Skip to content

add STFT_COMPLEX, derive stft_magnitude via graph composition#395

Merged
HenryNdubuaku merged 3 commits intocactus-compute:mainfrom
yujonglee:stft-complex
Feb 24, 2026
Merged

add STFT_COMPLEX, derive stft_magnitude via graph composition#395
HenryNdubuaku merged 3 commits intocactus-compute:mainfrom
yujonglee:stft-complex

Conversation

@yujonglee
Copy link
Copy Markdown
Contributor

@yujonglee yujonglee commented Feb 24, 2026

STFT Magnitude: 0.615ms
STFT Complex: 0.603ms

in my testing.

Signed-off-by: Yujong Lee <[email protected]>
@HenryNdubuaku
Copy link
Copy Markdown
Collaborator

@yujonglee thanks so much for this! maybe lets swap in the internal logic, of existing fused stft, so we only have one function, to avoid duplicates.

@yujonglee
Copy link
Copy Markdown
Contributor Author

Hi @HenryNdubuaku thanks for the review

Done in fe40d27.

Internal API:

  • cactus_stft_complex_f16 is the single kernel
  • stft_complex maps directly to it at the graph level
  • stft_magnitude is now just a graph-level composition
    • stft_complex + slice + multiply + add + sqrt (no separate kernel)

Public API:

  • Unchanged so nothing upstream breaks.

Let me know if I missed something!

@yujonglee yujonglee changed the title add STFT_COMPLEX add STFT_COMPLEX, derive stft_magnitude via graph composition Feb 24, 2026
Signed-off-by: HenryNdubuaku <[email protected]>
@HenryNdubuaku HenryNdubuaku merged commit cf4ea0b into cactus-compute:main Feb 24, 2026
2 of 4 checks passed
@yujonglee yujonglee deleted the stft-complex branch February 24, 2026 01:26
ncylich pushed a commit that referenced this pull request Feb 24, 2026
* add STFT_COMPLEX

Signed-off-by: Yujong Lee <[email protected]>

* change based on review

* Cleanup

Signed-off-by: HenryNdubuaku <[email protected]>

---------

Signed-off-by: Yujong Lee <[email protected]>
Signed-off-by: HenryNdubuaku <[email protected]>
Co-authored-by: HenryNdubuaku <[email protected]>
cattermelon1234 pushed a commit to cattermelon1234/cactus that referenced this pull request Feb 28, 2026
…-compute#395)

* add STFT_COMPLEX

Signed-off-by: Yujong Lee <[email protected]>

* change based on review

* Cleanup

Signed-off-by: HenryNdubuaku <[email protected]>

---------

Signed-off-by: Yujong Lee <[email protected]>
Signed-off-by: HenryNdubuaku <[email protected]>
Co-authored-by: HenryNdubuaku <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants