Workaround for cuFFT strides bug

Author: vlad-perevezentsev

dpnp/fft/dpnp_utils_fft.py

@@ -400,13 +400,26 @@ def _fft(a, norm, out, forward, in_place, c2c, axes, batch_fft=True):
         if dpnp.is_cuda_backend(a) and not a.flags.c_contiguous:
             a = dpnp.ascontiguousarray(a)
 
+    # w/a for cuFFT to avoid "Invalid strides" error when
+    # the last dimension is 1 and there are multiple axes
+    # by swapping the last two axes to correct the input.
+    # TODO: Remove this ones the OneMath issue is resolved
+    # https://github.com/uxlfoundation/oneMath/issues/631
+    cufft_wa = dpnp.is_cuda_backend(a) and a.shape[-1] == 1 and len(axes) > 1
+    if cufft_wa:
+        a = dpnp.moveaxis(a, -1, -2)
+
     a_strides = _standardize_strides_to_nonzero(a.strides, a.shape)
     dsc, out_strides = _commit_descriptor(
         a, forward, in_place, c2c, a_strides, index, batch_fft
     )
     res = _compute_result(dsc, a, out, forward, c2c, out_strides)
     res = _scale_result(res, a.shape, norm, forward, index)
 
+    # Revert swapped axes
+    if cufft_wa:
+        res = dpnp.moveaxis(res, -1, -2)
+
     if batch_fft:
         tmp_shape = a_shape_orig[:-1] + (res.shape[-1],)
         res = dpnp.reshape(res, tmp_shape)