Merge b2315a5 into 5e2f2da

doc/source/brng/dsfmt.rst

@@ -1,4 +1,4 @@
-Double SIMD Mersenne Twister (dSFMT) 
+Double SIMD Mersenne Twister (dSFMT)
 ------------------------------------
 
 .. module:: randomgen.dsfmt

doc/source/brng/index.rst

@@ -1,11 +1,11 @@
 Basic Random Number Generators
 ------------------------------
 
-The random values produced by :class:`~randomgen.generator.RandomGenerator` 
+The random values produced by :class:`~randomgen.generator.RandomGenerator`
 are produced by a basic RNG.  These basic RNGs do not directly provide
 random numbers and only contains methods used for seeding, getting or
-setting the state, jumping or advancing the state, and for accessing 
-low-level wrappers for consumption by code that can efficiently 
+setting the state, jumping or advancing the state, and for accessing
+low-level wrappers for consumption by code that can efficiently
 access the functions provided, e.g., `numba <https://numba.pydata.org>`_.
 
 Stable RNGs

doc/source/brng/mt19937.rst

@@ -1,4 +1,4 @@
-Mersenne Twister (MT19937) 
+Mersenne Twister (MT19937)
 --------------------------
 
 .. module:: randomgen.mt19937

doc/source/change-log.rst

@@ -3,6 +3,16 @@
 Change Log
 ----------
 
+v1.16.6
+=======
+- Changed the default jump step size to phi times the period of the generator for
+  :class:`~randomgen.pcg32.PCG32` and :class:`~randomgen.pcg64.PCG64`.
+- Improved the performance of :class:`~randomgen.pcg64.PCG64` on Windows.
+- Improved performance of :func:`~randomgen.dsfmt.DSFMT.jump` and
+  :func:`~randomgen.dsfmt.DSFMT.jumped`.
+- Improves backward compatability of :class:`~randomgen.mtrand.RandomState`
+
+
 v1.16.5
 =======
 - Fixed bugs in :func:`~randomgen.mtrand.RandomState.laplace`,

doc/source/generator.rst

@@ -1,16 +1,16 @@
 Random Generator
 ----------------
-The :class:`~randomgen.generator.RandomGenerator` provides access to 
-a wide range of distributions, and served as a replacement for 
-:class:`~numpy.random.RandomState`.  The main difference between 
+The :class:`~randomgen.generator.RandomGenerator` provides access to
+a wide range of distributions, and served as a replacement for
+:class:`~numpy.random.RandomState`.  The main difference between
 the two is that :class:`~randomgen.generator.RandomGenerator` relies
-on an additional basic RNG to manage state and generate the random 
-bits which are then transformed into random values from useful 
-distributions. The default basic RNG used by 
-:class:`~randomgen.generator.RandomGenerator` is 
-:class:`~randomgen.xoroshiro128.Xoroshiro128`.  The basic RNG can be 
-changed by passing an instantized basic RNG to 
-:class:`~randomgen.generator.RandomGenerator`. 
+on an additional basic RNG to manage state and generate the random
+bits which are then transformed into random values from useful
+distributions. The default basic RNG used by
+:class:`~randomgen.generator.RandomGenerator` is
+:class:`~randomgen.xoroshiro128.Xoroshiro128`.  The basic RNG can be
+changed by passing an instantized basic RNG to
+:class:`~randomgen.generator.RandomGenerator`.
 
 .. currentmodule:: randomgen.generator
 

doc/source/index.rst

@@ -7,8 +7,8 @@ generator be be changed.
 Quick Start
 -----------
 
-Like :mod:`numpy.random`, RandomGen can be used at the module level. 
-This uses the default :class:`~randomgen.generator.RandomGenerator` which 
+Like :mod:`numpy.random`, RandomGen can be used at the module level.
+This uses the default :class:`~randomgen.generator.RandomGenerator` which
 uses normals provided by :class:`~randomgen.xoroshiro128.Xoroshiro128`.
 
 .. code-block:: python
@@ -17,10 +17,10 @@ uses normals provided by :class:`~randomgen.xoroshiro128.Xoroshiro128`.
   import randomgen.generator as random
   random.standard_normal()
 
-:class:`~randomgen.generator.RandomGenerator` can also be used as a 
+:class:`~randomgen.generator.RandomGenerator` can also be used as a
 replacement for :class:`~numpy.random.RandomState`, although the random
-values are generated by :class:`~randomgen.xoroshiro128.Xoroshiro128`. It 
-also isn't possible to directly seed a 
+values are generated by :class:`~randomgen.xoroshiro128.Xoroshiro128`. It
+also isn't possible to directly seed a
 :class:`~randomgen.generator.RandomGenerator`.
 
