diff options
Diffstat (limited to 'numpy')
-rw-r--r-- | numpy/core/fromnumeric.py | 2 | ||||
-rw-r--r-- | numpy/core/numeric.py | 107 | ||||
-rw-r--r-- | numpy/core/src/umath/loops.c.src | 5 | ||||
-rw-r--r-- | numpy/core/src/umath/simd.inc.src | 166 | ||||
-rw-r--r-- | numpy/core/tests/test_numeric.py | 110 |
5 files changed, 344 insertions, 46 deletions
diff --git a/numpy/core/fromnumeric.py b/numpy/core/fromnumeric.py index a2937c5c5..67d2c5b48 100644 --- a/numpy/core/fromnumeric.py +++ b/numpy/core/fromnumeric.py @@ -518,7 +518,7 @@ def transpose(a, axes=None): See Also -------- - rollaxis + moveaxis argsort Notes diff --git a/numpy/core/numeric.py b/numpy/core/numeric.py index 4f3d418e6..a18b38072 100644 --- a/numpy/core/numeric.py +++ b/numpy/core/numeric.py @@ -1,6 +1,7 @@ from __future__ import division, absolute_import, print_function import sys +import operator import warnings import collections from numpy.core import multiarray @@ -15,8 +16,10 @@ from ._internal import TooHardError if sys.version_info[0] >= 3: import pickle basestring = str + import builtins else: import cPickle as pickle + import __builtin__ as builtins loads = pickle.loads @@ -31,15 +34,15 @@ __all__ = [ 'ascontiguousarray', 'asfortranarray', 'isfortran', 'empty_like', 'zeros_like', 'ones_like', 'correlate', 'convolve', 'inner', 'dot', 'einsum', 'outer', 'vdot', 'alterdot', 'restoredot', 'roll', - 'rollaxis', 'cross', 'tensordot', 'array2string', 'get_printoptions', - 'set_printoptions', 'array_repr', 'array_str', 'set_string_function', - 'little_endian', 'require', 'fromiter', 'array_equal', 'array_equiv', - 'indices', 'fromfunction', 'isclose', 'load', 'loads', 'isscalar', - 'binary_repr', 'base_repr', 'ones', 'identity', 'allclose', - 'compare_chararrays', 'putmask', 'seterr', 'geterr', 'setbufsize', - 'getbufsize', 'seterrcall', 'geterrcall', 'errstate', 'flatnonzero', - 'Inf', 'inf', 'infty', 'Infinity', 'nan', 'NaN', 'False_', 'True_', - 'bitwise_not', 'CLIP', 'RAISE', 'WRAP', 'MAXDIMS', 'BUFSIZE', + 'rollaxis', 'moveaxis', 'cross', 'tensordot', 'array2string', + 'get_printoptions', 'set_printoptions', 'array_repr', 'array_str', + 'set_string_function', 'little_endian', 'require', 'fromiter', + 'array_equal', 'array_equiv', 'indices', 'fromfunction', 'isclose', 'load', + 'loads', 'isscalar', 'binary_repr', 'base_repr', 'ones', 'identity', + 'allclose', 'compare_chararrays', 'putmask', 'seterr', 'geterr', + 'setbufsize', 'getbufsize', 'seterrcall', 'geterrcall', 'errstate', + 'flatnonzero', 'Inf', 'inf', 'infty', 'Infinity', 'nan', 'NaN', 'False_', + 'True_', 'bitwise_not', 'CLIP', 'RAISE', 'WRAP', 'MAXDIMS', 'BUFSIZE', 'ALLOW_THREADS', 'ComplexWarning', 'full', 'full_like', 'matmul', 'shares_memory', 'may_share_memory', 'MAY_SHARE_BOUNDS', 'MAY_SHARE_EXACT', 'TooHardError', @@ -1422,6 +1425,7 @@ def rollaxis(a, axis, start=0): See Also -------- + moveaxis : Move array axes to new positions. roll : Roll