Skips implementing it for the x86 JIT because that's a bit of a
nightmare to think about.
The ARM64 implementation requires sequential registers which means if
the incoming sources aren't sequential then we need to move the sources
in to the two vector temporaries. This is fine since we have zero-cycle
vector renames and the alternative is slower.
This is taking steps to start fixing RCLSE which was started by #2700.
Same situation as that PR, since #2170 when we converted
{Load,Store}Context in to {Load,Store}Register we broke this pass
entirely. It hasn't been doing anything for redundant GPRs and FPRs
since at least November of last year.
Technically it was potentially still optimizing redundant MMX
accesses, but it is so broken that it doesn't matter.
Instead of going all in like #2700 did, tear down the pass and start
again. We are now /only/ optimizing redundant context/register loads.
This fixes an issue that comes up commonly where the same register used
as sources was getting loaded twice, causing redundant moves.
`packsswb xmm0, xmm0` for example was generating a four instruction
sequence instead of three instructions because we weren't eliminating
the redundant load.
Going to take reimplementing all the optimizations that this pass does
in steps. This way we can track any regression in the independent steps
unlike what happened in #2700.
Confirmed that Proton/Sonic Mania still works after this.
It turns out that pure SSA isn't a great choice for the sort of emulation we do.
On one hand, it discards information from the guest binary's register allocation
that would let us skip stuff. On the other hand, it doesn't have nearly as many
benefits in this setting as in a traditional compiler... We really *don't* want
to do global RA or really any global optimization. We assume the guest optimizer
did its job for x86, we just need to clean up the mess left from going x86 ->
arm. So we just need enough SSA to peephole optimize.
My concrete IR proposals are that:
* SSA values must be killed in the same block that they are defined.
* Explicit LoadGPR/StoreGPR instructions can be used for global persistence.
* LoadGPR/StoreGPR are eliminated in favour of SSA within a block.
This has a lot of nice properties for our setting:
* Except for some internal REP instruction emulation (etc), we already have
registers for everything that escapes block boundaries, so this form is very
easy to go into -- straightforward local value numbering, not a full into
SSA pass.
* Spilling is entirely local (if it happens at all), since everything is in
registers at block boundaries. This is excellent, because Belady's algorithm
lets us spill nearly optimally in linear-time for individual blocks. (And
the global version of Belady's algorithm is massively more complicated...)
A nice fit for a JIT.
Relatedly, it turns out allowing spilling is probably a decent decision,
since the same spiller code can be used to rematerialize constants in a
straightforward way. This is an issue with the current RA.
* Register assignment is entirely local. For the same reason, we can assign
registers "optimally" in linear time & memory (e.g. with linear scan). And
the impl is massively simpler than a full blown SSA-based tree scan RA. For
example, we don't have to worry about parallel copies or coalescing phis or
anything. Massively nicer algorithm to deal with.
* SSA value names can be block local which makes the validation implicit :~)
It also has remarkably few drawbacks, because we didn't want to do CFG global
optimization anyway given our time budget and the diminishng returns. The few
global optimizations we might want (flag escape analysis?) don't necessarily
benefit from pure SSA anyway.
Anyway, we explicitly don't want phi nodes in any of this. They're currently
unused. Let's just remove them so nobody gets the bright idea of changing that.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Now that PF calculation is deferred, the cost of calculating PF correctly should
be tolerable. Remove the speed hack to skip PF. It's fundamentally broken, and
there are enough broken things in FEX as it is that we don't need to maintain
this one ;-)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
FEX has a problem with large blocks that uses a ton of constants spread
throughout the block. Once a block gets large enough with enough
constants that have large live ranges, FEX slows down to unusable speeds
due to the register allocator spending more time calculating node
interferences than anything else in the program.
This adds a little heuristic to ensure that constants aren't reused if
the previous value is past a certain distance threshold. This threshold
works well enough that XeSS's pedantic initialization code doesn't have
issues now. See https://github.com/FEX-Emu/FEX/issues/2688 for more
information about that.
FEX itself should work to remove bad constant usages to make this pass
less necessary anyway. In most cases we are materializing duplicated 0,
1, and masks which could be done without a constant entirely.
Maybe once we've improve that enough we could remove this constant
pooling entirely.
To note, this doesn't fix the issue that XeSS causes our register
allocator, this is purely a heuristic workaround.
With Or, Orlshl, Bfe, and Bfi there were some assumptions made that i8
and i16 operations made sense. Which required us to disable the IR
validation for these operations when it was just added.
This removes the final assumptions about these IR operations supporting
these small operating sizes allowing us to enable the IR validation.
Also a very minor optimization by moving a couple extracts from source
before trying to BFI from it, making RA more optimal.
The FXSAVE and FSAVE tag words are written out in different formats,
with FXSAVE using an abridged version that lacks the zero/special/valid
distinction. Switch to using this abridged version internally for
simplicity, and to allow the calculation of zero/special/valid
distinction to be deferred until an fxsave instruction (in the future,
currently the distinction is ignored and only valid/empty states are
possible).
AddNZCV is a new op to return the NZCV for an addition directly, which lets us
skip software flag calculation in some cases. In the future it would be nice to
fuse this into the Add itself as a second destination to avoid repeating the
addition, but that's a very involved change and right now I'm building FEX on an
old Chromebook because my M1 kernel is FUBAR.
Similarly, SubNZCV returns flags for Sub. This has the extra twist of needing to
invert the carry bit due to the inverted definition between arm64 and x86_64.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
A bunch of the AES operations take a zero register upfront and we
currently materialize it for each instruction.
