Implements dpctl.tensor.where #1147

ndgrigorian · 2023-03-30T15:08:37Z

This PR implements dpt.where function, which takes a condition array and two arrays of data, x1 and x2. The output is an array with elements from x1 where condition evaluates to true and x2 where condition is false. The output is of type determined by type promotion of x1 and x2.

i.e.,

In [4]: condition = dpt.asarray([False, True, True, True])

In [5]: x1 = dpt.asarray([1])

In [6]: x2 = dpt.asarray([2])

In [7]: dpt.where(condition, x1, x2)
Out[7]: usm_ndarray([2, 1, 1, 1])

Have you provided a meaningful PR description?
Have you added a test, reproducer or referred to an issue with a reproducer?
Have you tested your changes locally for CPU and GPU devices?
Have you made sure that new changes do not introduce compiler warnings?
Have you checked performance impact of proposed changes?
If this PR is a work in progress, are you opening the PR as a draft?

github-actions · 2023-03-30T15:31:34Z

View rendered docs @ https://intelpython.github.io/dpctl/pulls/1147/index.html

coveralls · 2023-03-30T15:34:37Z

Coverage: 83.204% (+0.2%) from 83.018% when pulling db43042 on where-impl-gh-1120 into 7bbfce1 on master.

github-actions · 2023-03-30T15:40:44Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_59 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-03-30T15:48:14Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_58 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-03-31T05:30:58Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_67 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

ndgrigorian · 2023-03-31T06:35:58Z

Significant performance gains from re-implemenation:

In [13]: %timeit where(x, x1, x2)
569 µs ± 30.4 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [14]: %timeit where2(x, x1, x2)
860 µs ± 39.5 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [15]: %timeit where(x, x1, x2)
524 µs ± 32.8 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [16]: %timeit where2(x, x1, x2)
851 µs ± 23.9 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Where x is only 60 elements and where2 is the original implementation.

For a larger array (on scale of 10^5 elements)

In [16]: %timeit wf.where2(x, x1, x2)
1.54 ms ± 71.6 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [17]: %timeit wf.where(x, x1, x2)
579 µs ± 59 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

oleksandr-pavlyk · 2023-03-31T13:41:30Z

An issue in type promotion logic:

import dpctl.tensor as dpt

m = dpt.zeros(100, dtype="?")
m[::2] = True; m[1::3] = True; m[2::5] = False

x1 = dpt.ones(100, dtype="i4")
x2 = dpt.zeros(100, dtype="f4")

dpt.where(m, x1, x2)  # raises ValueError

The error is ValueError: Device Intel(R) Graphics [0x9a49] does not provide native support for double-precision floating point type..

github-actions · 2023-04-03T15:28:40Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_68 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-04-03T15:39:58Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_69 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

dpctl/tensor/_search_functions.py

github-actions · 2023-04-04T06:10:35Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_74 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-04-05T03:51:56Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_79 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-04-05T07:13:50Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_82 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

oleksandr-pavlyk · 2023-04-05T17:20:58Z

Great to see coverage experience such an upward jump 👍

dpctl/tensor/_search_functions.py

dpctl/tensor/libtensor/source/where.cpp

oleksandr-pavlyk · 2023-04-05T18:34:40Z

dpctl/tensor/libtensor/source/where.cpp

+    // destination must be ample enough to accomodate all elements
+    {
+        size_t range =
+            static_cast<size_t>(dst_offsets.second - dst_offsets.first);
+        if (range + 1 < static_cast<size_t>(nelems)) {
+            throw py::value_error(
+                "Memory addressed by the destination array can not "
+                "accomodate all the "
+                "array elements.");
+        }
+    }


Sometime down the road it would be good to modularize this check into a utility as well. Perhaps dpctl::tensor::utils::can_accomodate(nelems, dst). This is best done in a separate PR.

I was thinking about this as well during refactoring. I'll make note of it.

dpctl/tensor/_search_functions.py

github-actions · 2023-04-05T18:46:14Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_83 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

oleksandr-pavlyk

Looks awesome! Thank you @ndgrigorian

github-actions · 2023-04-05T21:07:16Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_84 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-04-06T02:35:34Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_86 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

dpctl/tensor/_search_functions.py

github-actions · 2023-04-06T18:35:33Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_89 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-04-06T20:45:36Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_90 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

- Utility functions are for finding an output type for universal and binary functions when the device of allocation lacks fp16 or fp64

- Where now outputs an F-contiguous array when all inputs are F-contiguous - Where now outputs a empty 0D array if any input is a 0D empty array - Added tests for these cases Fixed incorrect logic in where test

github-actions · 2023-04-07T17:57:21Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_93 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-04-07T18:09:33Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_86 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

- Asymmetric dtype test to improve coverage

oleksandr-pavlyk

Thank you, @ndgrigorian !

oleksandr-pavlyk · 2023-04-07T19:10:32Z

Please wait till github-action CI has finished and merge. This way artifacts would be ready to get published on dppy/label/dev right away once the PR is merged.

