* git: Ignore uncaught linux build artifacts
* onepad-legacy: Return legitimate key event on `PADkeyEvent` call
Onepad-legacy was the plugin that was building and being used on my linux environment. I'm not sure if t hat's expected and normal 'onepad' has yet to take over.
* onepad: Return legitimate key event on `PADkeyEvent` call
* recording: Remove unnecessary `PadKeyDispatch(ev)` wrapping func
* recording: Cleanup key event handling when GSFrame CoreThread is paused
* recording: Refactor recording status check to be more explicit
* recording: Define additional key bindings for capitalized varient
Despite the advice of the docstrings in these files, on linux this seems to be required for the bindings to consistently fire.
* onepad: Declare `AnalyzeKeyEvent` not static
* recording: Ensure file extension is appended to recording file on linux
* recording: Correct typo in comment
* recording: Better wording around linux keybinding handling
* Remove unneeded `extern`
* IPU: Perform an interrupt on IPU_CTRL Reset.
Doing this reset performs a forced command end, meaning the IPU will run through the procedure of ending any command running and telling the core that it is done, even if there is no command currently active.
Fixes Fightbox
* Remove Fightbox patch
* VIF MSCAL instructions wait on VU T-Bit stop
Fixes:
The Sims
Spiderman 3
Sharks Tale
Maybe some bits of Over the Hedge
* Removed Spiderman 3 patches
* Fix VIF DMA resuming when stalled on tag
* Remove warning which doesn't really need to be there anymore
Don't interrupt if compare/overflow flag is already set
Removed patches for Harry Potter Prisoner of Azkeban and Chamber of Secrets as they should no longer be required.
Note - Some harry potter games still have the EE Timing Fix which may also not be required
* macOS compile
* Fix memprotect error on macOS
* Fix semaphore wait + thread cancel on macOS
* Fix timedlock timeout calculation
* spu2-x macOS
* onepad macOS support
* Add MacOS game controller db
* Disable onepad_legacy on macOS
* Fix spu2-x configuration crashes
* Make recompiler work on 32-bit macOS
* Use dylib extension for plugins on macOS
* Make app bundle on macOS
* Use git info for Info.plist version
This change makes the EE recompiler not hardcoded to working with 32 MB of RAM, and instead work with the amount of RAM set in Ps2MemSize::MainRam. The rest of PCSX2 seems to work fine with more than 32 MB of RAM - it is only the EE recompiler that has trouble. If the Ps2MemSize::MainRam value is not changed from the default 32 MB, there should be no change: 32 MB / 0x10000 = 0x200, the value that was there previously.
This may be helpful if anybody else in the future wants to emulate a PS2 dev kit with 128 MB or RAM, or maybe the PSX dvr thing which I think has 64 MB of RAM. I've confirmed that with the change, you could set Ps2MemSize::MainRam to 128 MB, and execute code with the recompiler that's above the first 32 MB of RAM, and do VIF and scratchpad DMA transfers from this upper memory as well.
* Modify VU addressing so it only multiplies by 8 before entering the program
Fixes issues with VU1 TPC being read multiplied by 8 (bad)
* Removed assert on SuperVU which no longer makes sense
* Change the minimum cmake version to a range. Uses the new policies of the newer versions if you are on them, taking care of CMP0054.
* Grab a newer version of FindHarfBuzz.cmake from Webkit that fixes the cmake warning the earlier one was giving.
* Add in -ftime-trace as an option on build.sh.
* Precompile PrecompiledHeader.h if you are using cmake 3.16+ and GCC. (Currently disabled on clang due to compilation issues.)
* Check if target_precompile_headers exists rather than by version.
Co-authored-by: scribam <scribam@users.noreply.github.com>
VS complains about errors so isolate the code in each file for TAS with macros.
Remove ifndef for m_capturingVideo as it is used for video capture menu.
Update all documentation for 1.6.
Source files for documentation are in md format and are kept in pcsx2\pcsx2\Docs, final exported/converted docs are in pdf format (Configuration_Guide and PCSX2_FAQ).
Update First time wizard to reflect recent documentation changes.
Update cmake to reflect recent documentation changes.
PO/POT/MO Localisation needs to be updated for the First time wizard to include the documentation changes.
NSIS installer needs to be updated to reflect the recent changes.
