Pos

Added

- Add parallel reads to GetFullProof. Thanks to marcoabreu!
- Print Target/Final Directory early in the logs refs and log process ID. Thanks to grayfallstown!

Changed

- Using Gulrak 1.5.6
- Optimize code in phase1.hpp. Thanks to 683280!
- Migrating to flags instead of booleans parameters for show_progress and nobitfield. Thanks to jespino and mrhacky!

Fixed

- reset write pointer on 0 write. Thanks to tinodj!

Added

- Reopen files on read or write failure. Thanks to mreid-moz!
- Prevent chiapos prover crashing for bad plots.

Changed

- Use system includes for open and write when ENABLE_LOGGING is on.
- Removed some redundant code logic. Thanks to BinacsLee!

Fixed

- Added missing space in logging. Thanks to fredericosilva!

Added

- We now measure time to copy from -2 to -d with thank for the PR to L3Sota.
- We now retry FileDisk Open for should-be-existing files. We know of at least on k32 this could have saved. Except for the first construction, `FileDisk` instances are only reading or writing files that should have already been created. If the file has magically disappeared, the user probably moved it elsewhere to free disk space, and will likely move it back (or quit plotting manually if they can't). This change copies the "wait 5 minutes and retry" behavior of read, write, copy, and rename failures in `FileDisk` and `DiskPlotter` to `FileDisk`'s `Open` member. Thanks for the PR go to L3Sota.

Changed

- Efficient plotting of k=32 with 2 threads is now possible with buffer size of 3389 MiB in bitfield mode by shrinking entry sizes to reduce memory and IO usage.
- We now use compiler intrinsics for byte swapping, instead of platform specific functions/macros for easier cross platform support. This was inspired by a PR from timkuijsten for BSD.

Added

- We now attempt to autodetect and use popcnt instructions, if available, on all platforms. Improves CPU performance of phase 2 by 10% when using bitfield. Check availability of popcnt instruction at runtime with cpuid on x86. Refuse to plot with bitfield if it's not available. Assume presence of optimized popcnt on other architectures (e.g. CNT on aarch64). With this and other changes below, bitfield is now close to the same speed as no bitfield on most machines.
- Initial changes to support a progress percentage display if enabled on the command line.
- There is now a docker install version. Thanks zmeyc.

Changed

- Increase size of bitfield index by 8x. Indexing every 1024 bits instead of every 8192 bits increases CPU performance of phase 2 about 1.5x.
- Use all allocated memory for sorting in phase 3 and 4. Previously we were using only a half of allocated memory for sorting in phases 3 and 4 due to a wrong assumption that another sort is taking place at the same time.
- Changed primary branch from master to main.
- uint128_t has replaced Bits in phase 2. Improves phase 3 CPU performance by 3%.
- Use uint64_t instead of Bits when rewriting left entries in phase 1. Improves phase 1 CPU performance by 5%.This should improve phase 1 CPU performance by 5%.
- The Hellman attacks example now correctly throws. Thanks xdustinface!
- The unused pybind11 submodule was removed.
- cxxopts are now fetched by CMake instead of statically in lib/.
- gulrak/filesystem was updated to the current release and is now fetched by CMake.

Fixed

- There was an edge case where harvester could wander off into garbage on certain proof of space challenges causing a crash.
- CI build was not correctly creating and testing bitfield test plots.

Changed

- We made various improvements in error handling to potentially address harvester crashes - especially when moving plots around.

Added

- We now build both MacOS x86_64 and MacOS universal wheels.

Changed

- We reverted to stop using build flags `-mtune=native`. This should speed the plotter up on certain modern CPUs.
- We enhanced performance with changes to BufferedDisk.