Threading VU1 took a lot of rewrites and new code to make possible (MTGS, microVU, gifUnit...), but we finally got to the point where it was feasible, and now we've done it! (so now everyone can stop complaining that pcsx2 only takes advantages of 2 cores :p).
The speedups in the games that benefit from it are great if you have a cpu with 3+ cores (generally a 10~45% speedup), however games that are GS limited can be a slowdown (especially on dual core cpu's).
The option can be found in the speedhacks section as "MTVU (Multi-Threaded microVU1)". And when enabled it should should show the VU thread-time percentage on the title bar window (Like we currently do for EE/GS/UI threads).
It is listed as a speedhack because in order for threading VU1 to have been a speedup, we need to assume that games will not send gif packets containing Signal/Finish/Label commands from path 1 (vu1's xgkick). The good news is very-few games ever do this, so the compatibility of MTVU is very high (a game that does do this will likely hang).
Note: vs2010 builds and Linux builds need to be updated to include "MTVU.h" and "MTVU.cpp".
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@4865 96395faa-99c1-11dd-bbfe-3dabce05a288
Pcsx2 now has a gifUnit class which acts like the ps2's gif and executes a single gif transfer at a time (and performs path arbitration according to priority and path3 slicing).
This new code is generally a speedup for most games. Particularly VU heavy games like GoW.
This revision breaks old saved state compatibility so don't update if you wish to keep playing with your old saved states.
Leave comments if this revision fixes or breaks anything...
Message to GS Plugin authors:
This new code now uses only 1 gif callback to transfer data to the gs plugin (GSgifTransfer).
pcsx2 also is garaunteed to send full GS primitives to the plugin. So you don't have to account for partial-transfers anymore.
Thanks goes out to shadowlady who tested around 500 games for me :D
Note 1: The old gif code is still in this revision, and can be enabled by the USE_OLD_GIF macro. The old code will be deleted soon.
Note 2: This revision also enables assertion dialogs in devel builds, and changed a lot of assume cases into assertions.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@4821 96395faa-99c1-11dd-bbfe-3dabce05a288
* VIF now sends a 128 bit tag instead of a 64 bit tag (lower 64 bits masked to 0 -- this should mimic real hardware behavior more closely)
* Added more GIFtag logging info
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3916 96395faa-99c1-11dd-bbfe-3dabce05a288
* All remaining code for handling partial/fragmented unpacks removed.
* vifRegs.NUM is now accurately simulated when queuing data from fragmented unpacks.
* Reduced the VIFunpack fragment buffer from 1MB to 4KB (max size of an unpack due to NUM being limited to 8 bits).
* Removed vif/vifRegs globals formally used by VIF interpreters (everything relies on the templated vifIdx now -- simpler and faster!)
* g_vifMask vars are integrated into vifStruct.
* All VIF mask register stuff uses the SSE-friendly vifStruct.MaskRow/Col vars now.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3762 96395faa-99c1-11dd-bbfe-3dabce05a288
* Writes via 16 and 8 bit ops now use 32-bit read/modify/write operations by default; which should enable nearly complete support for all such operations (instead of the formerly spotty coverage before).
* Eliminated almost all former 8/16-bit specific register operations. All code shares the same 32 bit handlers now.
* Completely revamped the developer trace logs for hardware registers! *ALL* registers are logged now, complete with address, name, and value being read/written (and nicely formatted!).
* Handlers are now fully page-based using templated functions (minor speedup)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3704 96395faa-99c1-11dd-bbfe-3dabce05a288
* Removed the hacky g_pVU1 pointer, which required VU1 cpu registers to be part of VU0. Replaced it with a standard VU1 variable (mimics all other CPU registers, which are standard static vars). We were using translation functions/tables for all VU0 memory operations anyway, so this was a no-brainer.
* Removed code from microVU that was only there to help deal with the fact that g_pVU1 was annoying.
* Turned eeMem->HW into a static global array eeHw [64k].
