diff --git a/release/c/wuffs-unsupported-snapshot.c b/release/c/wuffs-unsupported-snapshot.c index e808ab05..9215a961 100644 --- a/release/c/wuffs-unsupported-snapshot.c +++ b/release/c/wuffs-unsupported-snapshot.c @@ -8648,7 +8648,7 @@ extern const char wuffs_jpeg__error__unsupported_precision[]; // ---------------- Public Consts -#define WUFFS_JPEG__DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE 17184063744 +#define WUFFS_JPEG__DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE 51552191232 // ---------------- Struct Declarations @@ -8811,7 +8811,7 @@ struct wuffs_jpeg__decoder__struct { uint8_t f_components_tq[4]; uint32_t f_components_workbuf_widths[4]; uint32_t f_components_workbuf_heights[4]; - uint64_t f_components_workbuf_offsets[5]; + uint64_t f_components_workbuf_offsets[9]; uint32_t f_scan_num_components; uint8_t f_scan_comps_cselector[4]; uint8_t f_scan_comps_td[4]; @@ -8826,12 +8826,16 @@ struct wuffs_jpeg__decoder__struct { uint8_t f_scan_comps_by_offset[16]; uint32_t f_mcu_num_blocks; uint32_t f_mcu_current_block; - uint8_t f_mcu_blocks_sselector[16]; uint32_t f_mcu_zig_index; + uint8_t f_mcu_blocks_sselector[16]; + uint64_t f_mcu_blocks_offset[10]; + uint32_t f_mcu_blocks_mx_mul[10]; + uint32_t f_mcu_blocks_my_mul[10]; uint16_t f_mcu_previous_dc_values[4]; uint16_t f_restart_interval; uint16_t f_saved_restart_interval; uint16_t f_restarts_remaining; + uint16_t f_eob_run; uint64_t f_frame_config_io_position; uint32_t f_payload_length; bool f_seen_dqt[4]; @@ -8861,6 +8865,8 @@ struct wuffs_jpeg__decoder__struct { uint32_t p_decode_sos[1]; uint32_t p_prepare_scan[1]; uint32_t p_skip_past_the_next_restart_marker[1]; + uint32_t (*choosy_decode_mcu)( + wuffs_jpeg__decoder* self); } private_impl; struct { @@ -36790,6 +36796,13 @@ wuffs_jpeg__decoder__decode_sof( wuffs_jpeg__decoder* self, wuffs_base__io_buffer* a_src); +static uint32_t +wuffs_jpeg__decoder__quantize_dimension( + const wuffs_jpeg__decoder* self, + uint32_t a_width, + uint8_t a_h, + uint8_t a_max_incl_h); + static wuffs_base__status wuffs_jpeg__decoder__do_decode_frame_config( wuffs_jpeg__decoder* self, @@ -36833,6 +36846,24 @@ wuffs_jpeg__decoder__fill_bitstream( wuffs_jpeg__decoder* self, wuffs_base__io_buffer* a_src); +static wuffs_base__empty_struct +wuffs_jpeg__decoder__clear_mcu_blocks( + wuffs_jpeg__decoder* self); + +static wuffs_base__empty_struct +wuffs_jpeg__decoder__load_mcu_blocks( + wuffs_jpeg__decoder* self, + uint32_t a_mx, + uint32_t a_my, + wuffs_base__slice_u8 a_workbuf); + +static wuffs_base__empty_struct +wuffs_jpeg__decoder__save_mcu_blocks( + wuffs_jpeg__decoder* self, + uint32_t a_mx, + uint32_t a_my, + wuffs_base__slice_u8 a_workbuf); + static wuffs_base__status wuffs_jpeg__decoder__skip_past_the_next_restart_marker( wuffs_jpeg__decoder* self, @@ -36854,6 +36885,26 @@ static uint32_t wuffs_jpeg__decoder__decode_mcu( wuffs_jpeg__decoder* self); +static uint32_t +wuffs_jpeg__decoder__decode_mcu__choosy_default( + wuffs_jpeg__decoder* self); + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_ac_high_bits( + wuffs_jpeg__decoder* self); + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_ac_low_bit( + wuffs_jpeg__decoder* self); + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_dc_high_bits( + wuffs_jpeg__decoder* self); + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_dc_low_bit( + wuffs_jpeg__decoder* self); + // ---------------- VTables const wuffs_base__image_decoder__func_ptrs @@ -36931,6 +36982,8 @@ wuffs_jpeg__decoder__initialize( } } + self->private_impl.choosy_decode_mcu = &wuffs_jpeg__decoder__decode_mcu__choosy_default; + self->private_impl.magic = WUFFS_BASE__MAGIC; self->private_impl.vtable_for__wuffs_base__image_decoder.vtable_name = wuffs_base__image_decoder__vtable_name; @@ -37950,43 +38003,46 @@ wuffs_jpeg__decoder__do_decode_image_config( label__2__break:; if (v_marker == 0) { goto label__0__continue; - } else if ((208 <= v_marker) && (v_marker <= 215)) { - goto label__0__continue; - } - { - WUFFS_BASE__COROUTINE_SUSPENSION_POINT(5); - uint32_t t_4; - if (WUFFS_BASE__LIKELY(io2_a_src - iop_a_src >= 2)) { - t_4 = ((uint32_t)(wuffs_base__peek_u16be__no_bounds_check(iop_a_src))); - iop_a_src += 2; - } else { - self->private_data.s_do_decode_image_config[0].scratch = 0; - WUFFS_BASE__COROUTINE_SUSPENSION_POINT(6); - while (true) { - if (WUFFS_BASE__UNLIKELY(iop_a_src == io2_a_src)) { - status = wuffs_base__make_status(wuffs_base__suspension__short_read); - goto suspend; - } - uint64_t* scratch = &self->private_data.s_do_decode_image_config[0].scratch; - uint32_t num_bits_4 = ((uint32_t)(*scratch & 0xFF)); - *scratch >>= 8; - *scratch <<= 8; - *scratch |= ((uint64_t)(*iop_a_src++)) << (56 - num_bits_4); - if (num_bits_4 == 8) { - t_4 = ((uint32_t)(*scratch >> 48)); - break; - } - num_bits_4 += 8; - *scratch |= ((uint64_t)(num_bits_4)); - } + } else if ((208 <= v_marker) && (v_marker <= 217)) { + if (v_marker <= 215) { + goto label__0__continue; } - self->private_impl.f_payload_length = t_4; + } else { + { + WUFFS_BASE__COROUTINE_SUSPENSION_POINT(5); + uint32_t t_4; + if (WUFFS_BASE__LIKELY(io2_a_src - iop_a_src >= 2)) { + t_4 = ((uint32_t)(wuffs_base__peek_u16be__no_bounds_check(iop_a_src))); + iop_a_src += 2; + } else { + self->private_data.s_do_decode_image_config[0].scratch = 0; + WUFFS_BASE__COROUTINE_SUSPENSION_POINT(6); + while (true) { + if (WUFFS_BASE__UNLIKELY(iop_a_src == io2_a_src)) { + status = wuffs_base__make_status(wuffs_base__suspension__short_read); + goto suspend; + } + uint64_t* scratch = &self->private_data.s_do_decode_image_config[0].scratch; + uint32_t num_bits_4 = ((uint32_t)(*scratch & 0xFF)); + *scratch >>= 8; + *scratch <<= 8; + *scratch |= ((uint64_t)(*iop_a_src++)) << (56 - num_bits_4); + if (num_bits_4 == 8) { + t_4 = ((uint32_t)(*scratch >> 48)); + break; + } + num_bits_4 += 8; + *scratch |= ((uint64_t)(num_bits_4)); + } + } + self->private_impl.f_payload_length = t_4; + } + if (self->private_impl.f_payload_length < 2) { + status = wuffs_base__make_status(wuffs_jpeg__error__bad_marker); + goto exit; + } + self->private_impl.f_payload_length -= 2; } - if (self->private_impl.f_payload_length < 2) { - status = wuffs_base__make_status(wuffs_jpeg__error__bad_marker); - goto exit; - } - self->private_impl.f_payload_length -= 2; if (v_marker < 192) { status = wuffs_base__make_status(wuffs_jpeg__error__unsupported_marker); goto exit; @@ -38321,6 +38377,7 @@ wuffs_jpeg__decoder__decode_sof( uint64_t v_wh1 = 0; uint64_t v_wh2 = 0; uint64_t v_wh3 = 0; + uint64_t v_progressive = 0; const uint8_t* iop_a_src = NULL; const uint8_t* io0_a_src WUFFS_BASE__POTENTIALLY_UNUSED = NULL; @@ -38537,24 +38594,8 @@ wuffs_jpeg__decoder__decode_sof( goto exit; } } - if (self->private_impl.f_max_incl_components_h == 1) { - self->private_impl.f_width_in_mcus = ((self->private_impl.f_width + 7) / 8); - } else if (self->private_impl.f_max_incl_components_h == 2) { - self->private_impl.f_width_in_mcus = ((self->private_impl.f_width + 15) / 16); - } else if (self->private_impl.f_max_incl_components_h == 3) { - self->private_impl.f_width_in_mcus = ((self->private_impl.f_width + 23) / 24); - } else { - self->private_impl.f_width_in_mcus = ((self->private_impl.f_width + 31) / 32); - } - if (self->private_impl.f_max_incl_components_v == 1) { - self->private_impl.f_height_in_mcus = ((self->private_impl.f_height + 7) / 8); - } else if (self->private_impl.f_max_incl_components_v == 2) { - self->private_impl.f_height_in_mcus = ((self->private_impl.f_height + 15) / 16); - } else if (self->private_impl.f_max_incl_components_v == 3) { - self->private_impl.f_height_in_mcus = ((self->private_impl.f_height + 23) / 24); - } else { - self->private_impl.f_height_in_mcus = ((self->private_impl.f_height + 31) / 32); - } + self->private_impl.f_width_in_mcus = wuffs_jpeg__decoder__quantize_dimension(self, self->private_impl.f_width, 1, self->private_impl.f_max_incl_components_h); + self->private_impl.f_height_in_mcus = wuffs_jpeg__decoder__quantize_dimension(self, self->private_impl.f_height, 1, self->private_impl.f_max_incl_components_v); v_upper_bound = 65544; self->private_impl.f_components_workbuf_widths[0] = wuffs_base__u32__min(v_upper_bound, (8 * self->private_impl.f_width_in_mcus * ((uint32_t)(self->private_impl.f_components_h[0])))); self->private_impl.f_components_workbuf_widths[1] = wuffs_base__u32__min(v_upper_bound, (8 * self->private_impl.f_width_in_mcus * ((uint32_t)(self->private_impl.f_components_h[1])))); @@ -38568,11 +38609,19 @@ wuffs_jpeg__decoder__decode_sof( v_wh1 = (((uint64_t)(self->private_impl.f_components_workbuf_widths[1])) * ((uint64_t)(self->private_impl.f_components_workbuf_heights[1]))); v_wh2 = (((uint64_t)(self->private_impl.f_components_workbuf_widths[2])) * ((uint64_t)(self->private_impl.f_components_workbuf_heights[2]))); v_wh3 = (((uint64_t)(self->private_impl.f_components_workbuf_widths[3])) * ((uint64_t)(self->private_impl.f_components_workbuf_heights[3]))); + v_progressive = 0; + if (self->private_impl.f_sof_marker >= 194) { + v_progressive = 2; + } self->private_impl.f_components_workbuf_offsets[0] = 0; self->private_impl.f_components_workbuf_offsets[1] = (self->private_impl.f_components_workbuf_offsets[0] + v_wh0); self->private_impl.f_components_workbuf_offsets[2] = (self->private_impl.f_components_workbuf_offsets[1] + v_wh1); self->private_impl.f_components_workbuf_offsets[3] = (self->private_impl.f_components_workbuf_offsets[2] + v_wh2); self->private_impl.f_components_workbuf_offsets[4] = (self->private_impl.f_components_workbuf_offsets[3] + v_wh3); + self->private_impl.f_components_workbuf_offsets[5] = (self->private_impl.f_components_workbuf_offsets[4] + (v_wh0 * v_progressive)); + self->private_impl.f_components_workbuf_offsets[6] = (self->private_impl.f_components_workbuf_offsets[5] + (v_wh1 * v_progressive)); + self->private_impl.f_components_workbuf_offsets[7] = (self->private_impl.f_components_workbuf_offsets[6] + (v_wh2 * v_progressive)); + self->private_impl.f_components_workbuf_offsets[8] = (self->private_impl.f_components_workbuf_offsets[7] + (v_wh3 * v_progressive)); goto ok; ok: @@ -38594,6 +38643,30 @@ wuffs_jpeg__decoder__decode_sof( return status; } +// -------- func jpeg.decoder.quantize_dimension + +static uint32_t +wuffs_jpeg__decoder__quantize_dimension( + const wuffs_jpeg__decoder* self, + uint32_t a_width, + uint8_t a_h, + uint8_t a_max_incl_h) { + uint32_t v_ratio = 0; + + v_ratio = 0; + if (a_h > 0) { + v_ratio = ((uint32_t)((a_max_incl_h / a_h))); + } + if (v_ratio == 1) { + return ((a_width + 7) / 8); + } else if (v_ratio == 2) { + return ((a_width + 15) / 16); + } else if (v_ratio == 3) { + return ((a_width + 23) / 24); + } + return ((a_width + 31) / 32); +} + // -------- func jpeg.decoder.decode_frame_config WUFFS_BASE__MAYBE_STATIC wuffs_base__status @@ -38832,6 +38905,8 @@ wuffs_jpeg__decoder__do_decode_frame( uint32_t v_pixfmt = 0; wuffs_base__status v_status = wuffs_base__make_status(NULL); + uint64_t v_i = 0; + uint64_t v_i_end = 0; uint8_t v_c = 0; uint8_t v_marker = 0; @@ -38890,6 +38965,16 @@ wuffs_jpeg__decoder__do_decode_frame( } goto ok; } + v_i = self->private_impl.f_components_workbuf_offsets[4]; + v_i_end = self->private_impl.f_components_workbuf_offsets[8]; + if (v_i_end > ((uint64_t)(a_workbuf.len))) { + status = wuffs_base__make_status(wuffs_base__error__bad_workbuf_length); + goto exit; + } + while (v_i < v_i_end) { + a_workbuf.ptr[v_i] = 0; + v_i += 1; + } label__0__continue:; while (true) { while (true) { @@ -38925,43 +39010,46 @@ wuffs_jpeg__decoder__do_decode_frame( label__2__break:; if (v_marker == 0) { goto label__0__continue; - } else if ((208 <= v_marker) && (v_marker <= 215)) { - goto label__0__continue; - } - { - WUFFS_BASE__COROUTINE_SUSPENSION_POINT(4); - uint32_t t_2; - if (WUFFS_BASE__LIKELY(io2_a_src - iop_a_src >= 2)) { - t_2 = ((uint32_t)(wuffs_base__peek_u16be__no_bounds_check(iop_a_src))); - iop_a_src += 2; - } else { - self->private_data.s_do_decode_frame[0].scratch = 0; - WUFFS_BASE__COROUTINE_SUSPENSION_POINT(5); - while (true) { - if (WUFFS_BASE__UNLIKELY(iop_a_src == io2_a_src)) { - status = wuffs_base__make_status(wuffs_base__suspension__short_read); - goto suspend; - } - uint64_t* scratch = &self->private_data.s_do_decode_frame[0].scratch; - uint32_t num_bits_2 = ((uint32_t)(*scratch & 0xFF)); - *scratch >>= 8; - *scratch <<= 8; - *scratch |= ((uint64_t)(*iop_a_src++)) << (56 - num_bits_2); - if (num_bits_2 == 8) { - t_2 = ((uint32_t)(*scratch >> 48)); - break; - } - num_bits_2 += 8; - *scratch |= ((uint64_t)(num_bits_2)); - } + } else if ((208 <= v_marker) && (v_marker <= 217)) { + if (v_marker <= 215) { + goto label__0__continue; } - self->private_impl.f_payload_length = t_2; + } else { + { + WUFFS_BASE__COROUTINE_SUSPENSION_POINT(4); + uint32_t t_2; + if (WUFFS_BASE__LIKELY(io2_a_src - iop_a_src >= 2)) { + t_2 = ((uint32_t)(wuffs_base__peek_u16be__no_bounds_check(iop_a_src))); + iop_a_src += 2; + } else { + self->private_data.s_do_decode_frame[0].scratch = 0; + WUFFS_BASE__COROUTINE_SUSPENSION_POINT(5); + while (true) { + if (WUFFS_BASE__UNLIKELY(iop_a_src == io2_a_src)) { + status = wuffs_base__make_status(wuffs_base__suspension__short_read); + goto suspend; + } + uint64_t* scratch = &self->private_data.s_do_decode_frame[0].scratch; + uint32_t num_bits_2 = ((uint32_t)(*scratch & 0xFF)); + *scratch >>= 8; + *scratch <<= 8; + *scratch |= ((uint64_t)(*iop_a_src++)) << (56 - num_bits_2); + if (num_bits_2 == 8) { + t_2 = ((uint32_t)(*scratch >> 48)); + break; + } + num_bits_2 += 8; + *scratch |= ((uint64_t)(num_bits_2)); + } + } + self->private_impl.f_payload_length = t_2; + } + if (self->private_impl.f_payload_length < 2) { + status = wuffs_base__make_status(wuffs_jpeg__error__bad_marker); + goto exit; + } + self->private_impl.f_payload_length -= 2; } - if (self->private_impl.f_payload_length < 2) { - status = wuffs_base__make_status(wuffs_jpeg__error__bad_marker); - goto exit; - } - self->private_impl.f_payload_length -= 2; if (v_marker < 192) { status = wuffs_base__make_status(wuffs_jpeg__error__unsupported_marker); goto exit; @@ -38986,9 +39074,11 @@ wuffs_jpeg__decoder__do_decode_frame( status = wuffs_base__make_status(wuffs_jpeg__error__bad_sof_marker); goto exit; } else if (v_marker < 224) { - if (v_marker < 218) { + if (v_marker < 217) { status = wuffs_base__make_status(wuffs_jpeg__error__bad_marker); goto exit; + } else if (v_marker == 217) { + goto label__0__break; } else if (v_marker == 218) { if (a_src) { a_src->meta.ri = ((size_t)(iop_a_src - a_src->data.ptr)); @@ -39001,7 +39091,7 @@ wuffs_jpeg__decoder__do_decode_frame( if (status.repr) { goto suspend; } - goto label__0__break; + goto label__0__continue; } else if (v_marker == 219) { if (a_src) { a_src->meta.ri = ((size_t)(iop_a_src - a_src->data.ptr)); @@ -39352,10 +39442,7 @@ wuffs_jpeg__decoder__decode_sos( uint32_t v_decode_mcu_result = 0; uint32_t v_bitstream_length = 0; uint32_t v_b = 0; - uint32_t v_i = 0; uint8_t v_csel = 0; - uint64_t v_h = 0; - uint64_t v_v = 0; uint64_t v_stride = 0; uint64_t v_offset = 0; wuffs_base__status v_status = wuffs_base__make_status(NULL); @@ -39368,21 +39455,18 @@ wuffs_jpeg__decoder__decode_sos( switch (coro_susp_point) { WUFFS_BASE__COROUTINE_SUSPENSION_POINT_0; - if (((uint64_t)(a_workbuf.len)) < self->private_impl.f_components_workbuf_offsets[4]) { - status = wuffs_base__make_status(wuffs_base__error__bad_workbuf_length); - goto exit; - } WUFFS_BASE__COROUTINE_SUSPENSION_POINT(1); status = wuffs_jpeg__decoder__prepare_scan(self, a_src); if (status.repr) { goto suspend; } self->private_impl.f_next_restart_marker = 0; - self->private_impl.f_restarts_remaining = self->private_impl.f_restart_interval; self->private_impl.f_mcu_previous_dc_values[0] = 0; self->private_impl.f_mcu_previous_dc_values[1] = 0; self->private_impl.f_mcu_previous_dc_values[2] = 0; self->private_impl.f_mcu_previous_dc_values[3] = 0; + self->private_impl.f_restarts_remaining = self->private_impl.f_restart_interval; + self->private_impl.f_eob_run = 0; self->private_impl.f_bitstream_bits = 0; self->private_impl.f_bitstream_n_bits = 0; self->private_impl.f_bitstream_ri = 0; @@ -39393,7 +39477,12 @@ wuffs_jpeg__decoder__decode_sos( v_mx = 0; while (v_mx < self->private_impl.f_scan_width_in_mcus) { self->private_impl.f_mcu_current_block = 0; - self->private_impl.f_mcu_zig_index = 0; + self->private_impl.f_mcu_zig_index = ((uint32_t)(self->private_impl.f_scan_ss)); + if (self->private_impl.f_sof_marker >= 194) { + wuffs_jpeg__decoder__load_mcu_blocks(self, v_mx, v_my, a_workbuf); + } else { + wuffs_jpeg__decoder__clear_mcu_blocks(self); + } while (true) { v_decode_mcu_result = wuffs_jpeg__decoder__decode_mcu(self); if (v_decode_mcu_result == 0) { @@ -39414,13 +39503,14 @@ wuffs_jpeg__decoder__decode_sos( label__fill_bitstream__break:; } label__decode_mcu__break:; + if (self->private_impl.f_sof_marker >= 194) { + wuffs_jpeg__decoder__save_mcu_blocks(self, v_mx, v_my, a_workbuf); + } v_b = 0; while (v_b < self->private_impl.f_mcu_num_blocks) { v_csel = self->private_impl.f_scan_comps_cselector[self->private_impl.f_mcu_blocks_sselector[v_b]]; - v_h = ((uint64_t)(self->private_impl.f_components_h[v_csel])); - v_v = ((uint64_t)(self->private_impl.f_components_v[v_csel])); v_stride = ((uint64_t)(self->private_impl.f_components_workbuf_widths[v_csel])); - v_offset = (self->private_impl.f_components_workbuf_offsets[v_csel] + (8 * (((v_h * ((uint64_t)(v_mx))) + ((uint64_t)(self->private_impl.f_scan_comps_bx_offset[v_b]))) + (((v_v * ((uint64_t)(v_my))) + ((uint64_t)(self->private_impl.f_scan_comps_by_offset[v_b]))) * v_stride)))); + v_offset = (self->private_impl.f_mcu_blocks_offset[v_b] + (((uint64_t)(self->private_impl.f_mcu_blocks_mx_mul[v_b])) * ((uint64_t)(v_mx))) + (((uint64_t)(self->private_impl.f_mcu_blocks_my_mul[v_b])) * ((uint64_t)(v_my)))); if (v_offset <= ((uint64_t)(a_workbuf.len))) { wuffs_jpeg__decoder__decode_idct(self, wuffs_base__slice_u8__subslice_i(a_workbuf, v_offset), @@ -39430,15 +39520,6 @@ wuffs_jpeg__decoder__decode_sos( } v_b += 1; } - v_b = 0; - while (v_b < self->private_impl.f_mcu_num_blocks) { - v_i = 0; - while (v_i < 64) { - self->private_data.f_mcu_blocks[v_b][v_i] = 0; - v_i += 1; - } - v_b += 1; - } if (self->private_impl.f_restarts_remaining > 0) { #if defined(__GNUC__) #pragma GCC diagnostic push @@ -39454,11 +39535,12 @@ wuffs_jpeg__decoder__decode_sos( if (status.repr) { goto suspend; } - self->private_impl.f_restarts_remaining = self->private_impl.f_restart_interval; self->private_impl.f_mcu_previous_dc_values[0] = 0; self->private_impl.f_mcu_previous_dc_values[1] = 0; self->private_impl.f_mcu_previous_dc_values[2] = 0; self->private_impl.f_mcu_previous_dc_values[3] = 0; + self->private_impl.f_restarts_remaining = self->private_impl.f_restart_interval; + self->private_impl.f_eob_run = 0; self->private_impl.f_bitstream_bits = 0; self->private_impl.f_bitstream_n_bits = 0; self->private_impl.f_bitstream_ri = 0; @@ -39524,6 +39606,7 @@ wuffs_jpeg__decoder__prepare_scan( uint32_t v_b = 0; uint32_t v_bx_offset = 0; uint32_t v_by_offset = 0; + uint8_t v_csel = 0; const uint8_t* iop_a_src = NULL; const uint8_t* io0_a_src WUFFS_BASE__POTENTIALLY_UNUSED = NULL; @@ -39693,11 +39776,23 @@ wuffs_jpeg__decoder__prepare_scan( if (self->private_impl.f_scan_se != 0) { status = wuffs_base__make_status(wuffs_jpeg__error__bad_sos_marker); goto exit; + } else if (self->private_impl.f_scan_ah == 0) { + self->private_impl.choosy_decode_mcu = ( + &wuffs_jpeg__decoder__decode_mcu_progressive_dc_high_bits); + } else { + self->private_impl.choosy_decode_mcu = ( + &wuffs_jpeg__decoder__decode_mcu_progressive_dc_low_bit); } } else { if (self->private_impl.f_scan_num_components != 1) { status = wuffs_base__make_status(wuffs_jpeg__error__bad_sos_marker); goto exit; + } else if (self->private_impl.f_scan_ah == 0) { + self->private_impl.choosy_decode_mcu = ( + &wuffs_jpeg__decoder__decode_mcu_progressive_ac_high_bits); + } else { + self->private_impl.choosy_decode_mcu = ( + &wuffs_jpeg__decoder__decode_mcu_progressive_ac_low_bit); } } } @@ -39709,13 +39804,17 @@ wuffs_jpeg__decoder__prepare_scan( } v_i += 1; } - self->private_impl.f_scan_width_in_mcus = self->private_impl.f_width_in_mcus; - self->private_impl.f_scan_height_in_mcus = self->private_impl.f_height_in_mcus; if (self->private_impl.f_scan_num_components == 1) { self->private_impl.f_scan_comps_bx_offset[0] = 0; self->private_impl.f_scan_comps_by_offset[0] = 0; self->private_impl.f_mcu_num_blocks = 1; self->private_impl.f_mcu_blocks_sselector[0] = 0; + v_csel = self->private_impl.f_scan_comps_cselector[0]; + self->private_impl.f_mcu_blocks_offset[0] = self->private_impl.f_components_workbuf_offsets[v_csel]; + self->private_impl.f_mcu_blocks_mx_mul[0] = 8; + self->private_impl.f_mcu_blocks_my_mul[0] = (8 * self->private_impl.f_components_workbuf_widths[v_csel]); + self->private_impl.f_scan_width_in_mcus = wuffs_jpeg__decoder__quantize_dimension(self, self->private_impl.f_width, self->private_impl.f_components_h[v_csel], self->private_impl.f_max_incl_components_h); + self->private_impl.f_scan_height_in_mcus = wuffs_jpeg__decoder__quantize_dimension(self, self->private_impl.f_height, self->private_impl.f_components_v[v_csel], self->private_impl.f_max_incl_components_v); } else { v_total_hv = 0; v_i = 0; @@ -39749,6 +39848,16 @@ wuffs_jpeg__decoder__prepare_scan( goto exit; } self->private_impl.f_mcu_num_blocks = v_total_hv; + v_b = 0; + while (v_b < self->private_impl.f_mcu_num_blocks) { + v_csel = self->private_impl.f_scan_comps_cselector[self->private_impl.f_mcu_blocks_sselector[v_b]]; + self->private_impl.f_mcu_blocks_offset[v_b] = (self->private_impl.f_components_workbuf_offsets[v_csel] + (8 * ((uint64_t)(self->private_impl.f_scan_comps_bx_offset[v_b]))) + (8 * ((uint64_t)(self->private_impl.f_scan_comps_by_offset[v_b])) * ((uint64_t)(self->private_impl.f_components_workbuf_widths[v_csel])))); + self->private_impl.f_mcu_blocks_mx_mul[v_b] = (8 * ((uint32_t)(self->private_impl.f_components_h[v_csel]))); + self->private_impl.f_mcu_blocks_my_mul[v_b] = (8 * ((uint32_t)(self->private_impl.f_components_v[v_csel])) * self->private_impl.f_components_workbuf_widths[v_csel]); + v_b += 1; + } + self->private_impl.f_scan_width_in_mcus = self->private_impl.f_width_in_mcus; + self->private_impl.f_scan_height_in_mcus = self->private_impl.f_height_in_mcus; } goto ok; @@ -39839,6 +39948,115 @@ wuffs_jpeg__decoder__fill_bitstream( return wuffs_base__make_empty_struct(); } +// -------- func jpeg.decoder.clear_mcu_blocks + +static wuffs_base__empty_struct +wuffs_jpeg__decoder__clear_mcu_blocks( + wuffs_jpeg__decoder* self) { + uint32_t v_b = 0; + uint32_t v_i = 0; + + v_b = 0; + while (v_b < self->private_impl.f_mcu_num_blocks) { + v_i = 0; + while (v_i < 64) { + self->private_data.f_mcu_blocks[v_b][v_i] = 0; + v_i += 1; + } + v_b += 1; + } + return wuffs_base__make_empty_struct(); +} + +// -------- func jpeg.decoder.load_mcu_blocks + +static wuffs_base__empty_struct +wuffs_jpeg__decoder__load_mcu_blocks( + wuffs_jpeg__decoder* self, + uint32_t a_mx, + uint32_t a_my, + wuffs_base__slice_u8 a_workbuf) { + uint32_t v_b = 0; + uint8_t v_csel = 0; + uint64_t v_h = 0; + uint64_t v_v = 0; + uint64_t v_stride16 = 0; + uint64_t v_offset = 0; + wuffs_base__slice_u8 v_s = {0}; + uint32_t v_i = 0; + + v_h = 1; + v_v = 1; + v_b = 0; + while (v_b < self->private_impl.f_mcu_num_blocks) { + v_csel = self->private_impl.f_scan_comps_cselector[self->private_impl.f_mcu_blocks_sselector[v_b]]; + if (self->private_impl.f_scan_num_components > 1) { + v_h = ((uint64_t)(self->private_impl.f_components_h[v_csel])); + v_v = ((uint64_t)(self->private_impl.f_components_v[v_csel])); + } + v_stride16 = ((uint64_t)((self->private_impl.f_components_workbuf_widths[v_csel] * 16))); + v_offset = (self->private_impl.f_components_workbuf_offsets[(v_csel | 4)] + (((v_h * ((uint64_t)(a_mx))) + ((uint64_t)(self->private_impl.f_scan_comps_bx_offset[v_b]))) * 128) + (((v_v * ((uint64_t)(a_my))) + ((uint64_t)(self->private_impl.f_scan_comps_by_offset[v_b]))) * v_stride16)); + if (v_offset <= ((uint64_t)(a_workbuf.len))) { + v_s = wuffs_base__slice_u8__subslice_i(a_workbuf, v_offset); + if (((uint64_t)(v_s.len)) >= 128) { + v_i = 0; + while (v_i < 64) { + self->private_data.f_mcu_blocks[v_b][v_i] = ((uint16_t)(((((uint16_t)(v_s.ptr[((2 * v_i) + 0)])) << 0) | (((uint16_t)(v_s.ptr[((2 * v_i) + 1)])) << 8)))); + v_i += 1; + } + } + } + v_b += 1; + } + return wuffs_base__make_empty_struct(); +} + +// -------- func jpeg.decoder.save_mcu_blocks + +static wuffs_base__empty_struct +wuffs_jpeg__decoder__save_mcu_blocks( + wuffs_jpeg__decoder* self, + uint32_t a_mx, + uint32_t a_my, + wuffs_base__slice_u8 a_workbuf) { + uint32_t v_b = 0; + uint8_t v_csel = 0; + uint64_t v_h = 0; + uint64_t v_v = 0; + uint64_t v_stride16 = 0; + uint64_t v_offset = 0; + wuffs_base__slice_u8 v_s = {0}; + uint32_t v_i = 0; + uint16_t v_m = 0; + + v_h = 1; + v_v = 1; + v_b = 0; + while (v_b < self->private_impl.f_mcu_num_blocks) { + v_csel = self->private_impl.f_scan_comps_cselector[self->private_impl.f_mcu_blocks_sselector[v_b]]; + if (self->private_impl.f_scan_num_components > 1) { + v_h = ((uint64_t)(self->private_impl.f_components_h[v_csel])); + v_v = ((uint64_t)(self->private_impl.f_components_v[v_csel])); + } + v_stride16 = ((uint64_t)((self->private_impl.f_components_workbuf_widths[v_csel] * 16))); + v_offset = (self->private_impl.f_components_workbuf_offsets[(v_csel | 4)] + (((v_h * ((uint64_t)(a_mx))) + ((uint64_t)(self->private_impl.f_scan_comps_bx_offset[v_b]))) * 128) + (((v_v * ((uint64_t)(a_my))) + ((uint64_t)(self->private_impl.f_scan_comps_by_offset[v_b]))) * v_stride16)); + if (v_offset <= ((uint64_t)(a_workbuf.len))) { + v_s = wuffs_base__slice_u8__subslice_i(a_workbuf, v_offset); + if (((uint64_t)(v_s.len)) >= 128) { + v_i = 0; + while (v_i < 64) { + v_m = self->private_data.f_mcu_blocks[v_b][v_i]; + v_s.ptr[((2 * v_i) + 0)] = ((uint8_t)((v_m >> 0))); + v_s.ptr[((2 * v_i) + 1)] = ((uint8_t)((v_m >> 8))); + v_i += 1; + } + } + } + v_b += 1; + } + return wuffs_base__make_empty_struct(); +} + // -------- func jpeg.decoder.skip_past_the_next_restart_marker static wuffs_base__status @@ -40205,7 +40423,7 @@ wuffs_jpeg__decoder__workbuf_len( return wuffs_base__utility__empty_range_ii_u64(); } - return wuffs_base__utility__make_range_ii_u64(self->private_impl.f_components_workbuf_offsets[4], self->private_impl.f_components_workbuf_offsets[4]); + return wuffs_base__utility__make_range_ii_u64(self->private_impl.f_components_workbuf_offsets[8], self->private_impl.f_components_workbuf_offsets[8]); } // -------- func jpeg.decoder.decode_mcu @@ -40213,6 +40431,12 @@ wuffs_jpeg__decoder__workbuf_len( static uint32_t wuffs_jpeg__decoder__decode_mcu( wuffs_jpeg__decoder* self) { + return (*self->private_impl.choosy_decode_mcu)(self); +} + +static uint32_t +wuffs_jpeg__decoder__decode_mcu__choosy_default( + wuffs_jpeg__decoder* self) { uint32_t v_ret = 0; uint64_t v_bits = 0; uint32_t v_n_bits = 0; @@ -40369,7 +40593,9 @@ wuffs_jpeg__decoder__decode_mcu( goto label__ac_components__break; } self->private_impl.f_mcu_zig_index = (v_z + 1); - } else if ((v_ac_rrrr < 15) || ((self->private_impl.f_mcu_zig_index + 16) > 63)) { + } else if (v_ac_rrrr < 15) { + goto label__ac_components__break; + } else if ((self->private_impl.f_mcu_zig_index + 16) > 63) { goto label__ac_components__break; } else { self->private_impl.f_mcu_zig_index = (self->private_impl.f_mcu_zig_index + 16); @@ -40400,6 +40626,608 @@ wuffs_jpeg__decoder__decode_mcu( return v_ret; } +// -------- func jpeg.decoder.decode_mcu_progressive_ac_high_bits + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_ac_high_bits( + wuffs_jpeg__decoder* self) { + uint32_t v_ret = 0; + uint64_t v_bits = 0; + uint32_t v_n_bits = 0; + wuffs_base__io_buffer u_r = wuffs_base__empty_io_buffer(); + wuffs_base__io_buffer* v_r = &u_r; + const uint8_t* iop_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io0_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io1_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io2_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + uint32_t v_pos = 0; + uint8_t v_ac_h = 0; + uint32_t v_ac_symbol = 0; + uint32_t v_ac_ht_fast = 0; + uint32_t v_ac_bl = 0; + uint32_t v_ac_code = 0; + uint32_t v_ac_blm1 = 0; + uint32_t v_ac_ht_slow = 0; + uint32_t v_ac_value = 0; + uint32_t v_ac_rrrr = 0; + uint32_t v_ac_ssss = 0; + uint32_t v_z = 0; + + if (self->private_impl.f_eob_run > 0) { +#if defined(__GNUC__) +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wconversion" +#endif + self->private_impl.f_eob_run -= 1; +#if defined(__GNUC__) +#pragma GCC diagnostic pop +#endif + return 0; + } + v_bits = self->private_impl.f_bitstream_bits; + v_n_bits = self->private_impl.f_bitstream_n_bits; + if (self->private_impl.f_bitstream_ri > self->private_impl.f_bitstream_wi) { + return 2; + } + { + wuffs_base__io_buffer* o_0_v_r = v_r; + const uint8_t *o_0_iop_v_r = iop_v_r; + const uint8_t *o_0_io0_v_r = io0_v_r; + const uint8_t *o_0_io1_v_r = io1_v_r; + const uint8_t *o_0_io2_v_r = io2_v_r; + v_r = wuffs_base__io_reader__set( + &u_r, + &iop_v_r, + &io0_v_r, + &io1_v_r, + &io2_v_r, + wuffs_base__make_slice_u8_ij(self->private_data.f_bitstream_buffer, + self->private_impl.f_bitstream_ri, + self->private_impl.f_bitstream_wi), + ((uint64_t)(self->private_impl.f_bitstream_ri))); + while (true) { + while (true) { + while (true) { + if (((uint64_t)(io2_v_r - iop_v_r)) < 8) { + v_ret = 1; + goto label__goto_done__break; + } + v_bits |= (wuffs_base__peek_u64be__no_bounds_check(iop_v_r) >> (v_n_bits & 63)); + iop_v_r += ((63 - (v_n_bits & 63)) >> 3); + v_n_bits |= 56; + v_ac_h = (4 | self->private_impl.f_scan_comps_ta[self->private_impl.f_mcu_blocks_sselector[0]]); + v_ac_ht_fast = ((uint32_t)(self->private_impl.f_huff_tables_fast[v_ac_h][(v_bits >> 56)])); + v_ac_bl = (v_ac_ht_fast >> 8); + if (v_n_bits >= v_ac_bl) { + v_ac_symbol = (255 & v_ac_ht_fast); + v_bits <<= (v_ac_bl & 63); + v_n_bits -= v_ac_bl; + } else { + v_ac_code = ((uint32_t)((v_bits >> 55))); + v_ac_blm1 = 8; + v_bits <<= 9; + v_n_bits -= 9; + while (true) { + v_ac_ht_slow = self->private_impl.f_huff_tables_slow[v_ac_h][v_ac_blm1]; + if (v_ac_code < (v_ac_ht_slow >> 8)) { + v_ac_symbol = ((uint32_t)(self->private_impl.f_huff_tables_symbols[v_ac_h][(255 & ((uint32_t)(v_ac_code + v_ac_ht_slow)))])); + goto label__0__break; + } + v_ac_code = (((uint32_t)(v_ac_code << 1)) | ((uint32_t)((v_bits >> 63)))); + v_bits <<= 1; + v_n_bits -= 1; + v_ac_blm1 = ((v_ac_blm1 + 1) & 15); + if (v_ac_blm1 == 0) { + v_ac_symbol = 0; + goto label__0__break; + } + } + label__0__break:; + } + v_ac_rrrr = (v_ac_symbol >> 4); + v_ac_ssss = (v_ac_symbol & 15); + if (v_ac_ssss > 0) { + v_ac_value = ((uint32_t)((v_bits >> (64 - v_ac_ssss)))); + if ((v_bits >> 63) == 0) { + v_ac_value += ((uint32_t)(1 + ((uint32_t)(((uint32_t)(4294967295)) << v_ac_ssss)))); + } + v_bits <<= v_ac_ssss; + v_n_bits -= v_ac_ssss; + v_z = (self->private_impl.f_mcu_zig_index + v_ac_rrrr); + self->private_data.f_mcu_blocks[0][WUFFS_JPEG__UNZIG[v_z]] = ((uint16_t)(((uint32_t)(v_ac_value << self->private_impl.f_scan_al)))); + if ((v_z + 1) > ((uint32_t)(self->private_impl.f_scan_se))) { + goto label__ac_components__break; + } + self->private_impl.f_mcu_zig_index = (v_z + 1); + } else if (v_ac_rrrr < 15) { + self->private_impl.f_eob_run = ((uint16_t)(((((uint16_t)(1)) << v_ac_rrrr) - 1))); + if (v_ac_rrrr > 0) { + self->private_impl.f_eob_run += ((uint16_t)((v_bits >> (64 - v_ac_rrrr)))); + v_bits <<= v_ac_rrrr; + v_n_bits -= v_ac_rrrr; + } + goto label__ac_components__break; + } else if ((self->private_impl.f_mcu_zig_index + 16) > ((uint32_t)(self->private_impl.f_scan_se))) { + goto label__ac_components__break; + } else { + self->private_impl.f_mcu_zig_index = (self->private_impl.f_mcu_zig_index + 16); + } + } + label__ac_components__break:; + self->private_impl.f_mcu_zig_index = 0; + goto label__block__break; + } + label__block__break:; + goto label__goto_done__break; + } + label__goto_done__break:; + v_pos = ((uint32_t)(wuffs_base__u64__sat_add((v_r ? v_r->meta.pos : 0), ((uint64_t)(iop_v_r - io0_v_r))))); + if (v_pos > self->private_impl.f_bitstream_wi) { + v_ret = 2; + } else { + self->private_impl.f_bitstream_ri = v_pos; + } + v_r = o_0_v_r; + iop_v_r = o_0_iop_v_r; + io0_v_r = o_0_io0_v_r; + io1_v_r = o_0_io1_v_r; + io2_v_r = o_0_io2_v_r; + } + self->private_impl.f_bitstream_bits = v_bits; + self->private_impl.f_bitstream_n_bits = v_n_bits; + return v_ret; +} + +// -------- func jpeg.decoder.decode_mcu_progressive_ac_low_bit + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_ac_low_bit( + wuffs_jpeg__decoder* self) { + uint32_t v_ret = 0; + uint64_t v_bits = 0; + uint32_t v_n_bits = 0; + uint16_t v_one_lshift_scan_al = 0; + wuffs_base__io_buffer u_r = wuffs_base__empty_io_buffer(); + wuffs_base__io_buffer* v_r = &u_r; + const uint8_t* iop_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io0_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io1_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io2_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + uint32_t v_pos = 0; + uint8_t v_ac_h = 0; + uint32_t v_ac_symbol = 0; + uint32_t v_ac_ht_fast = 0; + uint32_t v_ac_bl = 0; + uint32_t v_ac_code = 0; + uint32_t v_ac_blm1 = 0; + uint32_t v_ac_ht_slow = 0; + uint32_t v_ac_value = 0; + uint32_t v_ac_rrrr = 0; + uint32_t v_ac_ssss = 0; + uint8_t v_unzig = 0; + bool v_bit = false; + + v_bits = self->private_impl.f_bitstream_bits; + v_n_bits = self->private_impl.f_bitstream_n_bits; + v_one_lshift_scan_al = (((uint16_t)(1)) << self->private_impl.f_scan_al); + if (self->private_impl.f_bitstream_ri > self->private_impl.f_bitstream_wi) { + return 2; + } + { + wuffs_base__io_buffer* o_0_v_r = v_r; + const uint8_t *o_0_iop_v_r = iop_v_r; + const