Fixes booting issues in the following games:
Jak X, Namco 50th anniversary, Spongebob the Movie, Spongebob Battle for Bikini Bottom,
The Incredibles, The Incredibles rize of the underminer, Soukou kihei armodyne, Garfield Saving Arlene, Tales of Fandom Vol. 2.
The games will no longer require a patch to boot.
Add forum and license links.
Remove Developers and Contributors names, replaced with:
"Big thanks to everyone who contributed to the project throughout the years."
Avoid some function overloads that can accept char*, wxChar* and
wxString as a parameter. wxString can be constructed with either of
those parameters, so the function overloads may actually result in
ambiguity errors. Keep the wxString parameter versions and remove the
rest.
Fixes some compile errors on FreeBSD.
The pack command converts from RGB32 to RGB16/INDX4, not RAW8 to
RGB16/INDX4.
Also, the VQ conversion (still unimplemented) only occurs when the
output format is 0, so fix that as well.
This fixes timing issues in Dynasty Warriors 3 Xtreme Legends
(fake save corruption), Jak II (random speedup issues) and Shadow of
Rome (FMV audio issues).
Also update the comments since a lot of them refer to past revisions of
the code.
* cdvd: Fix off-by-one end of file checks
* cdvd: Fix loading for games that attempt to read non-existent sectors
Some games will hang when attempting to read non-existent sectors.
Just do nothing when it occurs instead of erroring out.
* cdvdgigaherz: Fix loading for games that attempt to read non-existent sectors
Some games will hang when attempting to read non-existent sectors.
Just do nothing when it occurs instead of erroring out.
* cdvd: Don't write non-existent sectors to blockdump
Setting the toggle variable is now at the vsync point. FMV toggle should still work as usual, but the F9 toggle should have to wait the full period of around 32 frames. This seems short enough while also depends on how fast those frames can rendered, avoiding having to set a timer in ms.
* Refactor build.sh. Restructure finding plugins. Remove legacy code. Add --clean-plugins option. Document --no-trans option.
Don't do setcap if dev9ghzdrk isn't being built. List dependencies if they are missing. Remove code related to zerospu2, and misc plugins we aren't building. Minor message changes.
2 pass fallback cases for the untitled scenario was mentioned in the TODO comment to be completed, added one of them. (Still doesn't seem reliable enough, but honestly when do we Also modified blank space indentation at a part to tab space to satisfy lightning boi.
Previously the blockdumps will be automatically stored in the root directory of PCSX2, added an INI option in "PCSX2_ui.ini" named "CurrentBlockdump" to modify the directory to whatever is preferred by the user. (Some people were requesting for this)
A GUI could also be added if anyone wants, but considering the popularity of this I'm not sure as blockdumps will only be used my less percentile of users and mostly by testers and developers for debugging.
in vs 2019, the system xhash used by unordered_map uses Target in its template definition. pcsx2 #defines this in the r5900 and r3000 headers which leak cause a cryptic compilation error. A quick fix here is just to undef it in gamebase.h. A better fix would be to encapsulate the definition or use something more c++ to avoid the pre-processor. Add a comment to explain the _Target_ undef.
tap.h did not have string defined. I guess this was included transitively somewhere with an earlier set of system headers.
Biostools, the file is opened in text mode which causes the bios to be partially read. I'm not sure that's vs2019 specific but it's wrong in any case. Maybe the default mode parameter changed at some point. Maybe wxFile would be a better choice here rather than WxFFile because wxFile always open s in binary mode.
Attention! Older savestates will be incompatible so please use a
previous older dev build (such as dev 3043) to save your progress on
memory cards instead!
Squashed commit:
[7955b42e3] recording: Throw errors on fread/fwrite errors.
[5a2160f9e] recording: Remove function implementation from header files
[f2937ab5f] recording: Fixed UndoCount metadata bug and will gracefully fail if savestate is missing
[d7f4d43e5] recording: Refactored code-style to be consistent
[0f77fbb71] recording: Refactor to use switch statements
[28d7945f6] recording: Resolve CMake warnings and use tagged github links for cross-linking to LilyPad
[7c01c6cb4] recording: corrected disparity between comment and code
[17a8bd8d6] recording: Remove all usages of #define
[3830f5a82] recording: Refactor enums and general cleanup
[569ef7d67] recording: Completely disable new console log sources when recording is disabled
Regressions were discovered after merging with master due to way the save state data was saved within the movie file.
This change uses the same functions used in the GUI to create savestates to create a compressed save-state file. Eventually this could be re-incorporated back into the recording file and could be backwards compatible.