the elements of an array by a number of positions along a given axis. @@ -1457,6 +1461,91 @@ def rollaxis(a, axis, start=0): return a.transpose(axes) +def _validate_axis(axis, ndim, argname): + try: + axis = [operator.index(axis)] + except TypeError: + axis = list(axis) + axis = [a + ndim if a < 0 else a for a in axis] + if not builtins.all(0 <= a < ndim for a in axis): + raise ValueError('invalid axis for this array in `%s` argument' % + argname) + if len(set(axis)) != len(axis): + raise ValueError('repeated axis in `%s` argument' % argname) + return axis + + +def moveaxis(a, source, destination): + """ + Move axes of an array to new positions. + + Other axes remain in their original order. + + .. versionadded::1.11.0 + + Parameters + ---------- + a : np.ndarray + The array whose axes should be reordered. + source : int or sequence of int + Original positions of the axes to move. These must be unique. + destination : int or sequence of int + Destination positions for each of the original axes. These must also be + unique. + + Returns + ------- + result : np.ndarray + Array with moved axes. This array is a view of the input array. + + See Also + -------- + transpose: Permute the dimensions of an array. + swapaxes: Interchange two axes of an array. + + Examples + -------- + + >>> x = np.zeros((3, 4, 5)) + >>> np.moveaxis(x, 0, -1).shape + (4, 5, 3) + >>> np.moveaxis(x, -1, 0).shape + (5, 3, 4) + + These all achieve the same result: + + >>> np.transpose(x).shape + (5, 4, 3) + >>> np.swapaxis(x, 0, -1).shape + (5, 4, 3) + >>> np.moveaxis(x, [0, 1], [-1, -2]).shape + (5, 4, 3) + >>> np.moveaxis(x, [0, 1, 2], [-1, -2, -3]).shape + (5, 4, 3) + + """ + try: + # allow duck-array types if they define transpose + transpose = a.transpose + except AttributeError: + a = asarray(a) + transpose = a.transpose + + source = _validate_axis(source, a.ndim, 'source') + destination = _validate_axis(destination, a.ndim, 'destination') + if len(source) != len(destination): + raise ValueError('`source` and `destination` arguments must have ' + 'the same number of elements') + + order = [n for n in range(a.ndim) if n not in source] + + for dest, src in sorted(zip(destination, source)): + order.insert(dest, src) + + result = transpose(order) + return result + + # fix hack in scipy which imports this function def _move_axis_to_0(a, axis): return rollaxis(a, axis, 0) diff --git a/numpy/core/src/umath/loops.c.src b/numpy/core/src/umath/loops.c.src index aff6180c7..fc9ffec94 100644 --- a/numpy/core/src/umath/loops.c.src +++ b/numpy/core/src/umath/loops.c.src @@ -1558,14 +1558,11 @@ NPY_NO_EXPORT void /**begin repeat1 * #kind = isnan, isinf, isfinite, signbit# * #func = npy_isnan, npy_isinf, npy_isfinite, npy_signbit# - * #isnan = 1, 0*3# **/ NPY_NO_EXPORT void @TYPE@_@kind@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func)) { - char * margs[] = {args[0], args[0], args[1]}; - npy_intp msteps[] = {steps[0], steps[0], steps[1]}; - if (!