Commit graph

56 commits

Author SHA1 Message Date
gdkchan
c8bb3cc50e
Fix register read after write on STREX implementation (#1801)
* Fix register read after write on STREX implementation

* PTC version update
2020-12-13 12:19:38 -03:00
sharmander
36f6bbf5b9
CPU: Implement VFNMA.F32 | F.64 (#1783)
* Implement VFNMA.F<32/64>

* Update PTC Version

* Update Implementation & Renames & Correct Order

* Fix alignment

* Update implementation to not trigger assert

* Actually use the intrinsic that makes sense :)
2020-12-07 21:04:01 -03:00
LDj3SNuD
567ea726e1
Add support for guest Fz (Fpcr) mode through host Ftz and Daz (Mxcsr) modes (fast paths). (#1630)
* Add support for guest Fz (Fpcr) mode through host Ftz and Daz (Mxcsr) modes (fast paths).

* Ptc.InternalVersion = 1630

* Nits.

* Address comments.

* Update Ptc.cs

* Address comment.
2020-12-07 10:37:07 +01:00
sharmander
b479a43939
CPU: Implement VFNMS.F32/64 (#1758)
* Add necessary methods / op-code

* Enable Support for FMA Instruction Set

* Add Intrinsics / Assembly Opcodes for VFMSUB231XX.

* Add X86 Instructions for VFMSUB231XX

* Implement VFNMS

* Implement VFNMS Tests

* Add special cases for FMA instructions.

* Update PPTC Version

* Remove unused Op

* Move Check into Assert / Cleanup

* Rename and cleanup

* Whitespace

* Whitespace / Rename

* Re-sort

* Address final requests

* Implement VFMA.F64

* Simplify switch

* Simplify FMA Instructions into their own IntrinsicType.

* Remove whitespace

* Fix indentation

* Change tests for Vfnms -- disable inf / nan

* Move args up, not description ;)

* Undo vfma

* Completely remove vfms code.,

* Fix order of instruction in assembler
2020-12-03 20:20:02 +01:00
riperiperi
9852cb9c9e
Use backup when PTC compression is corrupt (#1704)
* Use backup when PTC compression is corrupt

* Apply suggestions from code review

Co-authored-by: LDj3SNuD <35856442+LDj3SNuD@users.noreply.github.com>

Co-authored-by: LDj3SNuD <35856442+LDj3SNuD@users.noreply.github.com>
2020-11-20 02:51:59 +01:00
LDj3SNuD
0679084f11
CPU (A64): Add FP16/FP32 fast paths (F16C Intrinsics) for Fcvt_S, Fcvtl_V & Fcvtn_V Instructions. Now HardwareCapabilities uses CpuId. (#1650)
* net5.0

* CPU (A64): Add FP16/FP32 fast paths (F16C Intrinsics) for Fcvt_S, Fcvtl_V & Fcvtn_V Instructions. Switch to .NET 5.0.

Nits.

Tests performed successfully in both debug and release mode (for all instructions involved).

* Address comment.

* Update appveyor.yml

* Revert "Update appveyor.yml"

This reverts commit 27cdd59e8b.

* Remove Assembler CpuId.

* Update appveyor.yml

* Address comment.
2020-11-18 19:35:54 +01:00
Mary
863edae328
shader cache: Fix Linux boot issues (#1709)
* shader cache: Fix Linux boot issues

This rollback the init logic back to previous state, and replicate the
way PTC handle initialization.

* shader cache: set default state of ready for translation event to false

* Fix cpu unit tests
2020-11-17 22:40:19 +01:00
riperiperi
b4d8d893a4
Memory Read/Write Tracking using Region Handles (#1272)
* WIP Range Tracking

- Texture invalidation seems to have large problems
- Buffer/Pool invalidation may have problems
- Mirror memory tracking puts an additional `add` in compiled code, we likely just want to make HLE access slower if this is the final solution.
- Native project is in the messiest possible location.
- [HACK] JIT memory access always uses native "fast" path
- [HACK] Trying some things with texture invalidation and views.

