Created starter files for the project.

venv/Lib/site-packages/numpy/random/_examples/numba/extending.py

@@ -0,0 +1,84 @@
+import numpy as np
+import numba as nb
+
+from numpy.random import PCG64
+from timeit import timeit
+
+bit_gen = PCG64()
+next_d = bit_gen.cffi.next_double
+state_addr = bit_gen.cffi.state_address
+
+def normals(n, state):
+    out = np.empty(n)
+    for i in range((n + 1) // 2):
+        x1 = 2.0 * next_d(state) - 1.0
+        x2 = 2.0 * next_d(state) - 1.0
+        r2 = x1 * x1 + x2 * x2
+        while r2 >= 1.0 or r2 == 0.0:
+            x1 = 2.0 * next_d(state) - 1.0
+            x2 = 2.0 * next_d(state) - 1.0
+            r2 = x1 * x1 + x2 * x2
+        f = np.sqrt(-2.0 * np.log(r2) / r2)
+        out[2 * i] = f * x1
+        if 2 * i + 1 < n:
+            out[2 * i + 1] = f * x2
+    return out
+
+# Compile using Numba
+normalsj = nb.jit(normals, nopython=True)
+# Must use state address not state with numba
+n = 10000
+
+def numbacall():
+    return normalsj(n, state_addr)
+
+rg = np.random.Generator(PCG64())
+
+def numpycall():
+    return rg.normal(size=n)
+
+# Check that the functions work
+r1 = numbacall()
+r2 = numpycall()
+assert r1.shape == (n,)
+assert r1.shape == r2.shape
+
+t1 = timeit(numbacall, number=1000)
+print('{:.2f} secs for {} PCG64 (Numba/PCG64) gaussian randoms'.format(t1, n))
+t2 = timeit(numpycall, number=1000)
+print('{:.2f} secs for {} PCG64 (NumPy/PCG64) gaussian randoms'.format(t2, n))
+
+# example 2
+
+next_u32 = bit_gen.ctypes.next_uint32
+ctypes_state = bit_gen.ctypes.state
+
+@nb.jit(nopython=True)
+def bounded_uint(lb, ub, state):
+    mask = delta = ub - lb
+    mask |= mask >> 1
+    mask |= mask >> 2
+    mask |= mask >> 4
+    mask |= mask >> 8
+    mask |= mask >> 16
+
+    val = next_u32(state) & mask
+    while val > delta:
+        val = next_u32(state) & mask
+
+    return lb + val
+
+
+print(bounded_uint(323, 2394691, ctypes_state.value))
+
+
+@nb.jit(nopython=True)
+def bounded_uints(lb, ub, n, state):
+    out = np.empty(n, dtype=np.uint32)
+    for i in range(n):
+        out[i] = bounded_uint(lb, ub, state)
+
+
+bounded_uints(323, 2394691, 10000000, ctypes_state.value)
+
+

venv/Lib/site-packages/numpy/random/_examples/numba/extending_distributions.py

@@ -0,0 +1,67 @@
+r"""
+Building the required library in this example requires a source distribution
+of NumPy or clone of the NumPy git repository since distributions.c is not
+included in binary distributions.
+
+On *nix, execute in numpy/random/src/distributions
+
+export ${PYTHON_VERSION}=3.8 # Python version
+export PYTHON_INCLUDE=#path to Python's include folder, usually \
+    ${PYTHON_HOME}/include/python${PYTHON_VERSION}m
+export NUMPY_INCLUDE=#path to numpy's include folder, usually \
+    ${PYTHON_HOME}/lib/python${PYTHON_VERSION}/site-packages/numpy/core/include
+gcc -shared -o libdistributions.so -fPIC distributions.c \
+    -I${NUMPY_INCLUDE} -I${PYTHON_INCLUDE}
+mv libdistributions.so ../../_examples/numba/
+
+On Windows
+
+rem PYTHON_HOME and PYTHON_VERSION are setup dependent, this is an example
+set PYTHON_HOME=c:\Anaconda
+set PYTHON_VERSION=38
+cl.exe /LD .\distributions.c -DDLL_EXPORT \
+    -I%PYTHON_HOME%\lib\site-packages\numpy\core\include \
+    -I%PYTHON_HOME%\include %PYTHON_HOME%\libs\python%PYTHON_VERSION%.lib
+move distributions.dll ../../_examples/numba/
+"""
+import os
+
+import numba as nb
+import numpy as np
+from cffi import FFI
+
+from numpy.random import PCG64
+
+ffi = FFI()
+if os.path.exists('./distributions.dll'):
+    lib = ffi.dlopen('./distributions.dll')
+elif os.path.exists('./libdistributions.so'):
+    lib = ffi.dlopen('./libdistributions.so')
+else:
+    raise RuntimeError('Required DLL/so file was not found.')
+
+ffi.cdef("""
+double random_standard_normal(void *bitgen_state);
+""")
+x = PCG64()
+xffi = x.cffi
+bit_generator = xffi.bit_generator
+
+random_standard_normal = lib.random_standard_normal
+
+
+def normals(n, bit_generator):
+    out = np.empty(n)
+    for i in range(n):
+        out[i] = random_standard_normal(bit_generator)
+    return out
+
+
+normalsj = nb.jit(normals, nopython=True)
+
+# Numba requires a memory address for void *
+# Can also get address from x.ctypes.bit_generator.value
+bit_generator_address = int(ffi.cast('uintptr_t', bit_generator))
+
+norm = normalsj(1000, bit_generator_address)
+print(norm[:12])