Optimize URL Decoding

[gaohao95]

This PR is intended to optimize the URL decoding performance, especially on large URLs. Additionally, a test case for large URLs has been added.

When tested on V100, baseline performance at 7521c3f

------------------------------------------------------------------------------------------------------------------
Benchmark                                                        Time             CPU   Iterations UserCounters...
------------------------------------------------------------------------------------------------------------------
UrlDecode<10>/url_decode_10pct/100000000/10/manual_time        111 ms          111 ms            6 bytes_per_second=11.7959G/s
UrlDecode<10>/url_decode_10pct/10000000/100/manual_time        107 ms          107 ms            7 bytes_per_second=9.0136G/s
UrlDecode<10>/url_decode_10pct/1000000/1000/manual_time        107 ms          107 ms            7 bytes_per_second=8.76755G/s
UrlDecode<50>/url_decode_50pct/100000000/10/manual_time        129 ms          129 ms            5 bytes_per_second=10.144G/s
UrlDecode<50>/url_decode_50pct/10000000/100/manual_time        126 ms          126 ms            6 bytes_per_second=7.70821G/s
UrlDecode<50>/url_decode_50pct/1000000/1000/manual_time        122 ms          122 ms            6 bytes_per_second=7.66783G/s

This PR

------------------------------------------------------------------------------------------------------------------
Benchmark                                                        Time             CPU   Iterations UserCounters...
------------------------------------------------------------------------------------------------------------------
UrlDecode<10>/url_decode_10pct/100000000/10/manual_time       97.5 ms         97.6 ms            7 bytes_per_second=13.3669G/s
UrlDecode<10>/url_decode_10pct/10000000/100/manual_time       28.8 ms         28.8 ms           24 bytes_per_second=33.6024G/s
UrlDecode<10>/url_decode_10pct/1000000/1000/manual_time       21.8 ms         21.8 ms           32 bytes_per_second=42.9686G/s
UrlDecode<50>/url_decode_50pct/100000000/10/manual_time        109 ms          109 ms            6 bytes_per_second=11.9786G/s
UrlDecode<50>/url_decode_50pct/10000000/100/manual_time       30.2 ms         30.3 ms           23 bytes_per_second=32.0311G/s
UrlDecode<50>/url_decode_50pct/1000000/1000/manual_time       22.7 ms         22.8 ms           31 bytes_per_second=41.1086G/s

close #8030

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.10@a153493). Click here to learn what that means.
The diff coverage is n/a.

❗ Current head 0d64559 differs from pull request most recent head 81af59c. Consider uploading reports for the commit 81af59c to get more accurate results

@@               Coverage Diff               @@
##             branch-21.10    #8622   +/-   ##
===============================================
  Coverage                ?   10.75%           
===============================================
  Files                   ?      114           
  Lines                   ?    18695           
  Branches                ?        0           
===============================================
  Hits                    ?     2010           
  Misses                  ?    16685           
  Partials                ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a153493...81af59c. Read the comment docs.

[shwina]

Hi @gaohao95 -- do you think this PR will be completed in time for 21.08 code freeze (July 22nd?). Wondering where this should go on the project board. Thanks!

[gaohao95]

Hi @gaohao95 -- do you think this PR will be completed in time for 21.08 code freeze (July 22nd?). Wondering where this should go on the project board. Thanks!

Hi @shwina, this PR is motivated by spark's string performance and I think the PR itself is more-or-less ready. But the last time I shared the code with @chenrui17, they reported a unit test failure on their ends. After that, I have discovered and fixed a bug, but I am unsure whether the fix solves their test failures. I think we should merge this only after we hear back from them.

Loop in @jlowe and @chenrui17 to see whether they have any feedbacks on priority.

[gaohao95]

@chenrui17 reported that the unit test can pass at their end now. I think this PR is ready for review.

[vyasr]

Nice work, the perf improvements look great. Some minor comments, a few of which are more for my edification than anything else.

[gaohao95]

Hi @vyasr @davidwendt, I think all comments have been addressed. Could you take another look to see whether it's ready to merge?

[davidwendt]

The kernel logic handling of null entries needs to be corrected.

[davidwendt]

Nice work.

[vyasr]

@gpucibot merge