imops 0.8.2__cp36-cp36m-win32.whl → 0.8.3__cp36-cp36m-win32.whl

imops/src/_numeric.pyx

@@ -8,10 +8,19 @@
 import numpy as np
 
 cimport numpy as np
+from libc.stdint cimport uint16_t
 
 from cython.parallel import prange
 
 
+# https://stackoverflow.com/questions/47421443/using-half-precision-numpy-floats-in-cython
+cdef extern from "numpy/halffloat.h":
+    ctypedef uint16_t npy_half
+
+    float npy_half_to_float(npy_half h) nogil
+    npy_half npy_float_to_half(float f) nogil
+
+
 ctypedef fused NUM:
     short
     int
@@ -20,45 +29,214 @@ ctypedef fused NUM:
     double
 
 
-def _parallel_sum(NUM[:] nums, Py_ssize_t num_threads) -> NUM:
-    cdef NUM res = 0
-    cdef Py_ssize_t i, len_nums = len(nums)
+ctypedef fused NUM_AND_NPY_HALF:
+    NUM
+    npy_half
 
-    for i in prange(len_nums, num_threads=num_threads, nogil=True):
-        res += nums[i]
 
-    return res
-
-
-def _parallel_pointwise_mul(
+# TODO: Generalize code below to n-d
+def _pointwise_add_array_3d(
     NUM[:, :, :] nums1,
     NUM[:, :, :] nums2,
-    Py_ssize_t[:] res_shape,
-    Py_ssize_t num_threads
+    NUM[:, :, :] out,
+    Py_ssize_t num_threads,
 ) -> np.ndarray:
-    cdef NUM[:, :, ::1] contiguous_nums1 = np.ascontiguousarray(nums1), contiguous_nums2 = np.ascontiguousarray(nums2)
-    cdef Py_ssize_t rows = res_shape[0], cols = res_shape[1], dims = res_shape[2]
+    cdef Py_ssize_t rows = out.shape[0], cols = out.shape[1], dims = out.shape[2]
+    cdef Py_ssize_t i, j, k
+
+    for i in prange(rows, nogil=True, num_threads=num_threads):
+        for j in prange(cols):
+            for k in prange(dims):
+                out[i, j, k] = nums1[i, j, k] + nums2[i, j, k]
+
+    return np.asarray(out)
 
-    cdef char[:] broadcast_mask1 = np.array([x == y for x, y in zip(res_shape, nums1.shape)], dtype=np.int8)
-    cdef char[:] broadcast_mask2 = np.array([x == y for x, y in zip(res_shape, nums2.shape)], dtype=np.int8)
 
-    cdef NUM[:, :, ::1] mul = np.empty_like(nums1, shape=res_shape)
+def _pointwise_add_array_4d(
+    NUM[:, :, :, :] nums1,
+    NUM[:, :, :, :] nums2,
+    NUM[:, :, :, :] out,
+    Py_ssize_t num_threads,
+) -> np.ndarray:
+    cdef Py_ssize_t dim1 = out.shape[0], dim2 = out.shape[1], dim3 = out.shape[2], dim4 = out.shape[3]
+    cdef Py_ssize_t i1, i2, i3, i4
+
+    for i1 in prange(dim1, nogil=True, num_threads=num_threads):
+        for i2 in prange(dim2):
+            for i3 in prange(dim3):
+                for i4 in prange(dim4):
+                    out[i1, i2, i3, i4] = nums1[i1, i2, i3, i4] + nums2[i1, i2, i3, i4]
+
+    return np.asarray(out)
+
+
+def _pointwise_add_value_3d(
+    NUM[:, :, :] nums,
+    NUM value,
+    NUM[:, :, :] out,
+    Py_ssize_t num_threads,
+) -> np.ndarray:
+    cdef Py_ssize_t rows = out.shape[0], cols = out.shape[1], dims = out.shape[2]
     cdef Py_ssize_t i, j, k
 
