[SM6.10] Add Matrix class methods and other LinAlg API to linalg.h by hekota · Pull Request #8299 · microsoft/DirectXShaderCompiler

hekota · 2026-03-26T20:07:14Z

Implements Matrix class methods and other API functions related to linear algebra as specified in LinAlg spec here.

The header disables warning when an argument is marked as groupshared which is reported when the language version is other than 202x. In order to do that the warning needed to be moved to a separate diagnostic group. Also fixes couple of places in SemaHLSL that did not account for the linalg matrix tyle.

Related to #7839

Implements Matrix class methods and other API function related to linear algebra as specified in LinAlg spec (here)[https://github.com/microsoft/hlsl-specs/blob/main/proposals/0035-linalg-matrix.md#appendix-1-hlsl-header]. The header disables warning for a groupshared argument when the language version is other than 202x. Related to microsoft#7839

…less input type interpretation is provided as a separate argument or in InterpretedVector.

…iler into linalg-matrix-header

llvm-beanz

This all looks sane to me. I have a few comments, but nothing that needs to block this PR.

llvm-beanz · 2026-03-30T18:18:58Z

+
+template <typename T> struct enable_if<true, T> {
+  using type = T;
+};


At some point we'll probably want to pull the traits and enable_if out into a separate header so that it can be reused without causing redeclaration issues. For now this is probably fine because I don't think we've really figured out the right way to handle implicit include paths.

…handle directly

tex3d

I think this looks reasonable, though I haven't thoroughly verified that the SFINAE code encodes all the required rules.

It looks like there's a missing use of Transpose in the Cast operation though.

V-FEXrt · 2026-03-30T21:54:39Z

+  Load(groupshared T Arr[Size], uint StartIdx, uint Stride,
+       MatrixLayoutEnum Layout) {
+    Matrix Result;
+    __builtin_LinAlg_MatrixLoadFromMemory(Result.__handle, Arr, StartIdx,


From the spec:

For the Load operations on groupshared arrays, the Stride argument is the count of elements in the groupshared array.

Should Stride be removed here and replace with Size?

There might be multiple matrices stored in one array, so Size of the array might not match the Stride. The StartIdx allows loading of different matrices from different places in the array.

I don’t think it should be replaced with Size, but perhaps it should be clearer what Stride we’re specifying. This isn’t the matrix stride, it’s the row/col stride (whichever is major), right?

…iler into linalg-matrix-header

github-actions · 2026-03-31T20:04:47Z

✅ With the latest revision this PR passed the C/C++ code formatter.

tex3d

I found that inconsistent naming for typename ...Ty vs. ComponentEnum ...Ty made this harder to read and review.

Matrix uses ComponentEnum MatrixDT, which helps, but DT isn't a convention used elsewhere, nor does it match the naming ComponentType which has a long history in HLSL and would better match the ComponentEnum naming.

Could we standardize on naming things with CT when referring to a ComponentType specified by the ComponentEnum and Ty when referring to an actual native type?

To be clear: for unblocking things, we can likely fix the issues I pointed out in a follow-up PR.

tex3d · 2026-03-31T20:53:23Z

  vector<OutputElTy, M> Result;
-  __builtin_LinAlg_MatrixVectorMultiply(Result, MatrixA.__handle, Vec,
+  __builtin_LinAlg_MatrixVectorMultiply(Result, MatrixA.__handle,
+                                        hlsl::is_signed<OutputElTy>::value, Vec,


I believe you need to construct the inputInterp input from the InputElTy , rather than passing in the MatrixDT.

I think for non-interpreted vectors the assumption is that the interpretation type matches the native scalar type. @llvm-beanz - can you confirm?

tex3d · 2026-03-31T20:53:58Z

-                                           MatrixDT, Bias, MatrixDT);
+  __builtin_LinAlg_MatrixVectorMultiplyAdd(Result, MatrixA.__handle,
+                                           hlsl::is_signed<OutputElTy>::value,
+                                           Vec, MatrixDT, Bias, MatrixDT);


I believe you need to construct the inputInterp and biasInterp inputs from the InputElTy and BiasElTy, rather than passing in the MatrixDT.