It works :)

Still a few hacks, messy things, slow things

More work in progress stuff (also move to memory project)

Quite a bit faster now.
- Unmapping GPU VA and CPU VA will now correctly update write tracking regions, and invalidate textures for the former.
- The Virtual range list is now non-overlapping like the physical one.
- Fixed some bugs where regions could leak.
- Introduced a weird bug that I still need to track down (consistent invalid buffer in MK8 ribbon road)

Move some stuff.

I think we'll eventually just put the dll and so for this in a nuget package.

Fix rebase.

[WIP] MultiRegionHandle variable size ranges

- Avoid reprotecting regions that change often (needs some tweaking)
- There's still a bug in buffers, somehow.
- Might want different api for minimum granularity

Fix rebase issue

Commit everything needed for software only tracking.

Remove native components.

Remove more native stuff.

Cleanup

Use a separate window for the background context, update opentk. (fixes linux)

Some experimental changes

Should get things working up to scratch - still need to try some things with flush/modification and res scale.

Include address with the region action.

Initial work to make range tracking work

Still a ton of bugs

Fix some issues with the new stuff.

* Fix texture flush instability

There's still some weird behaviour, but it's much improved without this. (textures with cpu modified data were flushing over it)

* Find the destination texture for Buffer->Texture full copy

Greatly improves performance for nvdec videos (with range tracking)

* Further improve texture tracking

* Disable Memory Tracking for view parents

This is a temporary approach to better match behaviour on master (where invalidations would be soaked up by views, rather than trigger twice)

The assumption is that when views are created to a texture, they will cover all of its data anyways. Of course, this can easily be improved in future.

* Introduce some tracking tests.

WIP

* Complete base tests.

* Add more tests for multiregion, fix existing test.

* Cleanup Part 1

* Remove unnecessary code from memory tracking

* Fix some inconsistencies with 3D texture rule.

* Add dispose tests.

* Use a background thread for the background context.

Rather than setting and unsetting a context as current, doing the work on a dedicated thread with signals seems to be a bit faster.

Also nerf the multithreading test a bit.

* Copy to texture with matching alignment

This extends the copy to work for some videos with unusual size, such as tutorial videos in SMO. It will only occur if the destination texture already exists at XCount size.

* Track reads for buffer copies. Synchronize new buffers before copying overlaps.

* Remove old texture flushing mechanisms.

Range tracking all the way, baby.

* Wake the background thread when disposing.

Avoids a deadlock when games are closed.

* Address Feedback 1

* Separate TextureCopy instance for background thread

Also `BackgroundContextWorker.InBackground` for a more sensible idenfifier for if we're in a background thread.

* Add missing XML docs.

* Address Feedback

* Maybe I should start drinking coffee.

* Some more feedback.

* Remove flush warning, Refocus window after making background context
2020-10-16 17:18:35 -03:00
LDj3SNuD
04e330cc00
Add Umaal & Vabd_I, Vabdl_I, Vaddl_I, Vhadd, Vqshrn, Vshll inst.s (slow paths). (#1577)
* Add Umaal & Vabd_I, Vabdl_I, Vaddl_I, Vhadd, Vqshrn, Vshll inst.s (slow paths).

No test provided (i.e. draft).

* Ptc InternalVersion = 1577
2020-10-13 22:41:33 +02:00
gdkchan
f2b12c9749
Remove old, unused CPU optimization (#1586) 2020-09-30 16:16:34 -03:00
gdkchan
6c9565693f
IPC refactor part 1: Use explicit separate threads to process requests (#1447)
* Changes to allow explicit management of service threads

* Remove now unused code

* Remove ThreadCounter, its no longer needed

* Allow and use separate server per service, also fix exit issues

* New policy change: PTC version now uses PR number
2020-09-22 14:50:40 +10:00
LDj3SNuD
66b799a6e4
Fix host stack overflow caused by some recursive guest methods. (#1528)
* Fix host stack overflow caused by some recursive guest methods.

