Extended Gemm interface to support mixed precision operations #500
Conversation
This file contains 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
Co-authored-by: pgorlani <[email protected]>
Co-authored-by: pgorlani <[email protected]>
Co-authored-by: pgorlani <[email protected]>
Co-authored-by: pgorlani <[email protected]>
Co-authored-by: pgorlani <[email protected]>
OuadiElfarouki
requested review from
muhammad-tanvir-1211,
aacostadiaz,
pgorlani and
s-Nick
and removed request for
aacostadiaz
hjabird
reviewed
Mar 26, 2024
s-Nick
reviewed
Mar 27, 2024
There was a problem hiding this comment.
Thank you for you work @OuadiElfarouki.
There is some confusion with element_in_t and element_out_t because you forget to update element_t to one or the other. Can you please fix it? @hjabird highlighted many of them. In general, I agree that something like in_element_t and out_element_t would be more readable, please consider it.
| if constexpr (is_half<element_in_t>::value || | ||
| !std::is_same_v<element_in_t, element_out_t>) { | ||
| #pragma unroll | ||
| for (int v = 0; v < VectorSize; ++v) { |
There was a problem hiding this comment.
Can you elaborate why this loop is necessary?
There was a problem hiding this comment.
@s-Nick AdaptiveCpp doesn't support sycl::vecsycl::half for mad operation, so had to break it down to its elements through the loop
Co-authored-by: HJA Bird <[email protected]>
Rbiessy
approved these changes
May 9, 2024
…xed_precision_gemm
hjabird
approved these changes
May 13, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Extend Gemm operator interface to support mixed precision operations, namely by decoupling matrix A and B type
element_in_tfrom output matrix C and scalars alpha and beta typeelement_out_t.Following oneMKL's spec notation for Gemm API : https://spec.oneapi.io/versions/latest/elements/oneMKL/source/domains/blas/gemm.html#onemkl-blas-gemm, this PR enables (Ta==Tb) to be set independently from (Tc==Ts). This feature has been enabled at a first stage for Ta=Tb=sycl::half and Tc=Ts=float. Thus enabling half support also enables the mixed precision case of (half, float) for gemm.
Changes include:
Note :
Following oneMKL expected Gemm API, Support of
bfloat16-floatandstd::int8_t-floatwould be straightforward afterwards, but the additional cases ofTa==Tb==TcwhileTs(alpha & beta) is separate will require additional decoupling & re-design work..