Implement StoreVectorNxM for Arm64 #94129

SwapnilGaikwad · 2023-10-28T05:37:08Z

Contribute towards #84510

// ST3 (multiple structures)
public static unsafe void StoreVector128x3(byte*   address, (Vector128<byte>   Value1, Vector128<byte>   Value2, Vector128<byte>   Value3) value);
public static unsafe void StoreVector128x3(sbyte*  address, (Vector128<sbyte>  Value1, Vector128<sbyte>  Value2, Vector128<sbyte>  Value3) value);
public static unsafe void StoreVector128x3(short*  address, (Vector128<short>  Value1, Vector128<short>  Value2, Vector128<short>  Value3) value);
public static unsafe void StoreVector128x3(ushort* address, (Vector128<ushort> Value1, Vector128<ushort> Value2, Vector128<ushort> Value3) value);
public static unsafe void StoreVector128x3(int*    address, (Vector128<int>    Value1, Vector128<int>    Value2, Vector128<int>    Value3) value);
public static unsafe void StoreVector128x3(uint*   address, (Vector128<uint>   Value1, Vector128<uint>   Value2, Vector128<uint>   Value3) value);
public static unsafe void StoreVector128x3(long*   address, (Vector128<long>   Value1, Vector128<long>   Value2, Vector128<long>   Value3) value);
public static unsafe void StoreVector128x3(ulong*  address, (Vector128<ulong>  Value1, Vector128<ulong>  Value2, Vector128<ulong>  Value3) value);
public static unsafe void StoreVector128x3(float*  address, (Vector128<float>  Value1, Vector128<float>  Value2, Vector128<float>  Value3) value);
public static unsafe void StoreVector128x3(double* address, (Vector128<double> Value1, Vector128<double> Value2, Vector128<double> Value3) value);

public static unsafe void StoreVector64x3(byte*   address, (Vector64<byte>   Value1, Vector64<byte>   Value2, Vector64<byte>   Value3) value);
public static unsafe void StoreVector64x3(sbyte*  address, (Vector64<sbyte>  Value1, Vector64<sbyte>  Value2, Vector64<sbyte>  Value3) value);
public static unsafe void StoreVector64x3(short*  address, (Vector64<short>  Value1, Vector64<short>  Value2, Vector64<short>  Value3) value);
public static unsafe void StoreVector64x3(ushort* address, (Vector64<ushort> Value1, Vector64<ushort> Value2, Vector64<ushort> Value3) value);
public static unsafe void StoreVector64x3(int*    address, (Vector64<int>    Value1, Vector64<int>    Value2, Vector64<int>    Value3) value);
public static unsafe void StoreVector64x3(uint*   address, (Vector64<uint>   Value1, Vector64<uint>   Value2, Vector64<uint>   Value3) value);
public static unsafe void StoreVector64x3(float*  address, (Vector64<float>  Value1, Vector64<float>  Value2, Vector64<float>  Value3) value);

// ST4 (multiple structures)
public static unsafe void StoreVector128x4(byte*   address, (Vector128<byte>   Value1, Vector128<byte>   Value2, Vector128<byte>   Value3, Vector128<byte>   Value4) value);
public static unsafe void StoreVector128x4(sbyte*  address, (Vector128<sbyte>  Value1, Vector128<sbyte>  Value2, Vector128<sbyte>  Value3, Vector128<sbyte>  Value4) value);
public static unsafe void StoreVector128x4(short*  address, (Vector128<short>  Value1, Vector128<short>  Value2, Vector128<short>  Value3, Vector128<short>  Value4) value);
public static unsafe void StoreVector128x4(ushort* address, (Vector128<ushort> Value1, Vector128<ushort> Value2, Vector128<ushort> Value3, Vector128<ushort> Value4) value);
public static unsafe void StoreVector128x4(int*    address, (Vector128<int>    Value1, Vector128<int>    Value2, Vector128<int>    Value3, Vector128<int>    Value4) value);
public static unsafe void StoreVector128x4(uint*   address, (Vector128<uint>   Value1, Vector128<uint>   Value2, Vector128<uint>   Value3, Vector128<uint>   Value4) value);
public static unsafe void StoreVector128x4(long*   address, (Vector128<long>   Value1, Vector128<long>   Value2, Vector128<long>   Value3, Vector128<long>   Value4) value);
public static unsafe void StoreVector128x4(ulong*  address, (Vector128<ulong>  Value1, Vector128<ulong>  Value2, Vector128<ulong>  Value3, Vector128<ulong>  Value4) value);
public static unsafe void StoreVector128x4(float*  address, (Vector128<float>  Value1, Vector128<float>  Value2, Vector128<float>  Value3, Vector128<float>  Value4) value);
public static unsafe void StoreVector128x4(double* address, (Vector128<double> Value1, Vector128<double> Value2, Vector128<double> Value3, Vector128<double> Value4) value);

