Fix for lost FastMathFlags in Reassociate optimization

[v_klochkov]

Summary

Hello,

Please review the fix for lost fast-math flags in Reassociate optimization.

If the input expression tree is made of FP MULs and FastMathFlags permit re-association, then the FastMathFlags must be propagated to the newly generated MUL expression tree.

Thank you,
Vyacheslav Klochkov

[hfinkel]

LGTM

[davide]

LGTM.

llvm/lib/Transforms/Scalar/Reassociate.cpp

Value *ReassociatePass::OptimizeMul(BinaryOperator *I,
   // Try to turn linear trees of multiplies without other uses of the
   // intermediate stages into minimal multiply DAGs with perfect sub-expression
   // re-use.
   SmallVector<Factor, 4> Factors;
   if (!collectMultiplyFactors(Ops, Factors))
     return nullptr; // All distinct factors, so nothing left for us to do.
 
   IRBuilder<> Builder(I);
+  // The reassociate transformation for FP operations is performed only
+  // if unsafe algebra is permitted by FastMathFlags. Propagate those flags
+  // to the newly generated operations.
+  if (auto FPI = dyn_cast<FPMathOperator>(I))
+    Builder.setFastMathFlags(FPI->getFastMathFlags());
+
   Value *V = buildMinimalMultiplyDAG(Builder, Factors);
   if (Ops.empty())
     return V;
 
   ValueEntry NewEntry = ValueEntry(getRank(V), V);
   Ops.insert(std::lower_bound(Ops.begin(), Ops.end(), NewEntry), NewEntry);
   return nullptr;
 }

llvm/test/Transforms/Reassociate/propagate-flags.ll

+; RUN: opt < %s -reassociate -S | FileCheck %s
+ 
+; CHECK-LABEL: func
+; CHECK:       fmul fast double
+; CHECK-NEXT:  fmul fast double
+; CHECK-NEXT:  ret
+
+define double @func(double %a, double %b) {
+entry:
+  %mul1 = fmul fast double %a, %a
+  %mul2 = fmul fast double %b, %b
+  %mul3 = fmul fast double %mul1, %mul2
+  ret double %mul3
+}