[FPEnv Core 13/14] Don't hoist FP-ops with side-effects in LICM

[sdmitrouk]

Summary

isSafeToSpeculativelyExecute() is not enough, because it's result
it taken into account only when it's positive and we need to forbid
the case where result is negative.

[majnemer]

lib/Transforms/Scalar/LICM.cpp

697–702

This doesn't make sense to me, can you please elaborate?

[sdmitrouk]

Rename flags.

[sdmitrouk]

lib/Transforms/Scalar/LICM.cpp

697–702

isSafeToSpeculativelyExecute() returns false, but code doesn't care and returns true from isGuaranteedToExecute() without this additional check.

[sdmitrouk]

Ping.

lib/Transforms/Scalar/LICM.cpp

static bool isSafeToExecuteUnconditionally(const Instruction &Inst,
                                            const DominatorTree *DT,
                                            const TargetLibraryInfo *TLI,
                                            const Loop *CurLoop,
                                            const LICMSafetyInfo *SafetyInfo,
                                            const Instruction *CtxI) {
   if (isSafeToSpeculativelyExecute(&Inst, CtxI, DT, TLI))
     return true;
 
+  // isSafeToSpeculativelyExecute() is not enough, because its result is
+  // taken into account only when it's positive and we need to forbid the case
+  // where its result is negative.
+  if (const FPMathOperator *MathOp = dyn_cast<FPMathOperator>(&Inst))
+    if (!MathOp->hasNoExceptions() || !MathOp->hasNoRounding())
+      return false;
+
   return isGuaranteedToExecute(Inst, DT, CurLoop, SafetyInfo);
 }
 
 static bool isGuaranteedToExecute(const Instruction &Inst,
                                   const DominatorTree *DT,
                                   const Loop *CurLoop,
                                   const LICMSafetyInfo * SafetyInfo) {
 

test/Transforms/LICM/fpenv.ll

+; RUN: opt < %s -O3 -S | FileCheck %s
+
+define void @fdiv-is-not-hoisted-out-of-loop(i32 %n) {
+; CHECK-LABEL: @fdiv-is-not-hoisted-out-of-loop(
+; CHECK: for.body:
+; CHECK: fdiv
+entry:
+  %cmp2 = icmp sgt i32 %n, 0
+  br i1 %cmp2, label %for.body.lr.ph, label %for.end
+
+for.body.lr.ph:
+  %0 = add i32 %n, -1
+  br label %for.body
+
+for.body:
+  %i.03 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
+  %div = fdiv double 0.000000e+00, 0.000000e+00
+  tail call void @process(double %div)
+  %inc = add nuw nsw i32 %i.03, 1
+  %exitcond = icmp eq i32 %i.03, %0
+  br i1 %exitcond, label %for.end.loopexit, label %for.body
+
+for.end.loopexit:
+  br label %for.end
+
+for.end:
+  ret void
+}
+
+declare void @process(double) nounwind