[Bug tree-optimization/126415] [17 regression] Wrong code at -O2/3/s on x86_64-pc-linux-gnu since r17-2078

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126415

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
   p1.0_1 = (unsigned char) p1_5(D);
   _2 = p1.0_1 + 159;
   _3 = (char) _2;
-  _7 = (unsigned int) _3;
-  _10 = _7 + 97;
+  _10 = (unsigned int) p1_5(D);

I'll note that this must be an error in the pattern conditions, that it
"misfires" outside of VN isn't the issue.

This is because even "inside VN" (op! @0 { inner_cst; }) isn't just
a hashtable lookup for the expression itself.  Even the hand-coded
variant with vn_nary_build_or_lookup will simplify.