https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125944
Bug ID: 125944
Summary: 3/4 different copyprops
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: compile-time-hog, internal-improvement
Severity: enhancement
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: pinskia at gcc dot gnu.org
Target Milestone: ---
>From IRC:
<richi> we have ccp, copyprop and sccopy (and forwprop to some extent), there's
quite some redundancy. Not to mention DOM. And FRE in non-iterating mode.
ccp (tree-ssa-ccp.cc)
copy_prop (tree-ssa-copy.cc)
sccopy (gimple-ssa-sccopy.cc)
Now ccp does some bitbased proping. This was useful/needed before ranger (and
prange). How useful is it after prange?
copy_prop is the traditional SSA copy prop pass.
While sc-copy is the strongly connected copy-prop pass.
We should figure out if copy_prop can be removed or replaced with sccopy.
Second does ccp bit get some stuff that ranger misses now? If so let's improve
ranger/vrp. Can the first ccp be replaced with sccopy and one that exports
"global" ranges? Or do that handling in forwprop (which has its own copy prop
too).
