[llvm] Optimize fptrunc(x)>=C1 --> x>=C2 (PR #99475)

[via llvm-commits]

================
@@ -7882,6 +7892,93 @@ static Instruction *foldFCmpReciprocalAndZero(FCmpInst &I, Instruction *LHSI,
   return new FCmpInst(Pred, LHSI->getOperand(1), RHSC, "", &I);
 }
 
+// Fold fptrunc(x) < constant --> x < constant if possible.
+static Instruction *foldFCmpFpTrunc(FCmpInst &I, Instruction *LHSI,
+                                    Constant *RHSC) {
+  FCmpInst::Predicate Pred = I.getPredicate();
+  bool RoundDown = false;
+
+  if (Pred == FCmpInst::FCMP_OGE || Pred == FCmpInst::FCMP_UGE ||
+      Pred == FCmpInst::FCMP_OLT || Pred == FCmpInst::FCMP_ULT)
+    RoundDown = true;
+  else if (Pred == FCmpInst::FCMP_OGT || Pred == FCmpInst::FCMP_UGT ||
+           Pred == FCmpInst::FCMP_OLE || Pred == FCmpInst::FCMP_ULE)
+    RoundDown = false;
+  else
+    return nullptr;
+
+  const APFloat *RValue;
+  if (!match(RHSC, m_APFloat(RValue)))
+    return nullptr;
+
+  // RHSC should not be nan or infinity.
+  if (RValue->isNaN() || RValue->isInfinity())
+    return nullptr;
+
+  Type *LType = LHSI->getOperand(0)->getType();
+  Type *RType = RHSC->getType();
+  Type *LEleType = LType->getScalarType();
+  Type *REleType = RType->getScalarType();
+
+  APFloat NextRValue = *RValue;
+  NextRValue.next(RoundDown);
+
+  // Round RValue to suitable value
+  APFloat ExtRValue = *RValue;
+  APFloat ExtNextRValue = NextRValue;
+  bool lossInfo;
+  ExtRValue.convert(LEleType->getFltSemantics(), APFloat::rmNearestTiesToEven,
+                    &lossInfo);
+  ExtNextRValue.convert(LEleType->getFltSemantics(),
+                        APFloat::rmNearestTiesToEven, &lossInfo);
+
+  // Binary search to find the maximal (or minimal) value after RValue promotion.
----------------
kissholic wrote:

std::lower_bound/std::upper_bound (or similar algorithm like std::binary_search) seems only accept [ForwardIt type](https://en.cppreference.com/w/cpp/algorithm/lower_bound) and there is likely no suitable substitution algorithm in LLVM too, which may requires constructing a new complex wrapper struct of APFloat, such as implementing name required iteration methods, calculating the mean of two APFloat (without constructing a new APFloat divisor), defining a comparation function and so on.

Considering the internal complexity of APFloat wrapper, it is simpler to keep the original one (i think)?

https://github.com/llvm/llvm-project/pull/99475