In a particular hashing heavy scene in Crazy Taxi the Murmur3 hash used 3.11% CPU time.
The new CRC32 hash in the same scene used 1.86%
This was tested on a Nvidia SHIELD Android TV with Cortex-A57s.
This will be a bit slower on the Nexus 9, the Denver CPU core is a bit slower with CRC32 texture hashing than Murmur3 texture hashing.
Fixes a bug in lfs where I was doing a rev64.16b when I needed to do a rev32.8b.
Change a ld1r.4s to a ld1r.2s.
Fix an issue where a fcvtl2 needed to be a fcvtl.
Re-enabled psq_l, issue with flickering can't be reproduced anymore, so whatever.
This cleans up some of the code between core and UI for disassembling and dumping code blocks.
Should help the QT UI in bringing up its debug UI since it won't have to deal with this garbage now.
Until loadstores are fully investigated, let's artificially extend the life of it by disabling all of them.
These instructions break in a very terrible way and eventually all loadstores end up attempting to access memory region (u32)-1.
Disable until someone has enough time to investigate further.
This is time based block profiling unlike the ARMv7 core.
I would like to add cycle counter based block profiling like the ARMv7 core, but it first requires writing a kernel module to enable the counters to userland applications
This affects enabling and disabling block profiling on the fly.
The block profiling pauses the CPU cores and then flushes the JIT's block cache and enables block profile.
The issue with this is that when we pause the CPU core, we don't have a way to tell if the JIT recompiler has actually left.
So if the secondary thread that is clearing the JIT block cache is too quick, it will clear the cache as a recompiler is still running that block that
has been cleared.
Using BindToRegister from the register caches causes the register bound to be marked dirty.
Using the regular R() function from the register caches loads the registers without being dirty.
When implementing new instructions make sure to BindToRegister registers that become dirty.
With my previous changes Dolphin would fail to create the user directory if it didn't exist, and would dump all the configuration options in to the cwdir.
This was a bit more complicated to fix in a clean fashion, so I took to moving around code concerning user directories.
Instead of having GetUserPath serve a dual purpose of both getting and setting our user directories, break out to a new SetUserPath function.
GetUserPath will know only get the configured user path.
SetUserPath will set our user paths and setup the internal user path state.
This ending up being a lot cleaner overall, which is nice. Also less mind bending when attempting to read the code.
So now we won't dump all of our configuration in to the cwdir if ~/.dolphin-emu isn't found.
Fixes issue 8371.
This came up from the discussion we were having prior about dumping half of a kilobyte of VFP registers to the stack is insanity.
This was due to me basically copying exactly what I did on ARMv7's paired loadstores, where the impact is less since we only use the bottom 64bits of
the VFP registers.
So I decided to think about how to improve upon this since I got called out on my terrible code.
The solution I have come up with is instead of jumping to the common ASM routine and having that check if it needs to take the fastmem or slowmem
routes, just inline the check in to the JIT block and jump to either a fastmem or slowmem handler.
Fairly simple, and this allows us to only flush the registers that are required when doing so. Should give a reasonable increase in performance for
games that use the slowmem path quite a lot.
We are declaring we require ARB_shader_image_load_store in the shader, this isn't an extension on GLES because it is part of the GLSL ES 3.1 spec.
If we are running as GLES then just not put it in the shaders.
Activity Monitor can start activities by using adb to invoke it.
This will allow us to set the user directory and autostart file from adb.
adb shell am start -n org.dolphinemu.dolphinemu/.gamelist.GameListActivity -e AutoStartFile /sdcard/AC.gcz -e UserDir /sdcard/dolphin-emu2/
This allows more automated testing to be done with Dolphin on Android.
A bit more efficient if we are only pushing two VFP registers.
We can probably be a bit more efficient in the future by mixing paired loadstores in to the other paths as well.
Previously on FPR pushing and popping we would do a single STR/LDR per quad FPR we wanted to push/pop.
In most of our cases when we are pushing and popping VFP registers they will be consecutive registers that will save more efficiently using the NEON
loadstores that can do up to four quad registers.
So this can potentially cutting instructions down to ~1/4th the amount of instructions if the registers are all consecutive.
On the Cortex-A57 this is basically just an icache improvement, but on the Nvidia Denver this may be optimized to be more efficient. Either way it's a
win.
Ryan Houdek