Considering that most AES operations are used back to back, we can
eliminate these materializations by caching it between instructions.
Additionally removes a move in the optimal case when destination matches
the state register, which is exactly what the SSE operation ends up
doing.
AESKeyGenAssist has an edge case that if the destination RA overlaps the
zero register then we still need to eat a move, hopefully doesn't happen
too frequently in practice. This is also the lesser used instruction so
it isn't a big deal. RA constraints could solve that still.
We can load the swizzle table from our constant pool now. This removes
the only usage of VTMP3 from our Arm64 JIT.
I would say the this is now optimal for the version without RCON set.
With RCON we could technically make some of the move of the constant
more optimal.
The range check and clamping is necessary in the cases of passing x86
shift amounts directly through VUSHL/VSSHR.
Some AVX operations are still using these with range clamping. A future
investigation task should be the check if they can be switched over to
the wide variants that we implemented for the SSE instructions.
When consuming our own controlled data, we don't want the range clamping
to be enabled.
The number of times the implicit size calculation in GPR operations has
bit us is immeasurable and was a mistake from the start of the project.
The vector based operations never had this problem since they were
explicitly sized for a long time now.
This converts the base IR operations to be explicitly sized, but adds
implicit sized helpers for the moment while we work on removing implicit
usage from the OpcodeDispatcher.
Should be NFC at this moment but it is a big enough change that I want
it in before the "real" work starts.
Noticed that we hadn't ever enabled this, which was a concern when our
GPR operations weren't as strict about leaving garbage in the upper bits
when operating as a 32-bit operation.
Now that our ALU operations are more strict about enforcing upper bit
zeroing we can enable this.
This causes Half-Life: Source FPS to get to > 200FPS finally. Causes
significant performance improvements for 32-bit games because we're no
longer redundantly moving registers before and after every operation.
Causing a bunch of 3-4 instruction sequences to convert to 1.
RAValidation was making an assumption that GPR register class would only
have up to 16 registers for either SRA or dynamic registers.
When running a 32-bit application we allow 17 GPRs to be dynamically
allocated, since we can take 8 back from SRA in that case.
Just split the two classes in the RAValidation pass since they will
never overlap their allocation.
Fixes validation in `32Bit_Secondary/15_XX_0.asm` locally that changed
behaviour due to tinkering.
Use a named constant for loading the sign inversion, then EOR the second
source and just FAdd it all.
In a vacuum it isn't a significant improvement, but as soon as more than
one instruction is in a block it will eventually get optimized with
named constant caching and be a significant win.
Thanks to @rygorous for the idea!
This takes the two independent VSXT{U}N{2,} operations and merges them
in to a single IR operations.
In some cases this can result in a more optimal implementation since
there is no need for moves inbetween.
VRev32 matches Arm64 semantics directly.
LoadNamedVectorConstant allows FEX to quickly load "named constants".
This will allow us to have specific hardcoded vector constant values
that we can load with a ldr(State)+ldr(Value) and will be more abused in
the future.
This also allows us to do a very simple optimization in the future where
we can optimize away redundant loads of these loads if they are used
multiple times in the same block. (Not implemented here).
This matches x86 vector shift behaviour closely for ps{rl,ra,ll}{w,d,q}
where the vector is shifted by a scalar value that is 64-bits wide.
Anything larger than the element size will set that element to zero.
With SVE we have some new wide element shifts that match this behaviour
exactly (except supports wide shift sources rather than scalar).
This is a significant improvement even on platforms that only support
128-bit SVE.
This is a bit of tricky operation where due to our our usage of SSA, the
incoming source isn't guaranteed to end its live-range at this
instruction.
This gives us a behaviour where to be optimal we need to take different
paths depending on if the incoming address register is the same as the
destination node.
Once we have form of RA constraints or non-SSA IR form that can
guarantee this restriction then this will go away.
Changes the idiom used for constant mask generation to a ternary.
This pattern is definitely used elsewhere in code but we can get rid of
all instances here.
Due to Intel dropping support for legacy segment registers[1] there is a
concern that this will break legacy 32-bit software that is doing some
magic segment register handling.
Adds some simple telemetry for 32-bit applications that when they
encounter an instruction that sets the segment register or uses a
segment register that the JIT will do a /relatively/ quick four
instruction check to see if it is not a null segment.
It's not enough to just check if the segment index is 0 or not, 32-bit
Linux software starts with non-zero segment register indexes but the LDT
for each segment index is a null-descriptor.
Once the segment address is loaded, the IR operation will do a quick
check against zero and if it /isn't/ zero then set the telemetry value.
A very minor optimization that segment registers only get checked once
per block to ensure overhead stays low.
[1] https://www.intel.com/content/www/us/en/developer/articles/technical/envisioning-future-simplified-architecture.html
- 3.6 - Restricted Subset of Segmentation
- `Bases are supported for FS, GS, GDT, IDT, LDT, and TSS
registers; the base for CS, DS, ES, and SS is ignored for 32-bit
mode, same as 64-bit mode (treated as zero).`
- 4.2.17 - MOV to Segment Register
- Will fault if SS is written (Breaking anything that writes to
SS).
- Will not fault if CS, DS, ES are written (Thus it sets the
segment but gets ignored due to 3.6).
It is not an external component, and it makes paths needlessly long.
Ryan seemed amenable to this when we discussed on IRC earlier.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Ryan Houdek