summaryrefslogtreecommitdiff
path: root/numpy
diff options
context:
space:
mode:
Diffstat (limited to 'numpy')
-rw-r--r--numpy/core/code_generators/generate_umath.py44
-rw-r--r--numpy/core/src/umath/loops.c.src71
-rw-r--r--numpy/core/src/umath/loops.h.src32
3 files changed, 90 insertions, 57 deletions
diff --git a/numpy/core/code_generators/generate_umath.py b/numpy/core/code_generators/generate_umath.py
index 1f066db69..21dab54c5 100644
--- a/numpy/core/code_generators/generate_umath.py
+++ b/numpy/core/code_generators/generate_umath.py
@@ -259,7 +259,7 @@ defdict = {
Ufunc(2, 1, Zero,
docstrings.get('numpy.core.umath.add'),
'PyUFunc_AdditionTypeResolver',
- TD(notimes_or_obj),
+ TD(notimes_or_obj, simd=[('avx2', ints)]),
[TypeDescription('M', FullTypeDescr, 'Mm', 'M'),
TypeDescription('m', FullTypeDescr, 'mm', 'm'),
TypeDescription('M', FullTypeDescr, 'mM', 'M'),
@@ -270,7 +270,7 @@ defdict = {
Ufunc(2, 1, None, # Zero is only a unit to the right, not the left
docstrings.get('numpy.core.umath.subtract'),
'PyUFunc_SubtractionTypeResolver',
- TD(notimes_or_obj),
+ TD(notimes_or_obj, simd=[('avx2', ints)]),
[TypeDescription('M', FullTypeDescr, 'Mm', 'M'),
TypeDescription('m', FullTypeDescr, 'mm', 'm'),
TypeDescription('M', FullTypeDescr, 'MM', 'm'),
@@ -281,7 +281,7 @@ defdict = {
Ufunc(2, 1, One,
docstrings.get('numpy.core.umath.multiply'),
'PyUFunc_MultiplicationTypeResolver',
- TD(notimes_or_obj),
+ TD(notimes_or_obj, simd=[('avx2', ints)]),
[TypeDescription('m', FullTypeDescr, 'mq', 'm'),
TypeDescription('m', FullTypeDescr, 'qm', 'm'),
TypeDescription('m', FullTypeDescr, 'md', 'm'),
@@ -328,7 +328,7 @@ defdict = {
Ufunc(1, 1, None,
docstrings.get('numpy.core.umath.conjugate'),
None,
- TD(ints+flts+cmplx),
+ TD(ints+flts+cmplx, simd=[('avx2', ints)]),
TD(P, f='conjugate'),
),
'fmod':
@@ -343,14 +343,14 @@ defdict = {
Ufunc(1, 1, None,
docstrings.get('numpy.core.umath.square'),
None,
- TD(ints+inexact),
+ TD(ints+inexact, simd=[('avx2', ints)]),
TD(O, f='Py_square'),
),
'reciprocal':
Ufunc(1, 1, None,
docstrings.get('numpy.core.umath.reciprocal'),
None,
- TD(ints+inexact),
+ TD(ints+inexact, simd=[('avx2', ints)]),
TD(O, f='Py_reciprocal'),
),
# This is no longer used as numpy.ones_like, however it is
@@ -388,7 +388,7 @@ defdict = {
Ufunc(1, 1, None,
docstrings.get('numpy.core.umath.negative'),
'PyUFunc_NegativeTypeResolver',
- TD(bints+flts+timedeltaonly),
+ TD(bints+flts+timedeltaonly, simd=[('avx2', ints)]),
TD(cmplx, f='neg'),
TD(O, f='PyNumber_Negative'),
),
@@ -402,57 +402,57 @@ defdict = {
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.greater'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(all, out='?'),
+ TD(all, out='?', simd=[('avx2', ints)]),
),
'greater_equal':
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.greater_equal'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(all, out='?'),
+ TD(all, out='?', simd=[('avx2', ints)]),
),
'less':
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.less'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(all, out='?'),
+ TD(all, out='?', simd=[('avx2', ints)]),
),
'less_equal':
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.less_equal'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(all, out='?'),
+ TD(all, out='?', simd=[('avx2', ints)]),
),
'equal':
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.equal'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(all, out='?'),
+ TD(all, out='?', simd=[('avx2', ints)]),
),
'not_equal':
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.not_equal'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(all, out='?'),
+ TD(all, out='?', simd=[('avx2', ints)]),
),
'logical_and':
Ufunc(2, 1, One,
