Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add FP8 rocblas gemm support #2473

Merged
merged 97 commits into from
Dec 6, 2023
Merged

Add FP8 rocblas gemm support #2473

merged 97 commits into from
Dec 6, 2023

Conversation

umangyadav
Copy link
Member

Depends on #2455

  • Checks if BETA API for FP8 rocblas gemm exists.
  • If it does and if the hardware is supported (>=gfx940) then calls into FP8 API.
  • If hardware is not supported or BETA API doesn't exist then it uses eliminate_fp8 pass that adds converts around unsupported operator to emulate them into Float and convert back result to FP8.
  • Adds verify tests for FP8
  • tested this on gfx942, it passes verification.

umangyadav and others added 30 commits November 9, 2023 23:23
@umangyadav
changes for the FP8 ref implementation
df7f8a3
@umangyadav
cppcheck fixes
9bc1828
@umangyadav
move FNUZ as template parameter
155a2b1
@umangyadav
Fix numeric limits
d9f11e3
@umangyadav
Working FNUZ and FN
4e9d51f
@umangyadav
use float equal
7639c28
@umangyadav
add test for fp8e5m2
a6372c5
@umangyadav
add test for fp8e5m2fnuz
439ea40
@umangyadav
refactor add some comments
183db78
@umangyadav
Review updates
ab653af
@umangyadav
Fix tidy
8319e01
@umangyadav
Fix test failure
9ee0418
@umangyadav
fix isfinite
355e4f6
@umangyadav
Merge remote-tracking branch 'origin/develop' into ref_fp8
ba471f4
@umangyadav
fix test for neg inf
6aec703
@umangyadav
fix warning
12aac37
@umangyadav
add tests
6009232
@umangyadav
Fix tests
03f7139
@umangyadav
add stringstream tests
1e220c0
@umangyadav
Remove clang diagnostics
a83e9dc
@umangyadav
Merge remote-tracking branch 'origin/develop' into ref_fp8
dfb35a6
@umangyadav
Remove NOLINTS
26956f1
@umangyadav
Bugfixes and additional tests
269ce6d
@umangyadav
Fix undoing
6414ee3
@umangyadav
Handle underflow case separately to avoid sanitization errors
cd26ada
@umangyadav
use std::min to avoid sanitization errors
1cf87ef
@umangyadav
Merge branch 'develop' into ref_fp8
e7e5ba2
@umangyadav
formatting
98a838f
@umangyadav
use 31 for min value
61e4e1d
@umangyadav
add note
a5c38eb
@umangyadav umangyadav mentioned this pull request Dec 4, 2023
Merged
pfultz2
pfultz2 reviewed Dec 5, 2023
View reviewed changes
src/targets/gpu/target.cpp Outdated Show resolved Hide resolved
pfultz2
pfultz2 reviewed Dec 5, 2023
View reviewed changes
src/eliminate_data_type.cpp Outdated Show resolved Hide resolved
pfultz2
pfultz2 reviewed Dec 5, 2023
View reviewed changes
src/eliminate_data_type.cpp Outdated Show resolved Hide resolved
pfultz2
pfultz2 reviewed Dec 5, 2023
View reviewed changes
src/eliminate_data_type.cpp Outdated Show resolved Hide resolved
@umangyadav umangyadav requested a review from pfultz2 December 5, 2023 15:05
@TedThemistokleous
Copy link
Collaborator

If we're not concerned about the hipRTC because thats expected right now. run looks good.

@umangyadav
Copy link
Member Author

Opend issue #2514 for the CK failures.

CK failure is not related to this PR. therefore disabling those for now.

@causten causten merged commit 6d0b6bc into develop Dec 6, 2023
44 checks passed
@causten causten deleted the rocblas_fp8 branch December 6, 2023 01:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@pfultz2 pfultz2 pfultz2 approved these changes

@TedThemistokleous TedThemistokleous TedThemistokleous approved these changes

Assignees

@umangyadav umangyadav

Labels
FP8 issues related to FP8 implemenation
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
5 participants
@umangyadav @migraphx-bot @TedThemistokleous @pfultz2 @causten