 
@@ -33,7 +33,7 @@ also isn't possible to directly seed a
 
 
 Seeds can be passed to any of the basic RNGs. Here :class:`~randomgen.mt19937.MT19937`
-is used and the :class:`~randomgen.generator.RandomGenerator` is accessed via 
+is used and the :class:`~randomgen.generator.RandomGenerator` is accessed via
 the property :attr:`~randomgen.mt19937.MT19937.generator`.
 
 .. code-block:: python
@@ -93,13 +93,13 @@ What's New or Different
 .. warning::
 
   The Box-Muller method used to produce NumPy's normals is no longer available
-  in :class:`~randomgen.generator.RandomGenerator`.  It is not possible to 
-  reproduce the random values using :class:`~randomgen.generator.RandomGenerator` 
+  in :class:`~randomgen.generator.RandomGenerator`.  It is not possible to
+  reproduce the random values using :class:`~randomgen.generator.RandomGenerator`
   for the normal distribution or any other distribution that relies on the
-  normal such as the gamma or student's t. If you require backward compatibility, a 
+  normal such as the gamma or student's t. If you require backward compatibility, a
   legacy generator, :class:`~randomgen.mtrand.RandomState`, has been created
   which can fully reproduce the sequence produced by NumPy.
-  
+
 * The normal, exponential and gamma generators use 256-step Ziggurat
   methods which are 2-10 times faster than NumPy's Box-Muller or inverse CDF
   implementations.

doc/source/legacy.rst

@@ -1,19 +1,19 @@
 Legacy Random Generation
 ------------------------
-The :class:`~randomgen.mtrand.RandomState` provides access to 
+The :class:`~randomgen.mtrand.RandomState` provides access to
 legacy generators.  These all depend on normals produced using a
 polar transformation or inverse CDF exponentials or gammas. This
 class should only be used  if it is essential to have randoms that
 are identical to what would have been produced by NumPy.
 
 :class:`~randomgen.mtrand.RandomState` add additional information
 to the state which is required when using Box-Muller normals since these
-are produced in pairs. It is important to use 
+are produced in pairs. It is important to use
 :attr:`~randomgen.mtrand.RandomState.get_state()`
-when accessing the state so that these extra values are saved. 
+when accessing the state so that these extra values are saved.
 
 .. code-block:: python
-  
+
    from randomgen import MT19937
    from randomgen.mtrand import RandomState
    from numpy.random import RandomState
@@ -31,7 +31,7 @@ when accessing the state so that these extra values are saved.
 
    rs.standard_exponential()
    lg.standard_exponential()
-   
+
 
 .. currentmodule:: randomgen.mtrand
 
@@ -46,7 +46,7 @@ Seeding and State
 
    ~RandomState.get_state
    ~RandomState.set_state
-   
+
 Simple random data
 ==================
 .. autosummary::

doc/source/multithreading.rst

@@ -20,41 +20,41 @@ that the same seed will produce the same outputs.
     import multiprocessing
     import concurrent.futures
     import numpy as np
-    
+
     class MultithreadedRNG(object):
         def __init__(self, n, seed=None, threads=None):
             rg = Xorshift1024(seed)
             if threads is None:
                 threads = multiprocessing.cpu_count()
             self.threads = threads
-    
+
             self._random_generators = []
             for _ in range(0, threads-1):
                 _rg = Xorshift1024()
                 _rg.state = rg.state
                 self._random_generators.append(_rg.generator)
                 rg.jump()
             self._random_generators.append(rg.generator)
-    
+
             self.n = n
             self.executor = concurrent.futures.ThreadPoolExecutor(threads)
             self.values = np.empty(n)
             self.step = np.ceil(n / threads).astype(np.int)
-    
+
         def fill(self):
             def _fill(random_state, out, first, last):
                 random_state.standard_normal(out=out[first:last])
-    
+
             futures = {}
             for i in range(self.threads):
-                args = (_fill, 
+                args = (_fill,
                         self._random_generators[i],
-                        self.values, 
-                        i * self.step, 
+                        self.values,
+                        i * self.step,
                         (i + 1) * self.step)
                 futures[self.executor.submit(*args)] = i
             concurrent.futures.wait(futures)
-    
+
         def __del__(self):
             self.executor.shutdown(False)
 
@@ -80,7 +80,7 @@ the time required to generate using a single thread.
 
