffman_groups : base.u32[..= 256], ht_n_symbols : base.u32[..= 2328], ht_code_lengths_remaining : base.u32, color_indexing_palette_size : base.u32[..= 256], color_indexing_width : base.u32[..= 0x4000], // The 0th element is for the Meta (Huffman Group) selectors. // The 1st element is for the Predictor transform. // The 2nd element is for the Cross Color transform. // The 3rd element is overall workbuf length. workbuf_offset_for_transform : array[4] base.u32[..= 0x4C00_0000], // If the Color Indexing transform is present then workbuf[i .. j] // holds the pre-transformed pixel data (the Green component of the // pixel's bgra data holds color indexes) and workbuf[0 .. j] holds the // post-transformed data: // // - i is workbuf_offset_for_color_indexing // - j is workbuf_offset_for_transform[0] // // i is calculated as: // // - if there are 8 bits per color index then i is 0. // - if there are 4 bits per color index then i is j*1/2, roughly. // - if there are 2 bits per color index then i is j*3/4, roughly. // - if there are 1 bits per color index then i is j*7/8, roughly. // // The two slices (workbuf[i .. j] and workbuf[0 .. j]) overlap. // apply_transform_color_indexing modifies the pixel buffer in-place. workbuf_offset_for_color_indexing : base.u32[..= 0x4000_0000], vp8 : vp8.decoder, swizzler : base.pixel_swizzler, util : base.utility, ) + ( palette : array[4 * 256] base.u8, color_cache : array[2048] base.u32, codes : array[2328] base.u16, code_lengths : array[2328] base.u16, // Each node is a base.u16 whose bit patterns are: // // - 0000_0000_0000_0000 Invalid. // - 0CCC_CCCC_CCCC_CCCC Branch node (children_offset is CCC). // - 1SSS_SSSS_SSSS_SSSS Leaf node (symbol is SSS, it may be zero). code_lengths_huffman_nodes : array[37] base.u16, // A Huffman group has five (5) Huffman trees. // // Start .. End Size // 0x064C .. 0x187B 0x122F (1) Green, back-ref length, color cache. // 0x0000 .. 0x01FF 0x01FF (2) Red. // 0x01FF .. 0x03FE 0x01FF (3) Blue. // 0x03FE .. 0x05FD 0x01FF (4) Alpha. // 0x05FD .. 0x064C 0x004F (5) Back-ref distance. // // The Green+etc tree is last (in terms of start offset) because it's // by far the largest (worst case) and its color cache component has // variable length. // // 0x122F = 4655 = ((2 * 2328) - 1) and the same 2328 turns up in func // decoder.decode_huffman_tree. // // 2328 is (256 + 24 + 2048), combining 256 Green values, 24 back-ref // lengths and up to 2048 = (1 << 11) color cache keys. // // 0x01FF = 511 = ((2 * 256) - 1) is for 256 Red (Blue, Alpha) values. // // 0x004F = 79 = ((2 * 40) - 1) is for 40 back-ref distances. // // 0x187B = 6267 and (0x187B * sizeof(u16)) = 12534. Overall, this // field takes 0x30_F600 = 3_208704 bytes of memory. // // The base.u16's bits are the same as for code_lengths_huffman_nodes. huffman_nodes : array[256] array[0x187B] base.u16, ) pub func decoder.get_quirk(key: base.u32) base.u64 { return 0 } pub func decoder.set_quirk!(key: base.u32, value: base.u64) base.status { return base."#unsupported option" } pub func decoder.decode_image_config?(dst: nptr base.image_config, src: base.io_reader) { var status : base.status while true { status =? this.do_decode_image_config?(dst: args.dst, src: args.src) if (status == base."$short read") and args.src.is_closed() { return "#truncated input" } yield? status } } pri func decoder.do_decode_image_config?(dst: nptr base.image_config, src: base.io_reader) { var c32 : base.u32 var r_mark : base.u64 var status : base.status if this.call_sequence <> 0x00 { return base."#bad call sequence" } c32 = args.src.read_u32le?() if c32 <> 'RIFF'le { return "#bad header" } this.riff_chunk_length = args.src.read_u32le?