[mono] Tracking: Intrinsics implementation

[imhameed]

.NET 6

General implementation:

• System.Runtime.Intrinsics.Vector{128,64} Assigned to imhameed
• System.Runtime.Intrinsics.Arm.Sha256 Assigned to fanyang-mono
• System.Runtime.Intrinsics.Arm.Sha1 Assigned to fanyang-mono
• System.Runtime.Intrinsics.Arm.Rdm Assigned to imhameed
• System.Runtime.Intrinsics.Arm.Dp Assigned to imhameed
• System.Runtime.Intrinsics.Arm.Aes Assigned to fanyang-mono
• System.Runtime.Intrinsics.Arm.AdvSimd Assigned to naricc/imhameed
• MultiplyHigh
• 4-arg ops
• Tablegen automation

Bugs:

• [Mono] Vector256 bug #43676
• [mono] Don't crash when new unsupported intrinsic groups are added #53078

.NET 7

General implementation:

• [Mono] Intrinsify API's of System.Numerics.Vector and new API's of System.Runtime.Intrinsics.Vector{64, 128} on Arm64 #64072
• [Mono] Add SIMD support for System.Numerics.Vector4 on Arm64 #69186
• [Addressed input arguments with LLVM enabled] [mono] Pass Vector64, Vector128, and Vector256 in SIMD registers whenever possible #60068
• LoadPairVector
  • See [Arm64] Implement LoadPairVector64 and LoadPairVector128 #52424 (comment)
  • An implementation is available at 82bb175

Runtime test coverage:

• [mono] Tracking: Intrinsics runtime tests #43842

Cleanup and refactoring:

• [Mono] Clean up the type checks for Vector64/128 #65318
• [Mono] Clean up the function calls #66525

.NET 8

• [mono] Add arm64/neon opcode macros to the codegen #80788
• [Mono] Intrinsify System.Numerics.Vector and System.Runtime.Intrinsics.Vector128 API's with mini JIT on Arm64 #80566
• [mono] Re-enable passing Vector128 arguments in SIMD registers #74583
• Intrinsify RuntimeHelpers.CreateSpan on Mono #80762
• [Mono] Intrinsify Vector128:ExtractMostSignificantBits on ARM64 and AMD64 #76025 - Will focus on Arm64 first. May not be able to get to the Amd64 one in .NET8
• [Mono] Intrinsify Quaternion and Plane with LLVM as backend #81483
• Add handling for SIMD constants in Mono #81482
• [Mono] Intrinsify Vector128 API's on Amd64 with mini JIT #86272
• [Mono] Intrinsify System.Numerics.Vector4 API's with mini JIT on Arm64 #86957
• [mono][jit] ArmBase methods are now intrinsic. #85458
• SIMD problems when mixing llvm and non-llvm code #73454
• Add VectorTableLookup/VectorTableLookupExtension support in mono #83891

.NET 9

• [Mono] Add SIMD support for System.Numerics.Vector2 on Arm64 #73462
• [Mono] Add missing SIMD support for some methods from System.Numerics.Vector2/3/4 and Quaternion on Arm64 #91394
• [Mono] AdvSimd.Arm64 [Load/Store]Vector64 and [Load/Store]Vector128 for 2,3 and 4 variants with LLVM #93081
• [Mono] [Arm64] Add SIMD support for vector constructors using other vectors #97815
• [mono] Intrinsify API's in SpanHelpers.ByteMemOps.cs #99161
• [mono] Intrinsify Unsafe API's #101495

.NET10

• [Mono] Intrinsify loading/storing, reordering, and extracting a per-element "mask" API's of Vector{64, 128, 256} #73261 - focus on Vector128 only.

Future:

• [Mono] Intrinsify API's of System.Numerics.Vector and new API's of System.Runtime.Intrinsics.{128,256} on Amd64 with LLVM backend #66392
• [Handle output argument with LLVM enabled and more] [mono] Pass Vector64, Vector128, and Vector256 in SIMD registers whenever possible #60068
• System.Runtime.Intrinsics.Vector256
• 512-bit vector support
• [mono] Tracking: Implement System.Runtime.Intrinsics.X86.Avx #60483
• [mono] Tracking: Implement System.Runtime.Intrinsics.X86.Avx2 #60484
• [mono] Tracking: Implement System.Runtime.Intrinsics.X86.AvxVnni #60485
• Add X86Serialize API and intrinsic for mono. #68916
• [Mono] Intrinsify Vector2/3 #81501
• [Mono] Intrinsify Quaternion and Plane #88207
• [Mono] Issues to prevent SIMD registers from being enabled on Arm64 #87576
• [mono][jit] Add test for arm64/neon opcodes #81420
• [Mono] Intrinsify Vector64.ExtractMostSignificantBits on ARM64 #90402
• [mono] Reenable disabled x86 SIMD unit tests #91392
• [Mono] Implement small types atomic intrinsics for CompareExchange and Exchange #93488
• [Mono] [x64] Add SIMD support for System.Numerics.Vector #97988
• [mono] Implement * and / quaternion intrinsics #82408

Size optimization:

• [mono] Share large fallback immediate-parameter workaround code sequences #43050

Cleanup and refactoring:

• [mono] Intrinsics cleanup #53032
• [Mono] Code duplication between amd64 and arm64 intrinsics #75144

[Dotnet-GitSync-Bot]

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area:
See info in area-owners.md if you want to be subscribed.

[lambdageek]

@imhameed @fanyang-mono I can't tell from the summary, does this also include marshalling support for Vector128<T>/Vector256<T> (ie so that pinvokes on x86-64 use the xmm registers, and on arm64 use the NEON registers for argument passing to native code). Would allow us to run these tests https://github.com/dotnet/runtime/tree/main/src/tests/Interop/PInvoke/Generics and implement mono/mono#17868

A related issue #9578 is tracking the CoreCLR work to unblock passing Vector128<T>/Vector256<T> by value to P/Invokes. It's currently blocked (with a bad error message about the types not being blittable) due to CoreCLR incorrectly marshalling vector return values on Windows x64.

[imhameed]

That work hasn't happened yet. It's come up a few times in PR comments but I previously had only been tracking this in my head. I just created an issue for this: #60068

[SamMonoRT]

moving tracking issue to 8.0.0

[fanyang-mono]

Moving tracking issue to 9.0.0