* PPTC flag up.

* Address comments.

Co-authored-by: gdkchan <gab.dark.100@gmail.com>
2020-09-19 20:16:30 -03:00
FICTURE7
f60033e0aa
Implement block placement (#1549)
* Implement block placement

Implement a simple pass which re-orders cold blocks at the end of the
list of blocks in the CFG.

* Set PPTC version

* Use Array.Resize

Address gdkchan's feedback
2020-09-19 20:00:24 -03:00
FICTURE7
36ec1bc6c0
Relax block ordering constraints (#1535)
* Relax block ordering constraints

Before `block.Next` had to follow `block.ListNext`, now it does not.
Instead `CodeGenerator` will now emit the necessary jump instructions
to ensure control flow.

This makes control flow and block order modifications easier. It also
eliminates some simple cases of redundant branches.

* Set PPTC version
2020-09-12 12:32:53 -03:00
FICTURE7
4c7bebf3e6
Do not emit StoreToContext before Return (#1537)
* Do not emit StoreToContext before Return

* Set PPTC version
2020-09-07 12:52:17 +10:00
gdkchan
6cc187da59
SIMD&FP load/store with scale > 4 should be undefined (#1522)
* SIMD&FP load/store with scale > 4 should be undefined

* Catch more invalid encodings for FP&SIMD LDR/STR (reg variant)

* Set PTC version to PR number
2020-09-01 17:02:23 -03:00
FICTURE7
92f7f163ef
Improve static branch prediction along fast path for memory accesses (#1484)
* Improve static branch prediction along fast path for memory accesses

* Set PPTC interval version
2020-08-31 20:55:15 -03:00
mageven
b9398f1f3a
Allow launching with custom data directories (#1505)
* Allow launching with custom data directories

Don't load alternate keys when using custom directory

* Address gdkchan's comments

* Misc fixes to log levels

Added more enabled log levels by default

Moved successful config updation to Notice as
1. It's not a warning
2. Warnings could've been disabled by the config load and hence message
   would be lost
2020-08-30 18:51:53 +02:00
LDj3SNuD
6938988427
Fix Vcvt_FI & Vcvt_RM; Add Vfma_S & Vfms_S. Add Tests. (#1471)
* Fix Vcvt_FI & Vcvt_RM; Add Vfma_S & Vfms_S. Add Tests.

* Address PR feedback & Nit.
2020-08-13 02:34:02 -03:00
mageven
9c6a3aacc2
Fix PTC version increment from #1433 (#1462) 2020-08-09 05:32:00 +02:00
LDj3SNuD
e36e97c64d
CPU: This PR fixes Fpscr, among other things. (#1433)
* CPU: This PR fixes Fpscr, among other things.

* Add Fpscr.Qc = 1 if sat. for Vqrshrn & Vqrshrun.

* Fix Vcmp & Vcmpe opcode table.

* Revert "Fix Vcmp & Vcmpe opcode table."

This reverts commit c117d9410d.

* Address PR feedbacks.
2020-08-08 17:18:51 +02:00
Ficture Seven
ee22517d92
Improve branch operations (#1442)
* Add Compare instruction

* Add BranchIf instruction

* Use test when BranchIf & Compare against 0

* Propagate Compare into BranchIfTrue/False use

- Propagate Compare operations into their BranchIfTrue/False use and
  turn these into a BranchIf.

- Clean up Comparison enum.

* Replace BranchIfTrue/False with BranchIf

* Use BranchIf in EmitPtPointerLoad

- Using BranchIf early instead of BranchIfTrue/False improves LCQ and
  reduces the amount of work needed by the Optimizer.

  EmitPtPointerLoader was a/the big producer of BranchIfTrue/False.