() if (this.riff_chunk_length & 1) <> 0 { return "#bad header" } while true { io_limit (io: args.src, limit: this.riff_chunk_length as base.u64) { r_mark = args.src.mark() status =? this.do_decode_image_config_limited?(dst: args.dst, src: args.src) this.riff_chunk_length ~sat-= (args.src.count_since(mark: r_mark) & 0xFFFF_FFFF) as base.u32 } if status.is_ok() { break } else if not status.is_suspension() { return status } else if (status == base."$short read") and (this.riff_chunk_length == 0) { return "#short chunk" } yield? status } this.frame_config_io_position = args.src.position() if (this.variant <> 0x01) and (args.dst <> nullptr) { args.dst.set!( pixfmt: base.PIXEL_FORMAT__BGRA_NONPREMUL, pixsub: 0, width: this.width, height: this.height, first_frame_io_position: this.frame_config_io_position, first_frame_is_opaque: false) } this.call_sequence = 0x20 } pri func decoder.do_decode_image_config_limited?(dst: nptr base.image_config, src: base.io_reader) { var c32 : base.u32 var r_mark : base.u64 var status : base.status c32 = args.src.read_u32le?() if c32 <> 'WEBP'le { return "#bad header" } c32 = args.src.read_u32le?() if c32 == 'VP8 'le { this.variant = 0x01 } else if c32 == 'VP8L'le { this.variant = 0x02 } else if c32 == 'VP8X'le { this.variant = 0x04 } else { return "#bad header" } this.sub_chunk_length = args.src.read_u32le?() if this.sub_chunk_length < 4 { return "#bad header" } this.sub_chunk_has_padding = (this.sub_chunk_length & 1) <> 0 if this.variant == 0x01 { this.vp8.set_quirk!( key: base.QUIRK_SOURCE_LENGTH, value: ((this.sub_chunk_length as base.u64) << 1) | 1) } else if this.variant == 0x02 { // No-op. } else if this.sub_chunk_length <> 10 { return "#bad header" } else { this.vp8.set_quirk!( key: base.QUIRK_SOURCE_LENGTH, value: (VP8X_MAX_INCL_VP8_CHUNK_LENGTH_DEFAULT << 1) | 1) } while true { io_limit (io: args.src, limit: this.sub_chunk_length as base.u64) { r_mark = args.src.mark() if this.variant == 0x01 { status =? this.vp8.decode_image_config?(dst: args.dst, src: args.src) } else if this.variant == 0x02 { status =? this.do_decode_image_config_limited_vp8l?(src: args.src) } else { status =? this.do_decode_image_config_limited_vp8x?(src: args.src) } this.sub_chunk_length ~sat-= (args.src.count_since(mark: r_mark) & 0xFFFF_FFFF) as base.u32 } if status.is_ok() { break } else if not status.is_suspension() { return status } else if (status == base."$short read") and (this.sub_chunk_length == 0) { return "#short chunk" } yield? status } } pri func decoder.do_decode_image_config_limited_vp8l?(src: base.io_reader) { var c32 : base.u32 var transform_size : base.u32[..= 0x400_0000] c32 = args.src.read_u8_as_u32?() if c32 <> 0x2F { return "#bad header" } c32 = args.src.read_u32le?() this.width = (c32 & 0x3FFF) + 1 c32 >>= 14 this.height = (c32 & 0x3FFF) + 1 c32 >>= 14 // In theory, (c32 & 1) would let us discriminate between // base.PIXEL_FORMAT__BGRX and base.PIXEL_FORMAT__BGRA_NONPREMUL, but the // VP8 Lossless spec says that this bit "is a hint only, and *should not // impact decoding* [emphasis added]". c32 >>= 1 if c32 <> 0 { return "#bad header" } transform_size = 4 * ((this.width + 3) >> 2) * ((this.height + 3) >> 2) this.workbuf_offset_for_transform[0] = (4 * this.width * this.height) + (0 * transform_size) this.workbuf_offset_for_transform[1] = (4 * this.width * this.height) + (1 * transform_size) this.workbuf_offset_for_transform[2] = (4 * this.width * this.height) + (2 * transform_size) this.workbuf_offset_for_transform[3] = (4 * this.width * this.height) + (3 * transform_size) } pri func decoder.do_decode_image_config_limited_vp8x?(src: base.io_reader) { var c32 : base.u32 var transform_size : base.u32[..= 0x400_0000] var status : base.status c32 = args.src.read_u32le?