[PATCH] D77076: [InstSimplify] Allow some arithmetic optimizations for add/sub/div/rem look through freeze

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added inline comments.


================
Comment at: llvm/lib/Analysis/InstructionSimplify.cpp:646
   Value *Y = nullptr;
-  if (match(Op1, m_Sub(m_Value(Y), m_Specific(Op0))) ||
-      match(Op0, m_Sub(m_Value(Y), m_Specific(Op1))))
+  if (match(Op1, m_Sub(m_Value(Y), m_Specific(Op0, true))) ||
+      match(Op0, m_Sub(m_Value(Y), m_Specific(Op1, true))))
----------------
aqjune wrote:
> xbolva00 wrote:
> > Maybe m_SpecificAllowFreeze or so is more readable than passing true?
> Removed the true argument
Hmm yeah, i agree with @xbolva00 here, `m_Specific(StripFreeze())` seems not ideal,
i'd suggest two more matchers - `m_SpecificMaybeFrozen()` and `m_DeferredMaybeFrozen()`
that would do that.


================
Comment at: llvm/lib/Analysis/InstructionSimplify.cpp:127
+/// If it is a freeze instruction, returns its operand
+static Value *StripFreeze(Value *V) {
+  if (auto *FI = dyn_cast<FreezeInst>(V))
----------------
Hm, do we have precedent naming scheme for things like this?
Alternative variants could include considered `IgnoreFreeze()` or `SkipFreeze()`.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77076/new/

https://reviews.llvm.org/D77076