ENG, SIMD: Dispatch bool (invert, add, fmax, fmin...)

  Following functions are defined by umath generator
  to enable runtime dispatching without the need
  to redefine them within dsipatch-able sources:

    BOOL_invert, BOOL_add, BOOL_bitwise_and
    BOOL_bitwise_or, BOOL_logical_xor
    BOOL_bitwise_xor, BOOL_multiply
    BOOL_maximum, BOOL_minimum, BOOL_fmax,
    BOOL_fmin

2 files changed, 62 insertions, 34 deletions

diff --git a/numpy/core/code_generators/generate_umath.py b/numpy/core/code_generators/generate_umath.py
index 114c743e2..bf1a05ee4 100644
--- a/numpy/core/code_generators/generate_umath.py
+++ b/numpy/core/code_generators/generate_umath.py
@@ -246,7 +246,8 @@ chartoname = {
     'P': 'OBJECT',
 }
 
-noobj = '?bBhHiIlLqQefdgFDGmM'
+no_obj_bool = 'bBhHiIlLqQefdgFDGmM'
+noobj = '?' + no_obj_bool
 all = '?bBhHiIlLqQefdgFDGOmM'
 
 O = 'O'
@@ -280,6 +281,7 @@ nocmplxO = nocmplx+O
 nocmplxP = nocmplx+P
 notimes_or_obj = bints + inexact
 nodatetime_or_obj = bints + inexact
+no_bool_times_obj = ints + inexact
 
