Merge pull request #1216 from MouseLand/fix_mps_gui

Fix GUI Issues with mac machines and NWB warning issue.

Author: carsen-stringer

suite2p/__main__.py

@@ -51,16 +51,7 @@ def main():
             logging.exception(f'fatal error in {"run_plane" if args.single_plane else "run_s2p"}:')
             raise
 
-    else:
-        # Check if the OS is macOS and the machine is Apple Silicon (ARM-based)
-        if platform.system() == "Darwin" and 'arm' in platform.processor().lower():
-            # Set the number of threads for OpenMP and OpenBLAS
-            os.environ["OMP_NUM_THREADS"] = "1"
-            os.environ["OPENBLAS_NUM_THREADS"] = "1"
-            print("Environment set to use 1 thread for OpenMP and OpenBLAS (Apple Silicon macOS).")
-        else:
-            print("Not macOS on Apple Silicon, proceeding without limiting threads.")
-            
+    else:   
         from suite2p import gui
         gui.run()#statfile="C:/DATA/exs2p/suite2p/plane0/stat.npy")
 

suite2p/extraction/extract.py

@@ -57,21 +57,22 @@ def extract_traces(f_in, cell_masks, neuropil_masks, batch_size=500,
     if device.type == 'mps':
         device = torch.device('cpu')
 
-    npix_neuropil = torch.Tensor([len(nm) for nm in neuropil_masks]).to(device)
-    # create coo tensor of neuropil and cell masks 
-    ccol_indices = [m for nm in neuropil_masks for m in nm]
-    row_indices = [k for k in range(len(neuropil_masks)) for m in neuropil_masks[k]]
-    inds = torch.Tensor([ccol_indices, row_indices]).to(device)
-    # convert to csc (tried creating csc directly but it was slow)
-    nmasks = torch.sparse_coo_tensor(inds, torch.ones(len(row_indices), device=device), 
-                                     size=(Ly*Lx, ncells))
-    nmasks = nmasks.to_sparse_csc()
+    if neuropil_masks is not None:
+        npix_neuropil = torch.Tensor([len(nm) for nm in neuropil_masks]).to(device)
+        # create coo tensor of neuropil masks
+        ccol_indices = [m for nm in neuropil_masks for m in nm]
+        row_indices = [k for k in range(len(neuropil_masks)) for m in neuropil_masks[k]]
+        inds = torch.Tensor([ccol_indices, row_indices]).to(device)
+        # convert to csc (tried creating csc directly but it was slow)
+        nmasks = torch.sparse_coo_tensor(inds, torch.ones(len(row_indices), device=device),
+                                         size=(Ly*Lx, ncells))
+        nmasks = nmasks.to_sparse_csc()
 
     ccol_indices = [m for cm in cell_masks for m in cm[0]]
     row_indices = [k for k in range(len(cell_masks)) for m in cell_masks[k][0]]
     cell_lam = torch.Tensor([l for cm in cell_masks for l in cm[1]]).to(device)
     inds = torch.Tensor([ccol_indices, row_indices]).to(device)
-    cmasks = torch.sparse_coo_tensor(inds, cell_lam, 
+    cmasks = torch.sparse_coo_tensor(inds, cell_lam,
                                      size=(Ly*Lx, ncells))
     cmasks = cmasks.to_sparse_csc()
 
@@ -87,11 +88,12 @@ def extract_traces(f_in, cell_masks, neuropil_masks, batch_size=500,
         tstart, tend = n * batch_size, min((n+1) * batch_size, n_frames)
         data = torch.from_numpy(f_in[tstart : tend]).to(device)
         data = data.reshape(-1, Ly*Lx).float()
-        
-        Fneu_batch = (data @ nmasks) / npix_neuropil
-        Fneu[:, tstart : tend] = Fneu_batch.T.cpu().numpy()
-        
-        F_batch = data @ cmasks 
+
+        if neuropil_masks is not None:
+            Fneu_batch = (data @ nmasks) / npix_neuropil
+            Fneu[:, tstart : tend] = Fneu_batch.T.cpu().numpy()
+
+        F_batch = data @ cmasks
         F[:, tstart : tend] = F_batch.T.cpu().numpy()
 
     return F, Fneu

suite2p/gui/menus.py

@@ -3,7 +3,7 @@
 """
 from qtpy import QtGui
 from qtpy.QtWidgets import QAction, QMenu
-from pkg_resources import iter_entry_points
+from importlib.metadata import entry_points
 
 from . import reggui, drawroi, merge, io, rungui, visualize, classgui
 from suite2p.io.nwb import save_nwb
@@ -166,7 +166,13 @@ def plugins(parent):
     main_menu = parent.menuBar()
     parent.plugins = {}
     plugin_menu = main_menu.addMenu("&Plugins")
-    for entry_pt in iter_entry_points(group="suite2p.plugin", name=None):
+    try:
+        # Works for python 3.12+
+        suite2p_plugins = entry_points(group="suite2p.plugin")
+    except TypeError:
+        # works for Python 3.9-3.11
+        suite2p_plugins = entry_points().get("suite2p.plugin", [])
+    for entry_pt in suite2p_plugins:
         plugin_obj = entry_pt.load()  # load the advertised class from entry_points
         parent.plugins[entry_pt.name] = plugin_obj(
             parent

suite2p/gui/rungui_utils.py

@@ -115,25 +115,67 @@ def setup_logger(name):
     return logger
 
 
-class Suite2pWorker(QtCore.QThread):
+class Suite2pWorker(QtCore.QObject):
+    """Worker that runs suite2p in a separate process to avoid QThread stack limitations on macOS."""
     finished = QtCore.Signal(str)
-    
+
     def __init__(self, parent, db_file, settings_file):
         super(Suite2pWorker, self).__init__()
         self.db_file = db_file
         self.settings_file = settings_file
         self.parent = parent
-        # self.logHandler = ThreadLogger()
-        
-    def run(self):
-        db = np.load(self.db_file, allow_pickle=True).item()
-        settings = np.load(self.settings_file, allow_pickle=True).item()
-        
-        try:
-            logger_setup(get_save_folder(db))
-            run_s2p(db=db, settings=settings)
+        self.process = None
+
+    def start(self):
+        """Start suite2p in a separate process using QProcess."""
+        self.process = QtCore.QProcess()
+        self.process.setProcessChannelMode(QtCore.QProcess.MergedChannels)
+        self.process.readyReadStandardOutput.connect(self._on_output)
+        self.process.finished.connect(self._on_finished)
+
+        # Create a Python script to run suite2p
+        script = f'''
+import numpy as np
+from suite2p.run_s2p import logger_setup, run_s2p, get_save_folder
+
+db = np.load("{self.db_file}", allow_pickle=True).item()
+settings = np.load("{self.settings_file}", allow_pickle=True).item()
+
+logger_setup(get_save_folder(db))
+run_s2p(db=db, settings=settings)
+'''
+        self.process.start(sys.executable, ["-c", script])
+
+    def _on_output(self):
+        """Handle output from the subprocess."""
+        if self.process:
+            data = self.process.readAllStandardOutput()
+            text = bytes(data).decode("utf-8", errors="replace")
+            print(text, end="")
+
+    def _on_finished(self, exit_code, exit_status):
+        """Handle process completion."""
+        if exit_code == 0:
             self.finished.emit("finished")
-        except Exception as e:
-            print("ERROR:", e)
-            traceback.print_exc()
-            self.finished.emit("error")
+        else:
+            self.finished.emit("error")
+
+    def terminate(self):
+        """Terminate the subprocess if running."""
+        if self.process and self.process.state() != QtCore.QProcess.NotRunning:
+            self.process.terminate()
+
+    def quit(self):
+        """Stop the subprocess (alias for terminate, for QThread compatibility)."""
+        self.terminate()
+
+    def wait(self):
+        """Wait for the process to finish (for compatibility)."""
+        if self.process:
+            self.process.waitForFinished(-1)
+
+    def isRunning(self):
+        """Check if the process is still running (for QThread compatibility)."""
+        if self.process:
+            return self.process.state() == QtCore.QProcess.Running
+        return False

suite2p/io/nwb.py

@@ -54,6 +54,8 @@ def _load_npy_cross_platform(path):
     NWB = True
 except ModuleNotFoundError:
     NWB = False
+    logger.warning("pynwb not installed, save_nwb, read_nwb, and nwb_to_binary "
+                   "will not work. Install with: pip install pynwb")
 
 
 def nwb_to_binary(settings):

suite2p/registration/nonrigid.py

@@ -443,11 +443,19 @@ def transform_data(data, nblocks, xblock, yblock, ymax1, xmax1,
     yxup = yxup.permute(0, 2, 3, 1)
 
     if device.type == "mps":
-        # Manually pad the input tensor with the border values
+        # Manually pad the input tensor with the border values.
         data_padded = F.pad(data.float().unsqueeze(1), (1, 1, 1, 1), mode="replicate")
-        height, width = data.shape[-2:]  # Get the height and width of the original data tensor
-        # Adjust the grid to account for the padding
-        adjusted_yxup = yxup + torch.tensor([[[[1 / width, 1 / height]]]]).to(yxup.device)  # Adjust grid
+        # Get the height and width of the original data tensor
+        height, width = data.shape[-2:] 
+        # Scale the grid to account for the padding. Padded data is now of shape (width + 2) x (height + 2). 
+        # Scale_x and scale_y adjust so we exclude the padding. Align_corner is set to true so original image width is width -1. Same for the height.
+        scale_x = (width - 1) / (width + 1) 
+        scale_y = (height - 1) / (height + 1)
+        # Scale the padded image to be within the right coordinates for sampling
+        adjusted_yxup = yxup * torch.tensor([[[[scale_x, scale_y]]]]).to(yxup.device)
+        # Clamp the grid before subsampling as all coordinate values must lie between [-1,1]. 
+        # Sampling should always be along the image (not include padding coordinates, which will exceed [-1,1] range).
+        adjusted_yxup = torch.clamp(adjusted_yxup, -1, 1)
         # Perform grid sampling on the padded tensor
         fr_shift = F.grid_sample(
             data_padded,
@@ -460,5 +468,5 @@ def transform_data(data, nblocks, xblock, yblock, ymax1, xmax1,
         fr_shift = F.grid_sample(data.float().unsqueeze(1), yxup[:,:,:,[1,0]], 
                              mode="bilinear", padding_mode="border", align_corners=True)
 
-    
+
     return fr_shift.squeeze().short()#.cpu().numpy()

suite2p/registration/register.py

@@ -275,16 +275,24 @@ def compute_filters_and_norm(refImg, norm_frames=True, spatial_smooth=1.15, spat
         maskMul, maskOffset, cfRefImg = rigid.compute_masks_ref_smooth_fft(refImg=rimg, maskSlope=spatial_taper,
                                                                  smooth_sigma=spatial_smooth)
         Ly, Lx = refImg.shape
+        # MPS backend does not support float64, convert to float32
+        if device.type == "mps":
+            maskMul, maskOffset = maskMul.to(torch.float32), maskOffset.to(torch.float32)
+            cfRefImg = cfRefImg.to(torch.complex64)
         maskMul, maskOffset = maskMul.to(device), maskOffset.to(device)
         cfRefImg = cfRefImg.to(device)
         blocks = []
         if block_size is not None:
             blocks = nonrigid.make_blocks(Ly=Ly, Lx=Lx, block_size=block_size,
                                           lpad=lpad, subpixel=subpixel)
             maskMulNR, maskOffsetNR, cfRefImgNR = nonrigid.compute_masks_ref_smooth_fft(
-                refImg0=rimg, maskSlope=spatial_taper, smooth_sigma=spatial_smooth, 
+                refImg0=rimg, maskSlope=spatial_taper, smooth_sigma=spatial_smooth,
                 yblock=blocks[0], xblock=blocks[1],
             )
+            # MPS backend does not support float64, convert to float32
+            if device.type == "mps":
+                maskMulNR, maskOffsetNR = maskMulNR.to(torch.float32), maskOffsetNR.to(torch.float32)
+                cfRefImgNR = cfRefImgNR.to(torch.complex64)
             maskMulNR, maskOffsetNR = maskMulNR.to(device), maskOffsetNR.to(device)
             cfRefImgNR = cfRefImgNR.to(device)
 
@@ -440,6 +448,10 @@ def shift_frames(fr_torch, yoff, xoff, yoff1=None, xoff1=None, blocks=None,
             if fr_torch.device.type == "cuda":
                 yoff1 = torch.from_numpy(yoff1).pin_memory().to(device)
                 xoff1 = torch.from_numpy(xoff1).pin_memory().to(device)
+            elif device.type == "mps":
+                # MPS backend does not support float64
+                yoff1 = torch.from_numpy(yoff1).to(torch.float32).to(device)
+                xoff1 = torch.from_numpy(xoff1).to(torch.float32).to(device)
             else:
                 yoff1 = torch.from_numpy(yoff1).to(device)
                 xoff1 = torch.from_numpy(xoff1).to(device)
@@ -580,7 +592,9 @@ def register_frames(f_align_in, refImg, f_align_out=None, batch_size=100,
     if upsample_meanImg:
         if not isinstance(upsample_meanImg, (np.ndarray, list, tuple)):
             upsample_meanImg = [upsample_meanImg, upsample_meanImg]
-        mean_img_ups = torch.zeros((int(Ly*upsample_meanImg[0]), int(Lx*upsample_meanImg[1])), dtype=torch.double, device=device)
+        # MPS backend does not support float64
+        ups_dtype = torch.float32 if device.type == "mps" else torch.double
+        mean_img_ups = torch.zeros((int(Ly*upsample_meanImg[0]), int(Lx*upsample_meanImg[1])), dtype=ups_dtype, device=device)
         counts_ups = torch.zeros((int(Ly*upsample_meanImg[0]), int(Lx*upsample_meanImg[1])), dtype=torch.int, device=device)
     else:
         mean_img_ups, counts_ups, meanImg_ups = None, None, None
@@ -890,7 +904,10 @@ def registration_wrapper(f_reg, f_raw=None, f_reg_chan2=None, f_raw_chan2=None,
 
     nchannels = 2 if f_alt_in is not None else 1
     logger.info(f"registering {nchannels} channels")
-    
+    if device.type == "mps":
+        logger.warning("MPS device does not support float64, using float32 for registration. "
+                       "If you encounter registration issues, try using cuda or cpu instead.")
+
     ### ----- compute reference image and bidiphase shift -------------- ###
     n_frames, Ly, Lx = f_align_in.shape
     badframes0 = np.zeros(n_frames, "bool") if badframes is None else badframes.copy()

tests/test_registration.py

@@ -1,5 +1,8 @@
 import numpy as np
+import pytest
+import torch
 from suite2p.registration import bidiphase
+from suite2p.registration.nonrigid import transform_data
 
 
 def test_positive_bidiphase_shift_shifts_every_other_line():
@@ -41,4 +44,34 @@ def test_negative_bidiphase_shift_shifts_every_other_line():
 
     shifted = orig.copy()
     bidiphase.shift(shifted, -2)
-    assert np.allclose(shifted, expected)
+    assert np.allclose(shifted, expected)
+
+
+@pytest.mark.skipif(not torch.backends.mps.is_available(), reason="MPS not available")
+def test_transform_data_mps_cpu_consistency():
+    """Test that MPS and CPU code paths in transform_data produce similar results."""
+    from suite2p.registration.nonrigid import make_blocks
+
+    np.random.seed(42)
+    torch.manual_seed(42)
+    Ly, Lx, n_frames = 128, 128, 2
+    yblock, xblock, nblocks, *_ = make_blocks(Ly, Lx, (32, 32))
+    data_np = np.random.rand(n_frames, Ly, Lx).astype(np.float32) * 100
+    ymax1 = torch.randn(nblocks[0] * nblocks[1], n_frames) * 2
+    xmax1 = torch.randn(nblocks[0] * nblocks[1], n_frames) * 2
+
+    result_cpu = transform_data(
+        torch.from_numpy(data_np), nblocks, xblock, yblock, ymax1.clone(), xmax1.clone()
+    )
+    result_mps = transform_data(
+        torch.from_numpy(data_np).to("mps"), nblocks, xblock, yblock,
+        ymax1.clone().to("mps"), xmax1.clone().to("mps")
+    )
+
+    cpu_np = result_cpu.numpy().astype(np.float32)
+    mps_np = result_mps.cpu().numpy().astype(np.float32)
+    correlation = np.corrcoef(cpu_np.flatten(), mps_np.flatten())[0, 1]
+    max_diff = np.abs(cpu_np - mps_np).max()
+
+    assert correlation > 0.99, f"Correlation: {correlation}"
+    assert max_diff < 2, f"Max diff: {max_diff}"