Adds a separate author line for pnach entries. This will allow to separate comments with other info such as pnach purposes.
GameDB could be updated with the new changes for a more slick and clean look.
VuClipFlag hack is only used for SuperVU1 Recompiler, let's avoid enabling it when it's not used.
Previously a console message would be provided stating that "(GameDB) Enabled Gamefix: VuClipFlagHack" even when the user is using the Interpreter (or) MicroVU Recompiler. The following patch prevents such confusions.
Old presets touched parts such as VU clamping for what seems to be no good reason at all, and applied some rather dodgy combinations of EE cyclerate and cycle skipping. Also added MTVU to presets 3-6.
Pnach had a limit that, while increased at some point to 2048, is still not enough for everyone. This uses a vector to avoid that limit, as there is no reason to keep it and people loading pnach with over 2048 patches most likely know what they are doing.
Inipatch group member was also unused in the whole codebase so I did some cleanup and removed it.
gui: Move the Off option to the top instead of bottom for FMV Aspect Ratio Override.
It's not ideal for it to be at the bottom.
Adjust the logic how the button works, there was an issue if both aspect ratios
were the same the aspect ratio chosen was selected to Fill.
Also adjust the code so the button better responds to realtime toggling.
If disabling the button realtime during a fmv then the normal aspect ratio
should be chosen.
There's still an issue left where if the fmv button is changed first and then
the aspect ratio button the button would would not work. Switching it off
seems to reset and make it work again.
Change "Switch to 4:3 aspect ratio when an FMV plays" to
"FMV Aspect Ratio Override" drop down menu.
The option temporarily changes the Aspect Ratio to either
16:9 or 4:3 depending on the option that is selected,
and depending on the Window Aspect ratio that is selected.
Original commit:
8d471e1f16
Co author: lightningterror.
This pull request is for the pending reorganization of the folder structure on GSdx,
making it better organized and easier to work with.
Also remove unused GSTextureFX.cpp file.
Currently, none of the EE kernel is viewable from the debugger, which poses issues when debugging code that makes use of syscalls. This PR should allow the memory region 0x80000000 - 0x800FFFFF to have readable disassembly.
If the entry separator (line of dashes) is found, assume the multiline
section end tag is missing or incorrect and stop parsing so that it
doesn't affect subsequent database entries.
Also improve the error messages so it is slightly more clear what the
error is.
With these changes, saveslots will be labeled as either empty or
with the date that the file was last updated on. The menu items
for loading them are also disabled if the slot is empty.
It's possible if you are very fast to access the menu before the
slots change. It updates when the crc changes. When you save a saveslot,
the menu item also changes to show the time you told it to save until the
file is actually done saving.
Also fix an issue with backup saveslots not working properly from the
gui on first load.
This reverts
93d5b52df3f3e78b826755155ca7f1.
Unfortunately wxString stuff is a PITA to deal with. Breaks FreeBSD
compile, but there are probably more issues that just haven't been
reported yet...
Avoid some function overloads that can accept char*, wxChar* and
wxString as a parameter. wxString can be constructed with either of
those parameters, so the function overloads may actually result in
ambiguity errors. Keep the wxString parameter versions and remove the
rest.
Fixes some compile errors on FreeBSD.
Fixes an issue where the first breakpoint after rebooting a game may
be skipped if the following sequence takes place:
- The first breakpoint after booting the game is triggered once.
- The user hits run to resume the game
- The user reboots the game without any other breakpoint being
triggered.
It doesn't provides much of a performance improvement over directly
using an unordered map. This change also means that if there are
duplicate GameDB entries then they'll be merged together instead of
having only the last entry take effect.
Also increase the unordered map reserve size.
Converting the string to lowercase is unnecessary when the actual entry
is still case sensitive.
Also just use std::hash of std::string and std::wstring instead, which
fixes a FreeBSD compile error (cannot convert to const char*).
There's no need to have a lot of overloads accepting wxChar*, char* cand
const wxString&. Keep only the const wxString& versions and remove the
rest. This fixes an infinite recursion warning on FreeBSD.
Also simplify sectionExists and getSection to avoid unnecessary
conversion to and from wxString.
There's no use for it now that the database editor has been removed.
Also remove the Game_Data POD comment because I don't think it makes
sense and remove an unused variable.