() if (c32 & 0xFFFF_FFC1) <> 0 { return "#unsupported WebP file" } this.vp8x_header_flags = ((c32 | 0x01) & 0xFF) as base.u8 c32 = args.src.read_u24le_as_u32?() if c32 >= 0x3FFF { return "#unsupported WebP file" } this.width = (c32 & 0x3FFF) + 1 c32 = args.src.read_u24le_as_u32?() if c32 >= 0x3FFF { return "#unsupported WebP file" } this.height = (c32 & 0x3FFF) + 1 // It's valid to call this.workbuf_len after this.decode_image_config // returns, for the caller to know how much dynamically-sized memory // (called workbuf memory) is needed to decode the image file. // // Unfortunately, having only seen the VP8X chunk, we don't know yet // whether it's ultimately a VP8 or VP8L image, but we wouldn't know until // after we've processed the ALPH chunk (if present), and processing the // ALPH chunk requires workbuf memory (as the amount of memory depends on // this.width and this.height). // // Wuffs' "decode an image" API assumes that the caller, not callee, is // responsible for dynamically allocating memory. But also, it only assumes // sequential access (not random access) to the io_reader. // // So, here, we have to cater for the worst case and set aside workbuf // memory just like plain VP8L, even if it might end up being unused, for // example: VP8X+VP8 instead of VP8X+VP8L or VP8X+ALPH+VP8. transform_size = 4 * ((this.width + 3) >> 2) * ((this.height + 3) >> 2) this.workbuf_offset_for_transform[0] = (4 * this.width * this.height) + (0 * transform_size) this.workbuf_offset_for_transform[1] = (4 * this.width * this.height) + (1 * transform_size) this.workbuf_offset_for_transform[2] = (4 * this.width * this.height) + (2 * transform_size) this.workbuf_offset_for_transform[3] = (4 * this.width * this.height) + (3 * transform_size) status = this.vp8.set_quirk!( key: vp8.QUIRK_WIDTH_AND_HEIGHT, value: ((this.width as base.u64) | ((this.height as base.u64) << 32))) if not status.is_ok() { return "#unsupported WebP file" } } pub func decoder.decode_frame_config?(dst: nptr base.frame_config, src: base.io_reader) { var status : base.status while true { if this.variant == 0x01 { status =? this.vp8.decode_frame_config?(dst: args.dst, src: args.src) } else { status =? this.do_decode_frame_config?(dst: args.dst, src: args.src) } if (status == base."$short read") and args.src.is_closed() { return "#truncated input" } yield? status } } pri func decoder.do_decode_frame_config?(dst: nptr base.frame_config, src: base.io_reader) { if this.call_sequence == 0x20 { // No-op. } else if this.call_sequence < 0x20 { this.do_decode_image_config?(dst: nullptr, src: args.src) } else if this.call_sequence == 0x28 { if this.frame_config_io_position <> args.src.position() { return base."#bad restart" } } else if this.call_sequence == 0x40 { this.call_sequence = 0x60 return base."@end of data" } else { return base."@end of data" } if (this.variant <> 0x01) and (args.dst <> nullptr) { args.dst.set!(bounds: this.util.make_rect_ie_u32( min_incl_x: 0, min_incl_y: 0, max_excl_x: this.width, max_excl_y: this.height), duration: 0, index: 0, io_position: this.frame_config_io_position, disposal: 0, opaque_within_bounds: false, overwrite_instead_of_blend: false, background_color: 0x0000_0000) } this.call_sequence = 0x40 } pub func decoder.decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reader, blend: base.pixel_blend, workbuf: slice base.u8, opts: nptr base.decode_frame_options) { var status : base.status while true { status =? this.do_decode_frame?(dst: args.dst, src: args.src, blend: args.blend, workbuf: args.workbuf, opts: args.opts) if (status == base."