steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-13 09:00:30 +02:00
Code Issues Packages Projects Releases 67 Wiki Activity
9,533 Commits 11 Branches 71 Tags
c28824f94de6cb3e83becbbcd9473cfc9d2bfd22
Commit Graph
1283 Commits
Author SHA1 Message Date
Alyssa Rosenzweig 2aaf957ad8 RedundantFlagCalculationElimination: DCE as we go 
This is required to ensure single-iteration convergence with a sequence like:

  write C
  whatever = load C
  rmif C, whatever
  invalidate C

avoids regressing the "DEC dead" case with future work in the series.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Alyssa Rosenzweig d459b2f9b5 IR: propagate 0 into sub 
now that we have to handle it, we may as well take advantage of it.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Alyssa Rosenzweig 22ab7f2b3e IR: add SubWithFlags op (arm64 subs) 
with 8/16-bit handling to keep everything uniform.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Alyssa Rosenzweig f7e32373ce JIT: allow #0 in sub 
turns into neg, this will be generated via SubWithFlags -> Sub opts.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Alyssa Rosenzweig 25d422e92b JIT: use GetZeroableReg for NZCVSelect 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Alyssa Rosenzweig cfbeece09f JIT: use GetZeroableReg for CondAddNZCV 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Alyssa Rosenzweig a597a09825 JIT: use GetZeroableReg for SubNZCV 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Alyssa Rosenzweig 99854ff310 JIT: add GetZeroableReg helper 
for inlining constant zeroes in applicable sources

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-24 15:54:49 -04:00
Ryan Houdek 0a64f8a9c5 Moves SignalDelegator TLS tracking to the frontend 
FEXCore doesn't need track the TLS state of the SignalDelegator, this is
a frontend concept.

Removes the tracking from the backend and keeps it in the frontend.
 2024-02-24 01:07:29 -08:00
Ryan Houdek 59ec88f48d Merge pull request #3438 from Sonicadvance1/move_tls_allocation 
Moves JITSymbol allocation
 2024-02-23 14:49:04 -08:00
Ryan Houdek 6ec628fa31 Merge pull request #3433 from bylaws/arm64ec-pt1 
Arm64Emitter: Introduce ARM64EC SRA mappings
 2024-02-23 14:48:43 -08:00
Alyssa Rosenzweig 5378ae2e76 Merge pull request #3436 from alyssarosenzweig/ir/af-simplify 
Simplify CalculateAF
 2024-02-22 08:17:07 -04:00
Ryan Houdek bd4a81a2a1 Moves JITSymbol allocation 
This isn't actually using TLS allocations. Instead it is an allocation
tied to the InternalThreadState object.
 2024-02-21 17:57:34 -08:00
Ryan Houdek d4be2dc636 Merge pull request #3434 from bylaws/arm64ec-pt3 
FEXCore: Expose AbsoluteLoopTopAddress to the frontend
 2024-02-21 14:31:04 -08:00
Alyssa Rosenzweig 2bcd285851 Merge pull request #3430 from Sonicadvance1/tsc_scale 
Implement small TSC scaling
 2024-02-21 13:16:27 -04:00
Alyssa Rosenzweig 8762bc1fa3 OpcodeDispatcher: simplify CalculateAF signature 
- Res is unused
- SrcSize doesn't matter since we ignore the high bits, might as well always use
  32-bit, it doesn't matter

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-21 12:48:15 -04:00
Billy Laws 5b4162b712 FEXCore: Expose AbsoluteLoopTopAddress to the frontend 
ARM64EC has a shared SRA mapping between ARM64 and X64 code, so there
needs to be a public way to enter the dispatcher without refilling SRA
from the in-memory context struct.
 2024-02-21 11:46:24 +00:00
Billy Laws cb5c07f4b1 Arm64Emitter: Introduce ARM64EC SRA mappings 
See https://learn.microsoft.com/en-us/cpp/build/arm64ec-windows-abi-conventions?view=msvc-170
note that since mm registers are volatile there is no need to match the
mapping for them when in JIT, so they can be used as scratch regs.
Disallowed regs are also wiped on context switches, so they cannot be
taken advantage of to e.g. avoid spilling.
 2024-02-21 11:18:10 +00:00
Ryan Houdek b902b8edab Implement small TSC scaling 
Games engines are expecting >1Ghz cycle counters. Scale them to work
around the issue.

Resolves the excessive busy waiting in Unreal Engine 5 games.
 2024-02-20 12:05:44 -08:00
Alyssa Rosenzweig 0503c89ff6 OpcodeDispatcher: use NZCV update helpers 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-19 14:12:54 -04:00
Alyssa Rosenzweig 6dd410698a OpcodeDispatcher: add helpers for updating NZCV metadata 
to reduce error-prone copypaste

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-19 14:12:54 -04:00
Ryan Houdek 9cab746aa7 Merge pull request #3407 from neobrain/feature_libfwd_arguments_on_guest_stack 
Library Forwarding: Allocate packed arguments on the guest stack if needed
 2024-02-12 16:31:34 -08:00
Alyssa Rosenzweig 68232366e4 OpcodeDispatcher: don't mask add/sub sources 
not needed in the new approach

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-12 12:36:28 -04:00
Alyssa Rosenzweig d7ff1b78fb IR: handle 8/16-bit AddNZCV/SubNZCV 
we can do it more effectively than the current s/w lowering.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-12 12:36:09 -04:00
Ryan Houdek df3d6938ae Merge pull request #3410 from alyssarosenzweig/opt/nzcv-pass-2 
Add NZCV+PF/AF optimization pass
 2024-02-10 05:03:12 -08:00
Ryan Houdek ba41da7da0 Merge pull request #3414 from Sonicadvance1/fix_one_mutex_hang 
Fixes one mutex hang
 2024-02-09 05:54:40 -08:00
Ryan Houdek 2480bab409 Fixes one mutex hang 
When code invalidation is happening we currently have the issue that a
thread can acquire the code invalidation mutex in the middle of
invalidation. This is due to us acquiring and releasing the mutex
between each thread's code invalidation.