     for i in prange(rows, nogil=True, num_threads=num_threads):
         for j in prange(cols):
             for k in prange(dims):
-                mul[i, j, k] = (
-                    contiguous_nums1[
-                        i * broadcast_mask1[0],
-                        j * broadcast_mask1[1],
-                        k * broadcast_mask1[2]
-                    ] *
-                    contiguous_nums2[
-                        i * broadcast_mask2[0],
-                        j * broadcast_mask2[1],
-                        k * broadcast_mask2[2]
-                    ]
-                )
-
-    return np.asarray(mul)
+                out[i, j, k] = nums[i, j, k] + value
+
+    return np.asarray(out)
+
+
+def _pointwise_add_value_4d(
+    NUM[:, :, :, :] nums,
+    NUM value,
+    NUM[:, :, :, :] out,
+    Py_ssize_t num_threads,
+) -> np.ndarray:
+    cdef Py_ssize_t dim1 = out.shape[0], dim2 = out.shape[1], dim3 = out.shape[2], dim4 = out.shape[3]
+    cdef Py_ssize_t i1, i2, i3, i4
+
+    for i1 in prange(dim1, nogil=True, num_threads=num_threads):
+        for i2 in prange(dim2):
+            for i3 in prange(dim3):
+                for i4 in prange(dim4):
+                    out[i1, i2, i3, i4] = nums[i1, i2, i3, i4] + value
+
+    return np.asarray(out)
+
+
+def _pointwise_add_array_3d_fp16(
+    npy_half[:, :, :] nums1,
+    npy_half[:, :, :] nums2,
+    npy_half[:, :, :] out,
+    Py_ssize_t num_threads,
+) -> np.ndarray:
+    cdef Py_ssize_t rows = out.shape[0], cols = out.shape[1], dims = out.shape[2]
+    cdef Py_ssize_t i, j, k
+
+    for i in prange(rows, nogil=True, num_threads=num_threads):
+        for j in prange(cols):
+            for k in prange(dims):
+                out[i, j, k] = (npy_float_to_half(npy_half_to_float(nums1[i, j, k]) +
+                                npy_half_to_float(nums2[i, j, k])))
+
+    return np.asarray(out)
+
+
+def _pointwise_add_array_4d_fp16(
+    npy_half[:, :, :, :] nums1,
+    npy_half[:, :, :, :] nums2,
+    npy_half[:, :, :, :] out,
+    Py_ssize_t num_threads,
+) -> np.ndarray:
+    cdef Py_ssize_t dim1 = out.shape[0], dim2 = out.shape[1], dim3 = out.shape[2], dim4 = out.shape[3]
+    cdef Py_ssize_t i1, i2, i3, i4
+
+    for i1 in prange(dim1, nogil=True, num_threads=num_threads):
+        for i2 in prange(dim2):
+            for i3 in prange(dim3):
+                for i4 in prange(dim4):
+                    out[i1, i2, i3, i4] = (npy_float_to_half(npy_half_to_float(nums1[i1, i2, i3, i4]) +
+                                           npy_half_to_float(nums2[i1, i2, i3, i4])))
+
+    return np.asarray(out)
+
+
+def _pointwise_add_value_3d_fp16(
+    npy_half[:, :, :] nums,
+    npy_half value,
+    npy_half[:, :, :] out,
+    Py_ssize_t num_threads,
+) -> np.ndarray:
+    cdef Py_ssize_t rows = out.shape[0], cols = out.shape[1], dims = out.shape[2]
+    cdef Py_ssize_t i, j, k
+
+    for i in prange(rows, nogil=True, num_threads=num_threads):
+        for j in prange(cols):
+            for k in prange(dims):
+                out[i, j, k] = npy_float_to_half(npy_half_to_float(nums[i, j, k]) + npy_half_to_float(value))
+
+    return np.asarray(out)
+
+
+def _pointwise_add_value_4d_fp16(
+    npy_half[:, :, :, :] nums,
+    npy_half value,
+    npy_half[:, :, :, :] out,
+    Py_ssize_t num_threads,
+) -> np.ndarray:
+    cdef Py_ssize_t dim1 = out.shape[0], dim2 = out.shape[1], dim3 = out.shape[2], dim4 = out.shape[3]
+    cdef Py_ssize_t i1, i2, i3, i4
+
+    for i1 in prange(dim1, nogil=True, num_threads=num_threads):
+        for i2 in prange(dim2):
+            for i3 in prange(dim3):
+                for i4 in prange(dim4):
+                    out[i1, i2, i3, i4] = (npy_float_to_half(npy_half_to_float(nums[i1, i2, i3, i4]) +
+                                           npy_half_to_float(value)))
+
+    return np.asarray(out)
+
+
+def _fill_3d(NUM_AND_NPY_HALF[:, :, :] nums, NUM_AND_NPY_HALF value, Py_ssize_t num_threads) -> None:
+    cdef Py_ssize_t rows = nums.shape[0], cols = nums.shape[1], dims = nums.shape[2]
+    cdef Py_ssize_t i, j, k
+
+    for i in prange(rows, nogil=True, num_threads=num_threads):
+        for j in prange(cols):
+            for k in prange(dims):
+                nums[i, j, k] = value
+
+
+def _fill_4d(NUM_AND_NPY_HALF[:, :, :, :] nums, NUM_AND_NPY_HALF value, Py_ssize_t num_threads) -> None:
+    cdef Py_ssize_t dim1 = nums.shape[0], dim2 = nums.shape[1], dim3 = nums.shape[2], dim4 = nums.shape[3]
+    cdef Py_ssize_t i1, i2, i3, i4
+
+    for i1 in prange(dim1, nogil=True, num_threads=num_threads):
+        for i2 in prange(dim2):
+            for i3 in prange(dim3):
+                for i4 in prange(dim4):
+                    nums[i1, i2, i3, i4] = value
+
+
+# FIXME: somehow `const NUM_AND_NPY_HALF` is not working
+cpdef void _copy_3d(const NUM[:, :, :] nums1, NUM[:, :, :] nums2, Py_ssize_t num_threads):
+    cdef Py_ssize_t rows = nums1.shape[0], cols = nums1.shape[1], dims = nums1.shape[2]
+    cdef Py_ssize_t i, j, k
+
+    for i in prange(rows, nogil=True, num_threads=num_threads):
+        for j in prange(cols):
+            for k in prange(dims):
+                nums2[i, j, k] = nums1[i, j, k]
+
+
+cpdef void _copy_4d(const NUM[:, :, :, :] nums1, NUM[:, :, :, :] nums2, Py_ssize_t num_threads):
+    cdef Py_ssize_t dim1 = nums1.shape[0], dim2 = nums1.shape[1], dim3 = nums1.shape[2], dim4 = nums1.shape[3]
+    cdef Py_ssize_t i1, i2, i3, i4
+
+    for i1 in prange(dim1, nogil=True, num_threads=num_threads):
+        for i2 in prange(dim2):
+            for i3 in prange(dim3):
+                for i4 in prange(dim4):
+                    nums2[i1, i2, i3, i4] = nums1[i1, i2, i3, i4]
+
+
+cpdef void _copy_3d_fp16(const npy_half[:, :, :] nums1, npy_half[:, :, :] nums2, Py_ssize_t num_threads):
+    cdef Py_ssize_t rows = nums1.shape[0], cols = nums1.shape[1], dims = nums1.shape[2]
+    cdef Py_ssize_t i, j, k
+
+    for i in prange(rows, nogil=True, num_threads=num_threads):
+        for j in prange(cols):
+            for k in prange(dims):
+                nums2[i, j, k] = nums1[i, j, k]
+
+
+cpdef void _copy_4d_fp16(const npy_half[:, :, :, :] nums1, npy_half[:, :, :, :] nums2, Py_ssize_t num_threads):
+    cdef Py_ssize_t dim1 = nums1.shape[0], dim2 = nums1.shape[1], dim3 = nums1.shape[2], dim4 = nums1.shape[3]
+    cdef Py_ssize_t i1, i2, i3, i4
+
+    for i1 in prange(dim1, nogil=True, num_threads=num_threads):
+        for i2 in prange(dim2):
+            for i3 in prange(dim3):
+                for i4 in prange(dim4):
+                    nums2[i1, i2, i3, i4] = nums1[i1, i2, i3, i4]

imops/utils.py

@@ -1,39 +1,92 @@
 import os
+from contextlib import contextmanager
 from itertools import permutations
 from typing import Callable, Optional, Sequence, Tuple, Union
 from warnings import warn
 
 import numpy as np
 
-from .backend import BACKEND2NUM_THREADS_VAR_NAME, SINGLE_THREADED_BACKENDS, Backend
+from .backend import BACKEND_NAME2ENV_NUM_THREADS_VAR_NAME, SINGLE_THREADED_BACKENDS, Backend
 
 
 AxesLike = Union[int, Sequence[int]]
 AxesParams = Union[float, Sequence[float]]
 
 ZOOM_SRC_DIM = 4
+# TODO: define imops-specific environment variable like `OMP_NUM_THREADS`?
+IMOPS_NUM_THREADS = None
 
 
-def normalize_num_threads(num_threads: int, backend: Backend):
+def set_num_threads(num_threads: int) -> int:
+    assert isinstance(num_threads, int) or num_threads is None, 'Number of threads must be int value or None.'
+    global IMOPS_NUM_THREADS
+    current = IMOPS_NUM_THREADS
+    IMOPS_NUM_THREADS = num_threads
+    return current
+
+
+@contextmanager
+def imops_num_threads(num_threads: int):
+    previous = set_num_threads(num_threads)
+    try:
+        yield
+    finally:
+        set_num_threads(previous)
+
+
+def normalize_num_threads(num_threads: int, backend: Backend, warn_stacklevel: int = 1) -> int:
+    """Calculate the effective number of threads"""
+
+    global IMOPS_NUM_THREADS
     if backend.name in SINGLE_THREADED_BACKENDS:
         if num_threads != -1:
-            warn(f'"{backend.name}" backend is single-threaded. Setting `num_threads` has no effect.')
+            warn(
+                f'"{backend.name}" backend is single-threaded. Setting `num_threads` has no effect.',
+                stacklevel=warn_stacklevel,
+            )
         return 1
+
+    env_num_threads_var_name = BACKEND_NAME2ENV_NUM_THREADS_VAR_NAME[backend.name]
+    # here we also handle the case `env_num_threads_var_name`=" " gracefully
+    env_num_threads = os.environ.get(env_num_threads_var_name, '').strip()
+    env_num_threads = int(env_num_threads) if env_num_threads else None
+    # TODO: maybe let user set the absolute maximum number of threads?
+    num_available_cpus = len(os.sched_getaffinity(0))
+
+    max_num_threads = min(filter(bool, [IMOPS_NUM_THREADS, env_num_threads, num_available_cpus]))
+
     if num_threads >= 0:
         # FIXME
         if backend.name == 'Numba':
             warn(
                 'Setting `num_threads` has no effect with "Numba" backend. '
-                'Use `NUMBA_NUM_THREADS` environment variable.'
+                'Use `NUMBA_NUM_THREADS` environment variable.',
+                stacklevel=warn_stacklevel,
             )
-        return num_threads
-
-    num_threads_var_name = BACKEND2NUM_THREADS_VAR_NAME[backend.name]
-    # here we also handle the case `num_threads_var`=" " gracefully
-    env_num_threads = os.environ.get(num_threads_var_name, '').strip()
-    max_threads = int(env_num_threads) if env_num_threads else len(os.sched_getaffinity(0))
-
-    return max_threads + num_threads + 1
+            return num_threads
+
+        if num_threads > max_num_threads:
+            if max_num_threads == IMOPS_NUM_THREADS:
+                warn(
+                    f'Required number of threads ({num_threads}) is greater than `IMOPS_NUM_THREADS` '
+                    f'({IMOPS_NUM_THREADS}). Using {IMOPS_NUM_THREADS} threads.',
+                    stacklevel=warn_stacklevel,
+                )
+            elif max_num_threads == env_num_threads:
+                warn(
+                    f'Required number of threads ({num_threads}) is greater than `{env_num_threads_var_name}` '
+                    f'({env_num_threads}). Using {env_num_threads} threads.',
+                    stacklevel=warn_stacklevel,
+                )
+            else:
+                warn(
+                    f'Required number of threads ({num_threads}) is greater than number of available CPU-s '
+                    f'({num_available_cpus}). Using {num_available_cpus} threads.',
+                    stacklevel=warn_stacklevel,
+                )
+        return min(num_threads, max_num_threads)
+
+    return max_num_threads + num_threads + 1
 
 
 def get_c_contiguous_permutaion(array: np.ndarray) -> Optional[np.ndarray]:

imops/zoom.py

@@ -208,7 +208,7 @@ def _zoom(
 
     See `https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.zoom.html`
     """
-    backend = resolve_backend(backend)
+    backend = resolve_backend(backend, warn_stacklevel=4)
     if backend.name not in ('Scipy', 'Numba', 'Cython'):
         raise ValueError(f'Unsupported backend "{backend.name}".')
 
@@ -216,7 +216,7 @@ def _zoom(
     dtype = input.dtype
     cval = np.dtype(dtype).type(cval)
     zoom = fill_by_indices(np.ones(input.ndim, 'float64'), zoom, range(input.ndim))
-    num_threads = normalize_num_threads(num_threads, backend)
+    num_threads = normalize_num_threads(num_threads, backend, warn_stacklevel=4)
 
     if backend.name == 'Scipy':
         return scipy_zoom(

{imops-0.8.2.dist-info → imops-0.8.3.dist-info}/METADATA

@@ -1,12 +1,12 @@
 Metadata-Version: 2.1
 Name: imops
-Version: 0.8.2
+Version: 0.8.3
 Summary: Efficient parallelizable algorithms for multidimensional arrays to speed up your data pipelines
 Home-page: https://github.com/neuro-ml/imops
 Author: maxme1, vovaf709, talgat
 Author-email: maxs987@gmail.com, vovaf709@yandex.ru, saparov2130@gmail.com
 License: MIT
-Download-URL: https://github.com/neuro-ml/imops/archive/v0.8.2.tar.gz
+Download-URL: https://github.com/neuro-ml/imops/archive/v0.8.3.tar.gz
 Keywords: image processing,fast,ndarray,data pipelines
 Platform: UNKNOWN
 Classifier: Development Status :: 5 - Production/Stable
@@ -21,6 +21,7 @@ Requires-Python: >=3.6
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: scipy (<2.0.0,>=1.0)
+Requires-Dist: scikit-image
 Requires-Dist: connected-components-3d
 Requires-Dist: fastremap
 Requires-Dist: dataclasses ; python_version < "3.7"

