[SM6.10][Bugfix] MultiplyAdd - Convert Bias vector to output vector type#8394
Open
hekota wants to merge 3 commits intomicrosoft:mainfrom
Open
[SM6.10][Bugfix] MultiplyAdd - Convert Bias vector to output vector type#8394hekota wants to merge 3 commits intomicrosoft:mainfrom
hekota wants to merge 3 commits intomicrosoft:mainfrom
Conversation
If the Bias vector type on MultiplytAdd is different than the output vector type, convert it first to the output vector type before calling `dx.op.linAlgMatVecMulAdd` op. The interpretation for the bias vector is always going to be set to the output vector interpretation on the op. To accommodate this, MultiplyAdd functions have been split into two variants - one where bias interpretation matches the output vector interpretation and no conversion is necessary, and second where the types don't match and the bias vector is first converted to the output vector type before passing it into the __builtin_LinAlg_MatrixVectorMultiplyAdd. Adds`hlsl::__detail::TypeTraits` struct to enable mapping of HLSL scalar types to component type enum values. Fixes microsoft#8390
Contributor
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
If the
Biasvector type/interpretation onMultiplytAdddiffers from the output vector type, convert it first to the output vector type before passing it into the__builtin_LinAlg_MatrixVectorMultiplyAddbuilt-in function (translared todx.op.linAlgMatVecMulAddop). The bias vector interpretation is always going to be the same as the output vector interpretation on the call.To accommodate this, each
MultiplyAddnow has two SFINAE-constrained variants:Adds
hlsl::__detail::TypeTraitsstruct to enable mapping of HLSL scalar types to component type enum values.Fixes #8390