- Fix asserts firing when assembling BitwiseAnd because of type
  mismatch in EmitStoreExclusive. This is harmless and should not
  cause any diffs.

* Increment PPTC interval version

* Improve IRDumper for BranchIf & Compare

* Use BranchIf in EmitNativeCall

* Clean up

* Do not emit test when immediately preceded by and
2020-08-05 08:52:33 +10:00
mageven
a33dc2f491
Improved Logger (#1292)
* Logger class changes only

Now compile-time checking is possible with the help of Nullable Value
types.

* Misc formatting

* Manual optimizations

PrintGuestLog
PrintGuestStackTrace
Surfaceflinger DequeueBuffer

* Reduce SendVibrationXX log level to Debug

* Add Notice log level

This level is always enabled and used to print system info, etc...
Also, rewrite LogColor to switch expression as colors are static

* Unify unhandled exception event handlers

* Print enabled LogLevels during init

* Re-add App Exit disposes in proper order

nit: switch case spacing

* Revert PrintGuestStackTrace to Info logs due to #1407

PrintGuestStackTrace is now called in some critical error handlers
so revert to old behavior as KThread isn't part of Guest.

* Batch replace Logger statements
2020-08-04 01:32:53 +02:00
gdkchan
9878fc2d3c
Implement inline memory load/store exclusive and ordered (#1413)
* Implement inline memory load/store exclusive

* Fix missing REX prefix on 8-bits CMPXCHG

* Increment PTC version due to bugfix

* Remove redundant memory checks

* Address PR feedback

* Increment PPTC version
2020-07-30 11:29:28 -03:00
Ficture Seven
b3c051bbec
Use movd,movq for i32/64 VectorExtract %x, 0x0 (#1439)
* Use movd,movq for i32/64 VectorExtract %x, 0x0

* Increment PPTC interval version

* Use else-if instead

- Address gdkchan's feedback.
- Clean up Debug.Assert calls

* Inline `count` expression into Debug.Assert

Apparently the CoreCLR JIT will not eliminate this. :(
2020-07-30 15:52:26 +10:00
gdkchan
2678bf0010
PPTC version increment (#1427) 2020-07-25 20:01:57 +02:00
Valentin PONS
3af2ce74ec
Implements some 32-bit instructions (VBIC, VTST, VSRA) (#1192)
* Added some 32 bits instructions:

* VBIC
* VTST
* VSRA

* Incremented the PTC

* Add tests and fix implementation

* Fixed VBIC immediate opcode mapping

* Hey hey!

* Nit.

Co-authored-by: gdkchan <gab.dark.100@gmail.com>
Co-authored-by: LDj3SNuD <dvitiello@gmail.com>
Co-authored-by: LDj3SNuD <35856442+LDj3SNuD@users.noreply.github.com>
2020-07-19 15:11:58 -03:00
LDj3SNuD
56a61a5758
CPU: A32: Fix Vabs_V & Vneg_V (S8, S16, S32 & F32); add Tests. (#1394)
* Fix Vabs_V & Vneg_V (S8, S16, S32 & F32); add Tests.

* Update Ptc.cs
2020-07-17 10:57:49 -03:00
LDj3SNuD
88619d71b8
CPU: A32: Add Vadd & Vsub Wide (S/U_8/16/32) Inst.s with Test. (#1390) 2020-07-17 14:21:40 +10:00
LDj3SNuD
a804db6eed
Add Fmax/minv_V & S/Ushl_S Inst.s with Tests. Fix Maxps/d & Minps/d d… (#1335)
* Add Fmax/minv_V & S/Ushl_S Inst.s with Tests. Fix Maxps/d & Minps/d double zero sign handling. Allows better handling of NaNs.

* Optimized EmitSse2VectorIsNaNOpF() for multiple uses per opF.
2020-07-13 21:08:47 +10:00
riperiperi
d7044b10a2
Add SSE4.2 Path for CRC32, add A32 variant, add tests for non-castagnoli variants. (#1328)
* Add CRC32 A32 instructions.

