Added ARM FFT Support #576
Merged
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
|
/blossom-ci |
cliffburdick
force-pushed
the
nvpl_fft
branch
2 times, most recently
from
efd570b to
5b059f2
Compare
|
/blossom-ci |
cliffburdick
force-pushed
the
nvpl_fft
branch
3 times, most recently
from
5f7628c to
b73be7d
Compare
|
/blossom-ci |
1 similar comment
|
/blossom-ci |
cliffburdick
force-pushed
the
nvpl_fft
branch
from
b73be7d to
50349a8
Compare
|
/blossom-ci |
cliffburdick
force-pushed
the
nvpl_fft
branch
from
50349a8 to
3435117
Compare
|
/blossom-ci |
cliffburdick
force-pushed
the
nvpl_fft
branch
3 times, most recently
from
4f4495e to
8996029
Compare
|
/blossom-ci |
tbensonatl
reviewed
Mar 11, 2024
tbensonatl
approved these changes
Mar 11, 2024
|
/blossom-ci |
|
/blossom-ci |
Switched all unit tests to use tuple with executor and type
|
/blossom-ci |
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Outdated suggestions cannot be applied.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.
This PR adds initial support for FFTs on ARM using NVPL. NVPL follows the FFTW interface, which prevents a few features cuFFT has. Most notable half-precision support is missing and FFTW requires the data pointer to be part of the plan. This makes caching somewhat pointless since the same size FFT with different pointers cannot reuse a plan.
Included in the PR is full unit test coverage for all FFT tests that had fp32/fp64 support. Once the NVPL API evolves to possible allow plans without pointers we can rewrite this to use the cuFFT method of having plans based on 1D/2D classes.
You may also notice a seemingly unrelated change in conv2d. This is an nvcc compiler bug that only affects ARM. This change will be reverted once the compiler bug is addressed
This commit also moves all unit tests to the tuple format so we can switch executors per test quickly.