Runtime: Implement AVX Intrinsics

[tannergooding]

Pending (a check means it is in a pending PR):

• [x] Blend
• [x] BroadcastScalarToVector128
• [x] BroadcastScalarToVector256
• [ ] BroadcastVector128ToVector256
• [x] Ceiling
• [x] ConvertToVector128Int32
• [x] ConvertToVector128Single
• [x] ConvertToVector256Int32
• [x] ConvertToVector256Single
• [x] ConvertToVector256Double
• [x] ConvertToVector128Int32WithTruncation
• [x] ConvertToVector256Int32WithTruncation
• [x] DotProduct
• [ ] Extract
• [ ] ExtractVector128
• [x] ExtendToVector256
• [x] Floor
• [ ] GetLowerHalf
• [ ] Insert
• [ ] MaskLoad
• [ ] MaskStore
• [ ] MoveMask
• [ ] Permute
• [ ] Permute2x128
• [ ] PermuteVar
• [x] RoundCurrentDirection
• [x] RoundToNearestInteger
• [x] RoundToNegativeInfinity
• [x] RoundToPositiveInfinity
• [x] RoundToZero
• [ ] SetAllVector256
• [ ] SetHIghLow
• [ ] Shuffle
• [x] StaticCast
• [x] TestC
• [x] TestNotZAndNotC
• [x] TestZ
• [ ] ZeroAll
• [ ] ZeroUpper
• [ ] ZeroExtendToVector256

[tannergooding]

This is a parent issue to: https://github.com/dotnet/coreclr/issues/16583

@fiigii, could you confirm which ones you are currently planning on implementing?

[fiigii]

which ones you are currently planning on implementing?

BroadcastScalarToVector128
BroadcastScalarToVector256
BroadcastVector128ToVector256
Extract
ExtractVector128
Insert
SetAllVector256
SetVector256

[tannergooding]

@fiigii, thanks.

I am working on the "simple" intrinsics as per our earlier conversation.

This currently includes:

• Blend
• Ceiling
• DotProduct
• Floor
• RoundCurrentDirection
• RoundToNearestInteger
• RoundToNegativeInfinity
• RoundToPositiveInfinity
• RoundToZero
• TestC
• TestNotZAndNotC
• TestZ

(this is basically the Vector256 versions of the SSE4.1 I implemented in https://github.com/dotnet/coreclr/pull/16558)

[tannergooding]

https://github.com/dotnet/coreclr/pull/16655 contains the intrinsics I mentioned above

[tannergooding]

@fiigii, I'm going to work on the remaining, in alphabetical order, until you finish the ones you are currently working on, then we can sync up again.

[fiigii]

I will implement Permute as well.

[fiigii]

@RussKeldorph I think we can finish this for 2.1.

[tannergooding]

Currently writing tests for ExtendTo, GetLowerHalf, and StaticCast. Should be up shortly.

[fiigii]

Can we close this issue?

[tannergooding]

@fiigii, the only ones not implemented are ZeroAll and ZeroUpper, correct?

[fiigii]

And MaskStore.

[tannergooding]

Can you log bugs tracking those three separately?

Happy to close this issue afterwards.

[fiigii]

Logged at https://github.com/dotnet/coreclr/issues/17127