Xenia Version Control System (February 14, 2026)

Xenia Git (2026/02/14) is an experimental Xbox 360 emulator that currently does not support running games. Recent updates include improvements for building binutils, better stack alignment for AVX instructions, and fixes for various PPC instructions. Key changes involved enhancements to register handling in x64/Linux, ensuring correct machine code pointer returns, and addressing issues with float packing and arithmetic. The update also includes performance optimizations and bug fixes, such as correcting operand size mismatches and improving NaN handling in AVX-512. For more details, the build is available for download.

(HEY YOU!! We hope you enjoy! We try not to run ads. So basically, this is a very expensive hobby running this site. Please consider joining us for updates, forums, and more. Network w/ us to make some cash or friends while retro gaming, and you can win some free retro games for posting. Okay, carry on 👍)

Xenia Git (2026/02/14) has been compiled. Xenia is an experimental emulator designed for the Xbox 360, though it currently does not run games.

Xenia Git Changelog:
* [Testing] Standardize binutils across Windows and Linux
* [Testing] Introduce support for building binutils with mingw64
* Reapply “[x64] Zero extend on mov to 8bit register”
* Ensure stack allocations are aligned to 16 bytes for AVX instructions
* [CPU/PPC] Include VX128_2 VC field support for vperm128
* [CPU/PPC] Implement and correct FPSCR-related instructions
* [CPU/PPC] Implement the mcrxr instruction
* [CPU/PPC] Fix the mfvscr/mtvscr instruction format
* [PPC] vsubcuw is now implemented, enable it
* [x64/Linux] More functions return __m128i by value in xmm0 instead of pointer
* [x64/Linux] Return __m128i by value in xmm0 instead of through a pointer
* [x64/Linux] Ensure EmitHostToGuestThunk saves rsi
* [CPU] Set VSCR NJ (non-Java) bit
* [CPU] Correct f16 pack rounding and SHORT_2 test input
* [CPU] Fix incorrect all_same detection in constant vector shift paths
* [CPU] Use RAII lock_guard in GuestTrampolineGroup
* [CPU] Correctly zero extend rather than sign extending
* [CPU] Check for null when returning machine code pointer
* [CPU] Fix off-by-one error in bsearch
* Revert “[x64] Zero extend on mov to 8bit register”
* [x64] Add a software fallback for PACK_FLOAT16_4
* [x64] Implement f32 arithmetic and related tests
* [x64] Remove AVX512 optimization for vrefp
* [x64] Fix AVX2 optimization path for VECTOR_SHL_V128
* [x64] Fix AVX512 optimization path for vadduws
* [x64] Fix handling of NaN in AVX-512 vctuxs
* [x64] Correct operand size mismatches in CNTLZ fallback