Re: [新聞] PCSX2 Playground rev 586 & GSDX 0.1.14

看板Emulator (模擬器)作者choan (八房龍の助)時間16年前 (2009/01/14 12:40)推噓1(1推 0噓 0→)

留言1則, 1人參與討論串1/1

PCSX2 Playground rev 586 rev 890 GSDX 0.1.14 http://re4rainbow.4shared.com/ Mirror: http://www.mediafire.com/re4rainbow Legend: '//blah-blah-blah//' or '/blah-blah-blah/' is re4rainbow's comment. Just FYI. Quoted from: http://code.google.com/p/pcsx2-playground/source/list rev 586: Disabled PCH for IPU.cpp -- it was causing problems that crashed FFXII vids among others. Moved Dynarec-related IPU const mem reads/writes to iIPU.cpp, and added some __fastcall directives to various relevant functions and function pointers (very small speedup). rev 585: Squashed an MTGS bug that I strongly suspect was responsible for MTGS problems reported in Issue 108. rev 584: Fixed crash bug when running Cpu Config before starting emulation of a game. Enforced correct propagation of the GS window handle to pad plugins on Esc/Resume. rev 583: Found something that might very well be a bugfix for some crashing problems stemming from changes made in r516. rev 582: Added some new highly-optimized memset alternatives, located in memzero.h (currently Win32-only). Since most uses of memset involve constant values for both the length and destination parameters, it's possible to use a function template to generate very precise inlined code. :) Additionally, the new memset alternatives correctly save the XMM0 register, which can be clobbered by MSVC's memset implementation. Changed some code in iR5900's eventTest to use EAX instead of ECX (very small speedup). rev 581: Just noticed I made a nasty typo in VifDma.cpp from r577, which probably made some games *very* unhappy. rev 580: Optimized the iR5900's event test code, undoing the slowdown from the previous revision and picking up another 1-2% on top of performance of earlier revisions. Applied gratuitous amounts of inlining to WinSysExec.cpp to ink out another 0.1% speed boost. heh Fixed the double-keystroke bug when using certain combinations of Pad plugins. Added JNS32 to the ix86 emitter. *IMPORTANT* VM builds are STILL broken. rev 579: fixed a typo from my last commit rev 578: *EXPERIMENTAL* Switched the code over to a new method of exception handlingwhich should be more efficient and easier to debug and, most importantly, fixes a lot of the weird errors we've been getting since r488. *NOTE: THIS IS REVISION WORKS ON VTLB ONLY -- VM BUILDS ARE CURRENTLY BROKEN* If you like VM, do not try to compile or use this revision. The VM build will crash anytime you try to load a savestate or exit/resume emulation. (Interpreters and some other stuff are also broken too, but will be fixed shortly) rev 577: Updated savestates to version 0x12 (0x02 VTLB) -- added a couple missing SIF vars and added support for cottonvibes' new VIregs structure format in r576. Fixed a few bugs in the GS-on-Close recovery system. It should work more and crash less now. Restructured the entire initialization process of Ps2 memory and cpu resources. All emulation system resources are allocated when the app first starts now, giving Pcsx2 the best possible chance to allocate needed Virtual Memory resources. And *Hopefully* this resolves some memory corruption issues that have persisted since r488. Added several new functions to System.cpp as partof the new API: SysAllocateMem(), SysAllocateDynarecs(), etc. Created some new files: VU0microInterp.cpp and VU1microInterp.cpp, and separated the VU cpus into their own "cpu" structure (they used to be grouped in with the R5900, which made it awkward to work with them). Turned off speed optimizations for the WinMain.cpp and WinSysExec.cpp files, due to a compiler bug in MSVC that causes it to completely break codegen for the RunExecute() function. rev 576: VI regs needed padding to make them 128bit VU0 maps VU1's VI regs as 128bits to addr 0x4xx0 in VU0 mem, with only lower 16 bits valid, and the upper 112bits are hardwired to 0. i changed the 2 functions that deal with VU mem addresses to reflect the changes (GET_VU_MEM() and recVUTransformAddr()) they also had a bug where they weren't mapping VU1's VF regs to 0x4000. i think i got all the VU mem functions that needed to be modified, if games break with this revision then we know i missed one >< rev 575: This'll at least get rid of the Linux compiler errors, though the tons of linker errors, half of which are in assembly files, are still there. rev 574: Various ZeroGS cleanups. A few things got mirrored from the dx version to the opengl version, and vice-versa. A few defines became inlined functions, and if statements became switch statements... rev 573: Repaired some more of the Linux-side of things. It's the least I can do for having constantly broken it. >_< rev 572: Moved the EE's various opcode subsets into their own namespaces, as per Cottonvibes' request (MMI, COP0, COP1), and fixed numerous compilation errors in the debug builds while I was at it. Improved the behavior of the VARG_PARAM thingie in release builds. It generates a little bit more compact compiled code now. rev 571: Added SysPrintfs to each VU opcode, to make debugging easier (phew..) :p rev 570: Re-enable a hack for the vm build. Fixes Fatal Frame 1 on that. VTLB builds don't need it :) rev 569: Removed some obsolete code regarding the EE's interrupt handler, and added some lines to the Linux code to help it build correctly (I hope) under the new namespace layout. rev 568: Fix a few compiler errors in Linux. I'm still trying to figure out the dependancy issues StringUtils is giving me, though, so doesn't compile in Linux right now... rev 567: Removed some obsolete #pragma warning() directives, and moved the one useful one into the precomiled header so that it doesn't need to be included into every file manually. rev 566: Code cleanup Revision: Implemented COP0 and COP1 (FPU) opcodes in the new opcode table system, rearranged namespaces to (hopefully!) make more sense, deleted a bunch of unused or duplicated prototypes in headers, and separated a few sections of general code into their own modules (such as XMMfreezeRegs/MMXfreezeRegs found in iCore and iR5900-32 modules, and some COP2-specific functions in VU0) rev 565: minor recUpdateFlags changes rev 564: minor VU recUpdateFlags() changes based on what Jake mentioned in r560 -- ※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 59.126.238.100 ※ 編輯: choan 來自: 59.126.238.100 (01/14 12:41)