docstrings.get('numpy.core.umath.logical_and'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(nodatetime_or_obj, out='?'),
+ TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)]),
TD(O, f='npy_ObjectLogicalAnd'),
),
'logical_not':
Ufunc(1, 1, None,
docstrings.get('numpy.core.umath.logical_not'),
None,
- TD(nodatetime_or_obj, out='?'),
+ TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)]),
TD(O, f='npy_ObjectLogicalNot'),
),
'logical_or':
Ufunc(2, 1, Zero,
docstrings.get('numpy.core.umath.logical_or'),
'PyUFunc_SimpleBinaryComparisonTypeResolver',
- TD(nodatetime_or_obj, out='?'),
+ TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)]),
TD(O, f='npy_ObjectLogicalOr'),
),
'logical_xor':
@@ -506,42 +506,42 @@ defdict = {
Ufunc(2, 1, AllOnes,
docstrings.get('numpy.core.umath.bitwise_and'),
None,
- TD(bints),
+ TD(bints, simd=[('avx2', ints)]),
TD(O, f='PyNumber_And'),
),
'bitwise_or':
Ufunc(2, 1, Zero,
docstrings.get('numpy.core.umath.bitwise_or'),
None,
- TD(bints),
+ TD(bints, simd=[('avx2', ints)]),
TD(O, f='PyNumber_Or'),
),
'bitwise_xor':
Ufunc(2, 1, Zero,
docstrings.get('numpy.core.umath.bitwise_xor'),
None,
- TD(bints),
+ TD(bints, simd=[('avx2', ints)]),
TD(O, f='PyNumber_Xor'),
),
'invert':
Ufunc(1, 1, None,
docstrings.get('numpy.core.umath.invert'),
None,
- TD(bints),
+ TD(bints, simd=[('avx2', ints)]),
TD(O, f='PyNumber_Invert'),
),
'left_shift':
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.left_shift'),
None,
- TD(ints),
+ TD(ints, simd=[('avx2', ints)]),
TD(O, f='PyNumber_Lshift'),
),
'right_shift':
Ufunc(2, 1, None,
docstrings.get('numpy.core.umath.right_shift'),
None,
- TD(ints),
+ TD(ints, simd=[('avx2', ints)]),
TD(O, f='PyNumber_Rshift'),
),
'degrees':
diff --git a/numpy/core/src/umath/loops.c.src b/numpy/core/src/umath/loops.c.src
index 157b30e70..60b5464f1 100644
--- a/numpy/core/src/umath/loops.c.src
+++ b/numpy/core/src/umath/loops.c.src
@@ -786,51 +786,71 @@ NPY_NO_EXPORT void
}
}
-NPY_NO_EXPORT NPY_GCC_OPT_3 void
-@TYPE@_square(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data))
+/**begin repeat1
+ * #isa = , _avx2#
+ * #ISA = , AVX2#
+ * #CHK = 1, HAVE_ATTRIBUTE_TARGET_AVX2#
+ * #ATTR = , NPY_GCC_TARGET_AVX2#
+ */
+
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_square@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data))
{
UNARY_LOOP_FAST(@type@, @type@, *out = in * in);
}
+#endif
-NPY_NO_EXPORT void
-@TYPE@_reciprocal(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_reciprocal@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data))
{
UNARY_LOOP_FAST(@type@, @type@, *out = 1.0 / in);
}
+#endif
-NPY_NO_EXPORT void
-@TYPE@_conjugate(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_conjugate@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = in);
}
+#endif
-NPY_NO_EXPORT NPY_GCC_OPT_3 void
-@TYPE@_negative(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_negative@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = -in);
}
+#endif
-NPY_NO_EXPORT NPY_GCC_OPT_3 void
-@TYPE@_logical_not(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_logical_not@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, npy_bool, *out = !in);
}
+#endif
-NPY_NO_EXPORT NPY_GCC_OPT_3 void
-@TYPE@_invert(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_invert@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
{
UNARY_LOOP_FAST(@type@, @type@, *out = ~in);
}
+#endif
-/**begin repeat1
+/**begin repeat2
* Arithmetic
* #kind = add, subtract, multiply, bitwise_and, bitwise_or, bitwise_xor,
* left_shift, right_shift#
* #OP = +, -,*, &, |, ^, <<, >>#
*/
-NPY_NO_EXPORT NPY_GCC_OPT_3 void
-@TYPE@_@kind@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_@kind@@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