* Fix CRC32 instructions.

* Add CRC intrinsic and fast path.

Loop is currently unrolled, will look into adding temp vars after tests are added.

* Begin work on Crc tests

* Fix SSE4.2 path for CRC32C, finialize tests.

* Remove unused IR path.

* Fix spacing between prefix checks.

* This should be Src.

* PTC Version

* OpCodeTable Order

* Integer check improvement. Value and Crc can be either 32 or 64 size.

* This wasn't necessary...

* If size is 3, value type must be I64.

* Fix same src+dest handling for non crc intrinsics.

* Pre-fix (ha) issue with vex encodings
2020-07-13 20:48:14 +10:00
Ficture Seven
30d4f752f4
Fix Node Uses/Assignments (#1376)
* Fix Node Uses/Assignments

* Bump PPTC Version Number

Co-authored-by: jduncanator <1518948+jduncanator@users.noreply.github.com>
2020-07-13 20:20:07 +10:00
Ficture Seven
b5127131c5
Fix folding of ConvertI64ToI32 imm64 (#1383)
* Fix folding of ConvertI64ToI32 imm64

* Increment PTC internal version

* Clean up
2020-07-13 19:10:33 +10:00
gdkchan
38b26cf424
Mask shift constants on x86 backend (#1382)
* Mask shift constants on x86 backendd

* Version bump
2020-07-11 15:52:38 +10:00
Ficture Seven
0195d0938d
Fold ZeroExtend8/16/32 imm32/64 (#1358)
* Fold ZeroExtend8/16/32 imm32/64

* Increment PTC version
2020-07-11 11:00:41 +10:00
Ficture Seven
7639bb2e86
Fold ConvertI64ToI32 imm64 (#1359)
* Fold ConvertI64ToI32 imm64

* Increment PTC version

* Bump PPTC InternalVersion

Co-authored-by: jduncanator <1518948+jduncanator@users.noreply.github.com>
2020-07-11 10:35:38 +10:00
LDj3SNuD
c050994995
Fix PPTC on Windows 7. (#1369)
* Fix PPTC on Windows 7.

* Address gdkchan comment.
2020-07-09 10:45:24 +10:00
riperiperi
9a49f8aec9
Fix VMVN (immediate), Add VPMIN, VPMAX, VMVN (register) (#1303)
* Add Vmvn (register), tests for both Vmvn variants.

* Add Vpmin, Vpmax, improve Non-FastFp accuracy for Vpadd

* Rebase on top of PTC.

* Add Nopcode

* Increment PTC version.

* Fix nits.
2020-06-24 10:43:44 +10:00
Thog
05c63f3061
Increment PTC version (#1311)
Fix issues caused by 2421186d97
2020-06-18 13:41:00 +02:00
Ficture Seven
2421186d97
Generalize tail continues (#1298)
* Generalize tail continues

* Fix DecodeBasicBlock

`Next` and `Branch` would be null, which is not the state expected by
the branch instructions. They end up branching or falling into a block
which is never populated by the `Translator`. This causes an assert to
be fired when building the CFG.

* Clean up Decode overloads

* Do not synchronize when branching into exit block

If we're branching into an exit block, that exit block will tail
continue into another translation which already has a synchronization.

* Remove A32 predicate tail continue

If `block` is not an exit block then the `block.Next` must exist (as
per the last instruction of `block`).

* Throw if decoded 0 blocks

Address gdkchan's feedback

* Rebuild block list instead of setting to null

Address gdkchan's feedback
2020-06-18 13:37:21 +10:00
LDj3SNuD
5e724cf24e
Add Profiled Persistent Translation Cache. (#769)
* Delete DelegateTypes.cs

* Delete DelegateCache.cs

* Add files via upload

* Update Horizon.cs

* Update Program.cs

* Update MainWindow.cs

* Update Aot.cs

* Update RelocEntry.cs

* Update Translator.cs

* Update MemoryManager.cs

* Update InstEmitMemoryHelper.cs

* Update Delegates.cs

* Nit.

