| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| README.md | 2025-08-17 | 1.2 kB | |
| v1.15_ Minor fixes and improvements source code.tar.gz | 2025-08-17 | 2.5 MB | |
| v1.15_ Minor fixes and improvements source code.zip | 2025-08-17 | 2.7 MB | |
| Totals: 3 Items | 5.2 MB | 0 | |
This release fixes some emulation issues and adds small improvements.
- Add new experimental feature
translate_use_syscall_clobbering_optimizationwhich reduces latency of system calls with JIT and register caching (disabled by default) - Improved performance of accessing virtual pages in binary translation
- Add missing FCVT.X.D emulation for 64-bit integers in binary translation
- Match N-bit arena size for other platforms without mmap()
- mremap() no longer has the wrong system call number (oops!)
- free_pages() now rounds up to the nearest page when freeing
- Add support for JIT on 64-bit RISC-V (that is, running libriscv on a real RISC-V machine)
- Fix missing zero-termination in guest C++ strings longer than SSO
- Minor improvements to binary translator codegen
As always, disabling register caching with JIT is still the lowest latency, however the new experimental feature reduces the gap a little bit. Having register caching enabled will improve JIT performance on heavy calculations.
WARNING: The ABI for binary translations has changed, and they need to be rebuilt to be compatible with the new version.
Full Changelog: https://github.com/libriscv/libriscv/compare/v1.14...v1.15