[VL] Remove the registry for Velox's prestosql scalar functions

[PHILO-HE]

What changes were proposed in this pull request?

Currently, Gluten registers Velox's prestosql scalar functions. Thus, some of them can be used for Spark SQL if there is no semantic difference. However, there are some conflict issues, which makes unexpected function calling. See below links. We plan to register used presto scalar functions (see below list) in sparksql registry.

#4759
#5150

Function list:
url_decode/url_encode
array_distinct
array_except
array_position
map_entries
map_keys
map_values
reverse
crc32
regexp_extract_all
like
abs (for long/short decimal)
asin
atan
cos
degrees

How was this patch tested?

Existing tests.

[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

[GlutenPerfBot]

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

query	log/native_5202_time.csv	log/native_master_04_09_2024_8e5f63c0c_time.csv	difference	percentage
q1	37.13	36.38	-0.750	97.98%
q2	23.82	23.87	0.045	100.19%
q3	37.27	36.63	-0.634	98.30%
q4	37.54	38.51	0.963	102.57%
q5	70.57	71.53	0.968	101.37%
q6	9.69	5.75	-3.939	59.35%
q7	81.58	85.53	3.950	104.84%
q8	86.21	85.59	-0.627	99.27%
q9	125.65	119.02	-6.626	94.73%
q10	47.97	43.75	-4.219	91.21%
q11	19.86	20.38	0.524	102.64%
q12	25.90	26.91	1.003	103.87%
q13	47.71	48.54	0.833	101.75%
q14	24.01	22.64	-1.373	94.28%
q15	31.92	31.84	-0.082	99.74%
q16	12.49	13.80	1.313	110.52%
q17	103.62	101.51	-2.105	97.97%
q18	142.55	143.19	0.640	100.45%
q19	15.68	13.58	-2.107	86.57%
q20	27.11	28.55	1.437	105.30%
q21	227.72	226.97	-0.753	99.67%
q22	13.85	13.92	0.069	100.50%
total	1249.85	1238.38	-11.469	99.08%

[Yohahaha]

Hi, could you share the current status?

[PHILO-HE]

Hi, could you share the current status?

@Yohahaha, this pr depends on a velox pr: facebookincubator/velox#9425. I will try to push the review progress.

[Yohahaha]

Hi, could you share the current status?

@Yohahaha, this pr depends on a velox pr: facebookincubator/velox#9425. I will try to push the review progress.

why not choose below usage? I think velox has already support register with prefix, the only thing we need to do is choose which prefixed-function is gluten needed. correct me if it's wrong, thank you!

  velox::functions::prestosql::registerAllScalarFunctions("presto_");
  velox::functions::sparksql::registerFunctions("spark_");

  prestoMap("array_distinct" -> "presto");
  // do naming conversion in cpp module

[PHILO-HE]

Hi, could you share the current status?

@Yohahaha, this pr depends on a velox pr: facebookincubator/velox#9425. I will try to push the review progress.

why not choose below usage? I think velox has already support register with prefix, the only thing we need to do is choose which prefixed-function is gluten needed. correct me if it's wrong, thank you!

  velox::functions::prestosql::registerAllScalarFunctions("presto_");
  velox::functions::sparksql::registerFunctions("spark_");

  prestoMap("array_distinct" -> "presto");
  // do naming conversion in cpp module

They are discussed internally. We feel it is more clear by using only one registry and we want to avoid maintaining such a map in Gluten.

[github-actions]

This PR is stale because it has been open 45 days with no activity. Remove stale label or comment or this will be closed in 10 days.

[rui-mo]

It seems some registrations in registerFunctionOverwrite() could be removed, including spark_rand, check_xx. After facebookincubator/velox@372abeb, we need to adapt to the new names checked_xx.

[PHILO-HE]

Yes, I also noted this. Will create another pr to fix it.

[rui-mo]

Thanks!