imops-0.8.3.dist-info/RECORD

@@ -0,0 +1,46 @@
+imops/__init__.py,sha256=QAqK6FnQnOjdDwSXzhnJNB2ur3cGUNbP3ZdLc_VoIEY,514
+imops/__version__.py,sha256=lPOlZg8rjdgN8DyBl7I1rkzxiXOhafW_7RzlMGnvNo8,23
+imops/_configs.py,sha256=6hBY2O4YSaQC_avvqI-dmhZkPMl4p39KQIPtqx5MAh0,734
+imops/backend.py,sha256=8x6Zq9hetOt9_RDcytNNBJZBLNXgQd7ob2dh4WQuuqM,2598
+imops/box.py,sha256=NbYsPKQ8fCGxSm5up0TyqERJ-5iWmnzHLo4Smv9eesk,1865
+imops/crop.py,sha256=hCgwxDgNdIo9m0gleMB3oOUNA7tQurhYNhE5S39tMlE,4055
+imops/interp1d.py,sha256=CSPJ77isqGOJTN-_C--iyz-3qJhpL4dE9i5VdJgSFDE,8400
+imops/measure.py,sha256=YJ5hBDH6iW1hgqm6fi6FIURBKmeZjVfr7LSINOHgFrM,8657
+imops/morphology.py,sha256=KG3RCCicsX7WHsBj4mH6iaJ0dy6tW9cGJJvD4fDisBQ,12723
+imops/numeric.py,sha256=yiSjPTUD2w177LOBD_39_BHh9n8uKbSsHJDchNVsKRY,13650
+imops/pad.py,sha256=OXy_r4X7hOPBU4NSH3B-8ZpxfcY1t6FFMgMSopULpv8,9314
+imops/radon.py,sha256=TWsgZvWRapBbU7MVr1fRdpVXCvi6-kt8fxroyQc44ek,9211
+imops/testing.py,sha256=Qiy_Qn7bd7MIXoaZr4T_6r0jQWVRJFwR2z12c_Zwt30,1708
+imops/utils.py,sha256=nfrMUU0OhPgK5UapD9GNC1StPxX3cKbob54kiuf4_qs,6890
+imops/zoom.py,sha256=wl8wt7yoX5V4Mn7sMefV1Fy9KEetqMEdRHLdhT90o6o,9955
+imops/src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+imops/src/_backprojection.cp36-win32.pyd,sha256=A6LH7d8sIvVJvWTlH2nLrJr3j97GcZzNxtFK9REinbI,155136
+imops/src/_backprojection.pyx,sha256=xngEJtJ6czjoBRBKL6tu5S7nt6ZgcsE82xDGdCIfagE,2724
+imops/src/_fast_backprojection.cp36-win32.pyd,sha256=_Ob4--N7IZW3kkI9RX75NbkTVI6FC6H08D1k9WwWd1U,155136
+imops/src/_fast_backprojection.pyx,sha256=xngEJtJ6czjoBRBKL6tu5S7nt6ZgcsE82xDGdCIfagE,2724
+imops/src/_fast_measure.cp36-win32.pyd,sha256=VfXOkRGYwmZpKbXbHtwOME631bxgbWcjAEitWnXMdNg,220672
+imops/src/_fast_measure.pyx,sha256=JzZ26PptkVT08OemSFaJakc4jf7xYocPqV8jfrYi_wA,2737
+imops/src/_fast_morphology.cp36-win32.pyd,sha256=zplrVcEH2uNgUD7f5iXuw4ctYrMcTiWbWesFovOUhIY,129536
+imops/src/_fast_morphology.pyx,sha256=QwFrPQ6M-bJDVQNCCtKK7FR3KOr5cZSx_CBvxh9THXM,9538
+imops/src/_fast_numeric.cp36-win32.pyd,sha256=h9BS_ygwyPFP0m9fSELRCwjbgLDbi5FHZsovZlKqnoQ,326144
+imops/src/_fast_numeric.pyx,sha256=c8q3KlL8mDe0wsz72ooeqhffk1iqvA0Org7NWW5RbYo,8221
+imops/src/_fast_radon.cp36-win32.pyd,sha256=bUb0OKw3YP18Tc0uh79bx9U_Ql1QesnlJ4NlLyep8Xw,202752
+imops/src/_fast_radon.pyx,sha256=pu8EZplCjIDz9HqBcAit8nj2lDgXpmdhU61_4CfdGWI,2987
+imops/src/_fast_zoom.cp36-win32.pyd,sha256=ZR7ExxrXxn07L3rK7tBgo8nGzLaxRvsADTv8lfXUUws,333312
+imops/src/_fast_zoom.pyx,sha256=Dd5MDelALAh4-3QhnzfzKTB5Gmi2lVERrbyYj-q8kVQ,23091
+imops/src/_measure.cp36-win32.pyd,sha256=bmsMA49jfJh4M5E9kSNgc4Q94U-aNsJkzQhdxovPm44,220160
+imops/src/_measure.pyx,sha256=JzZ26PptkVT08OemSFaJakc4jf7xYocPqV8jfrYi_wA,2737
+imops/src/_morphology.cp36-win32.pyd,sha256=JjwzaXLa30gxKIezYbX10zuj1fcfTtwXif7fVoiT_5w,129536
+imops/src/_morphology.pyx,sha256=QwFrPQ6M-bJDVQNCCtKK7FR3KOr5cZSx_CBvxh9THXM,9538
+imops/src/_numba_zoom.py,sha256=OmMj7mCFkV2P0FaWVXFuYpfzMBQ3Ks60YiHQURes5wQ,16857
+imops/src/_numeric.cp36-win32.pyd,sha256=PkrARdgAlTZ9byV3z80f45_XtwhGdeo5WJL6yEieNWc,325120
+imops/src/_numeric.pyx,sha256=c8q3KlL8mDe0wsz72ooeqhffk1iqvA0Org7NWW5RbYo,8221
+imops/src/_radon.cp36-win32.pyd,sha256=6XLWRXNN1bGFOsy98ddqxFjU_cWKy31dvdq2mrYG2VY,202752
+imops/src/_radon.pyx,sha256=pu8EZplCjIDz9HqBcAit8nj2lDgXpmdhU61_4CfdGWI,2987
+imops/src/_zoom.cp36-win32.pyd,sha256=0--K2pLZ7ctBXJdmlUgXPsn5xoKvnYfY_qKCz0VC8Ls,333312
+imops/src/_zoom.pyx,sha256=Dd5MDelALAh4-3QhnzfzKTB5Gmi2lVERrbyYj-q8kVQ,23091
+imops-0.8.3.dist-info/LICENSE,sha256=x7oyYZPKDTt8KoTitcNJLBWhueke5YEOopdr0XjyvBc,1085
+imops-0.8.3.dist-info/METADATA,sha256=NdA1fnC5rCLIg-kvfInPxqlWIckur-inz1uUF89mft0,7391
+imops-0.8.3.dist-info/WHEEL,sha256=ZX-iFFGFiHZRWPwXowAnONYwAL2TBxDHVbhC1HaSV9Q,97
+imops-0.8.3.dist-info/top_level.txt,sha256=kEVHvrc8A4lkLQP8xew2pWdYcyQAMvAwqnlZXtoWwnc,6
+imops-0.8.3.dist-info/RECORD,,

