On 10/4/26 1:53 PM, Aldy Hernandez wrote:
Most of what record_conditions is doing is a NOP, as ranger is picking
it up. We can nuke the derivations.
I tested that there is no difference in the assembly at -O2 for my .ii
bootstrap files, as well as the preprocessed Fortran files I keep from
a LAPACK build (for floating point stuff). I also confirmed that
LAPACK internal tests pass. And of course, the usual regression
tests. So I think this is dead code, and can be removed.
Tested on ppc64le Linux.
OK for trunk?
gcc/ChangeLog:
* tree-ssa-scopedtables.cc (build_and_record_new_cond): Remove.
(record_conditions): Do not derive the implied conditions.
gcc/testsuite/ChangeLog:
* gcc.dg/tree-ssa/ssa-dom-simplify-5.c: New test.
This likely interacts with the 90036 work that I want to come back to in
the very near future. I think the code you're nuking is right where the
infinite loops were occurring when the new functionality was added. And
to be 100% clear, I'm all for what you're doing. I'd rather start over
with 90036 work after your change than hold up your change.
OK
jeff
