[RFC]: Improve sorted arrays generation in C benchmarks for sorted-input packages

[LoayAhmed304]

Description

This RFC proposes to generalize the generation of random arrays in C benchmarks for sorted-input packages. It mainly tries to add a random factor for the generated values and make use of the already existing rand_double() function and time.h header for consistency between packages.

For example, for double precision packages, to be changed from: (e.g., from stats/base/strided/dminsorted)

x = (double *) malloc( len * sizeof( double ) );
for ( i = 0; i < len; i++ ) {
	x[ i ] = i;
}

To:

x = (double *) malloc( len * sizeof( double ) );
for ( i = 0; i < len; i++ ) {
    x[ i ] = (double)i + rand_double();
}

Packages that are directly affected by this (add more if found):

• stats/strided/dmediansorted
• stats/strided/dmaxsorted
• stats/strided/dminsorted
• stats/strided/smediansorted
• stats/strided/smaxsorted
• stats/strided/sminsorted

Related Issues

None.

Questions

Question 1

I suggest that we either enforce this or refactor the already existing packages that don't use any randomness by removing the time.h header and rand_double() function since they're not used anywhere in the benchmark.

Question 2

What about the absolute sorted-input packages? Current approach is: (from stats/base/strided/dmaxabssorted)

x = (double *) malloc( len * sizeof( double ) );
for ( i = 0; i < len; i++ ) {
		x[ i ] = i - (len/2);
	}

Other

No.

Checklist

• I have read and understood the Code of Conduct.
• Searched for existing issues and pull requests.
• The issue name begins with RFC:.

[stdlib-bot]

👋 Hi there! 👋

And thank you for opening your first issue! We will get back to you shortly. 🏃 💨

If you have any further questions while waiting for a response, please join our Zulip community to chat with project maintainers and other community members.

[pranavthakur0-0]

Hi,
I've opened a PR addressing this RFC: #10705

The PR covers all 6 packages listed in the RFC, plus the two absolute sorted-input packages (stats/strided/dmaxabssorted and stats/strided/smaxabssorted) mentioned in Question 2.

For the abs packages, I used the same approach while preserving the offset:

• x[ i ] = (double)( i - (len/2) ) + rand_double();
• x[ i ] = (float)( i - (len/2) ) + rand_float();

All C benchmarks were compiled and tested successfully.