I think the AArch64 JIT has come far enough that it doesn't have to
be called experimental anymore.
I'm also labeling the x86-64 JIT as x86-64 for consistence with the
AArch64 JIT. This will especially be helpful if we start supporting
AArch64 on macOS, as AArch64 macOS can run both the x86-64 JIT and
the AArch64 JIT depending on whether you enable Rosetta 2.
I haven't observed this breaking any game, but it didn't match
the behavior of the interpreter as far as I could tell from
reading the code, in that denormals weren't being flushed.
If we can prove that FCVT will provide a correct conversion,
we can use FCVT. This makes the common case a bit faster
and the less likely cases (unfortunately including zero,
which FCVT actually can convert correctly) a bit slower.
Preparation for following commits.
This commit intentionally doesn't touch paired stores,
since paired stores are supposed to flush to zero.
(Consistent with Jit64.)
This simplifies some of the following commits. It does require
an extra register, but hey, we have 32 of them.
Something I think would be nice to add to the register cache
in the future is the ability to keep both the single and double
version of a guest register in two different host registers
when that is useful. That way, the extra register we write to
here can be read by a later instruction, saving us from
having to perform the same conversion again.
Fixes https://bugs.dolphin-emu.org/issues/12388. Might also fix
other games that have problems with float/paired instructions
in JitArm64, but I haven't tested any.
-They might have never drawn if DrawMessages wasn't called before they actually expired
-Their fade was wrong if the duration of the message was less than the fade time
This makes them much more useful for debugging, I know there might be other means
of debugging like logs and imgui, but this was the simplest so that's what I used.
If you want to print the same message every frame, but with a slightly different value
to see the changes, it now work.
To compensate for the fact that they are now always rendered once,
so on start up a lot of old messages (printed while the emulation was off) could show up,
I've added a "drop" time, which means if a msg isn't rendered for the first
time within that time, it will be dropped and never rendered.
When the interpreter writes to a discarded register, its type
must be changed so that it is no longer considered discarded.
Fixes a 62ce1c7 regression.
We normally check for division by zero to know if we should set the
destination register to zero with a XOR. However, when the divisor and
destination registers are the same the explicit zeroing can be omitted.
In addition, some of the surrounding branching can be simplified as
well.
Before:
45 85 FF test r15d,r15d
75 05 jne normal_path
45 33 FF xor r15d,r15d
EB 0C jmp done
normal_path:
B8 5A 00 00 00 mov eax,5Ah
99 cdq
41 F7 FF idiv eax,r15d
44 8B F8 mov r15d,eax
done:
After:
45 85 FF test r15d,r15d
74 0C je done
B8 5A 00 00 00 mov eax,5Ah
99 cdq
41 F7 FF idiv eax,r15d
44 8B F8 mov r15d,eax
done:
Division by a power of two can be slightly improved when the
destination and dividend registers are the same.
Before:
8B C6 mov eax,esi
85 C0 test eax,eax
8D 70 03 lea esi,[rax+3]
0F 49 F0 cmovns esi,eax
C1 FE 02 sar esi,2
After:
85 F6 test esi,esi
8D 46 03 lea eax,[rsi+3]
0F 48 F0 cmovs esi,eax
C1 FE 02 sar esi,2
Repeated erase() + iteration on a std::multimap is extremely slow.
Slow enough that it causes a 7 second long stutter during some
transitions in F-Zero X (a N64 VC game that triggers many, many icache
invalidations).
And slow enough that JitBaseBlockCache::DestroyBlock shows up on a
flame graph as taking >50% of total CPU time on the CPU-GPU thread:
https://i.imgur.com/vvqiFL6.png
This commit optimises those block link queries by replacing the
std::multimap (which is typically implemented with red-black trees)
with hash tables.
Master: https://i.imgur.com/vvqiFL6.png / 7s stutters
(starting from 5.0-2021 and with branch following disabled)
This commit: https://i.imgur.com/hAO74fy.png / ~0.7s stutters, which
is pretty close to 5.0 stable. (5.0-2021 introduced the performance
regression and it is especially noticeable when branch following
is disabled, which is the case for all N64 VC games since 5.0-8377.)
VideoCommon: Change the type of BPMemory.scissorOffset to 10bit signed: S32X10Y10
VideoBackends: Fix Software Clipper.PerspectiveDivide function, use BPMemory.scissorOffset instead of hard code 342
Oversight from #9545, which moved the "new game has been loaded" logic
to a separate OnNewTitleLoad function that has to be called explicitly
*after* a title has loaded.
Coupled with the commit that makes Dolphin not clobber 0x1800-0x3000
when using MIOS, this fixes Wind Waker and other MIOS-patched games
when they are launched from the System Menu.
MIOS puts patch data in low MEM1 (0x1800-0x3000) for its own use.
Overwriting data in this range can cause the IPL to crash when
launching games that get patched by MIOS.
See https://bugs.dolphin-emu.org/issues/11952 for more info.
Not applying the Gecko HLE patches means that Gecko codes will not work
under MIOS, but this is better than the alternative of having specific
games crash.
This particular range is kind of bizarre, and would only interpret
interleave mode 2 as a valid mode, while rejecting interleave mode 1 and
the extension byte mode.
As far as I know, based off the information on Wiibrew, we should be
considering all three values within this range as valid.
texture serialization and deserialization used to involve many memory
allocations and deallocations, along with many copies to and from
those allocations. avoid those by reserving a memory region inside the
output and writing there directly, skipping the allocation and copy to
an intermediate buffer entirely.
We don't actually need to do this until we bump targetSdkVersion
to Android 12 (which we can't do yet since we're not compatible
with scoped storage), but I figured I'd get it out of the way early.
Not tested on Android 12, but tested to not break stuff on Android 10.
JosJuice
