[PATCH] D50222: [CodeGen] [SelectionDAG] More efficient code for X % C == 0

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added inline comments.


================
Comment at: lib/CodeGen/SelectionDAG/DAGCombiner.cpp:3372-3392
+  // fold (rem lhs, rhs) -> lhs when |lhs| < |rhs|
+  // FIXME: this optimization is otherwise missed if BuildREMEqFold succeeds.
+  // Ref: test1 in test/CodeGen/X86/jump_sign.ll
+  if (N1C) {
+    KnownBits LHSBits;
+    DAG.computeKnownBits(N0, LHSBits);
+    LHSBits.makeNonNegative();
----------------
lebedev.ri wrote:
> I think this is independent from the rest?
> Can you split it into a separate review?
You could put this fold into the middle-end, into instsimplify, if we don't do it already.
But since it needs `computeKnownBits()`, some measurements would be needed
(how many times does it fire on some codebase? possibly, measure compile-time impact)


Repository:
  rL LLVM

https://reviews.llvm.org/D50222