the idea is that with SSE, doing operations that don't modify all the vectors can give false-dependencies, and in some situations prevent CPU's out-of-order execution (slows things down).
so the solution is to only use stuff like "movss" when you need to preserve the upper 3 vectors, if not, its always better(faster) to use movaps.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@751 96395faa-99c1-11dd-bbfe-3dabce05a288
note: most of the microVU EFU opcodes are implemented with completely different algorithms than zerorecs. this might prove to be more accurate, but i mainly did it to avoid using x87 FPU instructions (since i'm using mmx regs for storage).
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@749 96395faa-99c1-11dd-bbfe-3dabce05a288
<drk||Razi>cotton2: write more helpful logs kthx ;p
<cotton2>doesn't matter, no-one cares what i write since it doesn't effect anything for now
<Krakatos>cotton, stop posting those useless commit comments XD
<feal87>no no, please continue. they're fun, i look forward to read them
<feal87>from time to time
<feal87>:°D
<cotton2>lol :D
<Dwarg>Please spend more time making them more amusing.
<drk||Razi>cotton2: it does matter
<cotton2>dwarg: yeah i know!
<Dwarg>drk||Raz: Why? They're useless for reversion testing
<cotton2>i have to think of good stuff to write :O
<drk||Razi>when the next vu coders (if any, ever ;p) want to look at your code .. and the history is full of crap
<drk||Razi>that won't be much useful
<drk||Razi>;p
<cotton2>there won't be a next vu coder!
<Dwarg>Comments in the actual code would be more useful, wouldn't they?
<cotton2>my code is perfect!
<drk||Razi>lol
<cotton2>>.> <.<
<cotton2>but seriously
<cotton2>i don't think it can get better
<cotton2>unless like
<Dwarg>Rather than going back and comparing SVN comments to changes made way back when
<cotton2>you want to support AVX
<cotton2>or w/e intel's new SSE thing is called
<feal87>and anyway AVX is like 3 years away
<feal87>in that timeline pcsx2 i think will be stabilized
<feal87>:°D
<cotton2>yeah
<drk||Razi>cotton2: just log kthx
<drk||Razi>implemented y,x,z
<drk||Razi>:>
<cotton2>thats what the diffs are for!
<feal87>and anyway really its useless for this stage
<drk||Razi>yea i know
<feal87>until microVU is active and used by PCSX2
<cotton2>i'm just implementing different opcodes though, its not like i'm chaning features
<drk||Razi>you can tell anything from the code -- no comments or changelog needed.Right ? ;p
<feal87>then every change should be documented for regression testing
<cotton2>drk: yup!
<cotton2>well
<cotton2>its not like i'm changing stuff
<cotton2>i'm just implementing things that havn't been coded
<cotton2>so theres nothing to regress to
<feal87>yep
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@742 96395faa-99c1-11dd-bbfe-3dabce05a288
Appended notes:
* ADD in iFPUd should be bit accurate (unless it isn't. needs TESTING)
* MUL in iFPUd with Software Emulate MUL is as much as I could get near bit accurate (not quite enough, probably. needs TESTING)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@728 96395faa-99c1-11dd-bbfe-3dabce05a288
* Fixed Memcard init so that Memcard1 isn't default in both slots (oops!)
* Fixed Memcard path logic so that cards outside your pcsx2 folder can be browsed/selected.
* Fixed CDVD-to-BIOS time sync (I simply forgot a function call!)
* Optimized yuvrgb_sse2, by using Mod/RM form instructions.
* Win32: Same optimization applied to FreezeXMMRegs and FreezeMMXRegs (linux already had this optimization)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@719 96395faa-99c1-11dd-bbfe-3dabce05a288
Pcsx2: Small speedup for the IOP's recClearMem (used a forceinline to cure MSVC's laxidazical inlining)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@714 96395faa-99c1-11dd-bbfe-3dabce05a288
Savestates: Pcsx2 now errors when it encounters a savestate made by a newer version of pcsx2.
Added a new SafeList type to the SafeArray collection (not well tested yet).