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3692 96395faa-99c1-11dd-bbfe-3dabce05a288
* Includes a minor tweak to DMAC.h - removed tDMA_TADR / tDMA_MADR / etc. and replaced them with a single tDMAC_ADDR class.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3644 96395faa-99c1-11dd-bbfe-3dabce05a288
* Added __fi and __ri, which are abbreviations for __forceinline and __releaseinline.
* Added some static qualifiers to functions in mVU, MMI ops, and others where appropriate.
* Removed some unnecessary __fastcall qualifiers (since GCC gets funny sometimes when you combine __fastcall and inlining).
* Made _1mb, _16mb, _1gb values common to all emulation code (moved from newVif/mvu to Common.h) -- they're useful! :)
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3624 96395faa-99c1-11dd-bbfe-3dabce05a288
Games
- Resident Evil Dead Aim flickering graphics fixed.
- Flatout games should now boot everytime (instead of randomly)
- Sega Superstar Tennis flashing fixed
- Destruction Derby graphics fixed (NTSC confirmed, PAL unsure)
- Tenchu Fatal Shadows flickering textures fixed, Issue 748
- Spyro Hero's Tail should boot again, Issue 746
General Info
- General reworking of the GS paths with all paths prioritizing more correctly.
- XGKICK can stall if another GS transfer is in progress
- Major reworking of Path3 masking, could still be combined with normal Path3 transfers, but lets get it working first :P
- Taken in to account a few conditions which could have caused VIF to lock up/crash when stalling on ends.
- Unreversed GIF and SPR and VIF so its the normal way round
- all DMA paths should now resume correctly after being stopped by hardware
- GIF FIFO now actually transfers stuff :P
- VIF CMD's simulated by the length of the interrupts at the end.
Phew, i think that's everything :S
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3274 96395faa-99c1-11dd-bbfe-3dabce05a288
There is a lot of code duplication in the vif0/vif1 files, I will probably work on getting them to use common functions.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@2489 96395faa-99c1-11dd-bbfe-3dabce05a288
Notes to Devs:
- Linux project files probably need to be updated since I deleted some files.
- In the vif0/vif1 Freeze() functions for saved states, I kept some dummy vars to keep saved state compatibility. We should remove them next time we decide to break saved state compatibility.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@2461 96395faa-99c1-11dd-bbfe-3dabce05a288
Compatibility is probably the same as the newVif interpreter code, but its faster.
Speedwise its similar to the old-vif unpack code (the one currently enabled by default in pcsx2).
Its about 0~2% slower on my machine from my limited testing, but I assume people with SSE4.1 cpus might have a lot better results since I added a lot of sse4.1 optimizations...
The SSE4.1 optimizations were also ported to the newVif interpreter code.
Also the "filling mode" should be fast compared to the old-vif unpack code since its sse optimized, but most games don't use this mode so it hasn't been tested much...
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@2358 96395faa-99c1-11dd-bbfe-3dabce05a288
Others: Fixed a couple of unpack bugs, tried to tackle an MFIFO bug with Tekken Tag. Also re-jiggled a few bits on my recent changed, please negative if it breaks anything.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@1400 96395faa-99c1-11dd-bbfe-3dabce05a288
Fixed split videos in Gradius V
Fixed Spyro hanging problem in Issue 112
Put in a hacky fix for FFX videos into IPU to compensate the spyro fix (which is actually correct).
Implementing unpack overflow protection (Guitar Hero 3 & Toni Hawks Project 8)
Writing XGKick to a temp buffer before sending to the GS (part of the GH3 / THP8 fix)
Note! THP8 and GH3 will STILL crash with any VUrecs on and MTGS on, these must all be OFF. Also use GSDX in software mode with the NLoop hack on for now. Slow i know, but it works :P hopefully we can fix the rec side of it soon.
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@989 96395faa-99c1-11dd-bbfe-3dabce05a288
TheTechnician27