{
if(IS_BINARY_REDUCE) {
BINARY_REDUCE_LOOP(@type@) {
@@ -842,17 +862,19 @@ NPY_NO_EXPORT NPY_GCC_OPT_3 void
BINARY_LOOP_FAST(@type@, @type@, *out = in1 @OP@ in2);
}
}
+#endif
-/**end repeat1**/
+/**end repeat2**/
-/**begin repeat1
+/**begin repeat2
* #kind = equal, not_equal, greater, greater_equal, less, less_equal,
* logical_and, logical_or#
* #OP = ==, !=, >, >=, <, <=, &&, ||#
*/
-NPY_NO_EXPORT NPY_GCC_OPT_3 void
-@TYPE@_@kind@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_@kind@@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
{
/*
* gcc vectorization of this is not good (PR60575) but manual integer
@@ -860,11 +882,13 @@ NPY_NO_EXPORT NPY_GCC_OPT_3 void
*/
BINARY_LOOP_FAST(@type@, npy_bool, *out = in1 @OP@ in2);
}
+#endif
-/**end repeat1**/
+/**end repeat2**/
-NPY_NO_EXPORT void
-@TYPE@_logical_xor(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
+#if @CHK@
+NPY_NO_EXPORT NPY_GCC_OPT_3 @ATTR@ void
+@TYPE@_logical_xor@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func))
{
BINARY_LOOP {
const int t1 = !!*(@type@ *)ip1;
@@ -872,6 +896,9 @@ NPY_NO_EXPORT void
*((npy_bool *)op1) = (t1 != t2);
}
}
+#endif
+
+/**end repeat1**/
/**begin repeat1
* #kind = maximum, minimum#
diff --git a/numpy/core/src/umath/loops.h.src b/numpy/core/src/umath/loops.h.src
index a6e775a3a..67d8d74c3 100644
--- a/numpy/core/src/umath/loops.h.src
+++ b/numpy/core/src/umath/loops.h.src
@@ -61,46 +61,52 @@ BOOL__ones_like(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UN
NPY_NO_EXPORT void
@S@@TYPE@__ones_like(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data));
+/**begin repeat2
+ * #isa = , _avx2#
+ */
+
NPY_NO_EXPORT void
-@S@@TYPE@_square(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data));
+@S@@TYPE@_square@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data));
NPY_NO_EXPORT void
-@S@@TYPE@_reciprocal(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data));
+@S@@TYPE@_reciprocal@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(data));
NPY_NO_EXPORT void
-@S@@TYPE@_conjugate(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+@S@@TYPE@_conjugate@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
NPY_NO_EXPORT void
-@S@@TYPE@_negative(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+@S@@TYPE@_negative@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
NPY_NO_EXPORT void
-@S@@TYPE@_logical_not(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+@S@@TYPE@_logical_not@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
NPY_NO_EXPORT void
-@S@@TYPE@_invert(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+@S@@TYPE@_invert@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
-/**begin repeat2
+/**begin repeat3
* Arithmetic
* #kind = add, subtract, multiply, bitwise_and, bitwise_or, bitwise_xor,
* left_shift, right_shift#
* #OP = +, -,*, &, |, ^, <<, >>#
*/
NPY_NO_EXPORT void
-@S@@TYPE@_@kind@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+@S@@TYPE@_@kind@@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
-/**end repeat2**/
+/**end repeat3**/
-/**begin repeat2
+/**begin repeat3
* #kind = equal, not_equal, greater, greater_equal, less, less_equal,
* logical_and, logical_or#
* #OP = ==, !=, >, >=, <, <=, &&, ||#
*/
NPY_NO_EXPORT void
-@S@@TYPE@_@kind@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
-/**end repeat2**/
+@S@@TYPE@_@kind@@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+
+/**end repeat3**/
NPY_NO_EXPORT void
-@S@@TYPE@_logical_xor(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+@S@@TYPE@_logical_xor@isa@(char **args, npy_intp *dimensions, npy_intp *steps, void *NPY_UNUSED(func));
+/**end repeat2**/
/**begin repeat2
* #kind = maximum, minimum#
View Lorry Controller Status