* Nit.

* Nit.

* 10 fewer MSIL bytes for us

* Add comment. Nits.

* Update Translator.cs

* Update Aot.cs

* Nits.

* Opt..

* Opt..

* Opt..

* Opt..

* Allow to change compression level.

* Update MemoryManager.cs

* Update Translator.cs

* Manage corner cases during the save phase. Nits.

* Update Aot.cs

* Translator response tweak for Aot disabled. Nit.

* Nit.

* Nits.

* Create DelegateHelpers.cs

* Update Delegates.cs

* Nit.

* Nit.

* Nits.

* Fix due to #784.

* Fixes due to #757 & #841.

* Fix due to #846.

* Fix due to #847.

* Use MethodInfo for managed method calls.

Use IR methods instead of managed methods about Max/Min (S/U).
Follow-ups & Nits.

* Add missing exception messages.

Reintroduce slow path for Fmov_Vi.
Implement slow path for Fmov_Si.

* Switch to the new folder structure.

Nits.

* Impl. index-based relocation information. Impl. cache file version field.

* Nit.

* Address gdkchan comments.

Mainly:
- fixed cache file corruption issue on exit; - exposed a way to disable AOT on the GUI.

* Address AcK77 comment.

* Address Thealexbarney, jduncanator & emmauss comments.

Header magic, CpuId (FI) & Aot -> Ptc.

* Adaptation to the new application reloading system.

Improvements to the call system of managed methods.
Follow-ups.
Nits.

* Get the same boot times as on master when PTC is disabled.

* Profiled Aot.

* A32 support (#897).

* #975 support (1 of 2).

* #975 support (2 of 2).

* Rebase fix & nits.

* Some fixes and nits (still one bug left).

* One fix & nits.

* Tests fix (by gdk) & nits.

* Support translations not only in high quality and rejit.

Nits.

* Added possibility to skip translations and continue execution, using `ESC` key.

* Update SettingsWindow.cs

* Update GLRenderer.cs

* Update Ptc.cs

* Disabled Profiled PTC by default as requested in the past by gdk.

* Fix rejit bug. Increased number of parallel translations. Add stack unwinding stuffs support (1 of 2).

Nits.

* Add stack unwinding stuffs support (2 of 2). Tuned number of parallel translations.

* Restored the ability to assemble jumps with 8-bit offset when Profiled PTC is disabled or during profiling.

Modifications due to rebase.
Nits.

* Limited profiling of the functions to be translated to the addresses belonging to the range of static objects only.

* Nits.

* Nits.

* Update Delegates.cs

* Nit.

* Update InstEmitSimdArithmetic.cs

* Address riperiperi comments.

* Fixed the issue of unjustifiably longer boot times at the second boot than at the first boot, measured at the same time or reference point and with the same number of translated functions.

* Implemented a simple redundant load/save mechanism.

Halved the value of Decoder.MaxInstsPerFunction more appropriate for the current performance of the Translator.
Replaced by Logger.PrintError to Logger.PrintDebug in TexturePool.cs about the supposed invalid texture format to avoid the spawn of the log.
Nits.

* Nit.

Improved Logger.PrintError in TexturePool.cs to avoid log spawn.
Added missing code for FZ handling (in output) for fp max/min instructions (slow paths).