public static unsafe void StoreVector64x4(byte*   address, (Vector64<byte>   Value1, Vector64<byte>   Value2, Vector64<byte>   Value3, Vector64<byte>   Value4) value);
public static unsafe void StoreVector64x4(sbyte*  address, (Vector64<sbyte>  Value1, Vector64<sbyte>  Value2, Vector64<sbyte>  Value3, Vector64<sbyte>  Value4) value);
public static unsafe void StoreVector64x4(short*  address, (Vector64<short>  Value1, Vector64<short>  Value2, Vector64<short>  Value3, Vector64<short>  Value4) value);
public static unsafe void StoreVector64x4(ushort* address, (Vector64<ushort> Value1, Vector64<ushort> Value2, Vector64<ushort> Value3, Vector64<ushort> Value4) value);
public static unsafe void StoreVector64x4(int*    address, (Vector64<int>    Value1, Vector64<int>    Value2, Vector64<int>    Value3, Vector64<int>    Value4) value);
public static unsafe void StoreVector64x4(uint*   address, (Vector64<uint>   Value1, Vector64<uint>   Value2, Vector64<uint>   Value3, Vector64<uint>   Value4) value);
public static unsafe void StoreVector64x4(float*  address, (Vector64<float>  Value1, Vector64<float>  Value2, Vector64<float>  Value3, Vector64<float>  Value4) value);

dotnet-issue-labeler · 2023-10-28T05:37:17Z

Note regarding the new-api-needs-documentation label:

This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change.

ghost · 2023-10-28T05:37:23Z

Tagging subscribers to this area: @dotnet/area-system-runtime-intrinsics
See info in area-owners.md if you want to be subscribed.

Issue Details

Contribute towards #84510

// ST3 (multiple structures)
public static unsafe void StoreVector128x3(byte*   address, (Vector128<byte>   Value1, Vector128<byte>   Value2, Vector128<byte>   Value3) value);
public static unsafe void StoreVector128x3(sbyte*  address, (Vector128<sbyte>  Value1, Vector128<sbyte>  Value2, Vector128<sbyte>  Value3) value);
public static unsafe void StoreVector128x3(short*  address, (Vector128<short>  Value1, Vector128<short>  Value2, Vector128<short>  Value3) value);
public static unsafe void StoreVector128x3(ushort* address, (Vector128<ushort> Value1, Vector128<ushort> Value2, Vector128<ushort> Value3) value);
public static unsafe void StoreVector128x3(int*    address, (Vector128<int>    Value1, Vector128<int>    Value2, Vector128<int>    Value3) value);
public static unsafe void StoreVector128x3(uint*   address, (Vector128<uint>   Value1, Vector128<uint>   Value2, Vector128<uint>   Value3) value);
public static unsafe void StoreVector128x3(long*   address, (Vector128<long>   Value1, Vector128<long>   Value2, Vector128<long>   Value3) value);
public static unsafe void StoreVector128x3(ulong*  address, (Vector128<ulong>  Value1, Vector128<ulong>  Value2, Vector128<ulong>  Value3) value);
public static unsafe void StoreVector128x3(float*  address, (Vector128<float>  Value1, Vector128<float>  Value2, Vector128<float>  Value3) value);
public static unsafe void StoreVector128x3(double* address, (Vector128<double> Value1, Vector128<double> Value2, Vector128<double> Value3) value);