imops/_numeric.py

@@ -1,124 +0,0 @@
-from warnings import warn
-
-import numpy as np
-
-from .backend import BackendLike, resolve_backend
-from .src._fast_numeric import (
-    _parallel_pointwise_mul as cython_fast_parallel_pointwise_mul,
-    _parallel_sum as cython_fast_parallel_sum,
-)
-from .src._numeric import _parallel_pointwise_mul as cython_parallel_pointwise_mul, _parallel_sum as cython_parallel_sum
-from .utils import normalize_num_threads
-
-
-def _sum(nums: np.ndarray, num_threads: int = -1, backend: BackendLike = None) -> float:
-    """
-    Parallel sum of flat numpy array
-
-    Parameters
-    ----------
-    nums: np.ndarray
-        1-dimensional array
-    num_threads: int
-        the number of threads to use for computation. Default = the cpu count. If negative value passed
-        cpu count + num_threads + 1 threads will be used
-    backend: BackendLike
-        which backend to use. `cython` and `scipy` (primarly for benchmarking) are available,
-        `cython` is used by default
-
-    Returns
-    -------
-    sum: float
-
-    Examples
-    --------
-    >>> s = _sum(x, num_threads=1)
-    >>> s = _sum(x, num_threads=8, backend=Cython(fast=True))  # ffast-math compiled version
-    """
-    ndim = nums.ndim
-
-    if ndim != 1:
-        raise ValueError(f'Input must be 1-dimensional instead of {ndim}-dimensional.')
-
-    backend = resolve_backend(backend)
-    if backend.name not in ('Cython', 'Scipy'):
-        raise ValueError(f'Unsupported backend "{backend.name}"')
-
-    num_threads = normalize_num_threads(num_threads, backend)
-
-    if backend.name == 'Scipy':
-        return nums.sum()
-
-    if backend.name == 'Cython':
-        src_parallel_sum = cython_fast_parallel_sum if backend.fast else cython_parallel_sum
-
-    return src_parallel_sum(nums, num_threads)
-
-
-def _mul(nums1: np.ndarray, nums2: np.ndarray, num_threads: int = -1, backend: BackendLike = None) -> np.ndarray:
-    """
-    Parallel pointwise multiplication of 2 numpy arrays (aka x * y). Works faster only for ndim <= 3.
-
-    Parameters
-    ----------
-    nums1: np.ndarray
-    nums2: np.ndarray
-    num_threads: int
-        the number of threads to use for computation. Default = the cpu count. If negative value passed
-        cpu count + num_threads + 1 threads will be used
-    backend: BackendLike
-        which backend to use. `cython` and `scipy` (primarly for benchmarking) are available,
-        `cython` is used by default
-
-    Returns
-    -------
-    multiplied: np.ndarray
-        result of nums1 * nums2
-
-    Examples
-    --------
-    >>> mul = _mul(nums1, nums2, num_threads=8)
-    >>> mul = _mul(np.ones((2, 3)), np.ones((1, 3)))  # broadcasting, mul.shape == (2, 3)
-    >>> mul = _mul(nums1, nums2, backend=Cython(fast=True))  # ffast-math compiled version
-    """
-    if not nums1.size and not nums2.size:
-        return np.array([], dtype=nums1.dtype)
-    if bool(nums1.size) ^ bool(nums2.size):
-        raise ValueError('One of the arrays is empty, hence pointwise multiplication cannot be performed.')
-    if len(nums1.shape) != len(nums2.shape):
-        raise ValueError('Both arrays must have the same number of dimensions for pointwise multiplication.')
-    for dim1, dim2 in zip(nums1.shape, nums2.shape):
-        if dim1 != dim2 and dim1 != 1 and dim2 != 1:
-            raise ValueError(f'Arrays of shapes {nums1.shape} and {nums2.shape} are not broadcastable.')
-
-    if nums1.ndim > 3:
-        warn('Parallel pointwise multiplication is only supported for ndim<=3. Falling back to naive x * y.')
-
-        return nums1 * nums2
-
-    backend = resolve_backend(backend)
-    if backend.name not in ('Cython', 'Scipy'):
-        raise ValueError(f'Unsupported backend "{backend.name}"')
-
-    num_threads = normalize_num_threads(num_threads, backend)
-
-    if backend.name == 'Scipy':
-        return nums1 * nums2
-
-    if backend.name == 'Cython':
-        src_parallel_pointwise_mul = (
-            cython_fast_parallel_pointwise_mul if backend.fast else cython_parallel_pointwise_mul
-        )
-
-    n_dummy = 3 - nums1.ndim
-
-    if n_dummy:
-        nums1 = nums1[(None,) * n_dummy]
-        nums2 = nums2[(None,) * n_dummy]
-
-    out = src_parallel_pointwise_mul(nums1, nums2, np.maximum(nums1.shape, nums2.shape), num_threads)
-
-    if n_dummy:
-        out = out[(0,) * n_dummy]
-
-    return out

