[llvm] [AggressiveInstCombine] Check profitability of low-bits mask table folds (PR #228736)

[Chris Kennelly via llvm-commits]

=?utf-8?q?Dávid_Bolvanský?= <david.bolvansky at gmail.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/228736 at github.com>


ckennelly wrote:

I put together some quick-and-dirty microbenchmarks for Zen 3, V2, and with a friend's help Apple M5: https://github.com/ckennelly/snippets/blob/main/llvm/pr228736/results/SUMMARY.md

In a throughput regime, tables can show a tiny edge but part of the address calculation can be hoisted in this and the microbenchmark is the most ideal set of circumstances. In a latency regime, arithmetic wins handily.

I'm looking at reproducing numbers for zstd now.




https://github.com/llvm/llvm-project/pull/228736