Skip to content

Releases

1.8.4-esphome.4

Choose a tag to compare

@clydebarrow clydebarrow released this 31 Oct 04:46
858912e

JPEGDEC 1.8.4 plus the changes ESPHome needs.

Added

  • Decode progressive images in full at every scale, in one call or in steps as their input arrives (#2). Before, only the first scan was drawn, at 1/8 size. A buffer for the DCT coefficients is allocated by decodeBegin(): 3 bytes per pixel for 4:2:0 colour, less at 1/4 and 1/8 scale. If it cannot be allocated, each block is drawn in one colour, without an error. On ESP32 it goes in PSRAM when possible.
  • Decode 8-bit extended sequential (SOF1) images, with up to 3 DC and 3 AC Huffman tables numbered 0-3 (#2). A third table of either kind takes 4.8 KB, allocated only for images that use one.
  • Run the tests in GitHub Actions with GCC, Clang and AddressSanitizer (#3).

Fixed

  • Fix wrong pixels from the AC bit refill in baseline images (#2).
  • Refuse frames with too many components rather than overwriting memory (#2).
  • Fix quarter-scale RGB8888 output of 4:4:0 images (#2).
  • Clip MCUs at the edge of a framebuffer: output whose width is not a multiple of the MCU width was wrong, and the last row could overrun the framebuffer (#1, bitbank2#131).
  • Fix the SSE marker filter's mask, which failed on x86_64 builds made with clang (#1).
  • Start a dither's errors at zero, so the output no longer depends on what the file held before it (#2).
  • Refuse rows too wide to dither, which overwrote memory and crashed (#2).
  • Close a file opened by name when the image is bad, on reopen and in the destructor; close() can be called more than once (#2).

Behaviour changes

  • A JPEGDEC must be made by its constructor, not malloc() alone, as open() frees what the object holds. It cannot be copied.
  • With the C API, call JPEG_close() before reopening a JPEGIMAGE, or its buffers leak. A JPEGIMAGE need not be cleared before it is opened.
  • Read and seek callbacks must keep iPos of the JPEGFILE up to date. A progressive image finds where its input ends by comparing it with iSize.
  • Now refused: sampling with no MCU layout, baseline colour images whose chroma is not 1x1, 12-bit progressive images, a bad first scan header, sequential images with a scan per component, and a scan using an undefined Huffman table.
  • decodeDither() fails with JPEG_INVALID_PARAMETER when a row of MCUs at the scale asked for is over 4088 pixels wide, or 2040 with JPEG_USES_DMA. Dithered output also differs slightly from before.
  • An image is decoded once per open(). A second decode() or decodeBegin() fails with JPEG_INVALID_PARAMETER.
  • During the scans of a progressive image, decodeRows() returns without calling the draw callback, and a single decode() does not call it until they are done.

Known

  • ESPHome still refuses progressive images. Using them needs a change there as well as this release.
  • JPEGDisplay has only been tested against stub classes, not on hardware. File-based decoding has not been tested with a real SD card.
  • Dither errors are kept in the framebuffer when one is set, so they do not start at zero there.

Checked on an ESP32-S3 at 800x480, before the last dither and iPos fixes: baseline decode 130 ms and progressive 383 ms.

1.8.4-esphome.3

Choose a tag to compare

@clydebarrow clydebarrow released this 22 Oct 10:15

JPEGDEC 1.8.4 plus the changes ESPHome needs. Pull requests for these are open or proposed upstream.

  • Add decodeBegin()/decodeRows() to decode a few MCU rows at a time.
  • Add JPEG_DECODE_NEED_DATA, so a decode can pause for input that has not arrived.
  • Fix filter faults on short reads: zero-length JPEGFilter, SIMD over-read, overlapping memcpy.
  • Fix a wild pointer for skipped blocks in progressive images (bitbank2#127).
  • Fix RGB8888 output for grayscale images (bitbank2#128).
  • Fix cropping, especially with scaling (bitbank2#129).
  • Fix swapped red and blue in the NEON 4:2:0 RGB8888 path (bitbank2#121).
  • Use the ESP32-S3 SIMD colour conversion in ESP-IDF builds too, not only Arduino ones. Checked on an ESP32-S3: within 2 levels of the C code for RGB8888 output and 4 for RGB565; full-size decodes 12-19% faster.
  • Build the library with -O2 (library.json build.flags). ESP-IDF builds default to -Os; JPEG decodes are 3-5% faster on an ESP32-S3.
  • Include upstream 81aa9f6 (NEON clipped-MCU overwrite) and 8628297 (skip AC Huffman tables for progressive images).

Known: framebuffer output is wrong for widths that are not a multiple of the MCU width (bitbank2#131).

1.8.4-esphome.2

Choose a tag to compare

@clydebarrow clydebarrow released this 22 Oct 09:11

JPEGDEC 1.8.4 plus the changes ESPHome needs. Pull requests for these are open or proposed upstream.

  • Add decodeBegin()/decodeRows() to decode a few MCU rows at a time.
  • Add JPEG_DECODE_NEED_DATA, so a decode can pause for input that has not arrived.
  • Fix filter faults on short reads: zero-length JPEGFilter, SIMD over-read, overlapping memcpy.
  • Fix a wild pointer for skipped blocks in progressive images (bitbank2#127).
  • Fix RGB8888 output for grayscale images (bitbank2#128).
  • Fix cropping, especially with scaling (bitbank2#129).
  • Fix swapped red and blue in the NEON 4:2:0 RGB8888 path (bitbank2#121).
  • Use the ESP32-S3 SIMD colour conversion in ESP-IDF builds too, not only Arduino ones. Checked on an ESP32-S3: within 2 levels of the C code for RGB8888 output and 4 for RGB565; full-size decodes 12-19% faster.
  • Include upstream 81aa9f6 (NEON clipped-MCU overwrite) and 8628297 (skip AC Huffman tables for progressive images).

Known: framebuffer output is wrong for widths that are not a multiple of the MCU width (bitbank2#131).

1.8.4-esphome.1

Choose a tag to compare

@clydebarrow clydebarrow released this 22 Oct 02:40

JPEGDEC 1.8.4 plus the changes ESPHome needs. Pull requests for these are open or proposed upstream.

  • Add decodeBegin()/decodeRows() to decode a few MCU rows at a time.
  • Add JPEG_DECODE_NEED_DATA, so a decode can pause for input that has not arrived.
  • Fix filter faults on short reads: zero-length JPEGFilter, SIMD over-read, overlapping memcpy.
  • Fix a wild pointer for skipped blocks in progressive images (bitbank2#127).
  • Fix RGB8888 output for grayscale images (bitbank2#128).
  • Fix cropping, especially with scaling (bitbank2#129).
  • Fix swapped red and blue in the NEON 4:2:0 RGB8888 path (bitbank2#121).
  • Include upstream 81aa9f6 (NEON clipped-MCU overwrite) and 8628297 (skip AC Huffman tables for progressive images).

Known: framebuffer output is wrong for widths that are not a multiple of the MCU width (bitbank2#131).