Add an additional method for advanced encodings scenarios. This methods allow to encode components with different
coding parameters like: near lossless, jpeg-ls preset coding parameters or scans with different interleave modes.
Extend the decoder to also be able to decode scans with different interleave modes.
To make it more clear to users and tooling that the project is a C++ project use the extension .hpp for pure C++ headers.
Add a new public header file charls.hpp. External C++ projects should this header file, C projects should still use charls.h.
Group the error IDs in errors that cannot be prevented and can be reported at runtime and in a set of errors that can be prevented.
Note 1: the error destination_too_small is considerd a runtime failure, but can in many cases be prevented.
Note 2: unexpected exceptions will now trigger a call to std::terminate (improved crash dumps)
* Add support to write mapping tables to the JPEG byte stream.
* Add API method to set the table ID for a component
* Add method to create JPEG-LS file with mapping tables only.
* Add support to read mapping tables
* Parse table-ids from the scan segment
* Extend API of charls_decoder_get_mapping_table_index to return -1 if index is not found
* Add support to read mapping table continuation segments
* Document new API methods
* Add sample image from appendix H.4.5 "Example of a palletised image" / Figure H.10 to unit test
* Update the get_mapping_table API to include size of the buffer
Background: a pixel consist of 1 or more samples. (RGB has 3 samples for example).
CharLS uses as SampleType a byte (uint8_t) or a 16 bit byte (uint16_t)
Invalid input data may trigger an invalid read.
This scenario will be hit for invalid input data that completes
a scan but hasn't a marker start byte at the expected position.
The current algorithm will add additional bits to the cache, but this requires an additional check for the position.
Note: found with LibFuzzer
- Use structured binding (when it makes sense)
- Add [[nodiscard]]
- Disable structured binding ReSharper warnings (it also advises to use structured binding for variables that will explode in too many parameters)
This was a Windows only option. Conversion is often not needed and Windows can do the conversion more efficient by itself. DirectX supports RGB encoded bitmaps.
CharLS doesn't support mixing interleave modes.
Add a unit test to ensure that CharLS will throw a not supported exception.
The actual checking was already done.
Note: also resolve some static analyzer warnings, reported with the latest version of ReSharper\clang-tidy.
The destructor of std::ofstream will close the stream, but not report any failures (by design).
Call close to trigger an exception when the close fails.
static_cast for constant values can be replaced with uniform initialization, this allows
to reduce the number of static_casts.
Change the bin+intermediate to be sub-directories of build for MSBuild builds.
This makes it easier to clean.
CharLS is cross-platform. -Wcast-align=strict will always warn for alignment issues, even when current platform can handle it (for example x86)
Use void*+static_cast to increase alignment. This is safe for dynamic allocated memory.
- Rename the log_2 function into log2_ceil as that is what the function performs.
- Add unit test and benchmark test to get insights if using it in more places has benefits.
Backport C++20 countl_zero
1) countl_zero can be used in peek_0_bits and is 2X faster.
Note: peek_0_bits is not on a hot path, overall gain is limited.
2) Apply updates for clang-tidy v13 (ships with VS 2022 17.1 Preview 2)
3) Increase upper range of supported CMake to 3.22
* Add an initial benchmark project to research optimizations.
* Use memchr to search for 0xFF during decoding.
* Rename function to make intent more clear.
* Use const pointers (decoding only need to read).
- Add missing API docs
- std::vector perform an initialization when its size is resized. This has a significant impact
on the performance measurements. Ensure that this step is done outside the timed loop.
* Add support to parse DRI segments
* Add support to decode restart markers
* Update CharLSTest to also decode color images none interleaved to ppm
* Replace static_cast<void> with std::ignore
* Update readme, changelog and refactor
Visual Studio 2015 cannot handle the /Zc:__cplusplus option and the new options in the manifest file. Exclude this things when build with Visual Studio 2015.
When encoding images ensure that only the valid bits are used. For bit size 16 and 8 there is no change and memcpy will be used.
Encoding with masking is not measurable slower.
Setting explicit InlineFunctionExpansion for debug needs to be removed as it prevents debugging. For release mode it is also not needed as MaxSpeed will already enable it.
Mask high input bits during encoding (interleave_mode::sample)
Mask high input bits during encoding