 # Find which code corresponds to int64.
 int64 = ''
@@ -299,7 +301,9 @@ defdict = {
     Ufunc(2, 1, Zero,
           docstrings.get('numpy.core.umath.add'),
           'PyUFunc_AdditionTypeResolver',
-          TD(notimes_or_obj, simd=[('avx2', ints)], dispatch=[('loops_arithm_fp', 'fdFD')]),
+          TD('?', cfunc_alias='logical_or', dispatch=[('loops_logical', '?')]),
+          TD(no_bool_times_obj, simd=[('avx2', ints)],
+                                dispatch=[('loops_arithm_fp', 'fdFD')]),
           [TypeDescription('M', FullTypeDescr, 'Mm', 'M'),
            TypeDescription('m', FullTypeDescr, 'mm', 'm'),
            TypeDescription('M', FullTypeDescr, 'mM', 'M'),
@@ -310,7 +314,8 @@ 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(ints + inexact, simd=[('avx2', ints)], dispatch=[('loops_arithm_fp', 'fdFD')]),
+          TD(no_bool_times_obj, simd=[('avx2', ints)],
+                                dispatch=[('loops_arithm_fp', 'fdFD')]),
           [TypeDescription('M', FullTypeDescr, 'Mm', 'M'),
            TypeDescription('m', FullTypeDescr, 'mm', 'm'),
            TypeDescription('M', FullTypeDescr, 'MM', 'm'),
@@ -321,7 +326,10 @@ defdict = {
     Ufunc(2, 1, One,
           docstrings.get('numpy.core.umath.multiply'),
           'PyUFunc_MultiplicationTypeResolver',
-          TD(notimes_or_obj, simd=[('avx2', ints)], dispatch=[('loops_arithm_fp', 'fdFD')]),
+          TD('?', cfunc_alias='logical_and',
+                  dispatch=[('loops_logical', '?')]),
+          TD(no_bool_times_obj, simd=[('avx2', ints)],
+                                dispatch=[('loops_arithm_fp', 'fdFD')]),
           [TypeDescription('m', FullTypeDescr, 'mq', 'm'),
            TypeDescription('m', FullTypeDescr, 'qm', 'm'),
            TypeDescription('m', FullTypeDescr, 'md', 'm'),
@@ -412,8 +420,8 @@ defdict = {
     Ufunc(1, 1, None,
           docstrings.get('numpy.core.umath.absolute'),
           'PyUFunc_AbsoluteTypeResolver',
-          TD(bints+flts+timedeltaonly, dispatch=[('loops_unary_fp', 'fd')]),
-          TD('?', dispatch=[('loops_logical', '?')]),
+          TD(bints+flts+timedeltaonly, dispatch=[('loops_unary_fp', 'fd'),
+                                                 ('loops_logical', '?')]),
           TD(cmplx, simd=[('avx512f', cmplxvec)], out=('f', 'd', 'g')),
           TD(O, f='PyNumber_Absolute'),
           ),
@@ -497,31 +505,33 @@ defdict = {
     Ufunc(2, 1, True_,
           docstrings.get('numpy.core.umath.logical_and'),
           'PyUFunc_SimpleBinaryComparisonTypeResolver',
-          TD('?', dispatch=[('loops_logical', '?')]),
-          TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)]),
+          TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)],
+                                dispatch=[('loops_logical', '?')]),
           TD(O, f='npy_ObjectLogicalAnd'),
           ),
 'logical_not':
     Ufunc(1, 1, None,
           docstrings.get('numpy.core.umath.logical_not'),
           None,
-          TD('?', dispatch=[('loops_logical', '?')]),
-          TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)]),
+          TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)],
+                                dispatch=[('loops_logical', '?')]),
           TD(O, f='npy_ObjectLogicalNot'),
           ),
 'logical_or':
     Ufunc(2, 1, False_,
           docstrings.get('numpy.core.umath.logical_or'),
           'PyUFunc_SimpleBinaryComparisonTypeResolver',
-          TD('?', dispatch=[('loops_logical', '?')]),
-          TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)]),
+          TD(nodatetime_or_obj, out='?', simd=[('avx2', ints)],
+                                dispatch=[('loops_logical', '?')]),
           TD(O, f='npy_ObjectLogicalOr'),
           ),
 'logical_xor':
     Ufunc(2, 1, False_,
           docstrings.get('numpy.core.umath.logical_xor'),
           'PyUFunc_SimpleBinaryComparisonTypeResolver',
-          TD(nodatetime_or_obj, out='?'),
+          TD('?', out='?', cfunc_alias='not_equal',
+                           dispatch=[('loops_comparison', '?')]),
+          TD(no_bool_times_obj, out='?'),
           # TODO: using obj.logical_xor() seems pretty much useless:
           TD(P, f='logical_xor'),
           ),
@@ -529,14 +539,17 @@ defdict = {
     Ufunc(2, 1, ReorderableNone,
           docstrings.get('numpy.core.umath.maximum'),
           'PyUFunc_SimpleUniformOperationTypeResolver',
-          TD(noobj, dispatch=[('loops_minmax', ints+'fdg')]),
+          TD('?', cfunc_alias='logical_or', dispatch=[('loops_logical', '?')]),
+          TD(no_obj_bool, dispatch=[('loops_minmax', ints+'fdg')]),
           TD(O, f='npy_ObjectMax')
           ),
 'minimum':
     Ufunc(2, 1, ReorderableNone,
           docstrings.get('numpy.core.umath.minimum'),
           'PyUFunc_SimpleUniformOperationTypeResolver',
-          TD(noobj, dispatch=[('loops_minmax', ints+'fdg')]),
+          TD('?', cfunc_alias='logical_and',
+                  dispatch=[('loops_logical', '?')]),
+          TD(no_obj_bool, dispatch=[('loops_minmax', ints+'fdg')]),
           TD(O, f='npy_ObjectMin')
           ),
 'clip':
@@ -550,14 +563,17 @@ defdict = {
     Ufunc(2, 1, ReorderableNone,
           docstrings.get('numpy.core.umath.fmax'),
           'PyUFunc_SimpleUniformOperationTypeResolver',
-          TD(noobj, dispatch=[('loops_minmax', 'fdg')]),
+          TD('?', cfunc_alias='logical_or', dispatch=[('loops_logical', '?')]),
+          TD(no_obj_bool, dispatch=[('loops_minmax', 'fdg')]),
           TD(O, f='npy_ObjectMax')
           ),
 'fmin':
     Ufunc(2, 1, ReorderableNone,
           docstrings.get('numpy.core.umath.fmin'),
           'PyUFunc_SimpleUniformOperationTypeResolver',
-          TD(noobj, dispatch=[('loops_minmax', 'fdg')]),
+          TD('?', cfunc_alias='logical_and',
+                  dispatch=[('loops_logical', '?')]),
+          TD(no_obj_bool, dispatch=[('loops_minmax', 'fdg')]),
           TD(O, f='npy_ObjectMin')
           ),
 'logaddexp':
@@ -576,28 +592,35 @@ defdict = {
     Ufunc(2, 1, AllOnes,
           docstrings.get('numpy.core.umath.bitwise_and'),
           None,
-          TD(bints, simd=[('avx2', ints)]),
+          TD('?', cfunc_alias='logical_and',
+                  dispatch=[('loops_logical', '?')]),
+          TD(ints, simd=[('avx2', ints)]),
           TD(O, f='PyNumber_And'),
           ),
 'bitwise_or':
     Ufunc(2, 1, Zero,
           docstrings.get('numpy.core.umath.bitwise_or'),
           None,
-          TD(bints, simd=[('avx2', ints)]),
+          TD('?', cfunc_alias='logical_or', dispatch=[('loops_logical', '?')]),
+          TD(ints, simd=[('avx2', ints)]),
           TD(O, f='PyNumber_Or'),
           ),
 'bitwise_xor':
     Ufunc(2, 1, Zero,
           docstrings.get('numpy.core.umath.bitwise_xor'),
           None,
-          TD(bints, simd=[('avx2', ints)]),
+          TD('?', cfunc_alias='not_equal',
+                  dispatch=[('loops_comparison', '?')]),
+          TD(ints, simd=[('avx2', ints)]),
           TD(O, f='PyNumber_Xor'),
           ),
 'invert':
     Ufunc(1, 1, None,
           docstrings.get('numpy.core.umath.invert'),
           None,
-          TD(bints, simd=[('avx2', ints)]),
+          TD('?', cfunc_alias='logical_not',
+                  dispatch=[('loops_logical', '?')]),
+          TD(ints, simd=[('avx2', ints)]),
           TD(O, f='PyNumber_Invert'),
           ),
 'left_shift':
diff --git a/numpy/core/src/umath/loops.h.src b/numpy/core/src/umath/loops.h.src
index 2e0ba9c40..411d53e94 100644
--- a/numpy/core/src/umath/loops.h.src
+++ b/numpy/core/src/umath/loops.h.src
@@ -10,24 +10,29 @@
     #define NPY_NO_EXPORT NPY_VISIBILITY_HIDDEN
 #endif
 
