Python simple cuda #685

MaximEremenko · 2025-05-23T20:48:52Z

This pull request introduces a simplified Python interface for Type 3 NUFFTs (nonuniform-to-nonuniform) in 1D, 2D, and 3D within the cufinufft library. The new functions—nufft1d3, nufft2d3, and nufft3d3—allow users to perform Type 3 transforms with minimal setup, similar to the existing Type 1 and Type 2 interfaces.
Changes

Added nufft1d3, nufft2d3, and nufft3d3 to cufinufft/_simple.py, providing user-friendly access to Type 3 NUFFTs.
Updated cufinufft/init.py to include the new functions in the module’s public API.
Modified _invoke_plan to correctly handle Type 3 parameters (s, t, u) and plan creation for Type 3 transforms.

With this update, all NUFFT types (1, 2, and 3) now offer a consistent and simplified Python interface.

DiamonDinoia · 2025-05-24T20:19:07Z

This looks great.

I would ask @janden to review.

My suggestion it to extend https://github.com/flatironinstitute/finufft/blob/master/python/cufinufft/tests/test_simple.py with type3

…e3 with CPU version gen_coef_ind

DiamonDinoia · 2025-05-28T14:36:05Z

Hi @MaximEremenko,

I noticed the tests failing do you need help with it?

…789 * n_target_pts),) keep in mind n_tr

MaximEremenko · 2025-05-28T15:00:53Z

Hi @DiamonDinoia ,
I have version that can pass tests. Problem is here:
in function: verify_type3

ind = (int(0.1789 * n_target_pts),) and assert type3_rel_err < 100 * tol

Original CPU code looks a little different.

I made verify_type3 to ind = tuple(n-1 for n in n_tr + (int(0.1789 *n_pts),)) and it works but it looks like GPU version gives bigger error compare to CPU and general test case tuple(np.random.randint(0, n) for n in n_tr + (n_pts,)) does not work correctly.

ahbarnett · 2025-05-28T15:21:36Z

Really the python tests should not rely on the error of a single output (ind), since this can fluctuate too much. In the C++ and matlab/octave tests I updated this (eg for the latter see https://github.com/flatironinstitute/finufft/blob/master/matlab/test/fullmathtest.m ). They use small tests where computing the answer directly is cheap, then using relative l2-norm of difference.

…

On Wed, May 28, 2025 at 11:01 AM Maxim Eremenko ***@***.***> wrote: *MaximEremenko* left a comment (flatironinstitute/finufft#685) <#685 (comment)> Hi, I have version that can pass tests. Problem is here: in function: verify_type3 ind = (int(0.1789 * n_target_pts),) and assert type3_rel_err < 100 * tol Original CPU code looks a little different. I made verify_type3 to ind = tuple(n-1 for n in n_tr + (int(0.1789 *n_pts),)) and it works but it looks like GPU version gives bigger error compere to CPU and general test case tuple(np.random.randint(0, n) for n in n_tr + (n_pts,)) does not work correctly. — Reply to this email directly, view it on GitHub <#685 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACNZRSSOZFHEX2Y4PVGOAS33AXFTXAVCNFSM6AAAAAB5ZTXJB6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDSMJWGY3TENZXGM> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- *-------------------------------------------------------------------~^`^~._.~' |\ Alex Barnett Center for Computational Mathematics, Flatiron Institute | \ http://users.flatironinstitute.org/~ahb 646-876-5942

DiamonDinoia · 2025-05-28T18:24:01Z

Hi @MaximEremenko,

Yes, let's leave something sensible that passes here.
@janden might want to change the tests to align to C++/matlab in a different PR.

janden

Thanks for putting this together!

Some things to discuss here.

janden · 2025-06-10T18:53:15Z

python/cufinufft/cufinufft/_simple.py

@@ -32,10 +41,13 @@ def _invoke_plan(dim, nufft_type, x, y, z, data, out, isign, eps,
        n_modes = out.shape[-dim:]
    if nufft_type == 2:
        n_modes = data.shape[-dim:]
+
+    if nufft_type == 3:
+        plan = Plan(nufft_type, dim, n_trans = n_trans,  eps=eps, isign=isign, dtype=dtype, **kwargs)


Any reason to specify the kwargs explicitly here?

