| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
For most scalars (except void ones) exposing the buffer interface
is trivial and does not require any fancy "format" handling.
Except for unicode, their buffer interface is also static and thus
can simply be hardcoded.
The main advantage, is that currently the buffer interface uses
a global dictionary to store these buffers and thus requires
probing that dictionary for every single scalar dealloc. This
results in an overhead of about 30% for simple operations (such
as slicing and simple math).
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
* ENH: Chain extensions in numpy and numpy/core
Co-authored-by: Ross Barnowski <rossbar@berkeley.edu>
Co-authored-by: Zuhair Ali-Khan <54608785+zalikh2@users.noreply.github.com>
|
|\ \ \ \ \ \
| | | | | | |
| | | | | | | |
BUG: Fix result when a gufunc output broadcasts the inputs.
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
In this case no error was given, but additional dimensions in the
output simply ignored. Thus the returned array was only partially
set.
Now it will be fully set, with the calculation being repeated as
often as necessary (typical broadcasting logic). This is consistent
with how normal ufuncs work.
Closes gh-16484
|
|\ \ \ \ \ \ \
| |_|_|_|/ / /
|/| | | | | | |
MAINT: use zip instead of range in piecewise
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
Minor cleanup while looking at something else.
|
|\ \ \ \ \ \ \
| | | | | | | |
| | | | | | | | |
DOC: Improve assert_warns docstring with example
|
| | | | | | | | |
|
| | | | | | | |
| | | | | | | |
| | | | | | | | |
* Neon implementation with intrinsic for bool argmax
|
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
* TST: Adding tests to validate strided np.ldexp and np.frexp
* ENH: Use AVX-512 for float and double np.ldexp
|
| |/ / / / / /
|/| | | | | |
| | | | | | |
| | | | | | | |
This requires some minor tweaks in `np.random` because there the two have different meanings, with `()` meaning 0d array and `None` meaning scalar.
|
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
* ENH: Use AVX-512 for np.isnan, np.infinite, np.isinf and np.signbit
* TST: Add tests to validate isnan, isfinite, signbit and isinf ufuncs
* BENCH: Adding benchmarks for isnan, isinf, isfinite and signbit
|
| |/ / / / /
|/| | | | |
| | | | | | |
Split from gh-15886.
|
|\ \ \ \ \ \
| | | | | | |
| | | | | | | |
DOC: make clearer that sinc is normalized by a factor pi
|
| | |_|_|_|/
| |/| | | |
| | | | | |
| | | | | | |
Addresses gh-13457
|
|/ / / / / |
|
| | | | | |
|
|\ \ \ \ \
| | | | | |
| | | | | | |
MAINT: support python 3.10
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Co-authored-by: Eric Wieser <wieser.eric@gmail.com>
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
To support the change to Py_SIZE and Py_TYPE the Py_SET_SIZE and
Py_SET_TYPE macros are provided in py39. This provides the same
macros falling back to the old method for < py39.
|
| |/ / / /
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
In https://github.com/python/cpython/pull/20290 CPython changed
`Py_TYPE` from a macro to an inline function. This requires a code
change to us `Py_SET_TYPE` instead when using `Py_TYPE()` as a lvalue
in c code.
In https://github.com/python/cpython/pull/20429 CPython changed
`Py_SIZE` from a macro to an inline function. This requires a code
change to us `Py_SET_SIZE` instead of using `Py_SIZE` as a lvalue in c
code.
|
|/ / / /
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
This was missing the extern, but pre GCC 10 it seemed to have worked
fine without.
Closes gh-16419
|
|\ \ \ \
| | | | |
| | | | | |
DOC: Update assert_warns parameter list
|
| | | | | |
|
|/ / / / |
|
|\ \ \ \
| | | | |
| | | | | |
DEP: Ensure indexing errors will be raised even on empty results
|
| |/ / /
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Previously, when the indexing result was empty, no check was done
for backward compatibility with pre 1.9 (assumingly).
This may have been only necessary when the outer iteration is empty
as opposed to when just the inner iteration is empty, though.
In any case, it is arguably buggy to ignore indexing errors in this
case. Since there may have been a reason back in the day, and this
is probably extremely rare, optiming for a brief deprecation period
for now.
Closes gh-15898
Co-authored-by: Eric Wieser <wieser.eric@gmail.com>
Co-authored-by: Matti Picus <matti.picus@gmail.com>
|
|\ \ \ \
| | | | |
| | | | | |
TST, MAINT: Fix detecting and testing armhf features
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
- fix detect and test Arm Advanced SIMD on aarch32
- fix testing armhf's rootfs on aarch64 kernel
|
|\ \ \ \ \
| | | | | |
| | | | | | |
API: Create Preliminary DTypeMeta class and np.dtype subclasses
|
| | | | | | |
|
| | | | | | |
|
| | | | | | |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Note that these attributes are *not* fixed yet, just a start to
have something.
|
| | | | | | |
|
| | | | | | |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
intention is
|
| | | | | | |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
(currently no public API changes, but the hash changed).
The type safety may not be important as such, but there used
to be issues, at least with MSVC about this, and clang warns.
|
| | |_|/ /
| |/| | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
This makes all dtypes being subclasses of a:
class DType(np.dtype):
is_abstract = False
is_flexible = False % True for void, strings, and datetimes
singleton = prototype_dtype_instance
# other information defined on the singleton currently
subclass of np.dtype.
Right now this happens by replacing the Type field of of the
singleton instance. This is valid for types/objects not created
in python (not HeapTypes). It is necessary to have this functionality
for continuing support of user defined legacy dtypes.
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
This docstring needs to be a raw string so the backslashes
in the example are not processed by Python or Sphinx.
Closes gh-16390.
|
|\ \ \ \ \
| | | | | |
| | | | | | |
BUG: `np.resize` negative shape and subclasses edge case fixes
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
When a negative entry is passed into `np.resize` shape, an error will
now be raised reliably. Previously the `-1` semantics of reshape allowed
edge cases to pass and return incorrect results.
In the corner case of an empty result shape or empty input shape, the
result type will now be preserved.
|
| | | | | | |
|
| |_|_|/ /
|/| | | | |
|
|\ \ \ \ \
| | | | | |
| | | | | | |
BUG: Fix small leaks in error path and `empty_like` with shape
|
| | | | | | |
|
| | | | | | |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
* Fix cpu detection for X86, GCC, and macOS
* Optimize feature detection
Co-authored-by: Matti Picus <matti.picus@gmail.com>
|
|\ \ \ \ \ \
| | | | | | |
| | | | | | | |
ENH: include dt64/td64 structs and funcs and tests in __init__.pxd.
|