* Add configuration migration for PTC

Co-authored-by: Thog <me@thog.eu>
2020-06-16 20:28:02 +02:00
LDj3SNuD
3b70a28087
Unwinding Follow-up. Fix a bug in JitUnwindWindows where ... (#1238)
... in case of "Vector" unwind codes the remaining unwind codes could be corrupted.
Nits.
2020-05-15 13:46:35 +02:00
gdkchan
96c7988671
Remove CpuId IR instruction (#1227) 2020-05-13 15:30:21 +10:00
Ficture Seven
4b79b78b8c
Fix tailcall case in EmitterContext (#1235) 2020-05-11 10:04:52 +10:00
Ficture Seven
180ad8605d
Improve IRDumper (#1135)
* Improve IRDumper

* Make Symbols.Add(ulong, ulong, ulong, string) thread safe

* Use a StringBuilder for MemoryOperand

* Add #if M_DEBUG guards

* Fix JMP_TABLE typo

* Fix using in Symbols

* Use Conditional("M_DEBUG") instead

Address gdkchan's feedback

* Use a struct instead of 4-tuple

Address gdkchan's feedback

* Place symbols in comments instead

Address gdkchan's feedback

* Use StringBuilder throughout

* Handle offsetted symbols

* Fix naming convention of Builder

* Avoid ArgumentException

* Remove unnecessary using

* Use switch expression instead

* Turn into a class

* Clean up

* Remove unnecessary using
2020-05-04 12:06:22 +10:00
gdkchan
f77694e4f7
Implement a new physical memory manager and replace DeviceMemory (#856)
* Implement a new physical memory manager and replace DeviceMemory

* Proper generic constraints

* Fix debug build

* Add memory tests

* New CPU memory manager and general code cleanup

* Remove host memory management from CPU project, use Ryujinx.Memory instead

* Fix tests

* Document exceptions on MemoryBlock

* Fix leak on unix memory allocation

* Proper disposal of some objects on tests

* Fix JitCache not being set as initialized

* GetRef without checks for 8-bits and 16-bits CAS

* Add MemoryBlock destructor

* Throw in separate method to improve codegen

* Address PR feedback

* QueryModified improvements

* Fix memory write tracking not marking all pages as modified in some cases

* Simplify MarkRegionAsModified

* Remove XML doc for ghost param

* Add back optimization to avoid useless buffer updates

* Add Ryujinx.Cpu project, move MemoryManager there and remove MemoryBlockWrapper

* Some nits

* Do not perform address translation when size is 0

* Address PR feedback and format NativeInterface class

* Remove ghost parameter description

* Update Ryujinx.Cpu to .NET Core 3.1

* Address PR feedback

* Fix build

* Return a well defined value for GetPhysicalAddress with invalid VA, and do not return unmapped ranges as modified

* Typo
2020-05-04 08:54:50 +10:00
Ficture Seven
71dbb38b9a
Do not compute dominance information when not in SSA (#1176) 2020-05-01 23:15:55 +02:00
Chenj168
561d64e5bf
Modify TranslatedFunction.GetPointer () to optimize performance (#995)
* add local var to reduce calling Marshal.GetFunctionPointerForDelegate

* modify code style
2020-03-20 09:11:20 +11:00
riperiperi
8226997bc7
CodeGen Optimisations (LSRA and Translator) (#978)
* Start of JIT garbage collection improvements

- thread static pool for Operand, MemoryOperand, Operation
- Operands and Operations are always to be constructed via their static
helper classes, so they can be pooled.
- removing LinkedList from Node for sources/destinations (replaced with
List<>s for now, but probably could do arrays since size is bounded)
- removing params constructors from Node
- LinkedList<> to List<> with Clear() for Operand assignments/uses
- ThreadStaticPool is very simple and basically just exists for the
purpose of our specific translation allocation problem. Right now it
will stay at the worst case allocation count for that thread (so far) -
the pool can never shrink.

- Still some cases of Operand[] that haven't been removed yet. Will need
to evaluate them (eg. is there a reasonable max number of params for
Calls?)