**kwargs are defined the same way as for nufft_type 1 and 2, matching the invoke_guru CPU implementation:
if tp == 3:
plan = Plan(tp, dim, n_trans, eps, isign, pdtype, **kwargs)
else:
plan = Plan(tp, n_modes, n_trans, eps, isign, pdtype, **kwargs)
Explicit specification of parameters (e.g., n_trans=n_trans, eps=eps, isign=isign, dtype=dtype) is optional. I can modify it if necessary

janden · 2025-06-10T18:53:30Z

python/cufinufft/cufinufft/_plan.py

@@ -332,7 +332,7 @@ def execute(self, data, out=None):

        _data, data_shape = _ensure_array_shape(_data, "data", req_data_shape,
                                                allow_reshape=True)
-        if self._type == 1:
+        if self._type in [1, 3]:


Good catch!

janden · 2025-06-10T18:55:33Z

python/cufinufft/cufinufft/_simple.py

@@ -46,7 +58,7 @@ def _invoke_plan(dim, nufft_type, x, y, z, data, out, isign, eps,


 def _get_ntrans(dim, nufft_type, data):
-    if nufft_type == 1:
+    if ((nufft_type == 1) | (nufft_type == 3)):


Would be preferable here to have if nufft_type == 1 or nufft_type == 3: or if nufft_type in [1, 3]: (see above). The | operator is a bitwise or.

Changed to if nufft_type in [1, 3]. Thanks

janden · 2025-06-10T18:55:56Z

python/cufinufft/tests/test_simple.py

+
+    target_coefs = to_cpu(target_coefs_gpu)
+
+    utils.verify_type3(source_pts, source_coefs, target_pts, target_coefs, tol)


Missing newline.

janden · 2025-06-10T18:57:21Z

python/cufinufft/tests/utils.py

@@ -22,6 +22,9 @@ def _real_dtype(complex_dtype):

    return real_dtype

+def gen_coef_ind(n_pts, n_tr):


I don't quite understand why this is needed here.

Consistent implementation with "def gen_coef_ind(n_pts, n_tr)" from python/finufft/tests/utils.py

janden · 2025-06-10T18:58:13Z

@janden might want to change the tests to align to C++/matlab in a different PR.

Agreed. Will make an issue.

…w line at end of file

MaximEremenko added 3 commits May 23, 2025 13:34

Add simplified interface for cuFINUFFT type 3

940aea8

Simplified type 3. Fixed syntex typo

eeb9d6a

Simplified type 3. Fixed 1D and 2D _invoke_plan input for type 3

a526c4e

DiamonDinoia requested a review from janden May 24, 2025 20:16

MaximEremenko added 2 commits May 27, 2025 16:04

Simplified type 3. Fix n_trans parameter for type 3 nufft

a439acf

Simplified type 3. Add test for cuda type 3 nufft. Unified verify_typ…

358c42e

…e3 with CPU version gen_coef_ind

Simplified type 3. Revert for GPU version tests verify_type3 (int(0.1…

180ce83

…789 * n_target_pts),) keep in mind n_tr

DiamonDinoia approved these changes May 28, 2025

View reviewed changes

DiamonDinoia assigned janden May 29, 2025

DiamonDinoia added this to the 2.5 milestone May 29, 2025

janden requested changes Jun 10, 2025

View reviewed changes

janden mentioned this pull request Jun 10, 2025

Align Python tests to C++/Matlab variants #695

Open

Fix logical condition in _get_ntrans function for cufinufft types. Ne…

5f12f5e

…w line at end of file

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Python simple cuda #685

Python simple cuda #685

MaximEremenko commented May 23, 2025

Uh oh!

DiamonDinoia commented May 24, 2025

Uh oh!

DiamonDinoia commented May 28, 2025

Uh oh!

MaximEremenko commented May 28, 2025 •

edited

Loading

Uh oh!

ahbarnett commented May 28, 2025 via email

Uh oh!

DiamonDinoia commented May 28, 2025

Uh oh!

janden left a comment

Uh oh!

janden Jun 10, 2025

Uh oh!

MaximEremenko Jun 10, 2025

Uh oh!

janden Jun 10, 2025

Uh oh!

janden Jun 10, 2025

Uh oh!

MaximEremenko Jun 10, 2025

Uh oh!

janden Jun 10, 2025

Uh oh!

janden Jun 10, 2025

Uh oh!

MaximEremenko Jun 10, 2025

Uh oh!

janden commented Jun 10, 2025

Uh oh!

Uh oh!


		target_coefs = to_cpu(target_coefs_gpu)

		utils.verify_type3(source_pts, source_coefs, target_pts, target_coefs, tol)

		@@ -22,6 +22,9 @@ def _real_dtype(complex_dtype):

		return real_dtype

		def gen_coef_ind(n_pts, n_tr):

Python simple cuda #685

Are you sure you want to change the base?

Python simple cuda #685

Conversation

MaximEremenko commented May 23, 2025

Uh oh!

DiamonDinoia commented May 24, 2025

Uh oh!

DiamonDinoia commented May 28, 2025

Uh oh!

MaximEremenko commented May 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ahbarnett commented May 28, 2025 via email

Uh oh!

DiamonDinoia commented May 28, 2025

Uh oh!

janden left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

janden commented Jun 10, 2025

Uh oh!

Uh oh!

MaximEremenko commented May 28, 2025 •

edited

Loading