* Enable logging from new source, PS2 system output
* Enable printf formatting with (up to only 7 currently)
* Fix fallthrough bug in SYSCALL switch/case
* Remove unnecessary memread32 call in sysPrintOut case
It has too many issues:
- It's in an incomplete state. The ListView isn't connected up and I'm
not sure what it was supposed to do.
- Comments are stripped from the database when changes are saved.
- Some key value pairs do not show (EE/VU clamp/rounding and
MemCardFilter).
- It doesn't work well on Linux (though this one seems easy to fix).
Fixes empty log and ini directory paths being passed to the plugins on
Windows when the pathnames contain characters not present in the current
system codepage.
Fixes the inability to create snapshots when the pathname contains
non-ASCII characters on Linux (Windows requires more pathname handling
fixes).
In games like ICO where the render and blanks both need to be rounded,
one was prioritized over the other one leading to a truncated value in
the total vsync cycles. This patch fixes it, it's more or less for a
trivial cycle accuracy rather than a signfifcant change which might
influence games. (Not sure, maybe it would?)
Potentially it might be nice to revamp the whole vertical blank
calculation system to make it more readable (discussed about this with
ref), but it's not of much priority right now.
Remove mention of Crash Tag Team Racing from VU I bit gamefix tooltip.
Not needed since the gamefix has been removed for the game as it causes
more issues.
Instead of binding all checkbox events to an event handler, bind only
the master trace log toggle checkbox event and continue processing the
event after it has been handled. This fixes the non-functional apply
button and is also more efficient.
Fixes performance drops and short freezes in adventure mode.
Alas, there is no Crash Tag Team Wrestling, so I replaced it with the
correct title.
Updates Gamefixes Panel entry of VU I bit Hack.
Closes#1785
I guess the intention of the older logic was to get the truncated part
by comparing with the original value, but that went wrong due to using
the pre-normalized value (before the division) for comparison.
Hence let's just do a modulo for rounding the render and blank.
Previously, the DVD variant NTSC/PAL modes used the horizontal blanking
interval calculation algorithm used by digital video modes, which
shouldn't be used and also rounding error check was neglected.
Added the DVD variant modes to the list in analog video mode finder
subroutine. This should impact timing/vertical synchronization in PSX
games significantly.
This adjusts the copy bytes command format from
5aaaaaaa nnnnnnnn
00000000 bbbbbbbb
to
5aaaaaaa nnnnnnnn
bbbbbbbb 00000000
so that it matches the copy bytes command format used by PS2 cheat
devices (GS/CB/XP/AR2).
Switch to using vector for the dump sector table, and also fix a bug
where memory was not allocated when writing to a v2 format block dump,
causing a null pointer dereference.
Also switch to using the v2 block dump format, which generally produces
smaller dump files (the dumps also seem smaller than the ones generated
by cdvdiso, which seems to repeat sectors).
- Adds tooltip for Convert memory card button in the memory card dialog
- Updates several memory card console writes.
- Adds 2018 copyrights to the updated files
Changes the file extension used by PSX memory cards to the common .mcr
instead of using the same (.ps2) extension used by PS2 memory cards.
Updates the extension listed in Create memory card dialog when a PSX or
PS2 memory card is selected.
Replaces the awkward checkbox for a PSX memory card in the bottom left
of the Create Memory Card dialog with a radio button like the other
memory card types.
The workaround was used to prevent macro substitution occuring on paths
containing "linux", but it's no longer necessary since "linux" is not
set as a predefined macro when a "-std=" option is passed to gcc.
Fixes#2268.
I found it might be useful for looking into issues at certain cases
where users provide a screenshot covering the titlebar, so we can easily
identify which video mode is being used.
Especially useful for looking into PCRTC issues.
The assembler was previously assembling instruction "b i" as an alias to "j I".
This caused unexpected behavior when attempting to assemble an
unconditional branch. The previous behavior would cause
position-independent code to fail due to the distinction between
the absolute address of a jump vs. the offset of a branch.
Fixes memcard access in Final Fantasy VIII, and likely in other PSX
games supporting the Pocketstation peripheral. This makes inserted PSX cards show up as PocketStation devices in the PS2 browser, which is okay for now.
doc: Update FAQ document
* Update system requirements to match README.md and remove references to v0.9.6
* Fix formatting issues, add link for DirectX and speed issues
It's not really used, and the OSD uses a different API.
The specified calling convention (stdcall) is also incorrect since
variadic functions are caller-clean, not callee-clean. The compilers
ignore the stdcall and just use cdecl (I think), though it does trigger
a -Wcast-calling-convention on clang.