@isnan@ || !run_binary_simd_not_equal_@TYPE@(margs, dimensions, msteps)) { + if (!run_@kind@_simd_@TYPE@(args, dimensions, steps)) { UNARY_LOOP { const @type@ in1 = *(@type@ *)ip1; *((npy_bool *)op1) = @func@(in1) != 0; diff --git a/numpy/core/src/umath/simd.inc.src b/numpy/core/src/umath/simd.inc.src index 84695f5d6..21ff97784 100644 --- a/numpy/core/src/umath/simd.inc.src +++ b/numpy/core/src/umath/simd.inc.src @@ -25,6 +25,7 @@ #endif #include <assert.h> #include <stdlib.h> +#include <float.h> #include <string.h> /* for memcpy */ /* Figure out the right abs function for pointer addresses */ @@ -259,6 +260,32 @@ run_binary_simd_@kind@_@TYPE@(char **args, npy_intp *dimensions, npy_intp *steps /**end repeat1**/ +/**begin repeat1 + * #kind = isnan, isfinite, isinf, signbit# + */ + +#if @vector@ && defined NPY_HAVE_SSE2_INTRINSICS + +static void +sse2_@kind@_@TYPE@(npy_bool * op, @type@ * ip1, npy_intp n); + +#endif + +static NPY_INLINE int +run_@kind@_simd_@TYPE@(char **args, npy_intp *dimensions, npy_intp *steps) +{ +#if @vector@ && defined NPY_HAVE_SSE2_INTRINSICS + if (steps[0] == sizeof(@type@) && steps[1] == 1 && + npy_is_aligned(args[0], sizeof(@type@))) { + sse2_@kind@_@TYPE@((npy_bool*)args[1], (@type@*)args[0], dimensions[0]); + return 1; + } +#endif + return 0; +} + +/**end repeat1**/ + /**end repeat**/ /* @@ -528,11 +555,104 @@ sse2_compress4_to_byte_@TYPE@(@vtype@ r1, @vtype@ r2, @vtype@ r3, @vtype@ * r4, #endif } +static void +sse2_signbit_@TYPE@(npy_bool * op, @type@ * ip1, npy_intp n) +{ + LOOP_BLOCK_ALIGN_VAR(ip1, @type@, 16) { + op[i] = npy_signbit(ip1[i]); + } + LOOP_BLOCKED(@type@, 16) { + @vtype@ a = @vpre@_load_@vsuf@(&ip1[i]); + int r = @vpre@_movemask_@vsuf@(a); + if (sizeof(@type@) == 8) { + op[i] = r & 1; + op[i + 1] = (r >> 1); + } + else { + op[i] = r & 1; + op[i + 1] = (r >> 1) & 1; + op[i + 2] = (r >> 2) & 1; + op[i + 3] = (r >> 3); + } + } + LOOP_BLOCKED_END { + op[i] = npy_signbit(ip1[i]); + } +} + +/**begin repeat1 + * #kind = isnan, isfinite, isinf# + * #var = 0, 1, 2# + */ + +static void +sse2_@kind@_@TYPE@(npy_bool * op, @type@ * ip1, npy_intp n) +{ +#if @var@ != 0 /* isinf/isfinite */ + /* signbit mask 0x7FFFFFFF after andnot */ + const @vtype@ mask = @vpre@_set1_@vsuf@(-0.@c@); + const @vtype@ ones = @vpre@_cmpeq_@vsuf@(@vpre@_setzero_@vsuf@(), + @vpre@_setzero_@vsuf@()); +#if @double@ + const @vtype@ fltmax = @vpre@_set1_@vsuf@(DBL_MAX); +#else + const @vtype@ fltmax = @vpre@_set1_@vsuf@(FLT_MAX); +#endif +#endif + LOOP_BLOCK_ALIGN_VAR(ip1, @type@, 16) { + op[i] = npy_@kind@(ip1[i]); + } + LOOP_BLOCKED(@type@, 64) { + @vtype@ a = @vpre@_load_@vsuf@(&ip1[i + 0 * 16 / sizeof(@type@)]); + @vtype@ b = @vpre@_load_@vsuf@(&ip1[i + 1 * 16 / sizeof(@type@)]); + @vtype@ c = @vpre@_load_@vsuf@(&ip1[i + 2 * 16 / sizeof(@type@)]); + @vtype@ d = @vpre@_load_@vsuf@(&ip1[i + 3 * 16 / sizeof(@type@)]); + @vtype@ r1, r2, r3, r4; +#if @var@ != 0 /* isinf/isfinite */ + /* fabs via masking of sign bit */ + r1 = @vpre@_andnot_@vsuf@(mask, a); + r2 = @vpre@_andnot_@vsuf@(mask, b); + r3 = @vpre@_andnot_@vsuf@(mask, c); + r4 = @vpre@_andnot_@vsuf@(mask, d); +#if @var@ == 1 /* isfinite */ + /* negative compare against max float, nan is always true */ + r1 = @vpre@_cmpnle_@vsuf@(r1, fltmax); + r2 = @vpre@_cmpnle_@vsuf@(r2, fltmax); + r3 = @vpre@_cmpnle_@vsuf@(r3, fltmax); + r4 = @vpre@_cmpnle_@vsuf@(r4, fltmax); +#else /* isinf */ + r1 = @vpre@_cmpnlt_@vsuf@(fltmax, r1); + r2 = @vpre@_cmpnlt_@vsuf@(fltmax, r2); + r3 = @vpre@_cmpnlt_@vsuf@(fltmax, r3); + r4 = @vpre@_cmpnlt_@vsuf@(fltmax, r4); +#endif + /* flip results to what we want (andnot as there is no sse not) */ + r1 = @vpre@_andnot_@vsuf@(r1, ones); + r2 = @vpre@_andnot_@vsuf@(r2, ones); + r3 = @vpre@_andnot_@vsuf@(r3, ones); + r4 = @vpre@_andnot_@vsuf@(r4, ones); +#endif +#if @var@ == 0 /* isnan */ + r1 = @vpre@_cmpneq_@vsuf@(a, a); + r2 = @vpre@_cmpneq_@vsuf@(b, b); + r3 = @vpre@_cmpneq_@vsuf@(c, c); + r4 = @vpre@_cmpneq_@vsuf@(d, d); +#endif + sse2_compress4_to_byte_@TYPE@(r1, r2, r3, &r4, &op[i]); + } + LOOP_BLOCKED_END { + op[i] = npy_@kind@(ip1[i]); + } + /* silence exceptions from comparisons */ + npy_clear_floatstatus(); +} + +/**end repeat1**/ + /**begin repeat1 * #kind = equal, not_equal, less, less_equal, greater, greater_equal# * #OP = ==, !=, <, <=, >, >=# * #VOP = cmpeq, cmpneq, cmplt, cmple, cmpgt, cmpge# - * #neq = 0, 1, 0*4# */ /* sets invalid fpu flag on QNaN for consistency with packed compare */ @@ -554,36 +674,20 @@ sse2_binary_@kind@_@TYPE@(npy_bool * op, @type@ * ip1, @type@ * ip2, npy_intp n) LOOP_BLOCK_ALIGN_VAR(ip1, @type@, 16) { op[i] = sse2_ordered_cmp_@kind@_@TYPE@(ip1[i], ip2[i]); } - /* isnan special unary case */ - if (@neq@ && ip1 == ip2) { - LOOP_BLOCKED(@type@, 64) { - @vtype@ a = @vpre@_load_@vsuf@(&ip1[i + 0 * 16 / sizeof(@type@)]); - @vtype@ b = @vpre@_load_@vsuf@(&ip1[i + 1 * 16 / sizeof(@type@)]); - @vtype@ c = @vpre@_load_@vsuf@(&ip1[i + 2 * 16 / sizeof(@type@)]); - @vtype@ d = @vpre@_load_@vsuf@(&ip1[i + 3 * 16 / sizeof(@type@)]); - @vtype@ r1 = @vpre@_@VOP@_@vsuf@(a, a); - @vtype@ r2 = @vpre@_@VOP@_@vsuf@(b, b); - @vtype@ r3 = @vpre@_@VOP@_@vsuf@(c, c); - @vtype@ r4 = @vpre@_@VOP@_@vsuf@(d, d); - sse2_compress4_to_byte_@TYPE@(r1, r2, r3, &r4, &op[i]); - } - } - else { - LOOP_BLOCKED(@type@, 64) { - @vtype@ a1 = @vpre@_load_@vsuf@(&ip1[i + 0 * 16 / sizeof(@type@)]); - @vtype@ b1 = @vpre@_load_@vsuf@(&ip1[i + 1 * 16 / sizeof(@type@)]); - @vtype@ c1 = @vpre@_load_@vsuf@(&ip1[i + 2 * 16 / sizeof(@type@)]); - @vtype@ d1 = @vpre@_load_@vsuf@(&ip1[i + 3 * 16 / sizeof(@type@)]); - @vtype@ a2 = @vpre@_loadu_@vsuf@(&ip2[i + 0 * 16 / sizeof(@type@)]); - @vtype@ b2 = @vpre@_loadu_@vsuf@(&ip2[i + 1 * 16 / sizeof(@type@)]); - @vtype@ c2 = @vpre@_loadu_@vsuf@(&ip2[i + 2 * 16 / sizeof(@type@)]); - @vtype@ d2 = @vpre@_loadu_@vsuf@(&ip2[i + 3 * 16 / sizeof(@type@)]); - @vtype@ r1 = @vpre@_@VOP@_@vsuf@(a1, a2); - @vtype@ r2 = @vpre@_@VOP@_@vsuf@(b1, b2); - @vtype@ r3 = @vpre@_@VOP@_@vsuf@(c1, c2); - @vtype@ r4 = @vpre@_@VOP@_@vsuf@(d1, d2); - sse2_compress4_to_byte_@TYPE@(r1, r2, r3, &r4, &op[i]); - } + LOOP_BLOCKED(@type@, 64) { + @vtype@ a1 = @vpre@_load_@vsuf@(&ip1[i + 0 * 16 / sizeof(@type@)]); + @vtype@ b1 = @vpre@_load_@vsuf@(&ip1[i + 1 * 16 / sizeof(@type@)]); + @vtype@ c1 = @vpre@_load_@vsuf@(&ip1[i + 2 * 16 / sizeof(@type@)]); + @vtype@ d1 = @vpre@_load_@vsuf@(&ip1[i + 3 * 16 / sizeof(@type@)]); + @vtype@ a2 = @vpre@_loadu_@vsuf@(&ip2[i + 0 * 16 / sizeof(@type@)]); + @vtype@ b2 = @vpre@_loadu_@vsuf@(&ip2[i + 1 * 16 / sizeof(@type@)]); + @vtype@ c2 = @vpre@_loadu_@vsuf@(&ip2[i + 2 * 16 / sizeof(@type@)]); + @vtype@ d2 = @vpre@_loadu_@vsuf@(&ip2[i + 3 * 16 / sizeof(@type@)]); + @vtype@ r1 = @vpre@_@VOP@_@vsuf@(a1, a2); + @vtype@ r2 = @vpre@_@VOP@_@vsuf@(b1, b2); + @vtype@ r3 = @vpre@_@VOP@_@vsuf@(c1, c2); + @vtype@ r4 = @vpre@_@VOP@_@vsuf@(d1, d2); + sse2_compress4_to_byte_@TYPE@(r1, r2, r3, &r4, &op[i]); } LOOP_BLOCKED_END { op[i] = sse2_ordered_cmp_@kind@_@TYPE@(ip1[i], ip2[i]); diff --git a/numpy/core/tests/test_numeric.py b/numpy/core/tests/test_numeric.py index d63118080..a114d5a5a 100644 --- a/numpy/core/tests/test_numeric.py +++ b/numpy/core/tests/test_numeric.py @@ -11,7 +11,8 @@ from numpy.core import umath from numpy.random import rand, randint, randn from numpy.testing import ( TestCase, run_module_suite, assert_, assert_equal, assert_raises, - assert_array_equal, assert_almost_equal, assert_array_almost_equal, dec + assert_raises_regex, assert_array_equal, assert_almost_equal, + assert_array_almost_equal, dec ) @@ -234,6 +235,31 @@ class TestBoolCmp(TestCase): self.nf[self.ef] = np.nan self.nd[self.ed] = np.nan + self.inff = self.f.copy() + self.infd = self.d.copy() + self.inff[::3][self.ef[::3]] = np.inf + self.infd[::3][self.ed[::3]] = np.inf + self.inff[1::3][self.ef[1::3]] = -np.inf + self.infd[1::3][self.ed[1::3]] = -np.inf + self.inff[2::3][self.ef[2::3]] = np.nan + self.infd[2::3][self.ed[2::3]] = np.nan + self.efnonan = self.ef.copy() + self.efnonan[2::3] = False + self.ednonan = self.ed.copy() + self.ednonan[2::3] = False + + self.signf = self.f.copy() + self.signd = self.d.copy() + self.signf[self.ef] *= -1. + self.signd[self.ed] *= -1. + self.signf[1::6][self.ef[1::6]] = -np.inf + self.signd[1::6][self.ed[1::6]] = -np.inf + self.signf[3::6][self.ef[3::6]] = -np.nan + self.signd[3::6][self.ed[3::6]] = -np.nan + self.signf[4::6][self.ef[4::6]] = -0. + self.signd[4::6][self.ed[4::6]] = -0. + + def test_float(self): # offset for alignment test for i in range(4): @@ -255,6 +281,10 @@ class TestBoolCmp(TestCase): # isnan on amd64 takes the same codepath assert_array_equal(np.isnan(self.nf[i:]), self.ef[i:]) + assert_array_equal(np.isfinite(self.nf[i:]), ~self.ef[i:]) + assert_array_equal(np.isfinite(self.inff[i:]), ~self.ef[i:]) + assert_array_equal(np.isinf(self.inff[i:]), self.efnonan[i:]) + assert_array_equal(np.signbit(self.signf[i:]), self.ef[i:]) def