public static unsafe void StoreVector64x3(byte*   address, (Vector64<byte>   Value1, Vector64<byte>   Value2, Vector64<byte>   Value3) value);
public static unsafe void StoreVector64x3(sbyte*  address, (Vector64<sbyte>  Value1, Vector64<sbyte>  Value2, Vector64<sbyte>  Value3) value);
public static unsafe void StoreVector64x3(short*  address, (Vector64<short>  Value1, Vector64<short>  Value2, Vector64<short>  Value3) value);
public static unsafe void StoreVector64x3(ushort* address, (Vector64<ushort> Value1, Vector64<ushort> Value2, Vector64<ushort> Value3) value);
public static unsafe void StoreVector64x3(int*    address, (Vector64<int>    Value1, Vector64<int>    Value2, Vector64<int>    Value3) value);
public static unsafe void StoreVector64x3(uint*   address, (Vector64<uint>   Value1, Vector64<uint>   Value2, Vector64<uint>   Value3) value);
public static unsafe void StoreVector64x3(float*  address, (Vector64<float>  Value1, Vector64<float>  Value2, Vector64<float>  Value3) value);

// ST4 (multiple structures)
public static unsafe void StoreVector128x4(byte*   address, (Vector128<byte>   Value1, Vector128<byte>   Value2, Vector128<byte>   Value3, Vector128<byte>   Value4) value);
public static unsafe void StoreVector128x4(sbyte*  address, (Vector128<sbyte>  Value1, Vector128<sbyte>  Value2, Vector128<sbyte>  Value3, Vector128<sbyte>  Value4) value);
public static unsafe void StoreVector128x4(short*  address, (Vector128<short>  Value1, Vector128<short>  Value2, Vector128<short>  Value3, Vector128<short>  Value4) value);
public static unsafe void StoreVector128x4(ushort* address, (Vector128<ushort> Value1, Vector128<ushort> Value2, Vector128<ushort> Value3, Vector128<ushort> Value4) value);
public static unsafe void StoreVector128x4(int*    address, (Vector128<int>    Value1, Vector128<int>    Value2, Vector128<int>    Value3, Vector128<int>    Value4) value);
public static unsafe void StoreVector128x4(uint*   address, (Vector128<uint>   Value1, Vector128<uint>   Value2, Vector128<uint>   Value3, Vector128<uint>   Value4) value);
public static unsafe void StoreVector128x4(long*   address, (Vector128<long>   Value1, Vector128<long>   Value2, Vector128<long>   Value3, Vector128<long>   Value4) value);
public static unsafe void StoreVector128x4(ulong*  address, (Vector128<ulong>  Value1, Vector128<ulong>  Value2, Vector128<ulong>  Value3, Vector128<ulong>  Value4) value);
public static unsafe void StoreVector128x4(float*  address, (Vector128<float>  Value1, Vector128<float>  Value2, Vector128<float>  Value3, Vector128<float>  Value4) value);
public static unsafe void StoreVector128x4(double* address, (Vector128<double> Value1, Vector128<double> Value2, Vector128<double> Value3, Vector128<double> Value4) value);

public static unsafe void StoreVector64x4(byte*   address, (Vector64<byte>   Value1, Vector64<byte>   Value2, Vector64<byte>   Value3, Vector64<byte>   Value4) value);
public static unsafe void StoreVector64x4(sbyte*  address, (Vector64<sbyte>  Value1, Vector64<sbyte>  Value2, Vector64<sbyte>  Value3, Vector64<sbyte>  Value4) value);
public static unsafe void StoreVector64x4(short*  address, (Vector64<short>  Value1, Vector64<short>  Value2, Vector64<short>  Value3, Vector64<short>  Value4) value);
public static unsafe void StoreVector64x4(ushort* address, (Vector64<ushort> Value1, Vector64<ushort> Value2, Vector64<ushort> Value3, Vector64<ushort> Value4) value);
public static unsafe void StoreVector64x4(int*    address, (Vector64<int>    Value1, Vector64<int>    Value2, Vector64<int>    Value3, Vector64<int>    Value4) value);
public static unsafe void StoreVector64x4(uint*   address, (Vector64<uint>   Value1, Vector64<uint>   Value2, Vector64<uint>   Value3, Vector64<uint>   Value4) value);
public static unsafe void StoreVector64x4(float*  address, (Vector64<float>  Value1, Vector64<float>  Value2, Vector64<float>  Value3, Vector64<float>  Value4) value);