uint8_t *o_0_io0_v_r = io0_v_r; + const uint8_t *o_0_io1_v_r = io1_v_r; + const uint8_t *o_0_io2_v_r = io2_v_r; + v_r = wuffs_base__io_reader__set( + &u_r, + &iop_v_r, + &io0_v_r, + &io1_v_r, + &io2_v_r, + wuffs_base__make_slice_u8_ij(self->private_data.f_bitstream_buffer, + self->private_impl.f_bitstream_ri, + self->private_impl.f_bitstream_wi), + ((uint64_t)(self->private_impl.f_bitstream_ri))); + while (true) { + while (true) { + while (true) { + if (self->private_impl.f_eob_run > 0) { + goto label__goto_do_eob__break; + } + while (true) { + if (((uint64_t)(io2_v_r - iop_v_r)) < 24) { + v_ret = 1; + goto label__goto_done__break; + } + v_bits |= (wuffs_base__peek_u64be__no_bounds_check(iop_v_r) >> (v_n_bits & 63)); + iop_v_r += ((63 - (v_n_bits & 63)) >> 3); + v_n_bits |= 56; + v_ac_h = (4 | self->private_impl.f_scan_comps_ta[self->private_impl.f_mcu_blocks_sselector[0]]); + v_ac_ht_fast = ((uint32_t)(self->private_impl.f_huff_tables_fast[v_ac_h][(v_bits >> 56)])); + v_ac_bl = (v_ac_ht_fast >> 8); + if (v_n_bits >= v_ac_bl) { + v_ac_symbol = (255 & v_ac_ht_fast); + v_bits <<= (v_ac_bl & 63); + v_n_bits -= v_ac_bl; + } else { + v_ac_code = ((uint32_t)((v_bits >> 55))); + v_ac_blm1 = 8; + v_bits <<= 9; + v_n_bits -= 9; + while (true) { + v_ac_ht_slow = self->private_impl.f_huff_tables_slow[v_ac_h][v_ac_blm1]; + if (v_ac_code < (v_ac_ht_slow >> 8)) { + v_ac_symbol = ((uint32_t)(self->private_impl.f_huff_tables_symbols[v_ac_h][(255 & ((uint32_t)(v_ac_code + v_ac_ht_slow)))])); + goto label__0__break; + } + v_ac_code = (((uint32_t)(v_ac_code << 1)) | ((uint32_t)((v_bits >> 63)))); + v_bits <<= 1; + v_n_bits -= 1; + v_ac_blm1 = ((v_ac_blm1 + 1) & 15); + if (v_ac_blm1 == 0) { + v_ac_symbol = 0; + goto label__0__break; + } + } + label__0__break:; + } + v_ac_rrrr = (v_ac_symbol >> 4); + v_ac_ssss = (v_ac_symbol & 15); + v_ac_value = 0; + if (v_ac_ssss > 0) { + v_ac_value = (((uint32_t)(1)) << self->private_impl.f_scan_al); + if ((v_bits >> 63) == 0) { + v_ac_value = ((uint32_t)(((uint32_t)(4294967295)) << self->private_impl.f_scan_al)); + } + v_bits <<= 1; + v_n_bits -= 1; + } else if (v_ac_rrrr < 15) { + self->private_impl.f_eob_run = (((uint16_t)(1)) << v_ac_rrrr); + if (v_ac_rrrr > 0) { + self->private_impl.f_eob_run += ((uint16_t)((v_bits >> (64 - v_ac_rrrr)))); + v_bits <<= v_ac_rrrr; + v_n_bits -= v_ac_rrrr; + } + goto label__goto_do_eob__break; + } + while (true) { + v_unzig = WUFFS_JPEG__UNZIG[self->private_impl.f_mcu_zig_index]; + if (self->private_data.f_mcu_blocks[0][v_unzig] != 0) { + if (v_n_bits == 0) { + if (((uint64_t)(io2_v_r - iop_v_r)) < 8) { + v_ret = 2; + goto label__goto_done__break; + } + v_bits |= (wuffs_base__peek_u64be__no_bounds_check(iop_v_r) >> (v_n_bits & 63)); + iop_v_r += ((63 - (v_n_bits & 63)) >> 3); + v_n_bits |= 56; + } + v_bit = ((v_bits >> 63) > 0); + v_bits <<= 1; + v_n_bits -= 1; + if (v_bit) { + if (self->private_data.f_mcu_blocks[0][v_unzig] < 32768) { +#if defined(__GNUC__) +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wconversion" +#endif + self->private_data.f_mcu_blocks[0][v_unzig] += v_one_lshift_scan_al; +#if defined(__GNUC__) +#pragma GCC diagnostic pop +#endif + } else { +#if defined(__GNUC__) +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wconversion" +#endif + self->private_data.f_mcu_blocks[0][v_unzig] -= v_one_lshift_scan_al; +#if defined(__GNUC__) +#pragma GCC diagnostic pop +#endif + } + } + } else if (v_ac_rrrr <= 0) { + goto label__refine_non_zeroes__break; + } else { + v_ac_rrrr -= 1; + } + if (self->private_impl.f_mcu_zig_index >= ((uint32_t)(self->private_impl.f_scan_se))) { + goto label__refine_non_zeroes__break; + } + self->private_impl.f_mcu_zig_index += 1; + } + label__refine_non_zeroes__break:; + if (v_ac_value != 0) { + self->private_data.f_mcu_blocks[0][WUFFS_JPEG__UNZIG[self->private_impl.f_mcu_zig_index]] = ((uint16_t)(v_ac_value)); + } + if (self->private_impl.f_mcu_zig_index >= ((uint32_t)(self->private_impl.f_scan_se))) { + goto label__ac_components__break; + } + self->private_impl.f_mcu_zig_index += 1; + } + label__ac_components__break:; + goto label__block__break; + } + label__goto_do_eob__break:; + if (self->private_impl.f_eob_run <= 0) { + v_ret = 2; + goto label__goto_done__break; + } + while (true) { + v_unzig = WUFFS_JPEG__UNZIG[self->private_impl.f_mcu_zig_index]; + if (self->private_data.f_mcu_blocks[0][v_unzig] != 0) { + if (v_n_bits == 0) { + if (((uint64_t)(io2_v_r - iop_v_r)) < 8) { + v_ret = 1; + goto label__goto_done__break; + } + v_bits |= (wuffs_base__peek_u64be__no_bounds_check(iop_v_r) >> (v_n_bits & 63)); + iop_v_r += ((63 - (v_n_bits & 63)) >> 3); + v_n_bits |= 56; + } + v_bit = ((v_bits >> 63) > 0); + v_bits <<= 1; + v_n_bits -= 1; + if (v_bit) { + if (self->private_data.f_mcu_blocks[0][v_unzig] < 32768) { +#if defined(__GNUC__) +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wconversion" +#endif + self->private_data.f_mcu_blocks[0][v_unzig] += v_one_lshift_scan_al; +#if defined(__GNUC__) +#pragma GCC diagnostic pop +#endif + } else { +#if defined(__GNUC__) +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wconversion" +#endif + self->private_data.f_mcu_blocks[0][v_unzig] -= v_one_lshift_scan_al; +#if defined(__GNUC__) +#pragma GCC diagnostic pop +#endif + } + } + } + if (self->private_impl.f_mcu_zig_index >= ((uint32_t)(self->private_impl.f_scan_se))) { + goto label__refine_eob_non_zeroes__break; + } + self->private_impl.f_mcu_zig_index += 1; + } + label__refine_eob_non_zeroes__break:; +#if defined(__GNUC__) +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wconversion" +#endif + self->private_impl.f_eob_run -= 1; +#if defined(__GNUC__) +#pragma GCC diagnostic pop +#endif + goto label__block__break; + } + label__block__break:; + goto label__goto_done__break; + } + label__goto_done__break:; + v_pos = ((uint32_t)(wuffs_base__u64__sat_add((v_r ? v_r->meta.pos : 0), ((uint64_t)(iop_v_r - io0_v_r))))); + if (v_pos > self->private_impl.f_bitstream_wi) { + v_ret = 2; + } else { + self->private_impl.f_bitstream_ri = v_pos; + } + v_r = o_0_v_r; + iop_v_r = o_0_iop_v_r; + io0_v_r = o_0_io0_v_r; + io1_v_r = o_0_io1_v_r; + io2_v_r = o_0_io2_v_r; + } + self->private_impl.f_bitstream_bits = v_bits; + self->private_impl.f_bitstream_n_bits = v_n_bits; + return v_ret; +} + +// -------- func jpeg.decoder.decode_mcu_progressive_dc_high_bits + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_dc_high_bits( + wuffs_jpeg__decoder* self) { + uint32_t v_ret = 0; + uint64_t v_bits = 0; + uint32_t v_n_bits = 0; + uint8_t v_csel = 0; + wuffs_base__io_buffer u_r = wuffs_base__empty_io_buffer(); + wuffs_base__io_buffer* v_r = &u_r; + const uint8_t* iop_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io0_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io1_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io2_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + uint32_t v_pos = 0; + uint8_t v_dc_h = 0; + uint32_t v_dc_symbol = 0; + uint32_t v_dc_ht_fast = 0; + uint32_t v_dc_bl = 0; + uint32_t v_dc_code = 0; + uint32_t v_dc_blm1 = 0; + uint32_t v_dc_ht_slow = 0; + uint32_t v_dc_value = 0; + + v_bits = self->private_impl.f_bitstream_bits; + v_n_bits = self->private_impl.f_bitstream_n_bits; + if (self->private_impl.f_bitstream_ri > self->private_impl.f_bitstream_wi) { + return 2; + } + { + wuffs_base__io_buffer* o_0_v_r = v_r; + const uint8_t *o_0_iop_v_r = iop_v_r; + const uint8_t *o_0_io0_v_r = io0_v_r; + const uint8_t *o_0_io1_v_r = io1_v_r; + const uint8_t *o_0_io2_v_r = io2_v_r; + v_r = wuffs_base__io_reader__set( + &u_r, + &iop_v_r, + &io0_v_r, + &io1_v_r, + &io2_v_r, + wuffs_base__make_slice_u8_ij(self->private_data.f_bitstream_buffer, + self->private_impl.f_bitstream_ri, + self->private_impl.f_bitstream_wi), + ((uint64_t)(self->private_impl.f_bitstream_ri))); + while (true) { + while (self->private_impl.f_mcu_current_block < self->private_impl.f_mcu_num_blocks) { + while (true) { + if (((uint64_t)(io2_v_r - iop_v_r)) < 8) { + v_ret = 1; + goto label__goto_done__break; + } + v_bits |= (wuffs_base__peek_u64be__no_bounds_check(iop_v_r) >> (v_n_bits & 63)); + iop_v_r += ((63 - (v_n_bits & 63)) >> 3); + v_n_bits |= 56; + v_dc_h = (0 | self->private_impl.f_scan_comps_td[self->private_impl.f_mcu_blocks_sselector[self->private_impl.f_mcu_current_block]]); + v_dc_ht_fast = ((uint32_t)(self->private_impl.f_huff_tables_fast[v_dc_h][(v_bits >> 56)])); + v_dc_bl = (v_dc_ht_fast >> 8); + if (v_n_bits >= v_dc_bl) { + v_dc_symbol = (15 & v_dc_ht_fast); + v_bits <<= (v_dc_bl & 63); + v_n_bits -= v_dc_bl; + } else { + v_dc_code = ((uint32_t)((v_bits >> 55))); + v_dc_blm1 = 8; + v_bits <<= 9; + v_n_bits -= 9; + while (true) { + v_dc_ht_slow = self->private_impl.f_huff_tables_slow[v_dc_h][v_dc_blm1]; + if (v_dc_code < (v_dc_ht_slow >> 8)) { + v_dc_symbol = (15 & ((uint32_t)(self->private_impl.f_huff_tables_symbols[v_dc_h][(255 & ((uint32_t)(v_dc_code + v_dc_ht_slow)))]))); + goto label__0__break; + } + v_dc_code = (((uint32_t)(v_dc_code << 1)) | ((uint32_t)((v_bits >> 63)))); + v_bits <<= 1; + v_n_bits -= 1; + v_dc_blm1 = ((v_dc_blm1 + 1) & 15); + if (v_dc_blm1 == 0) { + v_dc_symbol = 0; + goto label__0__break; + } + } + label__0__break:; + } + v_dc_value = ((uint32_t)(((v_bits >> 32) >> (32 - v_dc_symbol)))); + if ((v_bits >> 63) == 0) { + v_dc_value += ((uint32_t)(1 + ((uint32_t)(((uint32_t)(4294967295)) << v_dc_symbol)))); + } + v_bits <<= v_dc_symbol; + v_n_bits -= v_dc_symbol; + v_csel = self->private_impl.f_scan_comps_cselector[self->private_impl.f_mcu_blocks_sselector[self->private_impl.f_mcu_current_block]]; + self->private_impl.f_mcu_previous_dc_values[v_csel] += ((uint16_t)(v_dc_value)); + self->private_data.f_mcu_blocks[self->private_impl.f_mcu_current_block][0] = ((uint16_t)(self->private_impl.f_mcu_previous_dc_values[v_csel] << self->private_impl.f_scan_al)); + goto label__dc_component__break; + } + label__dc_component__break:; + self->private_impl.f_mcu_current_block += 1; + } + self->private_impl.f_mcu_current_block = 0; + goto label__goto_done__break; + } + label__goto_done__break:; + v_pos = ((uint32_t)(wuffs_base__u64__sat_add((v_r ? v_r->meta.pos : 0), ((uint64_t)(iop_v_r - io0_v_r))))); + if (v_pos > self->private_impl.f_bitstream_wi) { + v_ret = 2; + } else { + self->private_impl.f_bitstream_ri = v_pos; + } + v_r = o_0_v_r; + iop_v_r = o_0_iop_v_r; + io0_v_r = o_0_io0_v_r; + io1_v_r = o_0_io1_v_r; + io2_v_r = o_0_io2_v_r; + } + self->private_impl.f_bitstream_bits = v_bits; + self->private_impl.f_bitstream_n_bits = v_n_bits; + return v_ret; +} + +// -------- func jpeg.decoder.decode_mcu_progressive_dc_low_bit + +static uint32_t +wuffs_jpeg__decoder__decode_mcu_progressive_dc_low_bit( + wuffs_jpeg__decoder* self) { + uint32_t v_ret = 0; + uint64_t v_bits = 0; + uint32_t v_n_bits = 0; + uint16_t v_one_lshift_scan_al = 0; + wuffs_base__io_buffer u_r = wuffs_base__empty_io_buffer(); + wuffs_base__io_buffer* v_r = &u_r; + const uint8_t* iop_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io0_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io1_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + const uint8_t* io2_v_r WUFFS_BASE__POTENTIALLY_UNUSED = NULL; + uint32_t v_pos = 0; + + v_bits = self->private_impl.f_bitstream_bits; + v_n_bits = self->private_impl.f_bitstream_n_bits; + v_one_lshift_scan_al = (((uint16_t)(1)) << self->private_impl.f_scan_al); + if (self->private_impl.f_bitstream_ri > self->private_impl.f_bitstream_wi) { + return 2; + } + { + wuffs_base__io_buffer* o_0_v_r = v_r; + const uint8_t *o_0_iop_v_r = iop_v_r; + const uint8_t *o_0_io0_v_r = io0_v_r; + const uint8_t *o_0_io1_v_r = io1_v_r; + const uint8_t *o_0_io2_v_r = io2_v_r; + v_r = wuffs_base__io_reader__set( + &u_r, + &iop_v_r, + &io0_v_r, + &io1_v_r, + &io2_v_r, + wuffs_base__make_slice_u8_ij(self->private_data.f_bitstream_buffer, + self->private_impl.f_bitstream_ri, + self->private_impl.f_bitstream_wi), + ((uint64_t)(self->private_impl.f_bitstream_ri))); + while (true) { + while (self->private_impl.f_mcu_current_block < self->private_impl.f_mcu_num_blocks) { + while (true) { + if (((uint64_t)(io2_v_r - iop_v_r)) < 8) { + v_ret = 1; + goto label__goto_done__break; + } + v_bits |= (wuffs_base__peek_u64be__no_bounds_check(iop_v_r) >> (v_n_bits & 63)); + iop_v_r += ((63 - (v_n_bits & 63)) >> 3); + v_n_bits |= 56; + if ((v_bits >> 63) != 0) { + self->private_data.f_mcu_blocks[self->private_impl.f_mcu_current_block][0] |= v_one_lshift_scan_al; + } + v_bits <<= 1; + v_n_bits -= 1; + goto label__dc_component__break; + } + label__dc_component__break:; + self->private_impl.f_mcu_current_block += 1; + } + self->private_impl.f_mcu_current_block = 0; + goto label__goto_done__break; + } + label__goto_done__break:; + v_pos = ((uint32_t)(wuffs_base__u64__sat_add((v_r ? v_r->meta.pos : 0), ((uint64_t)(iop_v_r - io0_v_r))))); + if (v_pos > self->private_impl.f_bitstream_wi) { + v_ret = 2; + } else { + self->private_impl.f_bitstream_ri = v_pos; + } + v_r = o_0_v_r; + iop_v_r = o_0_iop_v_r; + io0_v_r = o_0_io0_v_r; + io1_v_r = o_0_io1_v_r; + io2_v_r = o_0_io2_v_r; + } + self->private_impl.f_bitstream_bits = v_bits; + self->private_impl.f_bitstream_n_bits = v_n_bits; + return v_ret; +} + #endif // !defined(WUFFS_CONFIG__MODULES) || defined(WUFFS_CONFIG__MODULE__JPEG) #if !defined(WUFFS_CONFIG__MODULES) || defined(WUFFS_CONFIG__MODULE__JSON) diff --git a/std/jpeg/decode_jpeg.wuffs b/std/jpeg/decode_jpeg.wuffs index d776ad4f..a7790849 100644 --- a/std/jpeg/decode_jpeg.wuffs +++ b/std/jpeg/decode_jpeg.wuffs @@ -35,7 +35,7 @@ pub status "#unsupported precision" pri status "#internal error: inconsistent decoder state" -pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0x4_0040_0100 +pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0xC_00C0_0300 pub struct decoder? implements base.image_decoder( width : base.u32[..= 0xFFFF], @@ -60,10 +60,11 @@ pub struct decoder? implements base.image_decoder( components_v : array[4] base.u8[..= 4], components_tq : array[4] base.u8[..= 3], - // components_workbuf_offsets, indexed by (b) and (b + 1), is the i and - // j bounds in the workbuf[i .. j] slice holding the post-IDCT values. - // components_workbuf_widths[b] is the row width, which is also the row - // stride (there is no slack). + // components_workbuf_offsets[0 .. 5], indexed by (csel) and (csel + + // 1), is the i and j bounds in the workbuf[i .. j] slice holding 8-bit + // samples (post-IDCT values) for the csel'th component. The row width + // is components_workbuf_widths[csel], which is also the row stride. + // There is no slack. // // For example, a 4:2:0 chroma-subsampled, 36 pixels wide × 28 pixels // high image has 6 blocks (4 Y, 1 Cb, 1 Cr; 8×8 samples each) per MCU. @@ -76,15 +77,44 @@ pub struct decoder? implements base.image_decoder( // 2: 0x18 = 24 // Cr // 3: 0x00 = 0 // Unused // - // The components_workbuf_offsets array is: - // 0: 0x000 = 0 - // 1: 0x600 = 1536 = previous + (48 * 32) - // 2: 0x790 = 1920 = previous + (24 * 16) - // 3: 0x900 = 2304 = previous + (24 * 16) - // 4: 0x900 = 2304 = previous + ( 0 * 0) + // The head of the components_workbuf_offsets array is: + // 0: 0x0000 = 0 + // 1: 0x0600 = 1536 = previous + (48 * 32) + // 2: 0x0780 = 1920 = previous + (24 * 16) + // 3: 0x0900 = 2304 = previous + (24 * 16) + // 4: 0x0900 = 2304 = previous + ( 0 * 0) + // + // That's components_workbuf_offsets[0 .. 5], for to a csel ranging in + // [0 .. 4]. For sequential JPEGs, components_workbuf_offsets[5 .. 9] + // are all equal to components_workbuf_offsets[4]. The four tail + // elements are unused, other than components_workbuf_offsets[8] (equal + // to components_workbuf_offsets[4]) being the workbuf_len. + // + // For progressive JPEGs, these hold the pre-IDCT pre-dequantization + // coefficients that persist between and are incrementally built by + // multiple Scans. There are 64 16-bit coefficients per block, just + // like there are 64 8-bit samples per block, but 16-bit coefficients + // obviously occupy twice as many bytes. Hence, the deltas between the + // components_workbuf_offsets tail elements are twice the deltas + // between its head elements. + // + // A block's 64 coefficients also have a different workbuf layout + // (placed in 128 consecutive bytes), compared to a block's 64 samples + // (placed in 8 rows; the stride is components_workbuf_widths[csel]). + // The 16-bit coefficients are stored little-endian. + // + // Continuing our 36 × 28 image example, if it was sequential encoded, + // the remaining four elements would all be 0x0900. If progressive + // encoded, the tail of the components_workbuf_offsets array is: + // 5: 0x1500 = 5376 = previous + (48 * 64) + // 6: 0x1800 = 6144 = previous + (24 * 32) + // 7: 0x1B00 = 6912 = previous + (24 * 32) + // 8: 0x1B00 = 6912 = previous + ( 0 * 0) + // + // The workbuf_len would be 0x0900 (baseline) or 0x1B00 (progressive). components_workbuf_widths : array[4] base.u32[..= 0x1_0008], components_workbuf_heights : array[4] base.u32[..= 0x1_0008], - components_workbuf_offsets : array[5] base.u64[..= 0x4_0040_0100], // 4 * 0x1_0008 * 0x1_0008. + components_workbuf_offsets : array[9] base.u64[..= 0xC_00C0_0300], // 12 * 0x1_0008 * 0x1_0008. scan_num_components : base.u32[..= 4], scan_comps_cselector : array[4] base.u8[..= 3], @@ -152,16 +182,58 @@ pub struct decoder? implements base.image_decoder( scan_ah : base.u8[..= 14], scan_al : base.u8[..= 13], + // These fields, the mx and my loop bounds in decode_sos, are usually + // measured "in MCUs" but, for progressive JPEGs' single-component + // scans, they're measured in blocks. scan_width_in_mcus : base.u32[..= 0x2000], scan_height_in_mcus : base.u32[..= 0x2000], + // The block's position within the MCU. For example, a 4:2:0 chroma- + // subsampled image has 6 blocks (4 Y, 1 Cb, 1 Cr) per MCU: + // 0: by = 0 bx = 0 Y + // 1: by = 0 bx = 1 Y + // 2: by = 1 bx = 0 Y + // 3: by = 1 bx = 1 Y + // 4: by = 0 bx = 0 Cb + // 5: by = 0 bx = 0 Cr + // + // The remaining (16 - mcu_num_blocks) elements are unused. scan_comps_bx_offset : array[16] base.u8[..= 3], scan_comps_by_offset : array[16] base.u8[..= 3], - mcu_num_blocks : base.u32[..= 10], - mcu_current_block : base.u32[..= 10], + mcu_num_blocks : base.u32[..= 10], + mcu_current_block : base.u32[..= 10], + mcu_zig_index : base.u32[..= 63], + mcu_blocks_sselector : array[16] base.u8[..= 3], - mcu_zig_index : base.u32[..= 63], + + // (offset[b] + (mx_mul[b] * mx) + (my_mul[b] * my)) is the workbuf + // offset of the top-left sample of the b'th block in the (mx, my) MCU. + // The block's 64 samples are arranged in 8 rows, each separated by a + // stride equal to components_workbuf_widths[csel]. + // + // For single-component scans, its calculation depends on: + // - components_workbuf_offsets + // - components_workbuf_widths + // + // For multi-component scans, its calculation depends on: + // - components_workbuf_offsets + // - components_workbuf_widths + // - scan_comps_bx_offset + // - scan_comps_by_offset + // - components_h + // - components_v + // + // Continuing our 4:2:0 chroma-subsampled, 36 × 28 sequential example: + // 0: my_mul = 768 mx_mul = 16 offset = 0x0000 + 0x0000 = 0 + // 1: my_mul = 768 mx_mul = 16 offset = 0x0000 + 0x0008 = 8 + // 2: my_mul = 768 mx_mul = 16 offset = 0x0000 + 0x0180 = 384 + // 3: my_mul = 768 mx_mul = 16 offset = 0x0000 + 0x0188 = 392 + // 4: my_mul = 192 mx_mul = 8 offset = 0x0600 + 0x0000 = 1536 + // 5: my_mul = 192 mx_mul = 8 offset = 0x0780 + 0x0000 = 1920 + mcu_blocks_offset : array[10] base.u64[..= 0xC_00D8_03D8], // (8 * 3 * (0x1_0008 + 1)) + 0xC_00C0_0300 + mcu_blocks_mx_mul : array[10] base.u32[..= 0x00_0020], // 8 * 4. + mcu_blocks_my_mul : array[10] base.u32[..= 0x20_0100], // 8 * 4 * 0x1_0008. mcu_previous_dc_values : array[4] base.u16, @@ -169,6 +241,12 @@ pub struct decoder? implements base.image_decoder( saved_restart_interval : base.u16, restarts_remaining : base.u16, + // End-Of-Band run count, per Section G.1.2.2: "Progressive encoding of + // AC coefficients with Huffman coding... An EOB run of length 5 means + // that the current block and the next four blocks have an end-of-band + // with no intervening non-zero coefficients". + eob_run : base.u16, + frame_config_io_position : base.u64, payload_length : base.u32[..= 0xFFFF], @@ -278,18 +356,20 @@ pri func decoder.do_decode_image_config?(dst: nptr base.image_config, src: base. if marker == 0x00 { // Ignore byte stuffing. continue - } else if (0xD0 <= marker) and (marker <= 0xD7) { - // RSTn (Restart) markers have no payload. - continue + } else if (0xD0 <= marker) and (marker <= 0xD9) { + // RSTn (Restart), SOI and EOI markers have no payload. + if marker <= 0xD7 { + continue + } + } else { + this.payload_length = args.src.read_u16be_as_u32?() + // Payload length includes the 2 bytes for the u16be just read. + if this.payload_length < 2 { + return "#bad marker" + } + this.payload_length -= 2 } - // Payload length includes the 2 bytes for the u16be payload length. - this.payload_length = args.src.read_u16be_as_u32?() - if this.payload_length < 2 { - return "#bad marker" - } - this.payload_length -= 2 - // Switch on the marker. if marker < 0xC0 { return "#unsupported marker" @@ -451,6 +531,8 @@ pri func decoder.decode_sof?(src: base.io_reader) { var wh2 : base.u64[..= 0x1_0010_0040] // 0x1_0008 * 0x1_0008. var wh3 : base.u64[..= 0x1_0010_0040] // 0x1_0008 * 0x1_0008. + var progressive : base.u64[..= 2] + if this.payload_length < 6 { return "#bad SOF marker" } @@ -555,25 +637,10 @@ pri func decoder.decode_sof?(src: base.io_reader) { } } - if this.max_incl_components_h == 1 { - this.width_in_mcus = (this.width + 0x07) / 0x08 - } else if this.max_incl_components_h == 2 { - this.width_in_mcus = (this.width + 0x0F) / 0x10 - } else if this.max_incl_components_h == 3 { - this.width_in_mcus = (this.width + 0x17) / 0x18 - } else { - this.width_in_mcus = (this.width + 0x1F) / 0x20 - } - - if this.max_incl_components_v == 1 { - this.height_in_mcus = (this.height + 0x07) / 0x08 - } else if this.max_incl_components_v == 2 { - this.height_in_mcus = (this.height + 0x0F) / 0x10 - } else if this.max_incl_components_v == 3 { - this.height_in_mcus = (this.height + 0x17) / 0x18 - } else { - this.height_in_mcus = (this.height + 0x1F) / 0x20 - } + this.width_in_mcus = this.quantize_dimension( + width: this.width, h: 1, max_incl_h: this.max_incl_components_h) + this.height_in_mcus = this.quantize_dimension( + width: this.height, h: 1, max_incl_h: this.max_incl_components_v) upper_bound = 0x1_0008 @@ -600,11 +667,39 @@ pri func decoder.decode_sof?