Scalarize select vector arguments when extracted

[arsenm]

Summary

When the elements are extracted from a select, do the select on the extracted scalars from the input.

With this, the vectorizer still isn't able to put a vector select back together. It is skipped because some of the insertelements are before the last select being vectorized.

Also fixes some missing CHECK-LABELs in the test that changed, but those are separate.

[arsenm]

Ping

[nadav]

Hi Matt,

I am sorry I missed your first email. Thanks for working on this.

When the elements are extracted from a select, do the select on the extracted scalars from the input.

I am not sure that this is always profitable or that it should be done in InstCombine. It is hard to know at IR-level (during inst-combine) if it is better to extract twice and perform a single select or if it is better to perform an vector select and extract once. I think that this kind of decision should be made in SelectionDAG.

With this, the vectorizer still isn't able to put a vector select back together. It is skipped because some of the insertelements are before the last select being vectorized.

I assume that you mean the SLP-vectorizer. What is the problem ? Do you have a test case ? I’d be happy to take a look.

Also fixes some missing CHECK-LABELs in the test that changed, but those are separate.

Can you commit the label fixes as a separate patch ?

Thanks,
Nadav

[arsenm]

Committed r194013

lib/Transforms/InstCombine/InstCombineVectorOps.cpp

if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
       // Canonicalize extractelement(cast) -> cast(extractelement)
       // bitcasts can change the number of vector elements and they cost nothing
       if (CI->hasOneUse() && (CI->getOpcode() != Instruction::BitCast)) {
         Value *EE = Builder->CreateExtractElement(CI->getOperand(0),
                                                   EI.getIndexOperand());
         Worklist.AddValue(EE);
         return CastInst::Create(CI->getOpcode(), EE, EI.getType());
       }
+    } else if (SelectInst *SI = dyn_cast<SelectInst>(I)) {
+      if (SI->hasOneUse()) {
+        // TODO: For a select on vectors, it might be useful to do this if it
+        // has multiple extractelement uses. For vector select, that seems to
+        // fight the vectorizer.
+
+        // If we are extracting an element from a vector select or a select on
+        // vectors, a select on the scalars extracted from the vector arguments.
+        Value *TrueVal = SI->getTrueValue();
+        Value *FalseVal = SI->getFalseValue();
+
+        Value *Cond = SI->getCondition();
+        if (Cond->getType()->isVectorTy()) {
+          Cond = Builder->CreateExtractElement(Cond,
+                                               EI.getIndexOperand(),
+                                               Cond->getName() + ".elt");
+        }
+
+        Value *V1Elem
+          = Builder->CreateExtractElement(TrueVal,
+                                          EI.getIndexOperand(),
+                                          TrueVal->getName() + ".elt");
+
+        Value *V2Elem
+          = Builder->CreateExtractElement(FalseVal,
+                                          EI.getIndexOperand(),
+                                          FalseVal->getName() + ".elt");
+        return SelectInst::Create(Cond,
+                                  V1Elem,
+                                  V2Elem,
+                                  SI->getName() + ".elt");
+      }
     }
   }
   return 0;
 }
 
 /// CollectSingleShuffleElements - If V is a shuffle of values that ONLY returns
 /// elements from either LHS or RHS, return the shuffle mask and true.
 /// Otherwise, return false.