Author:	SwapnilGaikwad
Assignees:	-
Labels:	`area-System.Runtime.Intrinsics`, `new-api-needs-documentation`, `community-contribution`
Milestone:	-

SwapnilGaikwad · 2023-10-28T05:40:51Z

Hi @kunalspathak , would you prefer an overloaded name for the interleaved multi-structure stores like we did for StoreSelectedScalar? If yes then what should we call such methods? Store or StoreVector may not convey the interleaving nature of underlying store instructions. 🤔

kunalspathak · 2023-10-29T01:28:03Z

for the interleaved multi-structure stores

sorry, but could you confirm which APIs are you asking about?

SwapnilGaikwad · 2023-10-29T22:35:08Z

for the interleaved multi-structure stores

sorry, but could you confirm which APIs are you asking about?

The StoreVectorNX2, StoreVectorNx3, StoreVectorNx3 that result in ST2, ST3 and ST4 instructions respectively.

SwapnilGaikwad · 2023-10-30T14:00:14Z

Hi @kunalspathak,

Could you please confirm if my understanding is correct?

LoadVector(N)x(M) is mapped to LD(M). E.g., LoadVector128x3 to LD3
StoreVector(N)x(M) should map to ST(M), E.g., StoreVector128x3 to ST3
StoreVector(N)x(M)AndZip should map to ST1 with M registers.

If this is correct, then aren't these names a little confusing?
LD2/3/4 would unzip/de-interleave while loading the input registers while LD1 will load given input vectors 2/3/4 consecutively.
Thus, wouldn't it make more sense for LoadVector(N)x(M) to emit LD1 with M input values and vice-versa for LoadVector(N)x(M)AndZip?

kunalspathak

Changes looks good, some minor fixes needed.

src/coreclr/jit/hwintrinsiccodegenarm64.cpp

src/tests/Common/GenerateHWIntrinsicTests/GenerateHWIntrinsicTests_Arm.cs

kunalspathak · 2023-10-31T14:09:19Z

Thus, wouldn't it make more sense for LoadVector(N)x(M) to emit LD1 with M input values and vice-versa for LoadVector(N)x(M)AndZip?

Looking carefully, yes, you are right. I swapped the two. I will send a PR to fix it. Thanks for spotting it.

SwapnilGaikwad · 2023-11-01T05:16:54Z

Thus, wouldn't it make more sense for LoadVector(N)x(M) to emit LD1 with M input values and vice-versa for LoadVector(N)x(M)AndZip?

Looking carefully, yes, you are right. I swapped the two. I will send a PR to fix it. Thanks for spotting it.

Cool, I'll update this PR accordingly 👍

SwapnilGaikwad · 2023-11-03T15:51:03Z

After merging #93223, I'll rebase/merge this PR and then mark it ready for review.

SwapnilGaikwad · 2023-11-09T08:38:27Z

After merging #93223, I'll rebase/merge this PR and then mark it ready for review.

As the #93223 taking longer, I'll mark this PR ready for review. If this can progress further quickly then we can merge this. I'll rebase/merge these changes wherever necessary.

kunalspathak

LGTM. Just a request to update the documentation for LoadVector equivalent.

src/libraries/System.Private.CoreLib/src/System/Runtime/Intrinsics/Arm/AdvSimd.cs

kunalspathak

LGTM

Implement StoreVectorNxM for Arm64

569debc

dotnet-issue-labeler bot added area-System.Runtime.Intrinsics new-api-needs-documentation labels Oct 28, 2023

ghost added the community-contribution Indicates that the PR has been added by a community member label Oct 28, 2023