(src: base.io_reader) { wh2 = (this.components_workbuf_widths[2] as base.u64) * (this.components_workbuf_heights[2] as base.u64) wh3 = (this.components_workbuf_widths[3] as base.u64) * (this.components_workbuf_heights[3] as base.u64) + progressive = 0 + if this.sof_marker >= 0xC2 { + // Pre-IDCT block coefficients (mcu_blocks elements) are 2 bytes each. + progressive = 2 + } + this.components_workbuf_offsets[0] = 0 this.components_workbuf_offsets[1] = this.components_workbuf_offsets[0] + wh0 this.components_workbuf_offsets[2] = this.components_workbuf_offsets[1] + wh1 this.components_workbuf_offsets[3] = this.components_workbuf_offsets[2] + wh2 this.components_workbuf_offsets[4] = this.components_workbuf_offsets[3] + wh3 + this.components_workbuf_offsets[5] = this.components_workbuf_offsets[4] + (wh0 * progressive) + this.components_workbuf_offsets[6] = this.components_workbuf_offsets[5] + (wh1 * progressive) + this.components_workbuf_offsets[7] = this.components_workbuf_offsets[6] + (wh2 * progressive) + this.components_workbuf_offsets[8] = this.components_workbuf_offsets[7] + (wh3 * progressive) +} + +pri func decoder.quantize_dimension(width: base.u32[..= 0xFFFF], h: base.u8[..= 4], max_incl_h: base.u8[..= 4]) base.u32[..= 0x2000] { + var ratio : base.u32 + + ratio = 0 + if args.h > 0 { + ratio = (args.max_incl_h / args.h) as base.u32 + } + + if ratio == 1 { + return (args.width + 0x07) / 0x08 + } else if ratio == 2 { + return (args.width + 0x0F) / 0x10 + } else if ratio == 3 { + return (args.width + 0x17) / 0x18 + } + return (args.width + 0x1F) / 0x20 } pub func decoder.decode_frame_config?(dst: nptr base.frame_config, src: base.io_reader) { @@ -668,6 +763,8 @@ pub func decoder.decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reader, 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 pixfmt : base.u32 var status : base.status + var i : base.u64 + var i_end : base.u64 var c : base.u8 var marker : base.u8 @@ -693,6 +790,23 @@ pri func decoder.do_decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reade return status } + // For progressive JPEGs, zero-initialize the saved pre-IDCT blocks. For + // sequential JPEGs, this is a no-op, other than checking that args.workbuf + // is long enough. + i = this.components_workbuf_offsets[4] + i_end = this.components_workbuf_offsets[8] + if i_end > args.workbuf.length() { + return base."#bad workbuf length" + } + while i < i_end, + inv i_end <= args.workbuf.length(), + { + assert i < 0xFFFF_FFFF_FFFF_FFFF via "a < b: a < c; c <= b"(c: i_end) + assert i < args.workbuf.length() via "a < b: a < c; c <= b"(c: i_end) + args.workbuf[i] = 0 + i += 1 + } endwhile + // Process chunks (markers and their payloads). while true { // Read the marker (a two-byte 0xFF 0x?? sequence). @@ -717,18 +831,20 @@ pri func decoder.do_decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reade if marker == 0x00 { // Ignore byte stuffing. continue - } else if (0xD0 <= marker) and (marker <= 0xD7) { - // RSTn (Restart) markers have no payload. - continue + } else if (0xD0 <= marker) and (marker <= 0xD9) { + // RSTn (Restart), SOI and EOI markers have no payload. + if marker <= 0xD7 { + continue + } + } else { + this.payload_length = args.src.read_u16be_as_u32?() + // Payload length includes the 2 bytes for the u16be just read. + if this.payload_length < 2 { + return "#bad marker" + } + this.payload_length -= 2 } - // Payload length includes the 2 bytes for the u16be payload length. - this.payload_length = args.src.read_u16be_as_u32?() - if this.payload_length < 2 { - return "#bad marker" - } - this.payload_length -= 2 - // Switch on the marker. if marker < 0xC0 { return "#unsupported marker" @@ -744,15 +860,17 @@ pri func decoder.do_decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reade return "#bad SOF marker" } else if marker < 0xE0 { - if marker < 0xDA { - // RSTn markers are already handled above. We either have 0xD8 - // (SOI: Start Of Image) or 0xD9 (EOI: End Of Image), neither - // of which are valid here. + if marker < 0xD9 { + // RSTn markers are already handled above. We have 0xD8 (SOI: + // Start Of Image), which is invalid here. return "#bad marker" + } else if marker == 0xD9 { // EOI (End Of Image). + break + } else if marker == 0xDA { // SOS (Start Of Scan). this.decode_sos?(dst: args.dst, src: args.src, workbuf: args.workbuf) - break + continue } else if marker == 0xDB { // DQT (Define Quantization Table). this.decode_dqt?(src: args.src) @@ -814,7 +932,7 @@ pri func decoder.decode_dht?(src: base.io_reader) { th = c & 0x0F tc4_th = ((tc * 4) | th) as base.u8 - // SOF0 (which is 0xC0) means a Baseline JPEG. The Baseline (th <= 1) + // SOF0 (which is 0xC0) means a baseline JPEG. The baseline (th <= 1) // restriction is specified in table B.5. if (this.sof_marker == 0xC0) and ((tc4_th & 3) > 1) { return "#bad DHT marker" @@ -1000,28 +1118,22 @@ pri func decoder.decode_sos?(dst: ptr base.pixel_buffer, src: base.io_reader, wo var bitstream_length : base.u32 var b : base.u32 - var i : base.u32 var csel : base.u8[..= 3] - var h : base.u64[..= 4] - var v : base.u64[..= 4] - var stride : base.u64[..= 0x4_0000] + var stride : base.u64[..= 0x1_0008] var offset : base.u64 var status : base.status - if args.workbuf.length() < this.components_workbuf_offsets[4] { - return base."#bad workbuf length" - } - this.prepare_scan?(src: args.src) // Reset. this.next_restart_marker = 0 - this.restarts_remaining = this.restart_interval this.mcu_previous_dc_values[0] = 0 this.mcu_previous_dc_values[1] = 0 this.mcu_previous_dc_values[2] = 0 this.mcu_previous_dc_values[3] = 0 + this.restarts_remaining = this.restart_interval + this.eob_run = 0 this.bitstream_bits = 0 this.bitstream_n_bits = 0 this.bitstream_ri = 0 @@ -1037,7 +1149,13 @@ pri func decoder.decode_sos?(dst: ptr base.pixel_buffer, src: base.io_reader, wo { assert mx < 0x2000 via "a < b: a < c; c <= b"(c: this.scan_width_in_mcus) this.mcu_current_block = 0 - this.mcu_zig_index = 0 + this.mcu_zig_index = this.scan_ss as base.u32 + + if this.sof_marker >= 0xC2 { + this.load_mcu_blocks!(mx: mx, my: my, workbuf: args.workbuf) + } else { + this.clear_mcu_blocks!() + } while.decode_mcu true, inv my < 0x2000, @@ -1063,6 +1181,10 @@ pri func decoder.decode_sos?(dst: ptr base.pixel_buffer, src: base.io_reader, wo } endwhile.fill_bitstream } endwhile.decode_mcu + if this.sof_marker >= 0xC2 { + this.save_mcu_blocks!(mx: mx, my: my, workbuf: args.workbuf) + } + // Apply IDCT. b = 0 while b < this.mcu_num_blocks, @@ -1071,12 +1193,10 @@ pri func decoder.decode_sos?(dst: ptr base.pixel_buffer, src: base.io_reader, wo { assert b < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) csel = this.scan_comps_cselector[this.mcu_blocks_sselector[b]] - h = this.components_h[csel] as base.u64 - v = this.components_v[csel] as base.u64 stride = this.components_workbuf_widths[csel] as base.u64 - offset = this.components_workbuf_offsets[csel] + (8 * - ((((h * (mx as base.u64)) + (this.scan_comps_bx_offset[b] as base.u64))) + - ((((v * (my as base.u64)) + (this.scan_comps_by_offset[b] as base.u64))) * stride))) + offset = this.mcu_blocks_offset[b] + + ((this.mcu_blocks_mx_mul[b] as base.u64) * (mx as base.u64)) + + ((this.mcu_blocks_my_mul[b] as base.u64) * (my as base.u64)) if offset <= args.workbuf.length() { this.decode_idct!( dst_buffer: args.workbuf[offset ..], @@ -1088,25 +1208,6 @@ pri func decoder.decode_sos?(dst: ptr base.pixel_buffer, src: base.io_reader, wo b += 1 } endwhile - // Clear this.mcu_blocks. - b = 0 - while b < this.mcu_num_blocks, - inv my < 0x2000, - inv mx < 0x2000, - { - assert b < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) - i = 0 - while i < 64, - inv my < 0x2000, - inv mx < 0x2000, - inv b < 10, - { - this.mcu_blocks[b][i] = 0 - i += 1 - } endwhile - b += 1 - } endwhile - // Check for a restart. if this.restarts_remaining > 0 { this.restarts_remaining -= 1 @@ -1114,11 +1215,12 @@ pri func decoder.decode_sos?(dst: ptr base.pixel_buffer, src: base.io_reader, wo this.skip_past_the_next_restart_marker?(src: args.src) // Reset. - this.restarts_remaining = this.restart_interval this.mcu_previous_dc_values[0] = 0 this.mcu_previous_dc_values[1] = 0 this.mcu_previous_dc_values[2] = 0 this.mcu_previous_dc_values[3] = 0 + this.restarts_remaining = this.restart_interval + this.eob_run = 0 this.bitstream_bits = 0 this.bitstream_n_bits = 0 this.bitstream_ri = 0 @@ -1150,6 +1252,7 @@ pri func decoder.prepare_scan?(src: base.io_reader) { var b : base.u32 var bx_offset : base.u32 var by_offset : base.u32 + var csel : base.u8[..= 3] if (this.payload_length < 6) or (this.payload_length > 12) { return "#bad SOS marker" @@ -1201,7 +1304,7 @@ pri func decoder.prepare_scan?(src: base.io_reader) { } this.scan_comps_td[i] = c >> 4 this.scan_comps_ta[i] = c & 0x0F - // SOF0 (which is 0xC0) means a Baseline JPEG. The Baseline (tx <= 1) + // SOF0 (which is 0xC0) means a baseline JPEG. The baseline (tx <= 1) // restriction is specified in table B.5. if this.sof_marker == 0xC0 { if (this.scan_comps_td[i] > 1) or (this.scan_comps_ta[i] > 1) { @@ -1251,11 +1354,19 @@ pri func decoder.prepare_scan?(src: base.io_reader) { // Progressive DC scans can have only one spectral element. if this.scan_se <> 0 { return "#bad SOS marker" + } else if this.scan_ah == 0 { + choose decode_mcu = [decode_mcu_progressive_dc_high_bits] + } else { + choose decode_mcu = [decode_mcu_progressive_dc_low_bit] } } else { // Progressive AC scans can have only one component. if this.scan_num_components <> 1 { return "#bad SOS marker" + } else if this.scan_ah == 0 { + choose decode_mcu = [decode_mcu_progressive_ac_high_bits] + } else { + choose decode_mcu = [decode_mcu_progressive_ac_low_bit] } } } @@ -1275,15 +1386,22 @@ pri func decoder.prepare_scan?(src: base.io_reader) { i += 1 } endwhile - this.scan_width_in_mcus = this.width_in_mcus - this.scan_height_in_mcus = this.height_in_mcus - if this.scan_num_components == 1 { this.scan_comps_bx_offset[0] = 0 this.scan_comps_by_offset[0] = 0 this.mcu_num_blocks = 1 this.mcu_blocks_sselector[0] = 0 + csel = this.scan_comps_cselector[0] + this.mcu_blocks_offset[0] = this.components_workbuf_offsets[csel] + this.mcu_blocks_mx_mul[0] = 8 + this.mcu_blocks_my_mul[0] = 8 * this.components_workbuf_widths[csel] + + this.scan_width_in_mcus = this.quantize_dimension( + width: this.width, h: this.components_h[csel], max_incl_h: this.max_incl_components_h) + this.scan_height_in_mcus = this.quantize_dimension( + width: this.height, h: this.components_v[csel], max_incl_h: this.max_incl_components_v) + } else { total_hv = 0 i = 0 @@ -1322,6 +1440,21 @@ pri func decoder.prepare_scan?(src: base.io_reader) { return "#bad SOS marker" } this.mcu_num_blocks = total_hv + + b = 0 + while b < this.mcu_num_blocks { + assert b < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) + csel = this.scan_comps_cselector[this.mcu_blocks_sselector[b]] + this.mcu_blocks_offset[b] = this.components_workbuf_offsets[csel] + + (8 * (this.scan_comps_bx_offset[b] as base.u64)) + + (8 * (this.scan_comps_by_offset[b] as base.u64) * (this.components_workbuf_widths[csel] as base.u64)) + this.mcu_blocks_mx_mul[b] = 8 * (this.components_h[csel] as base.u32) + this.mcu_blocks_my_mul[b] = 8 * (this.components_v[csel] as base.u32) * this.components_workbuf_widths[csel] + b += 1 + } endwhile + + this.scan_width_in_mcus = this.width_in_mcus + this.scan_height_in_mcus = this.height_in_mcus } } @@ -1382,6 +1515,111 @@ pri func decoder.fill_bitstream!