On 7/9/2026 2:03 AM, Richard Biener wrote:
On Wed, Jun 24, 2026 at 4:28 PM Jovan Dmitrovic
<[email protected]> wrote:
Function ivopts_estimate_reg_pressure is taxing candidate sets
by the number of candidates included in set. This makes for an
uneven playing field, making it very difficult for larger sets
to be chosen.
In try_improve_iv_set to candidate set another candidate is added
only if the cost improves (therefore prefering fewer candidates
per set), but having reg pressure cost factor in the number of
candidates makes a disbalance when choosing the best option.
This is OK (as it's simpler than 2/3 it should be re-ordered), it
can be pushed independently.
Looking at the interactions with 2/3, I'd prefer to see a v3 posted
which handles the reordering Richi mentions and makes this patch
independent. While I could fix things up so that 3/3 could go in now,
it'd create unnecessary churn.
jeff
