[egz] Nuance 0.6.7 30/05/26

05/30/2026 version 0.6.7:

• FINALLY, T3K can be played with no hangs on music transitions, and with proper music looping!   Also the level select hang (with DynamicCompiler enabled) is resolved (see GameCompatibility.txt/MPE3PacketHack-option though for the caveats)!
• Merlin Racing's AI drivers now follow the course as intended (with DynamicCompiler enabled).
• Add Linux port with CMake build system, X11/GLX backend and miniaudio audio output.
• Add libretro port.
• 32-bit build supports the x86 JIT dynamic recompiler on Linux via `__attribute__((fastcall))`.
• 64-bit builds employ asmjit instead.
• Static linking of glew (no more `glew32.dll` needed).
• Replace FMOD 3.75 with miniaudio. No more outdated FMOD SDK / `fmod.dll` dependency.
• Due to this, decouple host audio from Nuon DMA via a ring buffer (which was implicitly done by FMOD under the hood).
• Audio interrupts are kept intact while muted (matches previous behavior, needs to be verified on real HW).
• Drag and drop support for the UI.
• Can load games directly from ZIP/ISO files on Windows.
• On Linux additionally supports RAR/7z via FUSE tools.
• All "with carry" opcodes (=addwc/subwc/cmpwc variants) should now behave as written in the ISA spec.
• MSB JIT bug fixed.
• Fix bug in JIT versions of MULScalarShiftAcshift, MULImmediateShiftAcshift, MULScalarShiftScalar and MULImmediateShiftScalar (shift broken if left shift >= 32 used).
• Fix bug in JIT version of MULImmediateShiftLeftImmediate (wrong dependency used).
• Fix bug in ANDImmediateShiftScalar, ANDScalarShiftScalar, FTSTImmediateShiftScalar, FTSTScalarShiftScalar, ORImmediateShiftScalar, ORScalarShiftScalar, EORImmediateShiftScalar, EORScalarShiftScalar (shift broken if left shift == 32 used).
• Fix wrong flag setting for large numbers (highest bit set) in JIT versions of DECRc0 and DECRc1.
• Fix missing JIT implementations for loading odmactl,odmacptr,mdmactl,mdmacptr,comminfo,commctl,commxmit0/1/2/3,commrecv0/1/2/3 and configa/configb/dcachectl/icachectl.
• BTST is still not same as ISA spec (as apparently a real HW bug, at least on Aries 2 (=tested there)).
• Multiple bugs in the "CompilerConstantPropagation"-enabled path fixed, so enabled by default now, leading to more performance.
• Fix "_MPEAlloc()" logic to match original BIOS.
• Correct Audio Buffer sizes used for all configurations (via GetBufferSize()).
• Unfortunately, Tetris now sounds even worse than before.
• Fix (re)loading of 2ndary games: reset audio + MPE states.
• To make (at least) T3K work better, the whole MediaRead (and partially MediaWrite) implementation has been redone
•   to better match the spec (incl. proper return values, proper callback mode handling, and not filling up buffers too early),
•   at the tradeoff that now the loads actually need some time to arrive (incl. potential stutter).
• Fix "_MediaGetInfo()" byte-swapped fields.
• Fix uninitialized framebuffer (e.g. when starting fullscreen from command line).
• Optimize most JIT code that needs to do bswap/endian change.
• Enforce rz/rzi1/rzi2 (matching the Nuon hardware) to have bit 0 as 0 (due to 16-bit alignment).
• Some JIT instructions now are implemented correctly on non-SSE4.1 capable hardware (MUL_SVScalarShiftSvshift, MUL_SVRuShiftSvshift and MUL_SVRvShiftSvshift had vector-lane swaps for 2 components).
• Fix JIT for PVectorShiftSvshift (had wrong z component).
• Handle ITAGS and CTRL tags correctly in the JIT versions of StoreScalarLinear & StoreVectorLinear.
• Handle vdma* loads in JIT.
• Fix a potential correctness issue in the JIT MULScalarShiftLeftImmediate and MULImmediateShiftLeftImmediate with shifts > 32.
• Fix JIT RangeOnly and ModuloOnly register indices.
• instructiontest.cof has been updated with more tests (verified to match real HW / Samsung DVD-N501).

03/21/2025 version 0.6.6:**

• Implement (bi)linear address mirroring properly and enable it.
• Optimize (bi)linear memory reads, and some memory stores.
• Optimize DMA transfers (especially all that are triggered by T3K).
• Fix linear DMA if Write & Dup & Direct mode.
• Fix linear DMA when writing to local control register.
• Fix missing 0x7F mask for shifts in certain mul/shift ops when using the dynamic recompiler.
• Implement BDMA_Type8_Read_0 (e.g. fixes Space Invaders 'Overlay' play mode).
• Add zero 'area' return to GetPointerToMemory(), as some games read 'invalid' mem regions (like $0 which returns 0 on Nuon HW).

01/05/2023 version 0.6.5:

• Fix loading of nuance.cfg when drag and drop is used
• Hide mouse cursor when in fullscreen mode
• Make rendering window behave like the others when trying to close it (e.g. clicking 'X' and Alt-F4 work now)
• Fix user breakpoints
• Fix 16/16z pixel format bilinear DMA
• Enable kprintf outputs (also new documentation about it within this readme file)
• Fix regression in code interpreter (shift+XXX opcodes)
• Slightly optimize code interpreter (ALU add+/sub+XXX opcodes)

10/01/2023 version 0.6.4:

• Fix command line file parsing, also drag and drop will now work properly
• Fix wrong handling of global overlay alpha values > 0
• Optimize OpenGL output shaders
• Fix CRT-like shader for very high resolutions (e.g. 4K)
• Retweak CRT-like shader to be a bit more realistic compared to common CRT TVs back in the days

04/01/2023 version 0.6.3:

27/11/2022 version 0.6.2:

• Tweak audio interrupts some more
• Correctly set DST
• Some internal buffer overflow protection
• Improve debugging output a bit

09/19/2021 version 0.6.1:

Merlin Racing (VM Labs, Inc., Dec 05, 2000). Niestety w tym momencie następuje zwis... a wygląda całkiem całkiem.

Ballistic (VM Labs, Inc., 2000). Coś dla fanów Zumy, czyli dla mnie.

Ballistic (VM Labs, Inc., 2000)

• Optimizations
• Fix some more opcode implementations
• Improve Video- and SysTimer precision

5/28/2020, version 0.6.0:

• Fix corrupted gfx in (at least) T3K
• Fix wrong DMAs in Ballistic and Merlin Racing (so that these display mostly (always?) correct now)
• (Mostly) fix distorted sound
•  Optimizations all over the place (memory usage and performance)
• Always enable Pixel Shaders and fully support pixel mode 2 for more performance
• Remove fields-per-second and always-update-video settings in the ‘nuance.cfg’
• Support Nuons VSyncing mechanism if a PC is emulating ‘too fast’
• Resolve a lot of hangs during gameplay, although this can still happen in some demanding situations (especially on low end PCs)
• Fix fullscreen toggle (via F1/ESC)
• Rudimentary support for x64/64bit compiles, needs to force-disable dynamic compiler though
• Enable Reshade support by using double buffering