We need to hold the mutex for the entire duration for all thread's code
invalidation.
This fixes a rare hang on proton startup and resolves a consistent hang
on Proton application shutdown.

This now puts us on par with FEX-2312.1 with hanging.

This does not fix a relatively rare hang on fork (which also existed with FEX-2312.1).

This also does not fix the issue that the intersection of our mutexes
between frontend and backend are very convoluted. In part of the work
that is going to fix the rare fork mutex hang will change more of this.
 2024-02-08 18:18:00 -08:00
Alyssa Rosenzweig ad7202e7d7 OpcodeDispatcher: optimize test -1 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-08 14:10:13 -04:00
Alyssa Rosenzweig 175a57dd27 OpcodeDispatcher: emit AndWithFlags directly for primary alu 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 14:22:28 -04:00
Alyssa Rosenzweig e2ce60148c OpcodeDispatcher: emit AndWithFlags directly for 2ndary alu 
rely on opt pass to drop the flags.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 14:22:28 -04:00
Alyssa Rosenzweig 99660129f3 IR: implement AndWithFlags for 8/16-bit 
easier to deal with in the JIT

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 14:22:28 -04:00
Alyssa Rosenzweig 308d9a751c RedundantFlagCalculationElimination: optimize rmif 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 13:06:26 -04:00
Alyssa Rosenzweig 4bd28c0ed8 RedundantFlagCalculationElimination: optimize condaddnzcv 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 13:06:26 -04:00
Alyssa Rosenzweig 8397f3ac99 RedundantFlagCalculationElimination: refine AXFLAG 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 13:06:26 -04:00
Alyssa Rosenzweig 0452bc7212 RedundantFlagCalculationElimination: optimize condjump 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 13:06:26 -04:00
Alyssa Rosenzweig 3d7ed89ffb RedundantFlagCalculationElimination: optimize NZCVSelect 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 13:06:25 -04:00
Alyssa Rosenzweig 23ab0a978e RedundantFlagCalculationElimination: also handle InvalidateFlags 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 13:06:25 -04:00
Alyssa Rosenzweig 7f47a9ef0e IR: add local dead flag elimination pass 
RCLSE ignores NZCV and doesn't optimize stores which doesn't help us with PF/AF
either. So, we add a new pass for dead flag elimination (cannibalizing the old
and broken dead flag elimination pass). This is a simple local optimizer that
walks each block backwards, converging in linear time & constant space in a
single iteration.

Right now, it doesn't do a ton (other than a nice reduction in silliness in
the hot Sonic block), but it provides the framework to fuse comparisons.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-06 13:06:25 -04:00
Alyssa Rosenzweig 4331753ca0 Merge pull request #3408 from alyssarosenzweig/opt/tst 
Optimize TST
 2024-02-06 11:28:02 -04:00
Paulo Matos fa8bcfd67a Clean up access to possible nullptr 
Patch suggested by @Sonicadvance1
 2024-02-06 12:31:38 +00:00
Alyssa Rosenzweig 235f32ce8c Merge pull request #3401 from Sonicadvance1/runtime_preserve_all 
HostFeatures: Supports runtime disabling of preserve_all
 2024-02-05 15:34:46 -04:00
Alyssa Rosenzweig 2e0cb2fbd4 OpcodeDispatcher: optimize TST 
it's just an AndWithFlags setting the PF.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-05 15:32:21 -04:00
Alyssa Rosenzweig 4790a7ba79 IR: add AndWithFlags 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-05 15:32:21 -04:00
Tony Wasserka df3e51fc8c Library Forwarding: Allocate packed arguments on the guest stack if needed 
This is required for host-side calls to guest functions on 32-bit guests.
Since the host stack is allocated before FEX blocks memory inaccessible to
the guest, the guest would otherwise fail to read the packed argument data.
 2024-02-05 18:10:34 +01:00
Ryan Houdek c437129ed8 Revert "Revert "FEXLoader: Moves thread management to the frontend"" 
This reverts commit 5358af7794.
 2024-02-03 00:57:36 -08:00
Alyssa Rosenzweig 8d3f0b6f02 OpcodeDispatcher: reassociate and sink W in sha1 
We only need each part of W extracted in the corresponding round, so sink the
extract into the round to reduce pressure.

Further, W and E are added and then never used again. So, by reassociating we
can do the add upfront, killing W and E at the start and further reducing
pressure.

Eliminates spilling in sha1rnds4.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-02 13:03:07 -04:00
Alyssa Rosenzweig 60f7b9bcc4 OpcodeDispatcher: optimze sha1's 2/3 expr 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-02 13:03:07 -04:00
Alyssa Rosenzweig a487557173 OpcodeDispatcher: extract BitwiseAtLeastTwo 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-02 13:03:07 -04:00
Alyssa Rosenzweig 394b4888bb OpcodeDispatcher: reassociate and remat C0, G0 
costs 2 moves and eliminates the rest of our spilling

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-02 13:03:07 -04:00
Alyssa Rosenzweig 142cbdd852 OpcodeDispatcher: expand, reassociate, and interleave sha256 calc 
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
 2024-02-02 13:03:07 -04:00