Skip to content

Commit 9f6415e

Browse files
hommhomm
committed
more info
1 parent 67e55b7 commit 9f6415e

File tree

1 file changed

+80
-36
lines changed

1 file changed

+80
-36
lines changed

‎README.md‎

Lines changed: 80 additions & 36 deletions
Original file line numberDiff line numberDiff line change
@@ -3,53 +3,87 @@
33
Pillow-SIMD is "following" Pillow fork (which is PIL fork itself).
44

55
For more information about original Pillow, please
6-
[read the documentation](http://pillow.readthedocs.io/),
7-
[check the changelog](https://github.com/python-pillow/Pillow/blob/master/CHANGES.rst) and
8-
[find out how to contribute](https://github.com/python-pillow/Pillow/blob/master/.github/CONTRIBUTING.md).
6+
[read the documentation][original-docs],
7+
[check the changelog][original-changelog] and
8+
[find out how to contribute][original-contribute].
99

1010

11-
## What is SIMD
11+
## Why SIMD
12+
13+
There are many ways to improve performance of image processing.
14+
You can use better algorithms for the same task, you can make better
15+
implementation for current algorithms, or you can use more processing unit
16+
resources. It is perfect, when you can just use more efficient algirithm like
17+
when gaussian blur based on convolutions [was replaced][gaussian-blur-changes]
18+
with sequential box filters. But the number of such improvements is very limited.
19+
It is also very tempting to use more processor unit resources
20+
(via parallelization), when they are available. But it is more handy just
21+
to make things faster on the same resources. And that is where SIMD works better.
22+
23+
SIMD stands for "single instruction, multiple data". This is a way to perform
24+
same operations against the huge amount of homogeneous data.
25+
Modern CPU have differnt SIMD instructions sets, like
26+
MMX, SSE-SSE4, AVX, AVX2, AVX512, NEON.
27+
28+
Currently, Pillow-SIMD can be [compiled](#installation) with SSE4 (default)
29+
and AVX2 support.
1230

1331

1432
## Status
1533

34+
Pillow-SIMD can be used in production. Pillow-SIMD operates on
35+
[Uploadcare](https://uploadcare.com/) serverse for more than 1 year.
36+
Uploadcare is SAAS for image storing and processing in the cloud
37+
and the main sponsor of Pillow-SIMD project.
38+
39+
Cuurently, following methods are accelerated:
40+
41+
- Resize (convolustion-based resample): SSE4, AVX2
42+
- Gaussian and box blur: SSE4
43+
1644

1745
## Benchmarks
1846

19-
Numbers are megapixels of source image per second. For example if resize of
20-
7712x4352 image is done in 0.5 seconds, the resoult will be 61.7 Mpx/s.
47+
The numbers are processed megapixels of source image per second.
48+
For example if resize of 7712x4352 image is done in 0.5 seconds,
49+
the result will be 61.7 Mpx/s.
2150

2251
- ImageMagick 6.9.3-8 Q8 x86_64
2352
- Pillow 3.2.0
2453
- Pillow-SIMD 3.2.0.post1
2554

