On 8/19/26 9:04 AM, Jovan Dmitrovic wrote:
The target_reg_cost plays a large part in the estimated cost for
register pressure, but it should be the eventual spilling that
has the most influence on the cost of storing/loading IV candidates and
invariants.
Also, the number of available registers is now not impacted by reserved
target registers, because there is no need to differentiate between cases
when those registers are used, as target_reg_cost is no longer a part
of the estimated cost.
gcc/ChangeLog:
* tree-ssa-loop-ivopts.cc (ivopts_estimate_reg_pressure):
Remove target_reg_cost from the computation of register pressure
and target_res_regs from the computation of available registers.
Signed-off-by: Jovan Dmitrović<[email protected]>
The only difference between this patch and the one Richi ACK'd was the
ChangeLog. I've pushed this one too.
jeff
