Visitar URL original
GH-100425: Improve accuracy of builtin sum() for float inputs by rhettinger · Pull Request #100426 · python/cpython · GitHub
Skip to content

GH-100425: Improve accuracy of builtin sum() for float inputs #100426

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
rhettinger merged 13 commits into from
Dec 23, 2022
Merged

GH-100425: Improve accuracy of builtin sum() for float inputs #100426

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
916ab7e
Neumaier compensated summation
rhettinger Dec 22, 2022
787f2e5
Kahan summation
rhettinger Dec 22, 2022
bc308aa
Revert "Kahan summation"
rhettinger Dec 22, 2022
0c2ba04
Restore the bogus sign tests
rhettinger Dec 22, 2022
89e46c8
Add tests and comments
rhettinger Dec 22, 2022
a9cb7d2
Add tests and docs
rhettinger Dec 22, 2022
839cc05
Add blurb
rhettinger Dec 22, 2022
66db937
Remove out of date example for fsum()
rhettinger Dec 22, 2022
af7613e
Update floating point tutorial
rhettinger Dec 22, 2022
4f4b731
Correct handling of overflowed or infinite sums
rhettinger Dec 23, 2022
4da0b8f
Move local declarations inside the block where they are used.
rhettinger Dec 23, 2022
6c7894c
For consistent results, apply compensation on the slow path as well.
rhettinger Dec 23, 2022
3cf9536
Mark as an implementation detail.
rhettinger Dec 23, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Revert "Kahan summation" 
This reverts commit 787f2e5.
  • Loading branch information
@rhettinger
rhettinger committed Dec 22, 2022
commit bc308aabca045f22e2607e2448b2c8608951d989
7 changes: 3 additions & 4 deletions Lib/test/test_builtin.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1614,10 +1614,9 @@ def test_sum(self):
self.assertEqual(sum((i / 2 for i in range(10)), 1000.25), 1022.75)
self.assertEqual(sum([0.5, 1]), 1.5)
self.assertEqual(sum([1, 0.5]), 1.5)
self.assertEqual(sum([0.1] * 10), 1.0) # Test accuracy of Kahan summation
self.assertEqual(repr(sum([-0.0])), '0.0')
self.assertEqual(repr(sum([-0.0], -0.0)), '-0.0')
self.assertEqual(repr(sum([], -0.0)), '-0.0')
# self.assertEqual(repr(sum([-0.0])), '0.0')
# self.assertEqual(repr(sum([-0.0], -0.0)), '-0.0')
# self.assertEqual(repr(sum([], -0.0)), '-0.0')

self.assertRaises(TypeError, sum)
self.assertRaises(TypeError, sum, 42)
Expand Down
15 changes: 9 additions & 6 deletions Python/bltinmodule.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2533,22 +2533,25 @@ builtin_sum_impl(PyObject *module, PyObject *iterable, PyObject *start)
if (PyFloat_CheckExact(result)) {
double f_result = PyFloat_AS_DOUBLE(result);
double c = 0.0;
double x, t;
Py_SETREF(result, NULL);
while(result == NULL) {
item = PyIter_Next(iter);
if (item == NULL) {
Py_DECREF(iter);
if (PyErr_Occurred())
return NULL;
return PyFloat_FromDouble(f_result);
return PyFloat_FromDouble(f_result + c);
}
if (PyFloat_CheckExact(item)) {
// Kahan compensated summation
double x, y, t;
// Neumaier compensated summation
x = PyFloat_AS_DOUBLE(item);
y = x - c;
t = f_result + y;
c = (t - f_result) - y;
t = f_result + x;
if (fabs(f_result) >= fabs(x)) {
c += (f_result - t) + x;
} else {
c += (x - t) + f_result;
}
f_result = t;
_Py_DECREF_SPECIALIZED(item, _PyFloat_ExactDealloc);
continue;
Expand Down