Skip to content

mlir, tvm: Add f16 and bf16 computation support - #165

Open
yadej wants to merge 1 commit into
xtc-tools:mainfrom
yadej:dev/rcesista/add-other-float-type
Open

yadej wants to merge 1 commit into
xtc-tools:mainfrom
yadej:dev/rcesista/add-other-float-type

Conversation

@yadej

@yadej yadej commented Oct 28, 2026

Copy link
Copy Markdown
Contributor

Motivation

There is no support for float other than f32/f64 so adding new type f16 and bf16

Currently the accumulation is done on the same type as the computation.
There is a problem with the error tolerance because of that accumulation

Future work

  • Add other type of float like float8 or float 128
  • Add support to mixed type computation to possibly have a f32 accumulator

@yadej
yadej requested a review from guillon October 28, 2026 12:36
@codecov

codecov Bot commented Oct 28, 2026

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 87.50000% with 4 lines in your changes missing coverage. Please review.

Files with missing lines Patch % Lines
src/xtc/utils/numpy.py 75.00% 1 Missing and 1 partial ⚠️
src/xtc/targets/accelerator/gpu/TVMGPUEvaluator.py 50.00% 1 Missing ⚠️
src/xtc/utils/evaluation.py 50.00% 0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

@yadej yadej added enhancement New feature or request mlir tvm labels Oct 28, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request mlir tvm

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant