Mac support

[albertgong1]

I tried installing falkon on a MacBook Pro 14 (M2 Pro) from source. It installs without any error, but during runtime, I run into the following error (see call stack) when running fit(). Is support for Mac planned?

---------------------------------------------------------------------------
OSError                                   Traceback (most recent call last)
[/Users/ag2435/repos/falkon/notebooks/FalkonRegression.ipynb](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/FalkonRegression.ipynb) Cell 11 line 1
----> [1](vscode-notebook-cell:/Users/ag2435/repos/falkon/notebooks/FalkonRegression.ipynb#X13sZmlsZQ%3D%3D?line=0) model.fit(Xtr, Ytr)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/models/falkon.py:229](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/models/falkon.py:229), in Falkon.fit(self, X, Y, Xts, Yts, warm_start)
    227     if self.weight_fn is not None:
    228         ny_weight_vec = self.weight_fn(Y[ny_indices], X[ny_indices], ny_indices)
--> 229     precond.init(ny_points, weight_vec=ny_weight_vec)
    231 if _use_cuda_mmv:
    232     # Cache must be emptied to ensure enough memory is visible to the optimizer
    233     torch.cuda.empty_cache()

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/preconditioner/flk_preconditioner.py:101](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/preconditioner/flk_preconditioner.py:101), in FalkonPreconditioner.init(self, X, weight_vec)
     99     else:  # If sparse tensor we need fortran for kernel calculation
    100         C = create_fortran((M, M), dtype=dtype, device=dev, pin_memory=self._use_cuda)
--> 101     self.kernel(X, X, out=C, opt=self.params)
    102 if not is_f_contig(C):
    103     C = C.T

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/kernels/kernel.py:173](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/kernels/kernel.py:173), in Kernel.__call__(self, X1, X2, diag, out, opt)
    171     params = dataclasses.replace(self.params, **dataclasses.asdict(opt))
    172 mm_impl = self._decide_mm_impl(X1, X2, diag, params)
--> 173 return mm_impl(self, params, out, diag, X1, X2)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:554](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:554), in fmm(kernel, opt, out, diag, X1, X2)
    551 import falkon.kernels
    553 if isinstance(kernel, falkon.kernels.DiffKernel):
--> 554     return KernelMmFnFull.apply(kernel, opt, out, diag, X1, X2, *kernel.diff_params.values())
    555 else:
    556     return KernelMmFnFull.apply(kernel, opt, out, diag, X1, X2)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/torch/autograd/function.py:506](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/torch/autograd/function.py:506), in Function.apply(cls, *args, **kwargs)
    503 if not torch._C._are_functorch_transforms_active():
    504     # See NOTE: [functorch vjp and autograd interaction]
    505     args = _functorch.utils.unwrap_dead_wrappers(args)
--> 506     return super().apply(*args, **kwargs)  # type: ignore[misc]
    508 if cls.setup_context == _SingleLevelFunction.setup_context:
    509     raise RuntimeError(
    510         'In order to use an autograd.Function with functorch transforms '
    511         '(vmap, grad, jvp, jacrev, ...), it must override the setup_context '
    512         'staticmethod. For more details, please see '
    513         'https://pytorch.org/docs/master/notes/extending.func.html')

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:480](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:480), in KernelMmFnFull.forward(ctx, kernel, opt, out, diag, X1, X2, *kernel_params)
    478     out = KernelMmFnFull.run_diag(X1, X2, out, kernel, False, is_sparse)
    479 elif comp_dev_type == "cpu" and data_dev.type == "cpu":
--> 480     out = KernelMmFnFull.run_cpu_cpu(X1, X2, out, kernel, comp_dtype, opt, False)
    481 elif comp_dev_type == "cuda" and data_dev.type == "cuda":
    482     out = KernelMmFnFull.run_gpu_gpu(X1, X2, out, kernel, comp_dtype, opt, False)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:354](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:354), in KernelMmFnFull.run_cpu_cpu(X1, X2, out, kernel, dtype, options, diff)
    342 @staticmethod
    343 def run_cpu_cpu(X1, X2, out, kernel, dtype, options, diff):
    344     args = ArgsFmm(
    345         X1=X1,
    346         X2=X2,
   (...)
    352         differentiable=diff,
    353     )
--> 354     out = _call_direct(mm_run_starter, (args, -1))
    355     return out