test_double(self): # offset for alignment test @@ -277,6 +307,10 @@ class TestBoolCmp(TestCase): # isnan on amd64 takes the same codepath assert_array_equal(np.isnan(self.nd[i:]), self.ed[i:]) + assert_array_equal(np.isfinite(self.nd[i:]), ~self.ed[i:]) + assert_array_equal(np.isfinite(self.infd[i:]), ~self.ed[i:]) + assert_array_equal(np.isinf(self.infd[i:]), self.ednonan[i:]) + assert_array_equal(np.signbit(self.signd[i:]), self.ed[i:]) class TestSeterr(TestCase): @@ -2029,6 +2063,80 @@ class TestRollaxis(TestCase): assert_(not res.flags['OWNDATA']) +class TestMoveaxis(TestCase): + def test_move_to_end(self): + x = np.random.randn(5, 6, 7) + for source, expected in [(0, (6, 7, 5)), + (1, (5, 7, 6)), + (2, (5, 6, 7)), + (-1, (5, 6, 7))]: + actual = np.moveaxis(x, source, -1).shape + assert_(actual, expected) + + def test_move_new_position(self): + x = np.random.randn(1, 2, 3, 4) + for source, destination, expected in [ + (0, 1, (2, 1, 3, 4)), + (1, 2, (1, 3, 2, 4)), + (1, -1, (1, 3, 4, 2)), + ]: + actual = np.moveaxis(x, source, destination).shape + assert_(actual, expected) + + def test_preserve_order(self): + x = np.zeros((1, 2, 3, 4)) + for source, destination in [ + (0, 0), + (3, -1), + (-1, 3), + ([0, -1], [0, -1]), + ([2, 0], [2, 0]), + (range(4), range(4)), + ]: + actual = np.moveaxis(x, source, destination).shape + assert_(actual, (1, 2, 3, 4)) + + def test_move_multiples(self): + x = np.zeros((0, 1, 2, 3)) + for source, destination, expected in [ + ([0, 1], [2, 3], (2, 3, 0, 1)), + ([2, 3], [0, 1], (2, 3, 0, 1)), + ([0, 1, 2], [2, 3, 0], (2, 3, 0, 1)), + ([3, 0], [1, 0], (0, 3, 1, 2)), + ([0, 3], [0, 1], (0, 3, 1, 2)), + ]: + actual = np.moveaxis(x, source, destination).shape + assert_(actual, expected) + + def test_errors(self): + x = np.random.randn(1, 2, 3) + assert_raises_regex(ValueError, 'invalid axis .* `source`', + np.moveaxis, x, 3, 0) + assert_raises_regex(ValueError, 'invalid axis .* `source`', + np.moveaxis, x, -4, 0) + assert_raises_regex(ValueError, 'invalid axis .* `destination`', + np.moveaxis, x, 0, 5) + assert_raises_regex(ValueError, 'repeated axis in `source`', + np.moveaxis, x, [0, 0], [0, 1]) + assert_raises_regex(ValueError, 'repeated axis in `destination`', + np.moveaxis, x, [0, 1], [1, 1]) + assert_raises_regex(ValueError, 'must have the same number', + np.moveaxis, x, 0, [0, 1]) + assert_raises_regex(ValueError, 'must have the same number', + np.moveaxis, x, [0, 1], [0]) + + def test_array_likes(self): + x = np.ma.zeros((1, 2, 3)) + result = np.moveaxis(x, 0, 0) + assert_(x.shape, result.shape) + assert_(isinstance(result, np.ma.MaskedArray)) + + x = [1, 2, 3] + result = np.moveaxis(x, 0, 0) + assert_(x, list(result)) + assert_(isinstance(result, np.ndarray)) + + class TestCross(TestCase): def test_2x2(self): u = [1, 2] |