Fix failure of InstCombine to propagate fast math flags when scalarizing vector arithmetic.

[resistor]

Summary

Most portions of InstCombine properly propagate fast math flags, but apparently the vector secularization section was overlooked.

[arsenm]

lib/Transforms/InstCombine/InstCombineVectorOps.cpp

108–110

Should a BinaryOperator::CreateWithFlags be added? It already has CreateNSW etc.

[resistor]

lib/Transforms/InstCombine/InstCombineVectorOps.cpp

108–110

It would have to be something like BinaryOperator::CreateWithCopiedFlags, because there's not simple "give me all the flags" methods exposed on BinaryOperator. Seems a little awkward, but maybe it's still better?

[arsenm]

lib/Transforms/InstCombine/InstCombineVectorOps.cpp

108–110

Probably. It seems like a mess to have all of these separate copies everywhere this is needed

[resistor]

Added new CreateWithCopiedFlags() on BinaryOperator.

[resistor]

Ping?

[hfinkel]

LGTM

[resistor]

r262376

include/llvm/IR/InstrTypes.h

   }
 #include "llvm/IR/Instruction.def"
 
+  static BinaryOperator *CreateWithCopiedFlags(BinaryOps Opc,
+                                               Value *V1, Value *V2,
+                                               BinaryOperator *CopyBO,
+                                               const Twine &Name = "") {
+    BinaryOperator *BO = Create(Opc, V1, V2, Name);
+    BO->copyIRFlags(CopyBO);
+    return BO;
+  }
+
   static BinaryOperator *CreateNSW(BinaryOps Opc, Value *V1, Value *V2,
                                    const Twine &Name = "") {
     BinaryOperator *BO = Create(Opc, V1, V2, Name);

lib/Transforms/InstCombine/InstCombineVectorOps.cpp

                                      B0->getOperand(opId)->getName() + ".Elt"),
           *B0);
       Value *newPHIUser = InsertNewInstWith(
-          BinaryOperator::Create(B0->getOpcode(), scalarPHI, Op), *B0);
+          BinaryOperator::CreateWithCopiedFlags(B0->getOpcode(),
+                                                scalarPHI, Op, B0), *B0);
       scalarPHI->addIncoming(newPHIUser, inBB);
     } else {
       // Scalarize PHI input:

         Value *newEI1 =
           Builder->CreateExtractElement(BO->getOperand(1), EI.getOperand(1),
                                         EI.getName()+".rhs");
-        return BinaryOperator::Create(BO->getOpcode(), newEI0, newEI1);
+        return BinaryOperator::CreateWithCopiedFlags(BO->getOpcode(),
+                                                     newEI0, newEI1, BO);
       }
     } else if (InsertElementInst *IE = dyn_cast<InsertElementInst>(I)) {
       // Extracting the inserted element?

test/Transforms/InstCombine/fast-math-scalarization.ll

+; RUN: opt -instcombine -S < %s | FileCheck %s
+
+; CHECK-LABEL: test_scalarize_phi
+; CHECK: fmul fast float
+define void @test_scalarize_phi(i32 * %n, float * %inout) {
+entry:
+  %0 = load volatile float, float * %inout, align 4
+  %splat.splatinsert = insertelement <4 x float> undef, float %0, i32 0
+  %splat.splat = shufflevector <4 x float> %splat.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer
+  %splat.splatinsert1 = insertelement <4 x float> undef, float 3.0, i32 0
+  br label %for.cond
+
+for.cond:
+  %x.0 = phi <4 x float> [ %splat.splat, %entry ], [ %mul, %for.body ]
+  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+  %1 = load i32, i32 * %n, align 4
+  %cmp = icmp ne i32 %i.0, %1
+  br i1 %cmp, label %for.body, label %for.end
+
+for.body:
+  %2 = extractelement <4 x float> %x.0, i32 1
+  store volatile float %2, float * %inout, align 4
+  %mul = fmul fast <4 x float> %x.0, <float 0x4002A3D700000000, float 0x4002A3D700000000, float 0x4002A3D700000000, float 0x4002A3D700000000>
+  %inc = add nsw i32 %i.0, 1
+  br label %for.cond
+
+for.end:
+  ret void
+}
+
+; CHECK-LABEL: test_extract_element_fastmath
+; CHECK: fadd fast float
+define float @test_extract_element_fastmath(<4 x float> %x) #0 {
+entry:
+  %add = fadd fast <4 x float> %x, <float 0x4002A3D700000000, float 0x4002A3D700000000, float 0x4002A3D700000000, float 0x4002A3D700000000>
+  %0 = extractelement <4 x float> %add, i32 2
+  ret float %0
+}
+