imops-0.8.2.dist-info/RECORD

@@ -1,46 +0,0 @@
-imops/__init__.py,sha256=DH8Dp0MIEq8j_taTwUMXKqoY3WjdVdBbMnNdKHx_qes,459
-imops/__version__.py,sha256=XhFj_ipJJJgnq-LMeWY_tqb7jCV62Abi98yRdQ0le8U,23
-imops/_configs.py,sha256=6hBY2O4YSaQC_avvqI-dmhZkPMl4p39KQIPtqx5MAh0,734
-imops/_numeric.py,sha256=XfKU5mNGRAAoEKtJ9xrYTxpxNFZv0abHvoolEUVeVdo,4383
-imops/backend.py,sha256=lt3BrosqmxFh7BA-rYwYAer3XMdBTPS-pBIR5hHDNa0,2368
-imops/box.py,sha256=NbYsPKQ8fCGxSm5up0TyqERJ-5iWmnzHLo4Smv9eesk,1865
-imops/crop.py,sha256=hDrPNVFU3vV758KTwZmZxHzm4XgUkLQLSqJOYQHG4P0,3446
-imops/interp1d.py,sha256=55NFTGgYUDlcndRvPqKLLPpFe6U0K8B-so8yUgBbK3c,8183
-imops/measure.py,sha256=NqCsj2yxuUEfpPmGmOL-YSvI_2rrYUWJyzaSCPoOe90,8575
-imops/morphology.py,sha256=GDVnLjGjTZJgcGNyFXPn3BMM7WHZZQeg-i24ci6uCdQ,12625
-imops/pad.py,sha256=SAPJQwnuKosPoiCU26YPE4cyRPNrskwfzNOrT3KqSuk,7447
-imops/radon.py,sha256=C643fB9tzqK0_RyhH9EzQBfHAeEqjbnbGKDT485k5pk,9008
-imops/testing.py,sha256=Qiy_Qn7bd7MIXoaZr4T_6r0jQWVRJFwR2z12c_Zwt30,1708
-imops/utils.py,sha256=9XoA8qywuuSTluGDf1aEa_zP2Ixu2pNYQnBbWhy92T4,4720
-imops/zoom.py,sha256=g7lgv8Z51tNCYfAyQnLlA187QgDzyQHxckEY9Uu9EKY,9917
-imops/src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-imops/src/_backprojection.cp36-win32.pyd,sha256=tEJrvPUW9F1WYW8jHcrdUGPyvjQLD7GFV0A5BV6xRU0,153600
-imops/src/_backprojection.pyx,sha256=xngEJtJ6czjoBRBKL6tu5S7nt6ZgcsE82xDGdCIfagE,2724
-imops/src/_fast_backprojection.cp36-win32.pyd,sha256=hrm_VudsN4KrzJ9YQ4sRyQlnbKCrce5zuJx_QsS_-Ks,154112
-imops/src/_fast_backprojection.pyx,sha256=xngEJtJ6czjoBRBKL6tu5S7nt6ZgcsE82xDGdCIfagE,2724
-imops/src/_fast_measure.cp36-win32.pyd,sha256=O15wuxD58QyycWg_DN1KICRlqwh_GN16SmpzJl9kOKo,219648
-imops/src/_fast_measure.pyx,sha256=JzZ26PptkVT08OemSFaJakc4jf7xYocPqV8jfrYi_wA,2737
-imops/src/_fast_morphology.cp36-win32.pyd,sha256=699Q1QmRTM7mNlv33EOmnEsG6PGHOk1SC6AwDtjzi5k,129024