tex3d · 2026-03-31T20:55:14Z

-      Result, MatrixA.__handle, InterpVec.Data, InterpVec.Interpretation, Bias,
-      MatrixDT);
+      Result, MatrixA.__handle, hlsl::is_signed<OutputElTy>::value,
+      InterpVec.Data, InterpVec.Interpretation, Bias, MatrixDT);


I believe you need to construct the biasInterp input from the BiasElTy, rather than passing in the MatrixDT.

tex3d · 2026-03-31T20:57:48Z

-                                           MatrixDT, BiasVec, BiasElTy);
+  __builtin_LinAlg_MatrixVectorMultiplyAdd(Result, MatrixA.__handle,
+                                           hlsl::is_signed<OutputElTy>::value,
+                                           Vec, MatrixDT, BiasVec, BiasElTy);


I believe you need to construct the inputInterp input from the InputElTy , rather than passing in the MatrixDT.

hekota · 2026-03-31T21:26:41Z

Could we standardize on naming things with CT when referring to a ComponentType specified by the ComponentEnum and Ty when referring to an actual native type?

I've filed a follow-up issue: #8318

github-project-automation Bot added this to HLSL Roadmap Mar 26, 2026

github-project-automation Bot moved this to New in HLSL Roadmap Mar 26, 2026

hekota force-pushed the linalg-matrix-header branch from af16148 to c63a4b5 Compare March 26, 2026 23:17

Allow vector multiply only for input vectors with arithmetic types un…

f314654

…less input type interpretation is provided as a separate argument or in InterpretedVector.

hekota marked this pull request as ready for review March 27, 2026 19:11

hekota requested review from V-FEXrt and llvm-beanz March 27, 2026 19:11

Update MultiplyAdd with vectors to operate on matrix 'A'

3c269aa

hekota linked an issue Mar 30, 2026 that may be closed by this pull request

Finalize Linalg HLSL Headers #7839

Closed

hekota added 2 commits March 30, 2026 10:49

Update vector API to use Matrix A

7ba62fc

Merge branch 'main' of https://github.com/microsoft/DirectXShaderComp…

9a1dd65

…iler into linalg-matrix-header

llvm-beanz approved these changes Mar 30, 2026

View reviewed changes

Update few return arguments to ref to enable passing in the matrix …

3848117

…handle directly

tex3d reviewed Mar 30, 2026

View reviewed changes

Comment thread tools/clang/lib/Headers/hlsl/dx/linalg.h Outdated

Comment thread tools/clang/test/CodeGenDXIL/hlsl/linalg/api/matrix-class.hlsl Outdated

V-FEXrt approved these changes Mar 30, 2026

View reviewed changes

Fix Transpose argument on Matrix::Cast not being passed in

18f7a75

tex3d approved these changes Mar 30, 2026

View reviewed changes

hekota added 2 commits March 31, 2026 10:10

Merge branch 'main' of https://github.com/microsoft/DirectXShaderComp…

f8f2d31

…iler into linalg-matrix-header

Update after isOutputSigned argument was added to MatVecMultiply*

4220939

clang-format

bae755a

V-FEXrt approved these changes Mar 31, 2026

View reviewed changes

hekota enabled auto-merge (squash) March 31, 2026 20:55

tex3d reviewed Mar 31, 2026

View reviewed changes

hekota mentioned this pull request Mar 31, 2026

[SM6.10] Use consistent naming for ComponentType parameters #8318

Open

hekota merged commit 834978b into microsoft:main Mar 31, 2026
12 checks passed

github-project-automation Bot moved this from New to Done in HLSL Roadmap Mar 31, 2026

pow2clk mentioned this pull request Apr 8, 2026

[0035] Bring godbolt and spec header representations up to date with latest microsoft/hlsl-specs#847

Closed

tex3d mentioned this pull request Apr 15, 2026

[LinAlg] ComponentEnum for native vector inputs should not be based on matrix input #8376

Open

Conversation

hekota commented Mar 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvm-beanz left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

tex3d left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tex3d Mar 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Mar 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tex3d left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

hekota commented Mar 31, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

hekota commented Mar 26, 2026 •

edited

Loading

tex3d Mar 30, 2026 •

edited

Loading

github-actions Bot commented Mar 31, 2026 •

edited

Loading

tex3d left a comment •

edited

Loading