$short read") and args.src.is_closed() { return "#truncated input" } yield? status } } pri func decoder.do_decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reader, blend: base.pixel_blend, workbuf: slice base.u8, opts: nptr base.decode_frame_options) { var chunk_type : base.u32 var chunk_length : base.u32 if this.call_sequence == 0x40 { // No-op. } else if this.call_sequence < 0x40 { this.do_decode_frame_config?(dst: nullptr, src: args.src) } else { return base."@end of data" } while (this.variant & 0x03) == 0 { chunk_type = args.src.read_u32le?() chunk_length = args.src.read_u32le?() if chunk_type == 'ICCP'le { if this.variant > 0x04 { return "#bad chunk order" } else if (this.vp8x_header_flags & 0x20) == 0 { return "#bad VP8X chunk" } this.variant = 0x08 } else if chunk_type == 'ANIM'le { if this.variant > 0x08 { return "#bad chunk order" } else if (this.vp8x_header_flags & 0x02) == 0 { return "#bad VP8X chunk" } this.variant = 0x10 } else if chunk_type == 'ANMF'le { if this.variant > 0x10 { return "#bad chunk order" } else if (this.vp8x_header_flags & 0x02) == 0 { return "#bad VP8X chunk" } this.variant = 0x20 } else if chunk_type == 'ALPH'le { if this.variant > 0x20 { return "#bad chunk order" } else if (this.vp8x_header_flags & 0x10) == 0 { return "#bad VP8X chunk" } this.variant = 0x40 } else if chunk_type == 'VP8 'le { if (chunk_length as base.u64) > VP8X_MAX_INCL_VP8_CHUNK_LENGTH_DEFAULT { return "#unsupported VP8X/VP8 chunk length" } this.vp8.set_quirk!( key: base.QUIRK_SOURCE_LENGTH, value: ((chunk_length as base.u64) << 1) | 1) this.variant = 0x81 break } else if chunk_type == 'VP8L'le { if (this.vp8x_header_flags & 0x10) <> 0 { return "#bad ALPH+VP8L combination" } this.variant = 0x82 break } else if chunk_type == 'XMP 'le { if (this.vp8x_header_flags & 0x04) == 0 { return "#bad VP8X chunk" } } else if chunk_type == 'EXIF'le { if (this.vp8x_header_flags & 0x08) == 0 { return "#bad VP8X chunk" } } if chunk_length > 0x7FFF_FFFF { return "#unsupported WebP file" } else if (chunk_length & 1) <> 0 { chunk_length += 1 } args.src.skip_u32?(n: chunk_length) } if (this.variant & 0x01) <> 0 { this.vp8.decode_frame?(dst: args.dst, src: args.src, blend: args.blend, workbuf: args.workbuf, opts: args.opts) } else { this.do_decode_frame_vp8l?(dst: args.dst, src: args.src, blend: args.blend, workbuf: args.workbuf, opts: args.opts) } } pri func decoder.do_decode_frame_vp8l?(dst: ptr base.pixel_buffer, src: base.io_reader, blend: base.pixel_blend, workbuf: slice base.u8, opts: nptr base.decode_frame_options) { var c8 : base.u8 var has_more : base.u32[..= 1] var width : base.u32[..= 0x4000] var dst : slice base.u8 var tile_data : roslice base.u8 var status : base.status var pix : slice base.u8 var which : base.u32[..= 4] var transform_type : base.u32[..= 3] var ti : base.u64 var tj : base.u64 this.seen_transform[0] = false this.seen_transform[1] = false this.seen_transform[2] = false this.seen_transform[3] = false this.n_transforms = 0 while true { if this.n_bits < 1 { c8 = args.src.read_u8?() this.bits = (c8 as base.u32) this.n_bits = 8 assert this.n_bits >= 1 } has_more = this.bits & 1 this.bits >>= 1 this.n_bits -= 1 if has_more == 0 { break } this.decode_transform?(src: args.src, workbuf: args.workbuf) } width = this.width if this.seen_transform[3] { width = this.color_indexing_width } this.decode_color_cache_parameters?(src: args.src) this.overall_color_cache_bits = this.color_cache_bits this.decode_hg_table?(src: args.src, width: width, workbuf: args.workbuf) this.color_cache_bits = this.overall_color_cache_bits this.decode_huffman_groups?(src: args.src, n_huffman_groups: this.overall_n_huffman_groups) while true { if ((this.workbuf_offset_for_color_indexing as base.u64) > (this.workbuf_offset_for_transform[0] as base.u64)) or ((this.workbuf_offset_for_transform[0] as base.u64) > (this.workbuf_offset_for_transform[1] as base.u64)) or ((this.workbuf_offset_for_transform[1] as base.u64) > args.workbuf.length()) { return base."