[VL] Upgrade simdjson to 3.9.3 in vcpkg build

[PHILO-HE]

What changes were proposed in this pull request?

See velox commit:
facebookincubator/velox@f9ae45a

How was this patch tested?

CI build.

[github-actions]

Thanks for opening a pull request!

Could you open an issue for this pull request on Github Issues?

https://github.com/apache/incubator-gluten/issues

Then could you also rename commit message and pull request title in the following format?

[GLUTEN-${ISSUES_ID}][COMPONENT]feat/fix: ${detailed message}

See also:

• Other pull requests

[zhouyuan]

👍

[PHILO-HE]

Just created a velox pr facebookincubator/velox#9997 to fix the below issue (vcpkg installed simdjson cannot be used by velox, instead velox still builds it from source):

-- Setting simdjson source to AUTO
2024-05-31T08:52:10.0849360Z CMake Warning at /__w/incubator-gluten/incubator-gluten/dev/vcpkg/.vcpkg/scripts/buildsystems/vcpkg.cmake:859 (_find_package):
2024-05-31T08:52:10.0850312Z   Could not find a configuration file for package "simdjson" that is
2024-05-31T08:52:10.0851050Z   compatible with requested version "3.8.0".
2024-05-31T08:52:10.0851548Z 
2024-05-31T08:52:10.0851888Z   The following configuration files were considered but not accepted:
2024-05-31T08:52:10.0852654Z 
2024-05-31T08:52:10.0853386Z     /__w/incubator-gluten/incubator-gluten/dev/vcpkg/vcpkg_installed/x64-linux-avx/share/simdjson/simdjson-config.cmake, version: 3.9.3
2024-05-31T08:52:10.0854100Z 
2024-05-31T08:52:10.0854222Z Call Stack (most recent call first):
2024-05-31T08:52:10.0854613Z   CMake/ResolveDependency.cmake:70 (find_package)
2024-05-31T08:52:10.0855048Z   CMakeLists.txt:451 (resolve_dependency)

[FelixYBW]

@PHILO-HE Can you take this oppotunity to do a perf test of json parse? Masha said there is huge perf gain. Let's confirm from Gluten.

[PHILO-HE]

@PHILO-HE Can you take this oppotunity to do a perf test of json parse? Masha said there is huge perf gain. Let's confirm from Gluten.

@FelixYBW, I just did a small benchmark test with Spark. Velox with simdjson can bring 3x perf. gain, compared with vanilla spark. Simdjson's upgrading to version 3.93 brings 10% perf. gain, compared with the old simdjson version.

[PHILO-HE]

This pr is ready to merge. The above warning only requires velox code change to fix.

[GlutenPerfBot]

===== Performance report for TPCH SF2000 with Velox backend, for reference only ====

query	log/native_5938_time.csv	log/native_master_06_02_2024_26ff58d3b_time.csv	difference	percentage
q1	33.91	33.06	-0.849	97.50%
q2	26.49	26.68	0.192	100.73%
q3	35.18	36.70	1.528	104.34%
q4	35.15	35.50	0.357	101.02%
q5	69.89	69.83	-0.066	99.91%
q6	7.19	5.99	-1.200	83.30%
q7	80.10	81.70	1.606	102.00%
q8	84.56	84.47	-0.088	99.90%
q9	121.20	119.12	-2.077	98.29%
q10	45.14	45.39	0.246	100.55%
q11	21.88	20.18	-1.697	92.25%
q12	23.64	25.30	1.668	107.06%
q13	36.92	37.41	0.484	101.31%
q14	17.32	20.07	2.752	115.89%
q15	29.54	30.21	0.668	102.26%
q16	12.98	14.14	1.166	108.98%
q17	103.73	101.99	-1.740	98.32%
q18	145.00	147.16	2.160	101.49%
q19	15.78	17.14	1.360	108.62%
q20	27.51	26.85	-0.665	97.58%
q21	267.87	259.27	-8.600	96.79%
q22	12.03	12.39	0.354	102.94%
total	1253.01	1250.57	-2.440	99.81%