Fix for lost FastMathFlags in SLPVectorizer (intrinsic calls)

[v_klochkov]

Summary

Hello,

Please review the fix for lost FastMathFlags in Call statements generated by SLPVectorizer.
This fix is very similar to https://reviews.llvm.org/D26543

Thank you,
Vyacheslav Klochkov

[mzolotukhin]

LGTM, thanks for following-up!

Michael

llvm/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll

413 ↗

(On Diff #77711)

Should it be %load2?

431 ↗

(On Diff #77711)

The same.

[v_klochkov]

Michael, thank you for the quick code-review!

llvm/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll

413 ↗

(On Diff #77711)

Yes, it was going to be %load2.
Even though the test worked well as is, I fixed it.

llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp

case Instruction::Call: {
 
       // The scalar argument uses an in-tree scalar so we add the new vectorized
       // call to ExternalUses list to make sure that an extract will be
       // generated in the future.
       if (ScalarArg && ScalarToTreeEntry.count(ScalarArg))
         ExternalUses.push_back(ExternalUser(ScalarArg, cast<User>(V), 0));
 
       E->VectorizedValue = V;
+      propagateIRFlags(E->VectorizedValue, E->Scalars);
       ++NumVectorInstructions;
       return V;
     }
     case Instruction::ShuffleVector: {
       ValueList LHSVL, RHSVL;
       assert(isa<BinaryOperator>(VL0) && "Invalid Shuffle Vector Operand");
       reorderAltShuffleOperands(E->Scalars, LHSVL, RHSVL);
       setInsertPointAfterBundle(E->Scalars);

llvm/trunk/test/Transforms/SLPVectorizer/X86/call.ll

entry:
   store double %call, double* %c, align 8
   %arrayidx5 = getelementptr inbounds double, double* %c, i64 1
   store double %call5, double* %arrayidx5, align 8
   ret void
 }
 
 
 ; CHECK: sqrt_libm
-; CHECK: call <2 x double> @llvm.sqrt.v2f64
+; CHECK: call nnan <2 x double> @llvm.sqrt.v2f64
 ; CHECK: ret void
 define void @sqrt_libm(double* %a, double* %b, double* %c) {
 entry:
   %i0 = load double, double* %a, align 8
   %i1 = load double, double* %b, align 8
   %mul = fmul double %i0, %i1
   %call = tail call nnan double @sqrt(double %mul) nounwind readnone
   %arrayidx3 = getelementptr inbounds double, double* %a, i64 1

llvm/trunk/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll

define void @fcmp_no_fast(double* %x) #1 {
   %sel2 = select i1 %cmp2, double %load2, double %sub2
 
   store double %sel1, double* %idx1, align 8
   store double %sel2, double* %idx2, align 8
 
   ret void
 }
 
-attributes #1 = { "target-features"="+avx" }
+declare double @llvm.fabs.f64(double) nounwind readnone
+
+;CHECK-LABEL: @call_fast(
+;CHECK: call fast <2 x double> @llvm.fabs.v2f64
+define void @call_fast(double* %x) {
+  %idx1 = getelementptr inbounds double, double* %x, i64 0
+  %idx2 = getelementptr inbounds double, double* %x, i64 1
+
+  %load1 = load double, double* %idx1, align 8
+  %load2 = load double, double* %idx2, align 8
+
+  %call1 = tail call fast double @llvm.fabs.f64(double %load1) nounwind readnone
+  %call2 = tail call fast double @llvm.fabs.f64(double %load2) nounwind readnone
+
+  store double %call1, double* %idx1, align 8
+  store double %call2, double* %idx2, align 8
+
+  ret void
+}
 
+;CHECK-LABEL: @call_no_fast(
+;CHECK: call <2 x double> @llvm.fabs.v2f64
+define void @call_no_fast(double* %x) {
+  %idx1 = getelementptr inbounds double, double* %x, i64 0
+  %idx2 = getelementptr inbounds double, double* %x, i64 1
+
+  %load1 = load double, double* %idx1, align 8
+  %load2 = load double, double* %idx2, align 8
+
+  %call1 = tail call fast double @llvm.fabs.f64(double %load1) nounwind readnone
+  %call2 = tail call double @llvm.fabs.f64(double %load2) nounwind readnone
+
+  store double %call1, double* %idx1, align 8
+  store double %call2, double* %idx2, align 8
+
+  ret void
+}
+
+attributes #1 = { "target-features"="+avx" }