-#define BOOL_invert BOOL_logical_not
-#define BOOL_add BOOL_logical_or
-#define BOOL_bitwise_and BOOL_logical_and
-#define BOOL_bitwise_or BOOL_logical_or
-#define BOOL_logical_xor BOOL_not_equal
-#define BOOL_bitwise_xor BOOL_logical_xor
-#define BOOL_multiply BOOL_logical_and
-#define BOOL_maximum BOOL_logical_or
-#define BOOL_minimum BOOL_logical_and
-#define BOOL_fmax BOOL_maximum
-#define BOOL_fmin BOOL_minimum
-
 /*
  *****************************************************************************
  **                             BOOLEAN LOOPS                               **
  *****************************************************************************
  */
 
+/*
+ * Following functions are defined by umath generator
+ * to enable runtime dispatching without the need
+ * to redefine them within dsipatch-able sources.
+ */
+// #define BOOL_invert BOOL_logical_not
+// #define BOOL_add BOOL_logical_or
+// #define BOOL_bitwise_and BOOL_logical_and
+// #define BOOL_bitwise_or BOOL_logical_or
+// #define BOOL_logical_xor BOOL_not_equal
+// #define BOOL_bitwise_xor BOOL_logical_xor
+// #define BOOL_multiply BOOL_logical_and
+// #define BOOL_maximum BOOL_logical_or
+// #define BOOL_minimum BOOL_logical_and
+// #define BOOL_fmax BOOL_maximum
+// #define BOOL_fmin BOOL_minimum
+
 #ifndef NPY_DISABLE_OPTIMIZATION
     #include "loops_comparison.dispatch.h"
 #endif
@@ -207,7 +212,7 @@ NPY_NO_EXPORT void
 
 /**end repeat**/
 
- 
+
 #ifndef NPY_DISABLE_OPTIMIZATION
     #include "loops_unary.dispatch.h"
 #endif