Optimize gcd for array and scalar case by avoiding make_scalar_function where has unnecessary conversion between scalar and array

[jayzhan211]

Which issue does this PR close?

• Closes #.

Rationale for this change

make_scalar_function convert scalar to array and call function kernel. However, we don't need to convert to array and can directly compute on it. This improve the gcd performance in array vs scalar case

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

[jayzhan211]

Gnuplot not found, using plotters backend
Benchmarking gcd both array: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 7.5s, enable flat sampling, or reduce sample count to 50.
gcd both array          time:   [1.4799 ms 1.4826 ms 1.4853 ms]
                        change: [+1.1198% +1.3952% +1.6734%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high mild

Benchmarking gcd array and scalar: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 8.9s, enable flat sampling, or reduce sample count to 50.
gcd array and scalar    time:   [1.7456 ms 1.7486 ms 1.7519 ms]
                        change: [-4.8528% -4.6518% -4.4540%] (p = 0.00 < 0.05)
                        Performance has improved.

gcd both scalar         time:   [23.373 ns 23.431 ns 23.487 ns]
                        change: [-93.342% -93.306% -93.272%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 18 outliers among 100 measurements (18.00%)
  5 (5.00%) low severe
  5 (5.00%) low mild
  5 (5.00%) high mild
  3 (3.00%) high severe

[alamb]

Thanks @jayzhan211 -- look like a nice improvement to me

[alamb]

I think you can use try_binary and make it even faster:

Suggested change

	fn compute_gcd_for_arrays(a: &ArrayRef, b: &ArrayRef) -> Result<ColumnarValue> {
	let result: Result<Int64Array> = a
	.as_primitive::<Int64Type>()
	.iter()
	.zip(b.as_primitive::<Int64Type>().iter())
	.map(|(a, b)| match (a, b) {
	(Some(a), Some(b)) => Ok(Some(compute_gcd(a, b)?)),
	_ => Ok(None),
	})
	.collect();
	result.map(|arr| ColumnarValue::Array(Arc::new(arr) as ArrayRef))
	}
	fn compute_gcd_for_arrays(a: &ArrayRef, b: &ArrayRef) -> Result<ColumnarValue> {
	let result: Result<Int64Array> = a
	.as_primitive::<Int64Type>()
	.iter()
	.zip(b.as_primitive::<Int64Type>().iter())
	.map(|(a, b)| match (a, b) {
	(Some(a), Some(b)) => Ok(Some(compute_gcd(a, b)?)),
	_ => Ok(None),
	})
	.collect();
	result.map(|arr| ColumnarValue::Array(Arc::new(arr) as ArrayRef))
	}

I'll run some quick numbers with your new benchmark.

[alamb]

With my proposal the two array version seems to go 25% faster than this PR. Here is a PR with that improvement:

• Use try_binary to make gcd even faster jayzhan211/datafusion#5

Gnuplot not found, using plotters backend
Benchmarking gcd both array: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 5.9s, enable flat sampling, or reduce sample count to 60.
gcd both array          time:   [1.1940 ms 1.2007 ms 1.2081 ms]
                        change: [-27.574% -27.080% -26.470%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  6 (6.00%) high mild
  3 (3.00%) high severe

gcd array and scalar    time:   [1.9644 ms 1.9705 ms 1.9781 ms]
                        change: [-1.3293% -0.9077% -0.4670%] (p = 0.00 < 0.05)
                        Change within noise threshold.
Found 6 outliers among 100 measurements (6.00%)
  3 (3.00%) high mild
  3 (3.00%) high severe

gcd both scalar         time:   [26.094 ns 26.253 ns 26.467 ns]
                        change: [-1.1099% -0.6987% -0.2362%] (p = 0.00 < 0.05)
                        Change within noise threshold.
Found 8 outliers among 100 measurements (8.00%)
  4 (4.00%) high mild
  4 (4.00%) high severe

[jayzhan211]

try_binary actually did the length check

[jayzhan211]

Thanks @alamb