- added gpujpeg_color_space_by_name
- addeg gpujpeg_print_pixel_format
- allowed setting subsampling 4:0:0 for color input image
- allowed passing GPUJPEG_SUBSAMPLING_UNKNOWN to gpujpeg_parameters_chroma_subsampling() to disable explicitly set subsampling and return to implicit (automatic) one on the encoder
- added gpujpeg_default[_image]_parameters() functions returning the structure instead of setting it through a pointer. This is better for conveninece to intitialize defined structs vars immediately.
- removed deprecated gpujpeg_parameters_chroma_subsampling_42{0,2}
- replaced GPUJPEG_444_U8_P012{A,Z} with GPUJPEG_4444_U8_P0123
- moved comp_count from gpujpeg_image_parameters to gpujpeg_parameters. Being in image parameter was from historic reasons but has been replaced by pixel format. The comp_count now represents rather count of components inside JPEG.
- support for ZLUDA
- updated API - provide more information about the decompressed image in gpujpeg_decoder_output, namely the image size
- support for larger images - until now, only pictures up to approx. 85 MP were supported, now it is something like 512 megapixels (pratical limit is the size of GPU RAM, however, 512 Mpix would currently require some 26.5 GB)
- updated API - all image sizes are now size_t
- added support for Y4M files (only one image per file)
- removed implicit synchronization issues when multiple encoders are run in parallel; also removed cudaMemcpy for the decoder (but not tested if there are not any other issues)
- very slight performance improvements
- support for encoding alpha
- Support for encoding arbitrary images in planar formats
- Decoding of arbitrary images into planar formats
- Support for SPIFF files (decode, encode limited-range BT.709, BT.601)
- Support for FFmpeg limited-range BT.601 JPEGs
- Support for PNM and PAM files
- Command-line application is now supported in MSW
- Support for RGBA buffers
- Raw RGBA and I420 files supported
- Decoder compatibility with more (foreign) JPEGs
- Support for RGB JPEG images
- Support for decoding to OpenGL texture or CUDA buffer
- Encoding different image sizes
- Support for encoding of planar images (limited)
- New IDCT computing kernel 43 % faster
Changes: -Added new parallel GPU Huffman encoder for compute capabilities >= 2.0 -Rewritten GPU Huffman decoder -Rewritten GPU forward DCT, preserving precision of fixed point implementation and gaining better performance than NPP implementation. (Both old implementations were thus removed.) -Rewritten GPU encoder preprocessor to remove unnecessary operations -Minor performance improvement in GPU encoder postprocessor
Changes: -Optionally the encoder input or decoder output can be loaded/store from/to OpenGL texture.
Changes: -Refactored color spaces transformations, added new color spaces -Added new implementation of DCT/IDCT on CUDA, it is slightly slower than the NPP implementation, but the new IDCT doesn't cause color space change. It can be turned on by Makefile option. By default is NPP version used.
Changes: -Slightly improved performance of encoder. -Correction of segment info for large images
Changes: -Added option to encoder for generating segment info into APP13 header that is used by decoder to perform fast stream parsing. -Renamed --interleving option to --interleaved and --chroma-subsampling to --subsampled.
Changes: -Added option for verbose output from console application. -Library now stores all coder time durations inside coder structure and it can be accessed from userspace. -Console application now prints GPU allocation info (when --verbose is used). -Correction of encoding/decoding for large images (e.g. 4320p or 8K)
-Improved preprocessor and postprocessor performance. -Added recovering from error inside JPEG stream to decoder. -Library provides function for listing CUDA devices. -Library provides option for decoder to output result right into OpenGL PBO resource which improves performance when the result should be displayed on the same GPU. -Correction of build warnings. -Correction of other errors.
Changes: -Added option to libgpujpeg Makefile for moving gpu huffman coder tables into constant memory. -Option is default set to true and makes better performance on older GPUs.
Changes: -Added interleaved mode (optionally). -Added subsampling (optionally) - better performance when used (because of smaller amount of encoded/decoded data).
First public release of gpujpeg library and console application. Features: -Baseline JPEG, non-interleaved mode on CUDA-enabled GPUs. -Performance - realtime HD and 4K encoding/decoding on NVIDIA GTX 580. -Using CUDA and NPP library.