-imops/src/_fast_morphology.pyx,sha256=QwFrPQ6M-bJDVQNCCtKK7FR3KOr5cZSx_CBvxh9THXM,9538
-imops/src/_fast_numeric.cp36-win32.pyd,sha256=6MK-dQ9elgfHJiOUFc-IWdIo596an2KuQ53vg_z0Ft4,209920
-imops/src/_fast_numeric.pyx,sha256=bXv4qBx_k6B6INNR0r49HU2a8UNnpt-mCS19TczcU38,1919
-imops/src/_fast_radon.cp36-win32.pyd,sha256=9RPxYgYAYktiNNscPk7meGuCkg7pwMiJ4kG-g2lvnEo,195584
-imops/src/_fast_radon.pyx,sha256=pu8EZplCjIDz9HqBcAit8nj2lDgXpmdhU61_4CfdGWI,2987
-imops/src/_fast_zoom.cp36-win32.pyd,sha256=A25ZjipbwrrAO_IhGAtlVOPWErzNyEf0Bdd1bsu4IJY,312832
-imops/src/_fast_zoom.pyx,sha256=Dd5MDelALAh4-3QhnzfzKTB5Gmi2lVERrbyYj-q8kVQ,23091
-imops/src/_measure.cp36-win32.pyd,sha256=y6waCfQlPCSOlYEj87cvFE7ntls8QFlrTq63WyluacU,219136
-imops/src/_measure.pyx,sha256=JzZ26PptkVT08OemSFaJakc4jf7xYocPqV8jfrYi_wA,2737
-imops/src/_morphology.cp36-win32.pyd,sha256=fP3kYsa6PAlXOzhXyV8vMGum0jrweX-3yDyr-DMCuTM,129024
-imops/src/_morphology.pyx,sha256=QwFrPQ6M-bJDVQNCCtKK7FR3KOr5cZSx_CBvxh9THXM,9538
-imops/src/_numba_zoom.py,sha256=OmMj7mCFkV2P0FaWVXFuYpfzMBQ3Ks60YiHQURes5wQ,16857
-imops/src/_numeric.cp36-win32.pyd,sha256=LTRV7_kGgzEd04XsHd9S3xOLBziizQ7G9d4Agr6PcBE,209920
-imops/src/_numeric.pyx,sha256=bXv4qBx_k6B6INNR0r49HU2a8UNnpt-mCS19TczcU38,1919
-imops/src/_radon.cp36-win32.pyd,sha256=0sBYbxT9lToKTYrYSyrnBfy0ctY6sE4hBbxwLKBxYx0,195584
-imops/src/_radon.pyx,sha256=pu8EZplCjIDz9HqBcAit8nj2lDgXpmdhU61_4CfdGWI,2987
-imops/src/_zoom.cp36-win32.pyd,sha256=1jLv5ImsyMs-fvLGz1W0cVjMUPRoqZm0Tu6sbuSKVo8,312320
-imops/src/_zoom.pyx,sha256=Dd5MDelALAh4-3QhnzfzKTB5Gmi2lVERrbyYj-q8kVQ,23091
-imops-0.8.2.dist-info/LICENSE,sha256=x7oyYZPKDTt8KoTitcNJLBWhueke5YEOopdr0XjyvBc,1085
-imops-0.8.2.dist-info/METADATA,sha256=B9oMqoxEC0eeetqk0bQqDGbKTN8qxq3vGrp0sXIq9tg,7363
-imops-0.8.2.dist-info/WHEEL,sha256=ZX-iFFGFiHZRWPwXowAnONYwAL2TBxDHVbhC1HaSV9Q,97
-imops-0.8.2.dist-info/top_level.txt,sha256=kEVHvrc8A4lkLQP8xew2pWdYcyQAMvAwqnlZXtoWwnc,6
-imops-0.8.2.dist-info/RECORD,,