ENH: do integer**2. inplace #8896
Conversation
juliantaylor
force-pushed
the
inplace-square
branch
2 times, most recently
from
02cbcc4
to
34c7c4d
Compare
numpy/core/src/multiarray/number.c
Outdated
| @@ -589,24 +589,24 @@ fast_scalar_power(PyArrayObject *a1, PyObject *o2, int inplace) | |||
| (a1, (PyObject *)a1, fastop); | |||
| } | |||
| else { | |||
| PyArray_Descr *dtype = NULL; | |||
| PyObject *res; | |||
|
Actually, a bigger comment after all: why is |
juliantaylor
force-pushed
the
inplace-square
branch
from
34c7c4d
to
970f4b9
Compare
|
good question, square should work just fine. |
juliantaylor
force-pushed
the
inplace-square
branch
from
9657b5b
to
94f224b
Compare
|
All looks good to me. Any reason not to merge? |
numpy/core/tests/test_umath.py
Outdated
| assert_((x**2.00001).dtype is (x**2.0).dtype) | ||
| res = x**2.0 | ||
| assert_((x**2.00001).dtype is res.dtype) | ||
| assert_(res is not x) |
There was a problem hiding this comment.
I think I'd like a not np.shares_memory(res, x) here too - far too easy to break identity while still having a view.
Also a check that x is still [1, 2, 3] would do the job
There was a problem hiding this comment.
good idea, done
Squaring integer arrays with float argument always casts to double so the squaring itself can be done inplace. Also use square fastop instead of multiply.
juliantaylor
force-pushed
the
inplace-square
branch
from
94f224b
to
9a2f0fe
Compare
|
Release note? |
|
this is really minor, not worth a note |
|
Was thinking it would fit in with the other comments about avoiding temporaries, but fair enough. |
Squaring integer arrays with float argument always casts to double so
the squaring itself can be done inplace.