#bad workbuf length" } assert (this.workbuf_offset_for_transform[0] as base.u64) <= args.workbuf.length() via "a <= b: a <= c; c <= b"(c: this.workbuf_offset_for_transform[1] as base.u64) dst = args.workbuf[ (this.workbuf_offset_for_color_indexing as base.u64) .. (this.workbuf_offset_for_transform[0] as base.u64)] tile_data = args.workbuf[ (this.workbuf_offset_for_transform[0] as base.u64) .. (this.workbuf_offset_for_transform[1] as base.u64)] status =? this.decode_pixels?( dst: dst, src: args.src, width: width, height: this.height, tile_data: tile_data, tile_size_log2: this.overall_tile_size_log2) if status.is_ok() { break } yield? status } if (this.workbuf_offset_for_transform[0] as base.u64) > args.workbuf.length() { return base."#bad workbuf length" } pix = args.workbuf[.. (this.workbuf_offset_for_transform[0] as base.u64)] width = this.width if this.seen_transform[3] and ((this.workbuf_offset_for_color_indexing as base.u64) <= pix.length()) { pix = pix[this.workbuf_offset_for_color_indexing as base.u64 ..] width = this.color_indexing_width } which = this.n_transforms while which > 0 { which -= 1 transform_type = this.transform_type[which] as base.u32 tile_data = this.util.empty_slice_u8() if transform_type < 2 { ti = this.workbuf_offset_for_transform[transform_type + 1] as base.u64 tj = this.workbuf_offset_for_transform[transform_type + 2] as base.u64 if (ti <= tj) and (tj <= args.workbuf.length()) { tile_data = args.workbuf[ti .. tj] } } if transform_type == 0 { this.apply_transform_predictor!(pix: pix, width: width, tile_data: tile_data) } else if transform_type == 1 { this.apply_transform_cross_color!(pix: pix, width: width, tile_data: tile_data) } else if transform_type == 2 { this.apply_transform_subtract_green!(pix: pix) } else if (this.workbuf_offset_for_transform[0] as base.u64) <= args.workbuf.length() { pix = args.workbuf[.. (this.workbuf_offset_for_transform[0] as base.u64)] width = this.width this.apply_transform_color_indexing!(pix: pix, width: width) } } status = this.swizzle!( dst: args.dst, src: pix, blend: args.blend) if not status.is_ok() { return status } this.call_sequence = 0x60 } pri func decoder.decode_transform?(src: base.io_reader, workbuf: slice base.u8) { var status : base.status var c8 : base.u8 var transform_type : base.u32[..= 3] var tile_size_log2 : base.u32[..= 9] var width : base.u32[..= 0x4000] var p : slice base.u8 if this.n_bits < 2 { c8 = args.src.read_u8?() if this.n_bits >= 2 { return "#internal error: inconsistent n_bits" } this.bits |= (c8 as base.u32) << this.n_bits this.n_bits += 8 assert this.n_bits >= 2 } transform_type = this.bits & 3 this.bits >>= 2 this.n_bits -= 2 if this.seen_transform[transform_type] or (this.n_transforms >= 4) { return "#bad transform" } this.seen_transform[transform_type] = true this.transform_type[this.n_transforms] = transform_type as base.u8 this.n_transforms += 1 if transform_type < 2 { // Predictor, Cross Color transform. if this.n_bits < 3 { c8 = args.src.read_u8?() if this.n_bits >= 3 { return "#internal error: inconsistent n_bits" } this.bits |= (c8 as base.u32) << this.n_bits this.n_bits += 8 assert this.n_bits >= 3 } tile_size_log2 = (this.bits & 7) + 2 this.transform_tile_size_log2[transform_type] = tile_size_log2 as base.u8 this.bits >>= 3 this.n_bits -= 3 this.decode_color_cache_parameters?(src: args.src) this.decode_huffman_groups?