This was referenced Oct 28, 2023

Timeout in System.Net.Quic.Functional.Tests #86019

Closed

CI error: System.Net.Quic.QuicException: The connection timed out from inactivity #91757

Closed

kunalspathak requested changes Oct 31, 2023

View reviewed changes

src/coreclr/jit/hwintrinsiccodegenarm64.cpp Show resolved Hide resolved

src/tests/Common/GenerateHWIntrinsicTests/GenerateHWIntrinsicTests_Arm.cs Outdated Show resolved Hide resolved

ghost added the needs-author-action An issue or pull request that requires more info or actions from the author. label Oct 31, 2023

kunalspathak mentioned this pull request Oct 31, 2023

Swap loadvector* and loadvectorandunzip* #94220

Merged

ghost removed the needs-author-action An issue or pull request that requires more info or actions from the author. label Nov 1, 2023

SwapnilGaikwad added 2 commits November 3, 2023 15:38

Merge main

5e4882e

Use ST1 instruction variants while for StoreVectorNxM API calls

4536244

Fix formatting

20e5ad6

This was referenced Nov 3, 2023

Test_EventSource_EtwManifestGeneration* tests failing in CI #48798

Closed

[wasm] Runtime tests' build gets terminated with exited with code 137 #94077

Closed

Add minor fixes

824e1ce

SwapnilGaikwad marked this pull request as ready for review November 9, 2023 08:38

build-analysis bot mentioned this pull request Nov 9, 2023

Crash in Microsoft.Extensions.Logging.Generators.Roslyn4.0.Tests.WorkItemExecution #90019

Open

kunalspathak reviewed Nov 9, 2023

View reviewed changes

src/libraries/System.Private.CoreLib/src/System/Runtime/Intrinsics/Arm/AdvSimd.cs Show resolved Hide resolved

Update comments for LoadVectorNxM & LoadVectorAndUnzipNxM

4782c32

kunalspathak approved these changes Nov 9, 2023

View reviewed changes

kunalspathak merged commit e138ff1 into dotnet:main Nov 9, 2023
192 of 195 checks passed

SwapnilGaikwad deleted the github-st3-st4 branch November 9, 2023 20:07

github-actions bot locked and limited conversation to collaborators Dec 10, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement StoreVectorNxM for Arm64 #94129

Implement StoreVectorNxM for Arm64 #94129

SwapnilGaikwad commented Oct 28, 2023

dotnet-issue-labeler bot commented Oct 28, 2023

ghost commented Oct 28, 2023

SwapnilGaikwad commented Oct 28, 2023

kunalspathak commented Oct 29, 2023

SwapnilGaikwad commented Oct 29, 2023

SwapnilGaikwad commented Oct 30, 2023 •

edited

Loading

kunalspathak left a comment

kunalspathak commented Oct 31, 2023

SwapnilGaikwad commented Nov 1, 2023

SwapnilGaikwad commented Nov 3, 2023

SwapnilGaikwad commented Nov 9, 2023

kunalspathak left a comment

kunalspathak left a comment

Implement StoreVectorNxM for Arm64 #94129

Implement StoreVectorNxM for Arm64 #94129

Conversation

SwapnilGaikwad commented Oct 28, 2023

dotnet-issue-labeler bot commented Oct 28, 2023

ghost commented Oct 28, 2023

SwapnilGaikwad commented Oct 28, 2023

kunalspathak commented Oct 29, 2023

SwapnilGaikwad commented Oct 29, 2023

SwapnilGaikwad commented Oct 30, 2023 • edited Loading

kunalspathak left a comment

Choose a reason for hiding this comment

kunalspathak commented Oct 31, 2023

SwapnilGaikwad commented Nov 1, 2023

SwapnilGaikwad commented Nov 3, 2023

SwapnilGaikwad commented Nov 9, 2023

kunalspathak left a comment

Choose a reason for hiding this comment

kunalspathak left a comment

Choose a reason for hiding this comment

SwapnilGaikwad commented Oct 30, 2023 •

edited

Loading