It seems not all DMA channels have the same set of 32 bit registers. Removed
addresses of registers which aren't actually present in the memory
space.
Example: Channel 0/1/2 have address stack registers but the other
channels lack it. According to documents, the remaining memory space of
the channels seems to be reserved. Which means, write access would be
disregarded and read access would return an unknown value.
Credit goes to Gregory and CK1 for notifying me about it, special mention to ssakash for actually pushing the change to github. Also I wasn't the one who introduced the non-existent registers into the code, these registers were present before under a different name.
The macro for address of channel 9 was wrongly having the address of
channel 8, fixed it. (Luckily MADR and QWC were unused so we should be
safe)
Thanks to Fireboyd78 for notifying us about this. (Closes#2091)
Also fixed some inconsistencies where some of the DMA channel register
addresses weren't defined for all the bitfields.
If emulation is paused and resumed, vsync may become enabled even if the
frame limiter is currently disabled. This state persists until the
settings are changed or the hotkeys are used.
Fix the inconsistent framelimiter/vsync behaviour so that vsync isn't
enabled whenever the framelimiter is disabled, which matches the
behaviour in the rest of the code.
The safest preset should ideally try to provide
the highest accuracy and stability when emulating the
PS2, allowing the manipulation of the MTVU hack could
just destroy both of these things, hence the following patch
force disables the hack on safest preset.
v2: allow all combinations of framelimiter and vsync options
v3:
* disable vsync when the user disable framelimiter with F4
* Use g_Conf->EmuOptions instead of EmuConfig
The scalar limit value was updated only during any turbo/slowmotion
toggle, let's also update it properly after any change in the emulation
settings.
This prevents the need of toggling from turbo/slowmotion to update to
your requested frame rate percentage.
Current behavior - The tilebar isn't updated when the user enters full screen mode and when the user returns back to windowed mode they have the older title bar values for a brief second, this sort of behavior is undesirable just in the cost of saving some overhead for updating title. (which is really negligent)
Hence reverting the code back to how it has been for the past 7 years (Yes, I did my research), I'm doing it only for the windows side at the moment as a code comment describes of some sort of Linux specific issue on wxWidgets side.
The default 12 is rather low and won't suffice for most cases, updating
it to 20 to give some extra space for additional ISOs. Incrementing it
to an even higher value might not be so good as it consumes lots of
vertical space, not a nice idea for people with smaller screens.
Thanks to @colepcsx2 (https://github.com/PCSX2/pcsx2/pull/1896#commitcomment-21858717) for pointing it out!
I also updated the prefix in the inferior video mode detection of GSdx, I'm not even sure why we need the videomode info on the plugin side, might be useful someday.
The title bar will display "progressive" for 1-2 seconds when booting an interlaced game at some cases, it's due to an uninitialized SMODE2 register, so let's wait till the rendering starts to allow time for the
SMODE2 init and get the proper values from games.
My guess is that GCC reports an warning because of the default case.
I don't think we support yet __builtin_unreachable on Linux
Nevertheless it will still be an issue in non-release build
This reverts commit 99180f5afb.
Unfortunately Mint/Arch got segmentation fault in Xlib mutex locking...
Sadly Xlib will be back to thread-unsafe mode.
I'm cooking another solution based on EGL
X11 isn't thread safe by default. It make sense in 1990 but it is ugly nowadays.
The trick is that is must called before any X11 function. So the only
safe place is at the start of the main. Pcsx2App::OnInit() is the
sooner that I've found.
Fixes an issue where the game will either boot the wrong CDVD source or
fail to boot if the current CDVD option doesn't match the option
specified in the command line.
Improved the video mode detection code by also detecting the DVD variant video modes of NTSC & PAL, PSX mode actually make use of these specific variants, as well as the BIOS. Previously, I just had them as a single bios video mode due to laziness. (I know, my bad)
After further research, it seems that these DVD variant modes have their own individual VSync timing values similar to the standard NTSC & PAL video modes, dealing with those timer codes might be essential in getting timing accuracy of the PSX mode games. (I kept it to default NTSC/PAL values for now, interested people can mess with it later)
I had planned to do this before but there some were concerns that two different video modes make use of 0x73 gate in SetGsCrt, which was rather weird (how the heck could two video modes be used in a single param value?)
