Layernorm optimization

[brentmaas]

Motivation

This PR provides performance improvements to the layernorm kernels via optimizations and additional tuning options, including an option for vectorized loading and storing of data.

Technical Details

• Add tuning options for vectorized loading and storing of data, for scheduling strides in a different dimension, and for incorporating stride into the local size.
• Update heuristics using new tuning options.
• Miscellaneous optimizations.
• Deduplicate cache and tuning keys.
• Print data throughput in layernorm MIOpenDriver.
• Report multiple failing buffers in layernorm MIOpenDriver.

Test Plan

The changes can be verified by building and running the test target test_layernorm.

Test Result

All tests pass.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.