github-actions · 2023-04-07T19:18:40Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_87 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

github-actions · 2023-04-07T20:38:59Z

Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞

github-actions · 2023-04-07T21:11:52Z

Array API standard conformance tests for dpctl=0.14.3dev0=py310h76be34b_87 ran successfully.
Passed: 47
Failed: 787
Skipped: 280

oleksandr-pavlyk reviewed Apr 4, 2023

View reviewed changes

dpctl/tensor/_search_functions.py Outdated Show resolved Hide resolved

ndgrigorian marked this pull request as ready for review April 5, 2023 17:19

ndgrigorian requested a review from oleksandr-pavlyk April 5, 2023 17:20

ndgrigorian force-pushed the where-impl-gh-1120 branch from 3cda10f to e0f08ad Compare April 5, 2023 18:09

ndgrigorian changed the title ~~Implements dpt.where~~ Implements dpctl.tensor.where Apr 5, 2023

oleksandr-pavlyk suggested changes Apr 5, 2023

View reviewed changes

oleksandr-pavlyk self-requested a review April 5, 2023 20:47

oleksandr-pavlyk previously approved these changes Apr 5, 2023

View reviewed changes

ndgrigorian dismissed oleksandr-pavlyk’s stale review via a01db12 April 6, 2023 01:56

ndgrigorian force-pushed the where-impl-gh-1120 branch from 5c3c5d1 to a01db12 Compare April 6, 2023 01:56

ndgrigorian requested a review from oleksandr-pavlyk April 6, 2023 14:51

oleksandr-pavlyk reviewed Apr 6, 2023

View reviewed changes

dpctl/tensor/_search_functions.py Outdated Show resolved Hide resolved

ndgrigorian force-pushed the where-impl-gh-1120 branch from 2e0bb63 to b4967a4 Compare April 7, 2023 17:22

ndgrigorian added 5 commits April 7, 2023 10:34

Implements dpt.where

a0e44d2

Added utility functions, basic where tests

9018a79

- Utility functions are for finding an output type for universal and binary functions when the device of allocation lacks fp16 or fp64

Where changed for empty and F-contiguous input

b35b083

- Where now outputs an F-contiguous array when all inputs are F-contiguous - Where now outputs a empty 0D array if any input is a 0D empty array - Added tests for these cases Fixed incorrect logic in where test

Added tests for where and type utility functions

be96aec

Made changes as per PR review by @oleksandr-pavlyk

d50d1c6

ndgrigorian force-pushed the where-impl-gh-1120 branch from b4967a4 to d50d1c6 Compare April 7, 2023 17:34

oleksandr-pavlyk self-requested a review April 7, 2023 17:53

Added where test

db43042

- Asymmetric dtype test to improve coverage

oleksandr-pavlyk approved these changes Apr 7, 2023

View reviewed changes

ndgrigorian merged commit a504e96 into master Apr 7, 2023

ndgrigorian deleted the where-impl-gh-1120 branch April 7, 2023 20:38

Implements dpctl.tensor.where #1147

Implements dpctl.tensor.where #1147

Uh oh!

Conversation

ndgrigorian commented Mar 30, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Mar 30, 2023

Uh oh!

coveralls commented Mar 30, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Mar 30, 2023

Uh oh!

github-actions bot commented Mar 30, 2023

Uh oh!

github-actions bot commented Mar 31, 2023

Uh oh!

ndgrigorian commented Mar 31, 2023

Uh oh!

oleksandr-pavlyk commented Mar 31, 2023

Uh oh!

github-actions bot commented Apr 3, 2023

Uh oh!

github-actions bot commented Apr 3, 2023

Uh oh!

Uh oh!

github-actions bot commented Apr 4, 2023

Uh oh!

github-actions bot commented Apr 5, 2023

Uh oh!

github-actions bot commented Apr 5, 2023

Uh oh!

oleksandr-pavlyk commented Apr 5, 2023

Uh oh!

Uh oh!

Uh oh!

Uh oh!

oleksandr-pavlyk Apr 5, 2023

Choose a reason for hiding this comment

Uh oh!

ndgrigorian Apr 5, 2023

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions bot commented Apr 5, 2023

Uh oh!

oleksandr-pavlyk left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Apr 5, 2023

Uh oh!

github-actions bot commented Apr 6, 2023

Uh oh!

Uh oh!

github-actions bot commented Apr 6, 2023

Uh oh!

github-actions bot commented Apr 6, 2023

Uh oh!

github-actions bot commented Apr 7, 2023

Uh oh!

github-actions bot commented Apr 7, 2023

Uh oh!

oleksandr-pavlyk left a comment

Choose a reason for hiding this comment

Uh oh!

oleksandr-pavlyk commented Apr 7, 2023

Uh oh!

github-actions bot commented Apr 7, 2023

Uh oh!

github-actions bot commented Apr 7, 2023

Uh oh!

github-actions bot commented Apr 7, 2023

Uh oh!

Uh oh!

ndgrigorian commented Mar 30, 2023 •

edited

Loading

coveralls commented Mar 30, 2023 •

edited

Loading