Thread: Re: [PATCH] Hex-coding optimizations using SVE on ARM.
On Wed, Jan 22, 2025 at 11:10:10AM +0000, Chiranmoy.Bhattacharya@fujitsu.com wrote: > I realized I didn't attach the patch. Thanks. Would you mind creating a commitfest entry for this one? -- nathan
I have marked the commitfest entry for this [0] as waiting-on-author because the patch needs to be rebased. [0] https://commitfest.postgresql.org/patch/5538/ -- nathan
Re: [PATCH] Hex-coding optimizations using SVE on ARM.
From
"Chiranmoy.Bhattacharya@fujitsu.com"
Date:
Here's the rebased patch with a few modifications.
The hand-unrolled hex encode performs better than the non-unrolled version on
r8g.4xlarge. No improvement on m7g.4xlarge.
Added line-by-line comments explaining the changes with an example.
Below are the results. Input size is in bytes, and exec time is in ms.
encode - r8g.4xlarge
Input | master | SVE | SVE-unrolled
-------+--------+--------+--------------
8 | 4.971 | 6.434 | 6.623
16 | 8.532 | 4.399 | 4.710
24 | 12.296 | 5.007 | 5.780
32 | 16.003 | 5.027 | 5.234
40 | 19.628 | 5.807 | 6.201
48 | 23.277 | 5.815 | 6.222
56 | 26.927 | 6.744 | 7.030
64 | 30.419 | 6.774 | 6.347
128 | 83.250 | 10.214 | 9.104
256 |112.158 | 17.892 | 16.313
512 |216.544 | 31.060 | 29.876
1024 |429.351 | 59.310 | 53.374
2048 |854.677 |116.769 | 101.004
4096 |1706.528|237.322 | 195.297
8192 |3723.884|499.520 | 385.424
---------------------------------------
encode - m7g.4xlarge
Input | master | SVE | SVE-unrolled
-------+--------+--------+--------------
8 | 5.503 | 7.986 | 8.053
16 | 9.881 | 9.583 | 9.888
24 | 13.854 | 9.212 | 10.138
32 | 18.056 | 9.208 | 9.364
40 | 22.127 | 10.134 | 10.540
48 | 26.214 | 10.186 | 10.550
56 | 29.718 | 10.197 | 10.428
64 | 33.613 | 10.982 | 10.497
128 | 66.060 | 12.460 | 12.624
256 |130.225 | 18.491 | 18.872
512 |267.105 | 30.343 | 31.661
1024 |515.603 | 54.371 | 55.341
2048 |1013.766|103.898 | 105.192
4096 |2018.705|202.653 | 203.142
8192 |4000.496|400.918 | 401.842
---------------------------------------
decode - r8g.4xlarge
Input | master | SVE
-------+--------+--------
8 | 7.641 | 8.787
16 | 14.301 | 14.477
32 | 28.663 | 6.091
48 | 42.940 | 17.604
64 | 57.483 | 10.549
80 | 71.637 | 19.194
96 | 85.918 | 15.586
112 |100.272 | 25.956
128 |114.740 | 19.829
256 |229.176 | 36.032
512 |458.295 | 68.222
1024 |916.741 |132.927
2048 |1833.422|262.741
4096 |3667.096|522.009
8192 |7333.886|1042.447
---------------------------------------
decode - m7g.4xlarge
Input | master | SVE
-------+--------+--------
8 | 8.194 | 9.433
16 | 14.397 | 15.606
32 | 26.669 | 29.006
48 | 45.971 | 48.984
64 | 58.468 | 12.388
80 | 70.820 | 22.295
96 | 84.792 | 43.470
112 | 98.992 | 54.282
128 |113.250 | 25.508
256 |218.743 | 45.165
512 |414.133 | 86.800
1024 |828.493 |174.670
2048 |1617.921|346.375
4096 |3259.159|689.391
8192 |6551.879|1376.195
--------
Chiranmoy