     In [4]: print(mrng.threads)
         ...: %timeit mrng.fill()
-    
+
     4
     32.8 ms ± 2.71 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
 

doc/source/new-or-different.rst

@@ -6,10 +6,10 @@ What's New or Different
 .. warning::
 
   The Box-Muller method used to produce NumPy's normals is no longer available
-  in :class:`~randomgen.generator.RandomGenerator`.  It is not possible to 
-  reproduce the random values using :class:`~randomgen.generator.RandomGenerator` 
+  in :class:`~randomgen.generator.RandomGenerator`.  It is not possible to
+  reproduce the random values using :class:`~randomgen.generator.RandomGenerator`
   for the normal distribution or any other distribution that relies on the
-  normal such as the gamma or student's t. If you require backward compatibility, a 
+  normal such as the gamma or student's t. If you require backward compatibility, a
   legacy generator, :class:`~randomgen.mtrand.RandomState`, has been created
   which can fully reproduce the sequence produced by NumPy.
 

doc/source/performance.rst

@@ -56,7 +56,7 @@ percentage. The overall performance was computed using a geometric mean.
 .. csv-table::
     :header: ,Xoroshiro128,Xoshiro256**,Xorshift1024,PCG64,MT19937,Philox,ThreeFry
     :widths: 14,14,14,14,14,14,14,14
-    
+
     64-bit Unsigned Ints,353,309,283,263,233,191,162
     Uniforms,271,283,276,288,232,188,173
     32-bit Unsigned Ints,83,76,78,70,76,64,56

randomgen/common.pxd

@@ -74,6 +74,8 @@ cdef object float_fill(void *func, brng_t *state, object size, object lock, obje
 
 cdef object float_fill_from_double(void *func, brng_t *state, object size, object lock, object out)
 
+cdef object wrap_int(object val, object bits)
+
 cdef np.ndarray int_to_array(object value, object name, object bits, object uint_size)
 
 cdef object cont(void *func, void *state, object size, object lock, int narg,

randomgen/common.pyx

@@ -172,6 +172,16 @@ cdef double kahan_sum(double *darr, np.npy_intp n):
         sum = t
     return sum
 
+
+cdef object wrap_int(object val, object bits):
+    """Wraparound to place an integer into the interval [0, 2**bits)"""
+    upper = int(2)**int(bits)
+    if not 0<= val < upper:
+        divisor = val // upper
+        val = val - upper * divisor
+    return val
+
+
 cdef np.ndarray int_to_array(object value, object name, object bits, object uint_size):
     """Convert a large integer to an array of unsigned integers"""
     len = bits // uint_size

randomgen/pcg32.pyx

@@ -297,14 +297,18 @@ cdef class PCG32:
           RNG.  For example, two 16-bit integer values can be simulated
           from a single draw of a 32-bit RNG.
         """
+        delta = wrap_int(delta, 64)
         pcg32_advance_state(&self.rng_state, <uint64_t>delta)
         return self
 
     def jump(self, np.npy_intp iter=1):
         """
         jump(iter=1)
 
-        Jumps the state as-if 2**32 random numbers have been generated
+        Jump the state a fixed increment
+
+        Jumps the state as-if 11400714819323198486 random numbers have been
+        generated.
 
         Parameters
         ----------
@@ -316,7 +320,8 @@ cdef class PCG32:
         self : PCG32
             RNG jumped iter times
         """
-        return self.advance(iter * 2**32)
+        step = int(0x9e3779b97f4a7c16)
+        return self.advance(iter * step)
 
     @property
     def ctypes(self):

randomgen/pcg64.pyx

@@ -326,6 +326,8 @@ cdef class PCG64:
         Advancing the RNG state resets any pre-computed random numbers.
         This is required to ensure exact reproducibility.
         """
+        delta = wrap_int(delta, 128)
+
         cdef np.ndarray d = np.empty(2, dtype=np.uint64)
         d[0] = delta // 2**64
         d[1] = delta % 2**64
@@ -337,7 +339,10 @@ cdef class PCG64:
         """
         jump(iter=1)
 
-        Jumps the state as-if 2**64 random numbers have been generated
+        Jump the state a fixed increment
+
+        Jumps the state as-if 210306068529402873165736369884012333108 random
+        numbers have been generated.
 
         Parameters
         ----------
@@ -354,7 +359,9 @@ cdef class PCG64:
         Jumping the rng state resets any pre-computed random numbers. This is required
         to ensure exact reproducibility.
         """
-        return self.advance(iter * 2**64)
+        step = 0x9e3779b97f4a7c15f39cc0605cedc834
+        step *= int(iter)
+        return self.advance(step)
 
     @property
     def ctypes(self):