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/utils.py:86](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/utils.py:86), in _call_direct(target, arg)
     84 args_queue.put(arg[0])
     85 new_args_tuple = (-1, args_queue, arg[1])
---> 86 return target(*new_args_tuple)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:131](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:131), in mm_run_starter(proc_idx, queue, device_id)
    129     return sparse_mm_run_thread(X1, X2, out, kernel, n, m, computation_dtype, dev, tid=proc_idx)
    130 else:
--> 131     return mm_run_thread(X1, X2, out, kernel, n, m, computation_dtype, dev, tid=proc_idx)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:291](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/mmv_ops/fmm.py:291), in mm_run_thread(m1, m2, out, kernel, n, m, comp_dt, dev, tid)
    288 c_dev_out.fill_(0.0)
    290 # Compute kernel sub-matrix
--> 291 kernel.compute(c_dev_m1, c_dev_m2, c_dev_out, diag=False)
    293 # Copy back to host
    294 if has_gpu_bufs:

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/kernels/diff_kernel.py:91](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/kernels/diff_kernel.py:91), in DiffKernel.compute(self, X1, X2, out, diag)
     90 def compute(self, X1: torch.Tensor, X2: torch.Tensor, out: torch.Tensor, diag: bool):
---> 91     return self.core_fn(X1, X2, out, **self.diff_params, diag=diag, **self._other_params)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/kernels/distance_kernel.py:163](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/kernels/distance_kernel.py:163), in rbf_core(mat1, mat2, out, diag, sigma)
    161 mat1_div_sig = mat1 [/](https://file+.vscode-resource.vscode-cdn.net/) sigma
    162 mat2_div_sig = mat2 [/](https://file+.vscode-resource.vscode-cdn.net/) sigma
--> 163 norm_sq_mat1 = square_norm(mat1_div_sig, -1, True)  # b*n*1 or n*1
    164 norm_sq_mat2 = square_norm(mat2_div_sig, -1, True)  # b*m*1 or m*1
    166 out = _sq_dist(mat1_div_sig, mat2_div_sig, norm_sq_mat1, norm_sq_mat2, out)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/la_helpers/wrapper.py:129](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/la_helpers/wrapper.py:129), in square_norm(mat, dim, keepdim)
    128 def square_norm(mat: torch.Tensor, dim: int, keepdim: Optional[bool] = None) -> torch.Tensor:
--> 129     return c_ext.square_norm(mat, dim, keepdim)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/c_ext/__init__.py:15](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/c_ext/__init__.py:15), in _make_lazy_cuda_func.<locals>.call_cuda(*args, **kwargs)
     14 def call_cuda(*args, **kwargs):
---> 15     from ._backend import _assert_has_ext
     17     _assert_has_ext()
     18     return getattr(torch.ops.falkon, name)(*args, **kwargs)

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/c_ext/_backend.py:86](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/c_ext/_backend.py:86)
     84 lib_path = _get_extension_path("_C")
     85 try:
---> 86     torch.ops.load_library(lib_path)
     87 except OSError as e:
     88     # Hack: usually ld can't find torch_cuda_linalg.so which is in TORCH_LIB_PATH
     89     # if we load it first, then load_library will work.
     90     # TODO: This will only work on linux.
     91     if (missing_lib := lib_from_oserror(e)).startswith("libtorch_cuda_linalg"):

File [~/anaconda3/envs/falkon/lib/python3.10/site-packages/torch/_ops.py:643](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/site-packages/torch/_ops.py:643), in _Ops.load_library(self, path)
    638 path = _utils_internal.resolve_library_path(path)
    639 with dl_open_guard():
    640     # Import the shared library into the process, thus running its
    641     # static (global) initialization code in order to register custom
    642     # operators with the JIT.
--> 643     ctypes.CDLL(path)
    644 self.loaded_libraries.add(path)

File [~/anaconda3/envs/falkon/lib/python3.10/ctypes/__init__.py:374](https://file+.vscode-resource.vscode-cdn.net/Users/ag2435/repos/falkon/notebooks/~/anaconda3/envs/falkon/lib/python3.10/ctypes/__init__.py:374), in CDLL.__init__(self, name, mode, handle, use_errno, use_last_error, winmode)
    371 self._FuncPtr = _FuncPtr
    373 if handle is None:
--> 374     self._handle = _dlopen(self._name, mode)
    375 else:
    376     self._handle = handle

OSError: dlopen(/Users/ag2435/anaconda3/envs/falkon/lib/python3.10/site-packages/falkon/c_ext/_C.so, 0x0006): symbol not found in flat namespace '__ZN2at6native14lapackCholeskyIdEEvciPT_iPi'

[Giodiro]

Hi @ag2435
I'm not sure what's up honestly! I'll try to find a mac to test on in the next couple of weeks (sorry for the long timeline).
In the meantime, what version of falkon and of pytorch are you using?

[albertgong1]

I'm using the following versions:
falkon 0.8.4 pypi_0 pypi
pytorch 2.0.0 py3.10_0 pytorch

Thank you so much for looking into this!