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@701 96395faa-99c1-11dd-bbfe-3dabce05a288
Removed various instances of legacy VM code that is no longer needed for reference purposes.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@695 96395faa-99c1-11dd-bbfe-3dabce05a288
Fixed up the Saveslot detection somewhat (File:Load > menu now greys out unavailable saveslots).
Applied ICC fixup patch from Issue 84. :)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@677 96395faa-99c1-11dd-bbfe-3dabce05a288
* Moved most of the shared code between the two GUIs into platform independent source modules (System.cpp, Saveslots.cpp, and RecoverySystem.cpp)
* Created two new namespaces which house functions that need to be implemented by OS/platforms: HostGui and HostSys. HostGui is useful -- HostSys I'm not sure I'll keep around in the long run.
* Moved keyEvent struct from PS2Edefs.h to PS2Etypes.h
* Many improvements to the logic flow of the GUI (should be a little less fallible).
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@675 96395faa-99c1-11dd-bbfe-3dabce05a288
Added a temporary gamefix option for God of War, which has been broken since r659. Anyway I'll fix this problem in a proper way.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@667 96395faa-99c1-11dd-bbfe-3dabce05a288
This fixes the Digimon menu for example (Gamefix will stay a while longer though, until we confirm ingame is fine as well).
He also updated the advanced dialog a bit.
Remember that this mode can break games if VU clamp is below "Extra + preserve sign"
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@665 96395faa-99c1-11dd-bbfe-3dabce05a288
Win32: Moved various debugger-related dialogs into a new debugger.rc file, to help reduce clutter between both the std resources and the debugger resources.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@660 96395faa-99c1-11dd-bbfe-3dabce05a288
EE Interpreters: Fixed some signed/unsigned mistakes in some instructions, namely DIVU, DIVU1, unsigned Traps, and a couple unsigned right shifts. (all of these were already emulated correctly in the recs)
Also: Removed the ThreadPriority stuff from Pcsx2, since it was a throwback to the days of Win95's unstable multitasker. If you really really feel like you need to change the thread priority of Pcsx2, use the Windows Task Manager or a third party util.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@654 96395faa-99c1-11dd-bbfe-3dabce05a288
Fixed new emitter so it compiles in Debug builds, cleaned up the header files a bit.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@652 96395faa-99c1-11dd-bbfe-3dabce05a288
the emitter is now 'templified' so that you can run multiple instances such as:
eMOVRtoR<0>(EAX, ESP);
this uses emitter instance #0
to use another instance you can simply change the number in the brackets like:
eMOVRtoR<1>(EAX, ESP);
will use instance #1.
all old-functions are mapped to instance #0 by macros.
like:
#define MOVRtoR eMOVRtoR<0>
why do this to the emitter?
so we can have thread safety, and eventually thread the recompilers using different emitter instances.
note: this took me forever to get working (around 12 hours of non-stop coding).
however for some reason debug build is being extra-picky and still giving compile errors.
hopefully Jake or someone else can fix this, because i tried a few stuff, and just got more compile errors ><
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@647 96395faa-99c1-11dd-bbfe-3dabce05a288
* Fixed the IOP's recExecute so that it correctly preserves X86 registers (EDI and EBX were not preserved!)
* Optimized the recExecute procedure, seems like a nice speedup in FMVs and some games that are IOP intensive.
* Renamed psxMemRead to iopMemRead, added new Virt/Phys functions, and fixed several instances of DMAs using translated addresses (DMAs are always physical maps). Our IOP doesn't really emulate the tlb so it won't fix anything, but it's more correct in cas stuff is better supported in the future.
* Removed unneeded FreezeMMXRegs, since the IOPrec doesn't use any mmx/xmm regs anyway.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@628 96395faa-99c1-11dd-bbfe-3dabce05a288
Developers: Changed the way PCSX2_ALIGNED16 macros work, so that they're more friendly to MSVC and Visual Assist X intellisense (more more red squigglies on vars like cpuRegs!)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@618 96395faa-99c1-11dd-bbfe-3dabce05a288
WinGUI: Disabled arguments menu in release build, for sure :p
Buildsystem: Made pcsx2 dependent on zlib, and changed post-build script a bit
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@591 96395faa-99c1-11dd-bbfe-3dabce05a288
tweaked the fpu compare clamping so now gt4 works again.
made the digimon rumble arena 2 fix into a gamefix.
tekken 5 doesn't need a gamefix anymore.
