• Client: make sure a local server is ACTUALLY killed before starting demo playback
• Client: don't warn about g_sync if autoRecord
• Vulkan: fix texture bind regression on flames map
• Common: use more explicit hunk memory control, remove CL_ClearMemory()
• FS: increase \fs_basegame folder limit from 4 to 8
• FS: always report the filename extension from FS_AllowedExtension()
• Client: avoid duplicate download URL separator
• botlib: reject missing jump candidates before midpoint calculation
• Renderers: give PNG chunk headers byte alignment
• Sound: use pthread-compatible ALSA callbacks
• QVM: initialize unused native call arguments
• Qcommon: match zlib allocator callback signatures
• FS: validate extension before pointer arithmetic
• botlib: initialize the complete movement result
• SDL: include Windows declarations independently of curl
• Common: validate affinity hex digits before widening
• Common: preserve affinity expression operators
• Botlib: preserve the signed chat offset sentinel
• Common: simplify and use more explicit hunk zone memory management
• Renderers: apply client-side CM + BSP hunk memory load optimization
• Win32: set WASAPI thread termination timeout to 10s
• Win32: fixed WIN32 ARM64 build
• Client: use named pipes for \video-pipe redirection on Win32
• Client: allow setting target file extension for \video-pipe via adding "|" suffix ("|mkv" for example) at the end of \cl_aviPipeFormat
• CM: fixed another tiny precision loss/misprediction on q3dm12
• CM: fixed precision for curve mesh collisions
• CM: fix root of MAX_PATCH_PLANES error, reduce definition MAX_PATCH_PLANES to the original value
• x86: enforce x87 double precision by default for all platforms to unify behavior/IEEE-754 compliance
• VM_x86: implemented new x87 JIT compiler
• VM_x86: use more explicit x87 precision/rounding control for better determination
• VM_x86: corrected NaN-checks for CPUs without fast FPU comparisons
• VM_x86: optimized some int->float SSE conversions
• FS: refined filesystem checks (e.g. check extensions of a short source file not whole ospath)
• QVM: explicitly mark all jump targets with unsafe bit
• VM_x86: add more const optimizations
• VM_x86: improve VM_FindSameLoad4()
• VM_x86: implement instruction length detection for RIP-related offsets instead of using hardcoded values
• VM_x86: call FlushInstructionCache() for generated bytecode in win32, allocate initial memory with RW permissions instead of RWX
• VM_x86: fixed out-of-bounds access with macro optimization enabled
• VM_x86: added more macro optimizations
• Win32: apply Win key hooks only if key is bound
• FS: keep qvm-referenced paks always locked
• FS: reduce simultaneously opened file handle limit to 250
• FS: refactor/cleanup FS_TouchFileInPak()
• Misc: raize description limit for Cvar_SetDescription2() from MAX_CVAR_VALUE_STRING (256) to MAX_STRING_CHARS (1024)
• Vulkan: disabled discard optimization for depth-write shaders
• Network: fixed incorrect timeout parameter for NET_FlushPacketQueue()
• Win32: added netmask detection function for ipv4 addresses
• Vulkan: corrected initial swapchain image layout transition (write-after-read hazard)
• QVM-JIT: fixed potential clobbering of aliased registers on sign-extension optimizations
• VM_x86: slightly refactored MACRO_OPTIMIZE and enabled by default
• ppc64-QVM: zero-extend cached OP_LOAD1/LOAD2 instead of sign-extend
• ppc64-QVM: honor IEEE unordered semantics for float <= and >=
• ppc64-QVM: do not clobber an aliased register when reducing a cached load
• ppc64-QVM: zero-extend cached OP_LOAD1/LOAD2 instead of sign-extend
• CI: Update deps & fix warns for brew
• Client: Add CG_KEY_CAPSLOCK_ON cgame trap for Caps Lock latch state
• Renderers: fixed compatibility with ut43_azurea_b1 map
• Vulkan: fixed recent SDK validation layer warnings
• QVM: fixed opStack flush conditions
• win32: improved forward compatibility for msvc2017 project files
• FS: block autoexec.cfg and q3config from *.pk3dir
• ppc64-QVM: debug-only emit-macro range checks + diagnostic data trap
• ppc64-QVM: tighten CONST_OPTIMIZE immediate-fit check for BAND/BOR/BXOR
• ppc64-QVM: drop branch-trap data check, always mask in emit_CheckReg
• aarch64-QVM: split fixed-size MOVXi into emit_MOVXi64
• ppc64-QVM: clear high bits before data-segment bounds check
• ppc64-QVM: direct mov_sx_rx via mtvsrws+xscvspdp on POWER9+
• botlib: fixed BotNextConsoleMessage() argument aligment issues reported by UBSan
• Sound: fixed heavy sound swapping on ut4_skipjumps_b2 with default soundmegs
• Sound: fixed \s_list output
• Add VSX Q_rsqrt with Newton-Raphson refinement
• Add PPC64 VSX SIMD compiler flags
• Server: fixed redundant gamestate retransmission
• Server: refactored/re-applied "delta snapshot developer print fix" + "don't delta from pre-map restart messages"
• Server: fixed client' gamestate acknowledge state after cURL downloads
• Server: improved gamestate acknowledge function to be more packet-loss friendly
• Client: reworked packet retransmission logic to use relative times
• Cmake: Update regex filter for QCOMMON_SRCS to avoid PowerPC code.