(src: base.io_reader) { this.bitstream_wi = wi } +pri func decoder.clear_mcu_blocks!() { + var b : base.u32 + var i : base.u32 + + b = 0 + while b < this.mcu_num_blocks { + assert b < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) + i = 0 + while i < 64, + inv b < 10, + { + this.mcu_blocks[b][i] = 0 + i += 1 + } endwhile + b += 1 + } endwhile +} + +pri func decoder.load_mcu_blocks!(mx: base.u32[..= 0x2000], my: base.u32[..= 0x2000], workbuf: slice base.u8) { + var b : base.u32 + var csel : base.u8[..= 3] + var h : base.u64[..= 4] + var v : base.u64[..= 4] + var stride16 : base.u64[..= 0x10_0080] + var offset : base.u64 + var s : slice base.u8 + var i : base.u32 + + h = 1 + v = 1 + b = 0 + while b < this.mcu_num_blocks { + assert b < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) + csel = this.scan_comps_cselector[this.mcu_blocks_sselector[b]] + if this.scan_num_components > 1 { + h = this.components_h[csel] as base.u64 + v = this.components_v[csel] as base.u64 + } + stride16 = (this.components_workbuf_widths[csel] * 16) as base.u64 + offset = this.components_workbuf_offsets[csel | 4] + + (((h * (args.mx as base.u64)) + (this.scan_comps_bx_offset[b] as base.u64)) * 128) + + (((v * (args.my as base.u64)) + (this.scan_comps_by_offset[b] as base.u64)) * stride16) + if offset <= args.workbuf.length() { + s = args.workbuf[offset ..] + if s.length() >= 128 { + i = 0 + while i < 64, + inv b < 10, + inv s.length() >= 128, + { + this.mcu_blocks[b][i] = ( + ((s[(2 * i) + 0] as base.u16) << 0) | + ((s[(2 * i) + 1] as base.u16) << 8)) as base.u16 + i += 1 + } endwhile + } + } + b += 1 + } endwhile +} + +pri func decoder.save_mcu_blocks!(mx: base.u32[..= 0x2000], my: base.u32[..= 0x2000], workbuf: slice base.u8) { + var b : base.u32 + var csel : base.u8[..= 3] + var h : base.u64[..= 4] + var v : base.u64[..= 4] + var stride16 : base.u64[..= 0x10_0080] + var offset : base.u64 + var s : slice base.u8 + var i : base.u32 + var m : base.u16 + + h = 1 + v = 1 + b = 0 + while b < this.mcu_num_blocks { + assert b < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) + csel = this.scan_comps_cselector[this.mcu_blocks_sselector[b]] + if this.scan_num_components > 1 { + h = this.components_h[csel] as base.u64 + v = this.components_v[csel] as base.u64 + } + stride16 = (this.components_workbuf_widths[csel] * 16) as base.u64 + offset = this.components_workbuf_offsets[csel | 4] + + (((h * (args.mx as base.u64)) + (this.scan_comps_bx_offset[b] as base.u64)) * 128) + + (((v * (args.my as base.u64)) + (this.scan_comps_by_offset[b] as base.u64)) * stride16) + if offset <= args.workbuf.length() { + s = args.workbuf[offset ..] + if s.length() >= 128 { + i = 0 + while i < 64, + inv b < 10, + inv s.length() >= 128, + { + m = this.mcu_blocks[b][i] + s[(2 * i) + 0] = ((m >> 0) & 0xFF) as base.u8 + s[(2 * i) + 1] = ((m >> 8) & 0xFF) as base.u8 + i += 1 + } endwhile + } + } + b += 1 + } endwhile +} + pri func decoder.skip_past_the_next_restart_marker?(src: base.io_reader) { var c : base.u8 @@ -1596,6 +1834,6 @@ pub func decoder.tell_me_more?(dst: base.io_writer, minfo: nptr base.more_inform pub func decoder.workbuf_len() base.range_ii_u64 { return this.util.make_range_ii_u64( - min_incl: this.components_workbuf_offsets[4], - max_incl: this.components_workbuf_offsets[4]) + min_incl: this.components_workbuf_offsets[8], + max_incl: this.components_workbuf_offsets[8]) } diff --git a/std/jpeg/decode_mcu_default.wuffs b/std/jpeg/decode_mcu_default.wuffs index 38963798..4eade286 100644 --- a/std/jpeg/decode_mcu_default.wuffs +++ b/std/jpeg/decode_mcu_default.wuffs @@ -12,12 +12,15 @@ // See the License for the specific language governing permissions and // limitations under the License. -pri func decoder.decode_mcu!() base.u32 { +pri func decoder.decode_mcu!() base.u32, + choosy, +{ var ret : base.u32 var bits : base.u64 var n_bits : base.u32 - var csel : base.u8[..= 3] + + var csel : base.u8[..= 3] var r : base.io_reader var pos : base.u32 @@ -45,7 +48,8 @@ pri func decoder.decode_mcu!() base.u32 { var ac_rrrr : base.u32[..= 15] var ac_ssss : base.u32[..= 15] - var z : base.u32[..= 78] // 78 = 63 + 15. + + var z : base.u32[..= 78] // 78 = 63 + 15. bits = this.bitstream_bits n_bits = this.bitstream_n_bits @@ -114,7 +118,8 @@ pri func decoder.decode_mcu!() base.u32 { n_bits ~mod-= dc_symbol csel = this.scan_comps_cselector[this.mcu_blocks_sselector[this.mcu_current_block]] this.mcu_previous_dc_values[csel] ~mod+= (dc_value & 0xFFFF) as base.u16 - this.mcu_blocks[this.mcu_current_block][0] = this.mcu_previous_dc_values[csel] + this.mcu_blocks[this.mcu_current_block][0] = + this.mcu_previous_dc_values[csel] this.mcu_zig_index = 1 break.dc_component @@ -178,12 +183,15 @@ pri func decoder.decode_mcu!() base.u32 { bits ~mod<<= ac_ssss n_bits ~mod-= ac_ssss z = this.mcu_zig_index + ac_rrrr - this.mcu_blocks[this.mcu_current_block][UNZIG[z]] = (ac_value & 0xFFFF) as base.u16 + this.mcu_blocks[this.mcu_current_block][UNZIG[z]] = + (ac_value & 0xFFFF) as base.u16 if (z + 1) > 63 { break.ac_components } this.mcu_zig_index = z + 1 - } else if (ac_rrrr < 15) or ((this.mcu_zig_index + 16) > 63) { + } else if ac_rrrr < 15 { + break.ac_components + } else if (this.mcu_zig_index + 16) > 63 { break.ac_components } else { this.mcu_zig_index = this.mcu_zig_index + 16 diff --git a/std/jpeg/decode_mcu_progressive_ac_high_bits.wuffs b/std/jpeg/decode_mcu_progressive_ac_high_bits.wuffs new file mode 100644 index 00000000..952b0676 --- /dev/null +++ b/std/jpeg/decode_mcu_progressive_ac_high_bits.wuffs @@ -0,0 +1,150 @@ +// Copyright 2023 The Wuffs Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +pri func decoder.decode_mcu_progressive_ac_high_bits!() base.u32 { + var ret : base.u32 + + var bits : base.u64 + var n_bits : base.u32 + + var r : base.io_reader + var pos : base.u32 + + var ac_h : base.u8[..= 7] + var ac_symbol : base.u32[..= 0xFF] + var ac_ht_fast : base.u32 + var ac_bl : base.u32 + var ac_code : base.u32 + var ac_blm1 : base.u32[..= 15] + var ac_ht_slow : base.u32 + var ac_value : base.u32 + + var ac_rrrr : base.u32[..= 15] + var ac_ssss : base.u32[..= 15] + + var z : base.u32[..= 78] // 78 = 63 + 15. + + if this.eob_run > 0 { + this.eob_run -= 1 + return 0 + } + + bits = this.bitstream_bits + n_bits = this.bitstream_n_bits + + if this.bitstream_ri > this.bitstream_wi { + return 2 // Internal error. + } + io_bind (io: r, data: this.bitstream_buffer[this.bitstream_ri .. this.bitstream_wi], history_position: this.bitstream_ri as base.u64) { + while.goto_done true {{ + + while.block true { + while.ac_components true { + // Load at least 31 bits, by keeping 56 .. 64 loaded. + if r.length() < 8 { + ret = 1 // Request another fill_bitstream call. + break.goto_done + } + bits |= r.peek_u64be() >> (n_bits & 63) + r.skip_u32_fast!(actual: (63 - (n_bits & 63)) >> 3, worst_case: 8) + n_bits |= 56 + + // Read the Huffman-encoded ac_symbol, up to 16 bits long. + ac_h = 4 | this.scan_comps_ta[this.mcu_blocks_sselector[0]] + ac_ht_fast = this.huff_tables_fast[ac_h][bits >> 56] as base.u32 + ac_bl = ac_ht_fast >> 8 + if n_bits >= ac_bl { + ac_symbol = 0xFF & ac_ht_fast + bits ~mod<<= (ac_bl & 63) + n_bits -= ac_bl + } else { + ac_code = (bits >> 55) as base.u32 + ac_blm1 = 8 + bits ~mod<<= 9 + n_bits ~mod-= 9 + while true { + ac_ht_slow = this.huff_tables_slow[ac_h][ac_blm1] + if ac_code < (ac_ht_slow >> 8) { + ac_symbol = this.huff_tables_symbols[ac_h][0xFF & (ac_code ~mod+ ac_ht_slow)] as base.u32 + break + } + ac_code = (ac_code ~mod<< 1) | ((bits >> 63) as base.u32) + bits ~mod<<= 1 + n_bits ~mod-= 1 + ac_blm1 = (ac_blm1 + 1) & 15 + if ac_blm1 == 0 { + ac_symbol = 0 + break + } + } endwhile + } + + // Split the 8-bit ac_symbol into two 4-bit halves, per section + // F.2.2.2 "Decoding procedure for AC coefficients". + ac_rrrr = ac_symbol >> 4 + ac_ssss = ac_symbol & 15 + + // Process the ac_value in the next ac_ssss (up to 15) bits or + // the eob_run in the next ac_rrrr (up to 15) bits. + if ac_ssss > 0 { + ac_value = ((bits >> (64 - ac_ssss)) & 0xFFFF_FFFF) as base.u32 + if (bits >> 63) == 0 { // EXTEND per section F.2.2.1. + ac_value ~mod+= 1 ~mod+ ((0xFFFF_FFFF as base.u32) ~mod<< ac_ssss) + } + bits ~mod<<= ac_ssss + n_bits ~mod-= ac_ssss + z = this.mcu_zig_index + ac_rrrr + this.mcu_blocks[0][UNZIG[z]] = + ((ac_value ~mod<< this.scan_al) & 0xFFFF) as base.u16 + if (z + 1) > (this.scan_se as base.u32) { + break.ac_components + } + assert (z + 1) <= 63 via "a <= b: a <= c; c <= b"(c: (this.scan_se as base.u32)) + this.mcu_zig_index = z + 1 + } else if ac_rrrr < 15 { + this.eob_run = (((1 as base.u16) << ac_rrrr) - 1) as base.u16 + if ac_rrrr > 0 { + this.eob_run ~mod+= ((bits >> (64 - ac_rrrr)) & 0xFFFF) as base.u16 + bits ~mod<<= ac_rrrr + n_bits ~mod-= ac_rrrr + } + break.ac_components + } else if (this.mcu_zig_index + 16) > (this.scan_se as base.u32) { + break.ac_components + } else { + assert (this.mcu_zig_index + 16) <= 63 via "a <= b: a <= c; c <= b"(c: (this.scan_se as base.u32)) + this.mcu_zig_index = this.mcu_zig_index + 16 + } + } endwhile.ac_components + + this.mcu_zig_index = 0 + break.block + } endwhile.block + + break.goto_done + }} endwhile.goto_done + + pos = (r.position() & 0xFFFF_FFFF) as base.u32 + if pos > this.bitstream_wi { + ret = 2 // Internal error. + } else { + assert pos <= 0x800 via "a <= b: a <= c; c <= b"(c: this.bitstream_wi) + this.bitstream_ri = pos + } + } + + this.bitstream_bits = bits + this.bitstream_n_bits = n_bits + return ret +} diff --git a/std/jpeg/decode_mcu_progressive_ac_low_bit.wuffs b/std/jpeg/decode_mcu_progressive_ac_low_bit.wuffs new file mode 100644 index 00000000..502460d4 --- /dev/null +++ b/std/jpeg/decode_mcu_progressive_ac_low_bit.wuffs @@ -0,0 +1,243 @@ +// Copyright 2023 The Wuffs Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +pri func decoder.decode_mcu_progressive_ac_low_bit!() base.u32 { + var ret : base.u32 + + var bits : base.u64 + var n_bits : base.u32 + + var one_lshift_scan_al : base.u16[..= 0x2000] + + var r : base.io_reader + var pos : base.u32 + + var ac_h : base.u8[..= 7] + var ac_symbol : base.u32[..= 0xFF] + var ac_ht_fast : base.u32 + var ac_bl : base.u32 + var ac_code : base.u32 + var ac_blm1 : base.u32[..= 15] + var ac_ht_slow : base.u32 + var ac_value : base.u32 + + var ac_rrrr : base.u32[..= 15] + var ac_ssss : base.u32[..= 15] + + var unzig : base.u8[..= 63] + var bit : base.bool + + bits = this.bitstream_bits + n_bits = this.bitstream_n_bits + + one_lshift_scan_al = (1 as base.u16) << this.scan_al + + if this.bitstream_ri > this.bitstream_wi { + return 2 // Internal error. + } + io_bind (io: r, data: this.bitstream_buffer[this.bitstream_ri .. this.bitstream_wi], history_position: this.bitstream_ri as base.u64) { + while.goto_done true {{ + + while.block true { + while.goto_do_eob true {{ + if this.eob_run > 0 { + break.goto_do_eob + } + + while.ac_components true { + // Load at least 31 bits, by keeping 56 .. 