0x73- DVDPAL ( 720 x 480 @ ??.???Hz)
0x73- DVD480P ( 720 x 480 @ ??.???Hz)
Hence, we had decided to use the CMOD bit from SMODE1 (AKA color subcarrier frequency) to detect whether it's an analog or digital video mode and update the necessary timing values but seems like it's no longer necessary, after further discussions from some PS2 developers, we've come to the conclusion that only DVDPAL is possible via 0x73 in SetGsCrt. (So I assume the DVD480p init possibility was fake info from Blue and those other GSM guys who were reverse engineering the PS2)
Remove full path from every plugin binding and only add it to the
preceding "Loading Plugins..." line.
Instead the revision date and version number are printed for each plugin
to make it easier to identify issues with plugin versions.
Previously, the OSD neglected to mention any sort of message when the savestate load is failed, the following patch now also prints a message on OSD when detecting such cases of loading an incomplete/corrupt savestate.
This prevents the internal state of the objects from becoming
inconsistent, which causes inflate() to fail with recent zlib versions
(1.2.9 and later).
EE interpeter: remove unused argument
rdd is neither used, nor needed. It appears it was there to pass the _Rd_ word to write to, but the writing was moved to PHMSBH() to have one "if (_Rd_)".
Add a note on undefined behavior
The ring buffer is composed of severals read/write by transaction.
Atomic operations are only required at the start/end of the full
transaction. In the middle, you can use normal variable (optimization
opportunity for the compiler)
Use acquire/release semantics on isBusy and vuCycles to remain 100% safe
(relaxed might be doable but better be safe than sorry)
Use dedicated cache line for atomic variable to avoid any conflict between CPU
The struct is copied in various ring buffer (hot path)
We only need the return status of the function so use a reference instead of
a state variable
Side note: if we align the struct to 16B maybe the compiler can use SSE to copy it.
Warning: it breaks save state compatibility
Hopefully this translates well to slower systems :)
Tekken Tag:
Before: 79-81fps
After: 82-84fps
Front Mission 4 intro (as it pans over the roofs)
Before: 158-159fps
After: 165-166fps
Previously, the seconds variable of the RTC was updated on progressive modes after every 50 Vsyncs, which was obviously wrong. The code has been adjusted to update the RTC with respect to the vertical frequencies of various other video modes.
16B alignment is now useless for nVifBlock (no more SSE)
However update the alignment of bucket to 64B. It will reduce cache miss
probability in the find loop
It avoids memory stalls and greatly reduces the overhead of the dVifUnpack function
Here a vtune summary of this branch (done on SotC init)
dVifUnpack<1> was 14.5% of effective VU thread time
dVifUnpack<1> is now 3.8% of effective VU thread time
I hope it will translate to better fps
Delete() deletes the menu item but keeps the sub menu. Remove() doesn't
delete the menu item.
Also use AppendSubMenu - using Append on a submenu is deprecated.
It allow to compare only 8B in the lookup so SSE could be replaced with general instruction
As a bonus, it allow to compute the hash key with a mov rather than modulo (which was an 'and')
Inline the execution part
Add a num parameter to dVifsetVUptr
Use a local variable for the nVifBlock instead of a global struct state
The goal is to ease future update of the nVifBlock struct
Previous implementation saved the both the chain pointer and the chain size
Rational: size is useful to add new element and to detect the end of the chain
Vif cache is rarely miss. So 'add' is barely called and the end of a chain is
barely reached.
New implementation will add a null cell at the end of the chain. As a
cell contains a x86 pointer, if is null you could conclude that you
reach the end of the chain.
The 'add' function will traverse the chain to get the current size. It is
a cold path besides the chain is often short (< 4).
The 'find' function only need to check the startPtr bytes to detect the end
of the loop.
Note: SizeChain was replaced with a std::array
Safety:
* check remaining space before compilation
* clear hash if recompiler is reset
Perf:
* don't research the hash after a miss
* reduce branching in Unpack/ExecuteUnpack
Note: a potential speed optimization for dVifsetVUptr
Precompute the length and store in the cache. However it need 2B on the
nVifBlock struct. Maybe we can compact cl/wl. Or merge aligned with upkType
(if some bits are useless)
I misses some early return in my first tentative. Now VTune shows me
properly the time in VU recompiler.
Note: It seem some block overlap (likely due to the branching mess). But it is still way better than no data
GS_Packet constructor calls memset which is quite slow and useless as data is overwritten
Vtune overhead of Gif_Unit::Execute goes from 5.8% to 3.0% (EE thread)
arcum42