* ConcurrentStack instead of ConcurrentQueue for Rejit

* Optimize some parts of LSRA

- BitMap now operates on 64-bit int rather than 32-bit
- BitMap is now pooled in a ThreadStatic pool (within lrsa)
- BitMap now is now its own iterator. Marginally speeds up iterating
through the bits.
- A few cases where enumerators were generated have been converted to
forms that generate less garbage.
- New data structure for sorting _usePositions in LiveIntervals. Much
faster split, NextUseAfter, initial insertion. Random insertion is
slightly slower.
- That last one is WIP since you need to insert the values backwards. It
would be ideal if it just flipped it for you, uncomplicating things on
the caller side.

* Use a static pool of thread static pools. (yes.)

Prevents each execution thread creating its own lowCq pool and making me cry.

* Move constant value to top, change naming convention.

* Fix iteration of memory operands.

* Increase max thread count.

* Address Feedback
2020-03-18 22:44:32 +11:00
riperiperi
d904706fc0
Use a Jump Table for direct and indirect calls/jumps, removing transitions to managed (#975)
* Implement Jump Table for Native Calls

NOTE: this slows down rejit considerably! Not recommended to be used
without codegen optimisation or AOT.

- Does not work on Linux
- A32 needs an additional commit.

* A32 Support

(WIP)

* Actually write Direct Call pointers to the table

That would help.

* Direct Calls: Rather than returning to the translator, attempt to keep within the native stack frame.

A return to the translator can still happen, but only by exceptionally
bubbling up to it.

Also:
- Always translate lowCq as a function. Faster interop with the direct
jumps, and this will be useful in future if we want to do speculative
translation.
- Tail Call Detection: after the decoding stage, detect if we do a tail
call, and avoid translating into it. Detected if a jump is made to an
address outwith the contiguous sequence of blocks surrounding the entry
point. The goal is to reduce code touched by jit and rejit.

* A32 Support

* Use smaller max function size for lowCq, fix exceptional returns

When a return has an unexpected value and there is no code block
following this one, we now return the value rather than continuing.

* CompareAndSwap (buggy)

* Ensure CompareAndSwap does not get optimized away.

* Use CompareAndSwap to make the dynamic table thread safe.

* Tail call for linux, throw on too many arguments.

* Combine CompareAndSwap 128 and 32/64.

They emit different IR instructions since their PreAllocator behaviour
is different, but now they just have one function on EmitterContext.

* Fix issues separating from optimisations.

* Use a stub to find and execute missing functions.

This allows us to skip doing many runtime comparisons and branches, and reduces the amount of code we need to emit significantly.

For the indirect call table, this stub also does the work of moving in the highCq address to the table when one is found.

* Make Jump Tables and Jit Cache dynmically resize

Reserve virtual memory, commit as needed.

* Move TailCallRemover to its own class.

* Multithreaded Translation (based on heuristic)

A poor one, at that. Need to get core count for a better one, which
means a lot of OS specific garbage.

* Better priority management for background threads.

* Bound core limit a bit more

Past a certain point the load is not paralellizable and starts stealing from the main thread. Likely due to GC, memory, heap allocation thread contention. Reduce by one core til optimisations come to improve the situation.

* Fix memory management on linux.

* Temporary solution to some sync problems.

This will make sure threads exit correctly, most of the time. There is a potential race where setting the sync counter to 0 does nothing (counter stays at what it was before, thread could take too long to exit), but we need to find a better way to do this anyways. Synchronization frequency has been tightened as we never enter blockwise segments of code. Essentially this means, check every x functions or loop iterations, before lowcq blocks existed and were worth just as much. Ideally it should be done in a better way, since functions can be anywhere from 1 to 5000 instructions. (maybe based on host timer, or an interrupt flag from a scheduler thread)

* Address feedback minus CompareAndSwap change.

* Use default ReservedRegion granularity.

* Merge CompareAndSwap with its V128 variant.

* We already got the source, no need to do it again.

* Make sure all background translation threads exit.

* Fix CompareAndSwap128

Detection criteria was a bit scuffed.

* Address Comments.
2020-03-12 14:20:55 +11:00