26-
Source | Operation | IM | Pillow | SIMD SSE4 | SIMD AVX2
27-
----------|-----------------------------|------|--------|-----------|-----------
28-
7712x4352 | Bilinear resize to 16x16 | 27.0 | 217 | 456 | 545
29-
| Bilinear resize to 320x180 | 32.0 | 166 | 354 | 410
30-
| Bilinear resize to 2048x1155| 20.7 | 87.6 | 202 | 217
31-
| Bicubic resize to 16x16 | 10.9 | 115 | 240 | 278
32-
| Bicubic resize to 320x180 | 16.5 | 92.3 | 198 | 204
33-
| Bicubic resize to 2048x1155 | 12.2 | 65.7 | 126 | 130
34-
| Lanczos resize to 16x16 | 6.6 | 76.1 | 162 | 194
35-
| Lanczos resize to 320x180 | 11.0 | 63.2 | 133 | 147
36-
| Lanczos resize to 2048x1155 | 8.7 | 41.3 | 88.2 | 95.6
37-
| Blur 1px | 8.1 | 17.1 | 37.8
38-
| Blur 10px | 2.6 | 17.4 | 39.0
39-
| Blur 100px | 0.3 | 17.2 | 39.0
40-
1920x1280 | Bilinear resize to 16x16 | 41.6 | 196 | 422 | 489
41-
| Bilinear resize to 320x180 | 27.6 | 111 | 196 | 197
42-
| Bilinear resize to 2048x1155| 9.1 | 20.7 | 71.3 | 72.6
43-
| Bicubic resize to 16x16 | 18.9 | 102 | 225 | 263
44-
| Bicubic resize to 320x180 | 14.5 | 66.3 | 154 | 162
45-
| Bicubic resize to 2048x1155 | 6.3 | 16.9 | 49.3 | 54.3
46-
| Lanczos resize to 16x16 | 13.7 | 68.6 | 118 | 167
47-
| Lanczos resize to 320x180 | 9.8 | 44.3 | 102 | 107
48-
| Lanczos resize to 2048x1155 | 4.7 | 14.6 | 36.8 | 40.6
49-
| Blur 1px | 8.7 | 16.2 | 35.7
50-
| Blur 10px | 2.8 | 16.7 | 35.4
51-
| Blur 100px | 0.4 | 16.4 | 36.2
52-
55+
Source | Operation | Filter | IM | Pillow | SIMD SSE4 | SIMD AVX2
56+
----------|---------------------|---------|------|--------|-----------|-----------
57+
7712x4352 | Resize to 16x16 | Bilinear| 27.0 | 217 | 456 | 545
58+
| | Bicubic | 10.9 | 115 | 240 | 278
59+
| | Lanczos | 6.6 | 76.1 | 162 | 194
60+
| Resize to 320x180 | Bilinear| 32.0 | 166 | 354 | 410
61+
| | Bicubic | 16.5 | 92.3 | 198 | 204
62+
| | Lanczos | 11.0 | 63.2 | 133 | 147
63+
| Resize to 2048x1155 | Bilinear| 20.7 | 87.6 | 202 | 217
64+
| | Bicubic | 12.2 | 65.7 | 126 | 130
65+
| | Lanczos | 8.7 | 41.3 | 88.2 | 95.6
66+
| Blur | 1px | 8.1 | 17.1 | 37.8
67+
| | 10px | 2.6 | 17.4 | 39.0
68+
| | 100px | 0.3 | 17.2 | 39.0
69+
1920x1280 | Resize to 16x16 | Bilinear| 41.6 | 196 | 422 | 489
70+
| | Bicubic | 18.9 | 102 | 225 | 263
71+
| | Lanczos | 13.7 | 68.6 | 118 | 167
72+
| Resize to 320x180 | Bilinear| 27.6 | 111 | 196 | 197
73+
| | Bicubic | 14.5 | 66.3 | 154 | 162
74+
| | Lanczos | 9.8 | 44.3 | 102 | 107
75+
| Resize to 2048x1155 | Bilinear| 9.1 | 20.7 | 71.3 | 72.6
76+
| | Bicubic | 6.3 | 16.9 | 49.3 | 54.3
77+
| | Lanczos | 4.7 | 14.6 | 36.8 | 40.6
78+
| Blur | 1px | 8.7 | 16.2 | 35.7
79+
| | 10px | 2.8 | 16.7 | 35.4
80+
| | 100px | 0.4 | 16.4 | 36.2
81+
82+
83+
### Some conclusions
84+
85+
Pillow is always faster than ImagMagick and
86+
Pillow-SIMD is always faster than Pillow.
5387

5488
### Methodology
5589

@@ -60,18 +94,28 @@ Following script were using for tests:
6094
https://gist.github.com/homm/f9b8d8a84a57a7e51f9c2a5828e40e63
6195

6296

63-
## Why Pillow is so fast itself
97+
## Why Pillow itself is so fast
6498

6599
There are no cheats. High-quality resize and blur methods are used for all
66100
benchmakrs. And result is
67101

68102

69-
## Why Pillow-SIMD even faster
103+
## Why Pillow-SIMD is even faster
104+
105+
106+
## Installation
70107

71108

72109
## Contributing to Pillow-SIMD
73110

74111
**Important.** Pillow-SIMD and Pillow are two separate projects.
75112
Please submit bugs and improvements not related to SIMD to
76-
[original Pillow](https://github.com/python-pillow/Pillow/issues/new).
113+
[original Pillow][original-issues].
77114
All bugs and fixes in Pillow will appear in next Pillow-SIMD version automatically.
115+
116+
117+
[original-docs]: http://pillow.readthedocs.io/
118+
[original-issues]: https://github.com/python-pillow/Pillow/issues/new
119+
[original-changelog]: https://github.com/python-pillow/Pillow/blob/master/CHANGES.rst
120+
[original-contribute]: https://github.com/python-pillow/Pillow/blob/master/.github/CONTRIBUTING.md
121+
[gaussian-blur-changes]: http://pillow.readthedocs.io/en/3.2.x/releasenotes/2.7.0.html#gaussian-blur-and-unsharp-mask

0 commit comments

Comments
 (0)