On 11/29/25 12:48 AM, [email protected] wrote:
Actully I'm thinking If there is a PGO-like feedback loop that automatically
tunes vector cost adjustments
by benchmarking different configurations would be very valuable.
This could help us catch cases where the current heuristics make poor choices
(e.g., bad LMUL selection or vectorizing when scalar is faster).
Right now I’m finding these issues manually, which is slow and doesn’t scale.
Is there any existing GCC infrastructure that could support automated
cost-model tuning,
or do you have recommendations on the best way to build such a system?
Nothing for automated cost model tuning. These things take considerable
time to discover, then chase down.
What usually happens is there's one or more benchmarks (real world code
you care about or industry standard benchmarks, the former is much
preferable) and you just have to dive in and understand where the
hotspots are and how they're behaving. linux-perf is your friend.
jeff