(src: args.src, n_huffman_groups: 1) while true, inv transform_type < 2, inv tile_size_log2 >= 2, { width = this.width if this.seen_transform[3] { width = this.color_indexing_width } if ((this.workbuf_offset_for_transform[transform_type + 1] as base.u64) > (this.workbuf_offset_for_transform[transform_type + 2] as base.u64)) or ((this.workbuf_offset_for_transform[transform_type + 2] as base.u64) > args.workbuf.length()) { return base."#bad workbuf length" } status =? this.decode_pixels?( dst: args.workbuf[ (this.workbuf_offset_for_transform[transform_type + 1] as base.u64) .. (this.workbuf_offset_for_transform[transform_type + 2] as base.u64)], src: args.src, width: (width + (((1 as base.u32) << tile_size_log2) - 1)) >> tile_size_log2, height: (this.height + (((1 as base.u32) << tile_size_log2) - 1)) >> tile_size_log2, tile_data: this.util.empty_slice_u8(), tile_size_log2: 0) if status.is_ok() { break } yield? status } } else if transform_type == 2 { // Subtract Green transform. // No-op. } else { // Color Indexing transform. if this.n_bits < 8 { c8 = args.src.read_u8?() if this.n_bits >= 8 { return "#internal error: inconsistent n_bits" } this.bits |= (c8 as base.u32) << this.n_bits this.n_bits += 8 assert this.n_bits >= 8 } this.color_indexing_palette_size = (this.bits & 0xFF) + 1 this.bits >>= 8 this.n_bits -= 8 if this.color_indexing_palette_size <= 2 { this.color_indexing_width = (this.width + 7) / 8 this.transform_tile_size_log2[3] = 3 } else if this.color_indexing_palette_size <= 4 { this.color_indexing_width = (this.width + 3) / 4 this.transform_tile_size_log2[3] = 2 } else if this.color_indexing_palette_size <= 16 { this.color_indexing_width = (this.width + 1) / 2 this.transform_tile_size_log2[3] = 1 } else { this.color_indexing_width = this.width this.transform_tile_size_log2[3] = 0 } if this.width >= this.color_indexing_width { this.workbuf_offset_for_color_indexing = 4 * (this.width - this.color_indexing_width) * this.height } this.decode_color_cache_parameters?(src: args.src) this.decode_huffman_groups?(src: args.src, n_huffman_groups: 1) this.decode_pixels?( dst: this.palette[.. 4 * this.color_indexing_palette_size], src: args.src, width: this.color_indexing_palette_size, height: 1, tile_data: this.util.empty_slice_u8(), tile_size_log2: 0) this.palette[4 * this.color_indexing_palette_size .. 1024].bulk_memset!(byte_value: 0) p = this.palette[.. 4 * this.color_indexing_palette_size] while p.length() >= 8 { p[4] ~mod+= p[0] p[5] ~mod+= p[1] p[6] ~mod+= p[2] p[7] ~mod+= p[3] p = p[4 ..] } } } pri func decoder.decode_color_cache_parameters?(src: base.io_reader) { var c8 : base.u8 var use_color_cache : base.u32[..= 1] var color_cache_bits : base.u32[..= 15] if this.n_bits < 1 { c8 = args.src.read_u8?() this.bits = (c8 as base.u32) this.n_bits = 8 assert this.n_bits >= 1 } use_color_cache = this.bits & 1 this.bits >>= 1 this.n_bits -= 1 this.color_cache_bits = 0 if use_color_cache <> 0 { if this.n_bits < 4 { c8 = args.src.read_u8?() if this.n_bits >= 4 { return "#internal error: inconsistent n_bits" } this.bits |= (c8 as base.u32) << this.n_bits this.n_bits += 8 assert this.n_bits >= 4 } color_cache_bits = this.bits & 15 this.bits >>= 4 this.n_bits -= 4 if (color_cache_bits < 1) or (11 < color_cache_bits) { return "#bad color cache" } this.color_cache_bits = color_cache_bits } } pri func decoder.decode_hg_table?(src: base.io_reader, width: base.u32[..= 0x4000], workbuf: slice base.u8) { var status : base.status var c8 : base.u8 var use_hg_table : base.u32[..= 1] var tile_size_log2 : base.u32[..= 9] var hg_pixels : slice base.u8 var n : base.u64 var p : roslice base.u8 var hg_plus_1 : base.u32[..= 256] if this.n_bits < 1 { c8 = args.src.read_u8?