VU fixes:
fixed 2 opcodes thanks to nneeve.
optimized FCOR a bit.
changed the way ICO gamefix works so its less hacky (just always sets VI to 1 instead of setting VI to the opposite of the 'correct' result)
General fix:
there was some odd bug with the autogenerated TEXTINCLUDE stuff.
if you edited a resource, it would generate
#include "afxresmw.h
instead of
#include "afxresmw.h"
(a quotation mark was missing so you'd get compile errors)
so i fixed that ;p
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@563 96395faa-99c1-11dd-bbfe-3dabce05a288
this fixes a bug with Digimon rumble arena 2.
thanks to Nneeve for figuring out the problem.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@547 96395faa-99c1-11dd-bbfe-3dabce05a288
Fixed a bug in the memorycard hotswapper. It wasn't reloading the cards correctly after changes.
Improved the new INTC hack slightly, and changed its description since it's not quite as universally awesome as Pseudonym and I had hoped when we worked on it last night. -_-
Added __fastcall and __forceinline to some of the VIF's unpack functions, where appropriate (very small speedup).
Removed some code I added to the MULT/DIV instructions, since it wasn't needed afterall, and fixed some typos in vtlb's API.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@538 96395faa-99c1-11dd-bbfe-3dabce05a288
This new code properly handles these situations now, removing the need
for the magna carta gamefix, and also fixing problems in dragon quest 8(jp).
Thanks again, tmkk :)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@536 96395faa-99c1-11dd-bbfe-3dabce05a288
Cleaned up some of the signed/unsigned ambiguity surrounding MULT/MULTU instructions.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@533 96395faa-99c1-11dd-bbfe-3dabce05a288
To make up for the speed loss in Tekken5 and Persona4 the vu cache size is now bigger.
This means another 10% more speed for Tekken, and a whooping 30% for Persona :)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@524 96395faa-99c1-11dd-bbfe-3dabce05a288
note to pcsx2 users: for those that still don't know, these are new VU recompilers i'm working on. They're not working yet, and won't be for at least a few months. So don't expect noticeable changes when i update microVU stuff, since currently they're not being used (when they're in a semi-working/usable state, i'll be sure to put a note in the revision message ;p)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@478 96395faa-99c1-11dd-bbfe-3dabce05a288
Added interface.cpp (plugin/pcsx2 interface) and savestate.cpp to SPU2ghz, to help clean up SPU2.cpp.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@463 96395faa-99c1-11dd-bbfe-3dabce05a288
We'll test if any game still needs DaZ off, and if not so, make that default to on.
DaZ is a more correct behavior for the VU's, and only bugs prevented it from working as expected.
Thanks to Nneeve for his hard work :)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@461 96395faa-99c1-11dd-bbfe-3dabce05a288
- made "framelimit" mode default, instead of "normal" mode (was discussed by team)
- some more microVU rec work...
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@452 96395faa-99c1-11dd-bbfe-3dabce05a288
Changed the Hw FiFo to use what should be a more correct register addressing scheme. The entire contents of each page of the FIFO is mirrored into the lowest register(s). For example, any address between 0x10006010 and 0x10006ff0 should always be treated as address 0x10006000.
Code cleanups and minor optimizations to hw.cpp, FiFo.cpp, and gs.cpp.
Added some additional utility functions to the Path namespace, for splitting paths into files/folders parts.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@451 96395faa-99c1-11dd-bbfe-3dabce05a288
Fixed COP0's recompiled branch instructions (BC0F, BC0T, etc) -- the conditional was not implemented correctly (thanks to Refraction for spotting that one).
Used a better method of clearing the errorlevel during the Pre/Post Build steps. Should be a little less error-prone (pun?)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@447 96395faa-99c1-11dd-bbfe-3dabce05a288
- made pcsx2 to default to MTGS mode "on" instead of "off"
- renamed "Mega VU" recs to "Micro VU" recs since the team thought "Mega" sounded lame ;p
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@439 96395faa-99c1-11dd-bbfe-3dabce05a288
cottonvibes