[PATCH] D145846: [InstCombine] enhance icmp with sub folds

[Nikita Popov via Phabricator via llvm-commits]

nikic added inline comments.


================
Comment at: llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp:4096
+  if (match(&I, m_c_ICmp(Pred, m_OneUse(m_Neg(m_Value(X))), m_Deferred(X))) &&
+      (Pred == ICmpInst::ICMP_NE || Pred == ICmpInst::ICMP_EQ)) {
+    Type *Ty = X->getType();
----------------
nikic wrote:
> Use `ICmpInst::isEquality()`.
isEquality() means "EQ" or "NE". You don't need to check for ICMP_NE separately.


================
Comment at: llvm/test/Transforms/InstCombine/icmp-sub.ll:596
-; CHECK-NEXT:    [[SUB:%.*]] = sub i32 0, [[A:%.*]]
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[SUB]], [[A]]
 ; CHECK-NEXT:    ret i1 [[CMP]]
----------------
These end up not actually commuted due to complexity-based canonicalization. Grep for `thwart complexity-based canonicalization` for how to avoid it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D145846/new/

https://reviews.llvm.org/D145846