() this.bits = (c8 as base.u32) this.n_bits = 8 assert this.n_bits >= 1 } use_hg_table = this.bits & 1 this.bits >>= 1 this.n_bits -= 1 if use_hg_table == 0 { this.overall_n_huffman_groups = 1 this.overall_tile_size_log2 = 0 if ((this.workbuf_offset_for_transform[0] as base.u64) > (this.workbuf_offset_for_transform[1] as base.u64)) or ((this.workbuf_offset_for_transform[1] as base.u64) > args.workbuf.length()) { return base."#bad workbuf length" } hg_pixels = args.workbuf[ this.workbuf_offset_for_transform[0] as base.u64 .. this.workbuf_offset_for_transform[1] as base.u64] if hg_pixels.length() >= 4 { hg_pixels[0] = 0x00 hg_pixels[1] = 0x00 hg_pixels[2] = 0x00 hg_pixels[3] = 0x00 } return ok } if this.n_bits < 3 { c8 = args.src.read_u8?() if this.n_bits >= 3 { return "#internal error: inconsistent n_bits" } this.bits |= (c8 as base.u32) << this.n_bits this.n_bits += 8 assert this.n_bits >= 3 } tile_size_log2 = (this.bits & 7) + 2 this.bits >>= 3 this.n_bits -= 3 this.overall_tile_size_log2 = tile_size_log2 this.decode_color_cache_parameters?(src: args.src) this.decode_huffman_groups?(src: args.src, n_huffman_groups: 1) while true, inv tile_size_log2 >= 2, { if ((this.workbuf_offset_for_transform[0] as base.u64) > (this.workbuf_offset_for_transform[1] as base.u64)) or ((this.workbuf_offset_for_transform[1] as base.u64) > args.workbuf.length()) { return base."#bad workbuf length" } status =? this.decode_pixels?( dst: args.workbuf[ (this.workbuf_offset_for_transform[0] as base.u64) .. (this.workbuf_offset_for_transform[1] as base.u64)], src: args.src, width: (args.width + (((1 as base.u32) << tile_size_log2) - 1)) >> tile_size_log2, height: (this.height + (((1 as base.u32) << tile_size_log2) - 1)) >> tile_size_log2, tile_data: this.util.empty_slice_u8(), tile_size_log2: 0) if status.is_ok() { break } yield? status } this.overall_n_huffman_groups = 1 if ((this.workbuf_offset_for_transform[0] as base.u64) > (this.workbuf_offset_for_transform[1] as base.u64)) or ((this.workbuf_offset_for_transform[1] as base.u64) > args.workbuf.length()) { return base."#bad workbuf length" } hg_pixels = args.workbuf[ this.workbuf_offset_for_transform[0] as base.u64 .. this.workbuf_offset_for_transform[1] as base.u64] n = (((args.width + (((1 as base.u32) << tile_size_log2) - 1)) >> tile_size_log2) * ((this.height + (((1 as base.u32) << tile_size_log2) - 1)) >> tile_size_log2) * 4) as base.u64 if n > hg_pixels.length() { return base."#bad workbuf length" } p = hg_pixels[.. n] while p.length() >= 4 { if p[2] <> 0 { return "#unsupported number of Huffman groups" } hg_plus_1 = (p[1] as base.u32) + 1 if this.overall_n_huffman_groups < hg_plus_1 { this.overall_n_huffman_groups = hg_plus_1 } p = p[4 ..] } } pri func decoder.decode_pixels?(dst: slice base.u8, src: base.io_reader, width: base.u32[..= 0x4000], height: base.u32[..= 0x4000], tile_data: roslice base.u8, tile_size_log2: base.u32[..= 9]) { var i : base.u32 var n : base.u32[..= 2048] i = 0 n = (1 as base.u32) << this.color_cache_bits while i < n { assert i < 2048 via "a < b: a < c; c <= b"(c: n) this.color_cache[i] = 0 i += 1 } this.decode_pixels_slow?( dst: args.dst, src: args.src, width: args.width, height: args.height, tile_data: args.tile_data, tile_size_log2: args.tile_size_log2) } pri func decoder.swizzle!(dst: ptr base.pixel_buffer, src: roslice base.u8, blend: base.pixel_blend) base.status { var status : base.status var dst_pixfmt : base.pixel_format var dst_bits_per_pixel : base.u32[..= 256] var dst_bytes_per_pixel : base.u32[..= 32] var dst_bytes_per_row : base.u64 var dst_palette : slice base.u8 var tab : table base.u8 var src_bytes_per_row : base.u64 var dst : slice base.u8 var y : base.u32 status = this.swizzler.prepare!