64 loaded (call it 8 + // bytes). Just like the other decode_mcu_etc functions, this + // is enough for the Huffman-encoded ac_symbol (up to 16 bits) + // and processing the ac_value (up to 15 bits). + // + // After that, it's simpler for this "AC low bit" flavor if we + // never suspend for the rest of the while.ac_components loop. + // Specifically, we want to have enough bits in the bitstream + // so that we don't suspend inside the while.refine_non_zeroes + // loop. In the worst case, we need another 8 bytes for + // refining non-zero coefficients plus 8 bytes slack so that we + // can always call peek_u64be. 8 + 8 + 8 = 24 bytes in total. + if r.length() < 24 { + ret = 1 // Request another fill_bitstream call. + break.goto_done + } + bits |= r.peek_u64be() >> (n_bits & 63) + r.skip_u32_fast!(actual: (63 - (n_bits & 63)) >> 3, worst_case: 8) + n_bits |= 56 + + // Read the Huffman-encoded ac_symbol, up to 16 bits long. + ac_h = 4 | this.scan_comps_ta[this.mcu_blocks_sselector[0]] + ac_ht_fast = this.huff_tables_fast[ac_h][bits >> 56] as base.u32 + ac_bl = ac_ht_fast >> 8 + if n_bits >= ac_bl { + ac_symbol = 0xFF & ac_ht_fast + bits ~mod<<= (ac_bl & 63) + n_bits -= ac_bl + } else { + ac_code = (bits >> 55) as base.u32 + ac_blm1 = 8 + bits ~mod<<= 9 + n_bits ~mod-= 9 + while true { + ac_ht_slow = this.huff_tables_slow[ac_h][ac_blm1] + if ac_code < (ac_ht_slow >> 8) { + ac_symbol = this.huff_tables_symbols[ac_h][0xFF & (ac_code ~mod+ ac_ht_slow)] as base.u32 + break + } + ac_code = (ac_code ~mod<< 1) | ((bits >> 63) as base.u32) + bits ~mod<<= 1 + n_bits ~mod-= 1 + ac_blm1 = (ac_blm1 + 1) & 15 + if ac_blm1 == 0 { + ac_symbol = 0 + break + } + } endwhile + } + + // Split the 8-bit ac_symbol into two 4-bit halves, per section + // F.2.2.2 "Decoding procedure for AC coefficients". + ac_rrrr = ac_symbol >> 4 + ac_ssss = ac_symbol & 15 + + // Process the ac_value in the next ac_ssss (up to 15) bits or + // the eob_run in the next ac_rrrr (up to 15) bits. + ac_value = 0 + if ac_ssss > 0 { + ac_value = (0x0000_0001 as base.u32) << this.scan_al + if (bits >> 63) == 0 { + ac_value = (0xFFFF_FFFF as base.u32) ~mod<< this.scan_al + } + bits ~mod<<= 1 + n_bits ~mod-= 1 + } else if ac_rrrr < 15 { + this.eob_run = (1 as base.u16) << ac_rrrr + if ac_rrrr > 0 { + this.eob_run ~mod+= ((bits >> (64 - ac_rrrr)) & 0xFFFF) as base.u16 + bits ~mod<<= ac_rrrr + n_bits ~mod-= ac_rrrr + } + break.goto_do_eob + } + + // Consume 1 bit per non-zero AC coefficient, up to 63 bits. + while.refine_non_zeroes true { + unzig = UNZIG[this.mcu_zig_index] + if this.mcu_blocks[0][unzig] <> 0 { + if n_bits == 0 { + if r.length() < 8 { + ret = 2 // Internal error. + break.goto_done + } + bits |= r.peek_u64be() >> (n_bits & 63) + r.skip_u32_fast!(actual: (63 - (n_bits & 63)) >> 3, worst_case: 8) + n_bits |= 56 + } + bit = (bits >> 63) > 0 + bits ~mod<<= 1 + n_bits ~mod-= 1 + if bit { + if this.mcu_blocks[0][unzig] < 0x8000 { + this.mcu_blocks[0][unzig] += one_lshift_scan_al + } else { + this.mcu_blocks[0][unzig] -= one_lshift_scan_al + } + } + } else if ac_rrrr <= 0 { + break.refine_non_zeroes + } else { + ac_rrrr -= 1 + } + + if this.mcu_zig_index >= (this.scan_se as base.u32) { + break.refine_non_zeroes + } + assert this.mcu_zig_index < 63 via "a < b: a < c; c <= b"(c: (this.scan_se as base.u32)) + this.mcu_zig_index += 1 + } endwhile.refine_non_zeroes + + if ac_value <> 0 { + this.mcu_blocks[0][UNZIG[this.mcu_zig_index]] = (ac_value & 0xFFFF) as base.u16 + } + + if this.mcu_zig_index >= (this.scan_se as base.u32) { + break.ac_components + } + assert this.mcu_zig_index < 63 via "a < b: a < c; c <= b"(c: (this.scan_se as base.u32)) + this.mcu_zig_index += 1 + } endwhile.ac_components + + break.block + }} endwhile.goto_do_eob + + if this.eob_run <= 0 { + ret = 2 // Internal error. + break.goto_done + } + + while.refine_eob_non_zeroes true, + inv this.eob_run > 0, + { + unzig = UNZIG[this.mcu_zig_index] + if this.mcu_blocks[0][unzig] <> 0 { + if n_bits == 0 { + // Load at least 1 bit, by keeping 56 .. 64 loaded. + if r.length() < 8 { + ret = 1 // Request another fill_bitstream call. + break.goto_done + } + bits |= r.peek_u64be() >> (n_bits & 63) + r.skip_u32_fast!(actual: (63 - (n_bits & 63)) >> 3, worst_case: 8) + n_bits |= 56 + } + bit = (bits >> 63) > 0 + bits ~mod<<= 1 + n_bits ~mod-= 1 + if bit { + if this.mcu_blocks[0][unzig] < 0x8000 { + this.mcu_blocks[0][unzig] += one_lshift_scan_al + } else { + this.mcu_blocks[0][unzig] -= one_lshift_scan_al + } + } + } + + if this.mcu_zig_index >= (this.scan_se as base.u32) { + break.refine_eob_non_zeroes + } + assert this.mcu_zig_index < 63 via "a < b: a < c; c <= b"(c: (this.scan_se as base.u32)) + this.mcu_zig_index += 1 + } endwhile.refine_eob_non_zeroes + + this.eob_run -= 1 + + break.block + } endwhile.block + + break.goto_done + }} endwhile.goto_done + + pos = (r.position() & 0xFFFF_FFFF) as base.u32 + if pos > this.bitstream_wi { + ret = 2 // Internal error. + } else { + assert pos <= 0x800 via "a <= b: a <= c; c <= b"(c: this.bitstream_wi) + this.bitstream_ri = pos + } + } + + this.bitstream_bits = bits + this.bitstream_n_bits = n_bits + return ret +} diff --git a/std/jpeg/decode_mcu_progressive_dc_high_bits.wuffs b/std/jpeg/decode_mcu_progressive_dc_high_bits.wuffs new file mode 100644 index 00000000..365f7d9f --- /dev/null +++ b/std/jpeg/decode_mcu_progressive_dc_high_bits.wuffs @@ -0,0 +1,128 @@ +// Copyright 2023 The Wuffs Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +pri func decoder.decode_mcu_progressive_dc_high_bits!() base.u32 { + var ret : base.u32 + + var bits : base.u64 + var n_bits : base.u32 + + var csel : base.u8[..= 3] + + var r : base.io_reader + var pos : base.u32 + + var dc_h : base.u8[..= 7] + var dc_symbol : base.u32[..= 0x0F] + var dc_ht_fast : base.u32 + var dc_bl : base.u32 + var dc_code : base.u32 + var dc_blm1 : base.u32[..= 15] + var dc_ht_slow : base.u32 + var dc_value : base.u32 + + bits = this.bitstream_bits + n_bits = this.bitstream_n_bits + + if this.bitstream_ri > this.bitstream_wi { + return 2 // Internal error. + } + io_bind (io: r, data: this.bitstream_buffer[this.bitstream_ri .. this.bitstream_wi], history_position: this.bitstream_ri as base.u64) { + while.goto_done true {{ + + while.block this.mcu_current_block < this.mcu_num_blocks { + assert this.mcu_current_block < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) + while.dc_component true, + inv this.mcu_current_block < 10, + { + // Load at least 31 bits, by keeping 56 .. 64 loaded. + if r.length() < 8 { + ret = 1 // Request another fill_bitstream call. + break.goto_done + } + bits |= r.peek_u64be() >> (n_bits & 63) + r.skip_u32_fast!(actual: (63 - (n_bits & 63)) >> 3, worst_case: 8) + n_bits |= 56 + + // Read the Huffman-encoded dc_symbol, up to 16 bits long. + dc_h = 0 | this.scan_comps_td[this.mcu_blocks_sselector[this.mcu_current_block]] + dc_ht_fast = this.huff_tables_fast[dc_h][bits >> 56] as base.u32 + dc_bl = dc_ht_fast >> 8 + if n_bits >= dc_bl { + dc_symbol = 0x0F & dc_ht_fast + bits ~mod<<= (dc_bl & 63) + n_bits -= dc_bl + } else { + dc_code = (bits >> 55) as base.u32 + dc_blm1 = 8 + bits ~mod<<= 9 + n_bits ~mod-= 9 + while true, + inv this.mcu_current_block < 10, + { + dc_ht_slow = this.huff_tables_slow[dc_h][dc_blm1] + if dc_code < (dc_ht_slow >> 8) { + dc_symbol = 0x0F & (this.huff_tables_symbols[dc_h][0xFF & (dc_code ~mod+ dc_ht_slow)] as base.u32) + break + } + dc_code = (dc_code ~mod<< 1) | ((bits >> 63) as base.u32) + bits ~mod<<= 1 + n_bits ~mod-= 1 + dc_blm1 = (dc_blm1 + 1) & 15 + if dc_blm1 == 0 { + dc_symbol = 0 + break + } + } endwhile + } + + // Process the dc_value in the next dc_symbol (up to 15) bits. + // + // The dc_value is shifted by (64 - dc_symbol) in two steps, + // because we want to shift by 64 (not 0) when dc_symbol is 0. + dc_value = (((bits >> 32) >> (32 - dc_symbol)) & 0xFFFF_FFFF) as base.u32 + if (bits >> 63) == 0 { // EXTEND per section F.2.2.1. + dc_value ~mod+= 1 ~mod+ ((0xFFFF_FFFF as base.u32) ~mod<< dc_symbol) + } + bits ~mod<<= dc_symbol + n_bits ~mod-= dc_symbol + csel = this.scan_comps_cselector[this.mcu_blocks_sselector[this.mcu_current_block]] + this.mcu_previous_dc_values[csel] ~mod+= (dc_value & 0xFFFF) as base.u16 + this.mcu_blocks[this.mcu_current_block][0] = + this.mcu_previous_dc_values[csel] ~mod<< this.scan_al + + break.dc_component + } endwhile.dc_component + + assert this.mcu_current_block < 10 + this.mcu_current_block += 1 + } endwhile.block + this.mcu_current_block = 0 + + break.goto_done + }} endwhile.goto_done + + pos = (r.position() & 0xFFFF_FFFF) as base.u32 + if pos > this.bitstream_wi { + ret = 2 // Internal error. + } else { + assert pos <= 0x800 via "a <= b: a <= c; c <= b"(c: this.bitstream_wi) + this.bitstream_ri = pos + } + } + + this.bitstream_bits = bits + this.bitstream_n_bits = n_bits + return ret +} diff --git a/std/jpeg/decode_mcu_progressive_dc_low_bit.wuffs b/std/jpeg/decode_mcu_progressive_dc_low_bit.wuffs new file mode 100644 index 00000000..f2485387 --- /dev/null +++ b/std/jpeg/decode_mcu_progressive_dc_low_bit.wuffs @@ -0,0 +1,81 @@ +// Copyright 2023 The Wuffs Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +pri func decoder.decode_mcu_progressive_dc_low_bit!() base.u32 { + var ret : base.u32 + + var bits : base.u64 + var n_bits : base.u32 + + var one_lshift_scan_al : base.u16[..= 0x2000] + + var r : base.io_reader + var pos : base.u32 + + bits = this.bitstream_bits + n_bits = this.bitstream_n_bits + + one_lshift_scan_al = (1 as base.u16) << this.scan_al + + if this.bitstream_ri > this.bitstream_wi { + return 2 // Internal error. + } + io_bind (io: r, data: this.bitstream_buffer[this.bitstream_ri .. this.bitstream_wi], history_position: this.bitstream_ri as base.u64) { + while.goto_done true {{ + + while.block this.mcu_current_block < this.mcu_num_blocks { + assert this.mcu_current_block < 10 via "a < b: a < c; c <= b"(c: this.mcu_num_blocks) + while.dc_component true, + inv this.mcu_current_block < 10, + { + // Load at least 31 bits, by keeping 56 .. 64 loaded. + if r.length() < 8 { + ret = 1 // Request another fill_bitstream call. + break.goto_done + } + bits |= r.peek_u64be() >> (n_bits & 63) + r.skip_u32_fast!(actual: (63 - (n_bits & 63)) >> 3, worst_case: 8) + n_bits |= 56 + + // Refine a bit. + if (bits >> 63) <> 0 { + this.mcu_blocks[this.mcu_current_block][0] |= one_lshift_scan_al + } + bits ~mod<<= 1 + n_bits ~mod-= 1 + + break.dc_component + } endwhile.dc_component + + assert this.mcu_current_block < 10 + this.mcu_current_block += 1 + } endwhile.block + this.mcu_current_block = 0 + + break.goto_done + }} endwhile.goto_done + + pos = (r.position() & 0xFFFF_FFFF) as base.u32 + if pos > this.bitstream_wi { + ret = 2 // Internal error. + } else { + assert pos <= 0x800 via "a <= b: a <= c; c <= b"(c: this.bitstream_wi) + this.bitstream_ri = pos + } + } + + this.bitstream_bits = bits + this.bitstream_n_bits = n_bits + return ret +} diff --git a/test/nia-checksums-of-data.txt b/test/nia-checksums-of-data.txt index ea00b599..bd7e711a 100644 --- a/test/nia-checksums-of-data.txt +++ b/test/nia-checksums-of-data.txt @@ -85,7 +85,7 @@ d42a13cd test/data/peacock.default.jpeg d49bab2d test/data/peacock.grayscale.jpeg d42a13cd test/data/peacock.optimize.jpeg bdf8cc70 test/data/peacock.ppm -db9b1efa test/data/peacock.progressive.jpeg +d42a13cd test/data/peacock.progressive.jpeg 951676cc test/data/peacock.q30.jpeg 56835223 test/data/peacock.q99.jpeg 1543f677 test/data/peacock.s-weird.jpeg