• QVM: moved x86-specific optimization to the common
• Win32: removed \in_lagged and related code
• QVM: fixed/improved MACRO_OPTIMIZE for x86, changed function align to 16 bytes
• arm64-QVM: implemented avoidance of long conditional branches, code cleanup
• QVM: corrected RCONST usage with forced data mask
• ppc-QVM: added more const-optimizations, implemented avoidance of long conditional branches
• x86-QVM: corrected error functions offsets, code cleanup
• win32: added description for EXCEPTION_INT_DIVIDE_BY_ZERO
• QVM: implemented FPU load/store and variable mapping optimizations for PowerPC
• QVM: various fixes/improvements
  • implemented jump expansions and load/store optimizations for PowerPC
  • implemented empty function optimization for armv7l
  • replaced explicit color codes with corresponding warning/error defitions
  • reset proc_len at empty function detection
  • fixed SX/RX definition typo in vm_optimize.h
• Renderers: set correct shader time on 3D->2D transition
• QVM: slightly optimize scalar loads from syscall returns
• Common: reworked/simplified free block search for zone memory allocator and reduced overhead
• Server: do not rate-limit local client
• Fix OpenGL1 renderer failed to init
• Server: correct entity number for BotImport_EntityTrace()
• Server: correct entity numner for SV_Trace() as well
• QVM: Skip VM syscall array bounds check if DLL is used
• Common: Simplify RotatePointAroundVector()
• QMV: Add missing CG_CM_MARKFRAGMENTS syscall bounds checks
• Sound: Validate entity number passed to sound functions
• FS: Forbid reading/writing negative amount of bytes
• FS: Check FS_Read() return values in various subsystems
• Common: Use buffer size safe string functions where possible
• Server: Fix potential integer overflow in sv_filter code
• Renderers: Fix some minor PVS handling issues
• Renderers: Fix broken MD3 LODs registration
• Server: Harden game syscall argument validation
• Renderers: Harden BSP structure validity checks
• Renderers: Harden IQM, MD3 header validity checks
• QVM: Harden QVM header validity checks
• Botlib: Validate AAS file structure, Validate botlib vars range
• Botlib: Validate AAS_UpdateEntity() argument
• Botlib: Clean up botlib memory allocation code.
• QVM: apply generic (OP_LOAD + sign extendsion + OP_CONST + OP_EQ/OP_NE) optimization instead of platform-specific ones
• CI: Fix warn to update actions to Node24
• x86-QVM: implemented literal pool for FP immediates, implemented per-function jump compression
• QVM: added more optimizations for PowerPC JIT
• QVM: added basic const-optimizations for powerpc, fixed reduce_map_size() for BE systems
• QVM: utilize shared optimization framework for PowerPC bytecode compiler, estimated speedup is ~2x for now
• QVM: moved all platform-independend opStack/register optimizations to a separate header file
• QVM: refactored x86/armv7l/aarch64 bytecode compilers to utilize new optimization framework
• Makefile: added detection for aarch64 over 32-bit runtime environments like RPi4b + Debian Trixie armhf
• Win32: somewhat improved support for arm32 platform
• Huffman: reduced internal buffer sizes
• OpenGL: use GL_ARB_texture_border_clamp unconditionally if available
• OpenGL: Do not use VBO or per-pixel lighting without ARB shaders
• OpenGL: Prevent using merged lightmaps without GL_ARB_texture_border_clamp
• Renderers: fix drawSurf->sort typo in RB_RenderDrawSurfList()
• Vulkan: print note about buggy drivers
• Misc: define S_COLOR_DEVEL/S_COLOR_WARNING/S_COLOR_ERROR, replace some definitions as well
• QVM: slightly optimize syscallCount handling
• x86-QVM: use divss instead of divps for OP_DIVF to avoid triggering "Invalid Operation" bit
• Server: Add syscall overflow detection to prevent VM infinite loops
• Wire up max_routingcache cvar to routing cache eviction logic
• Renderers: Fix NULL pointer dereference in RE_Shutdown when renderer interface is not initialized
• Fix CMake renderer library naming to omit 'lib' prefix
• QVM: Add Power ISA 2.07/3.0 JIT optimizations to vm_powerpc.c
• Misc: Fix RENDERER_DEFAULT double-quoting in Makefile and CMakeLists.txt
• QVM: Add ppc64 big-endian (ELFv1 ABI) JIT support to vm_powerpc.c
• Make, CMake: Add ppc64 (PowerPC 64-bit BE) architecture support
• QVM: Add PPC64le JIT compiler (vm_powerpc.c)
• Make, Cmake: Add ppc64le (PowerPC 64-bit LE) architecture support
• x86-QVM: use movss instead of movd for explicit FP register initialization on old SSE1-only CPUs
• Renderers: fix dlight regression and optimize shaders on simpsons_q3 map
• Renderers: unlock \r_flareSize and limit its range to [1..40]
• SDL: sync modifier keys with SDL state
• Win32: use GetKeyState() instead of GetAsyncKeyState()
• SDL: add crash workaround for SDL_WarpMouseGlobal()