( dst_pixfmt: args.dst.pixel_format(), dst_palette: args.dst.palette_or_else(fallback: this.palette[..]), src_pixfmt: this.util.make_pixel_format(repr: base.PIXEL_FORMAT__BGRA_NONPREMUL), src_palette: this.util.empty_slice_u8(), blend: args.blend) if not status.is_ok() { return status } // TODO: the dst_pixfmt variable shouldn't be necessary. We should be able // to chain the two calls: "args.dst.pixel_format().bits_per_pixel()". dst_pixfmt = args.dst.pixel_format() dst_bits_per_pixel = dst_pixfmt.bits_per_pixel() if (dst_bits_per_pixel & 7) <> 0 { return base."#unsupported option" } dst_bytes_per_pixel = dst_bits_per_pixel / 8 dst_bytes_per_row = (this.width * dst_bytes_per_pixel) as base.u64 dst_palette = args.dst.palette_or_else(fallback: this.palette[..]) tab = args.dst.plane(p: 0) src_bytes_per_row = (this.width * 4) as base.u64 while src_bytes_per_row <= args.src.length() { dst = tab.row_u32(y: y) if dst_bytes_per_row < dst.length() { dst = dst[.. dst_bytes_per_row] } this.swizzler.swizzle_interleaved_from_slice!( dst: dst, dst_palette: dst_palette, src: args.src[.. src_bytes_per_row]) args.src = args.src[src_bytes_per_row ..] y ~mod+= 1 } return ok } pub func decoder.frame_dirty_rect() base.rect_ie_u32 { if this.variant == 0x01 { return this.vp8.frame_dirty_rect() } return this.util.make_rect_ie_u32( min_incl_x: 0, min_incl_y: 0, max_excl_x: this.width, max_excl_y: this.height) } pub func decoder.num_animation_loops() base.u32 { return 0 } pub func decoder.num_decoded_frame_configs() base.u64 { if this.variant == 0x01 { return this.vp8.num_decoded_frame_configs() } if this.call_sequence > 0x20 { return 1 } return 0 } pub func decoder.num_decoded_frames() base.u64 { if this.variant == 0x01 { return this.vp8.num_decoded_frames() } if this.call_sequence > 0x40 { return 1 } return 0 } pub func decoder.restart_frame!(index: base.u64, io_position: base.u64) base.status { var status : base.status if this.variant == 0x01 { status = this.vp8.restart_frame!(index: args.index, io_position: args.io_position) return status } if this.call_sequence < 0x20 { return base."#bad call sequence" } if (args.index <> 0) or (args.io_position <> this.frame_config_io_position) { return base."#bad argument" } this.call_sequence = 0x28 return ok } pub func decoder.set_report_metadata!(fourcc: base.u32, report: base.bool) { // No-op. } pub func decoder.tell_me_more?(dst: base.io_writer, minfo: nptr base.more_information, src: base.io_reader) { return base."#no more information" } pub func decoder.workbuf_len() base.range_ii_u64 { var r : base.range_ii_u64 var s : base.range_ii_u64 if this.variant == 0x01 { return this.vp8.workbuf_len() } r = this.util.make_range_ii_u64( min_incl: this.workbuf_offset_for_transform[3] as base.u64, max_incl: this.workbuf_offset_for_transform[3] as base.u64) if this.variant <> 0x02 { s = this.vp8.workbuf_len() r = this.util.make_range_ii_u64( min_incl: s.get_min_incl().max(no_less_than: r.get_min_incl()), max_incl: s.get_max_incl().max(no_less_than: r.get_max_incl())) } return r } pri const DISTANCE_MAP : roarray[120] base.u8 = [ 0x18, 0x07, 0x17, 0x19, 0x28, 0x06, 0x27, 0x29, 0x16, 0x1A, 0x26, 0x2A, 0x38, 0x05, 0x37, 0x39, 0x15, 0x1B, 0x36, 0x3A, 0x25, 0x2B, 0x48, 0x04, 0x47, 0x49, 0x14, 0x1C, 0x35, 0x3B, 0x46, 0x4A, 0x24, 0x2C, 0x58, 0x45, 0x4B, 0x34, 0x3C, 0x03, 0x57, 0x59, 0x13, 0x1D, 0x56, 0x5A, 0x23, 0x2D, 0x44, 0x4C, 0x55, 0x5B, 0x33, 0x3D, 0x68, 0x02, 0x67, 0x69, 0x12, 0x1E, 0x66, 0x6A, 0x22, 0x2E, 0x54, 0x5C, 0x43, 0x4D, 0x65, 0x6B, 0x32, 0x3E, 0x78, 0x01, 0x77, 0x79, 0x53, 0x5D, 0x11, 0x1F, 0x64, 0x6C, 0x42, 0x4E, 0x76, 0x7A, 0x21, 0x2F, 0x75, 0x7B, 0x31, 0x3F, 0x63, 0x6D, 0x52, 0x5E, 0x00, 0x74, 0x7C, 0x41, 0x4F, 0x10, 0x20, 0x62, 0x6E, 0x30, 0x73, 0x7D, 0x51, 0x5F, 0x40, 0x72, 0x7E, 0x61, 0x6F, 0x50, 0x71, 0x7F, 0x60, 0x70, ]