bidiphase.shift: copy the source slice before the overlapping in-place assignment

bidiphase.shift shifts the odd scan lines with an in-place assignment whose
source and destination slices overlap. numpy handles overlapping assignments
by making a temporary copy, but torch.Tensor.copy_ does not, so since the
registration path started passing torch tensors (v1.0.0.1) every odd line
was corrupted instead of shifted: with bidiphase=3, 34% of odd-line pixels
in register_frames' output differ from the intended shift (they contain
repeated copies of the first columns). The reference image, which still
goes through the numpy path in registration_wrapper, was shifted correctly,
so frames and reference disagreed.

Copy the source slice first (clone for torch, copy for numpy), which
restores the numpy semantics for both array types.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012d2sJAD5EUp4GqAStqoBX7

Author: claude

suite2p/registration/bidiphase.py

@@ -3,6 +3,7 @@
 """
 import numpy as np
 from numpy import fft
+import torch
 
 
 def compute(frames: np.ndarray) -> int:
@@ -60,8 +61,16 @@ def shift(frames: np.ndarray, bidiphase: int) -> None:
     frames : np.ndarray
         The input frames with odd lines shifted.
     """
+    # The source and destination slices overlap. numpy makes a temporary copy
+    # for overlapping assignments, but torch.Tensor.copy_ does not, so on a
+    # torch tensor the in-place assignment reads already-overwritten pixels
+    # and corrupts the odd lines. Copy the source first for both array types.
     if bidiphase > 0:
-        frames[:, 1::2, bidiphase:] = frames[:, 1::2, :-bidiphase]
+        src = frames[:, 1::2, :-bidiphase]
+        src = src.clone() if isinstance(frames, torch.Tensor) else src.copy()
+        frames[:, 1::2, bidiphase:] = src
     elif bidiphase < 0:
-        frames[:, 1::2, :bidiphase] = frames[:, 1::2, -bidiphase:]
+        src = frames[:, 1::2, -bidiphase:]
+        src = src.clone() if isinstance(frames, torch.Tensor) else src.copy()
+        frames[:, 1::2, :bidiphase] = src
     return frames