test/Transforms/InstCombine/select-extractelement.ll

+; RUN: opt -S -instcombine < %s | FileCheck %s
+
+declare void @v4float_user(<4 x float>) #0
+
+
+
+define float @extract_one_select(<4 x float> %a, <4 x float> %b, i32 %c) #0 {
+; CHECK-LABEL: @extract_one_select(
+; CHECK-NOT: select i1 {{.*}}, <4 x float>
+  %cmp = icmp ne i32 %c, 0
+  %sel = select i1 %cmp, <4 x float> %a, <4 x float> %b
+  %extract = extractelement <4 x float> %sel, i32 2
+  ret float %extract
+}
+
+; Multiple extractelements
+define <2 x float> @extract_two_select(<4 x float> %a, <4 x float> %b, i32 %c) #0 {
+; CHECK-LABEL: @extract_two_select(
+; CHECK: select i1 {{.*}}, <4 x float>
+  %cmp = icmp ne i32 %c, 0
+  %sel = select i1 %cmp, <4 x float> %a, <4 x float> %b
+  %extract1 = extractelement <4 x float> %sel, i32 1
+  %extract2 = extractelement <4 x float> %sel, i32 2
+  %build1 = insertelement <2 x float> undef, float %extract1, i32 0
+  %build2 = insertelement <2 x float> %build1, float %extract2, i32 1
+  ret <2 x float> %build2
+}
+
+; Select has an extra non-extractelement user, don't change it
+define float @extract_one_select_user(<4 x float> %a, <4 x float> %b, i32 %c) #0 {
+; CHECK-LABEL: @extract_one_select_user(
+; CHECK: select i1 {{.*}}, <4 x float>
+  %cmp = icmp ne i32 %c, 0
+  %sel = select i1 %cmp, <4 x float> %a, <4 x float> %b
+  %extract = extractelement <4 x float> %sel, i32 2
+  call void @v4float_user(<4 x float> %sel)
+  ret float %extract
+}
+
+define float @extract_one_vselect_user(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
+; CHECK-LABEL: @extract_one_vselect_user(
+; CHECK: select <4 x i1> {{.*}}, <4 x float>
+  %cmp = icmp ne <4 x i32> %c, zeroinitializer
+  %sel = select <4 x i1> %cmp, <4 x float> %a, <4 x float> %b
+  %extract = extractelement <4 x float> %sel, i32 2
+  call void @v4float_user(<4 x float> %sel)
+  ret float %extract
+}
+
+; Extract from a vector select
+define float @extract_one_vselect(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
+; CHECK-LABEL: @extract_one_vselect(
+; CHECK-NOT: select <4 x i1>
+  %cmp = icmp ne <4 x i32> %c, zeroinitializer
+  %select = select <4 x i1> %cmp, <4 x float> %a, <4 x float> %b
+  %extract = extractelement <4 x float> %select, i32 0
+  ret float %extract
+}
+
+; Multiple extractelements from a vector select
+define <2 x float> @extract_two_vselect(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
+; CHECK-LABEL: @extract_two_vselect(
+; CHECK-NOT: select i1 {{.*}}, <4 x float>
+  %cmp = icmp ne <4 x i32> %c, zeroinitializer
+  %sel = select <4 x i1> %cmp, <4 x float> %a, <4 x float> %b
+  %extract1 = extractelement <4 x float> %sel, i32 1
+  %extract2 = extractelement <4 x float> %sel, i32 2
+  %build1 = insertelement <2 x float> undef, float %extract1, i32 0
+  %build2 = insertelement <2 x float> %build1, float %extract2, i32 1
+  ret <2 x float> %build2
+}
+
+; All the vector selects should be decomposed into scalar selects
+; Test multiple extractelements
+define <4 x float> @simple_vector_select(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
+; CHECK-LABEL: @simple_vector_select(
+; CHECK-NOT: select i1 {{.*}}, <4 x float>
+entry:
+  %0 = extractelement <4 x i32> %c, i32 0
+  %tobool = icmp ne i32 %0, 0
+  %a.sink = select i1 %tobool, <4 x float> %a, <4 x float> %b
+  %1 = extractelement <4 x float> %a.sink, i32 0
+  %2 = insertelement <4 x float> undef, float %1, i32 0
+  %3 = extractelement <4 x i32> %c, i32 1
+  %tobool1 = icmp ne i32 %3, 0
+  %a.sink1 = select i1 %tobool1, <4 x float> %a, <4 x float> %b
+  %4 = extractelement <4 x float> %a.sink1, i32 1
+  %5 = insertelement <4 x float> %2, float %4, i32 1
+  %6 = extractelement <4 x i32> %c, i32 2
+  %tobool6 = icmp ne i32 %6, 0
+  %a.sink2 = select i1 %tobool6, <4 x float> %a, <4 x float> %b
+  %7 = extractelement <4 x float> %a.sink2, i32 2
+  %8 = insertelement <4 x float> %5, float %7, i32 2
+  %9 = extractelement <4 x i32> %c, i32 3
+  %tobool11 = icmp ne i32 %9, 0
+  %a.sink3 = select i1 %tobool11, <4 x float> %a, <4 x float> %b
+  %10 = extractelement <4 x float> %a.sink3, i32 3
+  %11 = insertelement <4 x float> %8, float %10, i32 3
+  ret <4 x float> %11
+}
+
+attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }

test/Transforms/LoopVectorize/minmax_reduction.ll

 ; vector for the reduction.
 ; CHECK-LABEL: @max_red(
 ; CHECK: %[[VAR:.*]] = insertelement <2 x i32> undef, i32 %max, i32 0
 ; CHECK: {{.*}} = shufflevector <2 x i32> %[[VAR]], <2 x i32> undef, <2 x i32> zeroinitializer
 ; CHECK: icmp sgt <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp sgt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @max_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; Turn this into a max reduction. The select has its inputs reversed therefore
 ; this is a max reduction.
 ; CHECK-LABEL: @max_red_inverse_select(
 ; CHECK: icmp slt <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp sgt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @max_red_inverse_select(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 }
 
 ; Turn this into a min reduction.
 ; CHECK-LABEL: @min_red(
 ; CHECK: icmp slt <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp slt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @min_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; Turn this into a min reduction. The select has its inputs reversed therefore
 ; this is a min reduction.
 ; CHECK-LABEL: @min_red_inverse_select(
 ; CHECK: icmp sgt <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp slt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @min_red_inverse_select(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 ; Unsigned tests.
 
 ; Turn this into a max reduction.
 ; CHECK-LABEL: @umax_red(
 ; CHECK: icmp ugt <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp ugt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @umax_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; Turn this into a max reduction. The select has its inputs reversed therefore
 ; this is a max reduction.
 ; CHECK-LABEL: @umax_red_inverse_select(
 ; CHECK: icmp ult <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp ugt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @umax_red_inverse_select(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 }
 
 ; Turn this into a min reduction.
 ; CHECK-LABEL: @umin_red(
 ; CHECK: icmp ult <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp ult <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @umin_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; Turn this into a min reduction. The select has its inputs reversed therefore
 ; this is a min reduction.
 ; CHECK-LABEL: @umin_red_inverse_select(
 ; CHECK: icmp ugt <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp ult <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @umin_red_inverse_select(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; SGE -> SLT
 ; Turn this into a min reduction (select inputs are reversed).
 ; CHECK-LABEL: @sge_min_red(
 ; CHECK: icmp sge <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp slt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @sge_min_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; SLE -> SGT
 ; Turn this into a max reduction (select inputs are reversed).
 ; CHECK-LABEL: @sle_min_red(
 ; CHECK: icmp sle <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp sgt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @sle_min_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; UGE -> ULT
 ; Turn this into a min reduction (select inputs are reversed).
 ; CHECK-LABEL: @uge_min_red(
 ; CHECK: icmp uge <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp ult <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @uge_min_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 
 ; ULE -> UGT
 ; Turn this into a max reduction (select inputs are reversed).
 ; CHECK-LABEL: @ule_min_red(
 ; CHECK: icmp ule <2 x i32>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: icmp ugt <2 x i32>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define i32 @ule_min_red(i32 %max) {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi i32 [ %max, %entry ], [ %max.red.0, %for.body ]

 ; Maximum.
 
 ; Turn this into a max reduction in the presence of a no-nans-fp-math attribute.
 ; CHECK-LABEL: @max_red_float(
 ; CHECK: fcmp ogt <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @max_red_float(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]
for.end:
   ret float %max.red.0
 }
 
 ; CHECK-LABEL: @max_red_float_ge(
 ; CHECK: fcmp oge <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @max_red_float_ge(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]
for.end:
   ret float %max.red.0
 }
 
 ; CHECK-LABEL: @inverted_max_red_float(
 ; CHECK: fcmp olt <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_max_red_float(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]
for.end:
   ret float %max.red.0
 }
 
 ; CHECK-LABEL: @inverted_max_red_float_le(
 ; CHECK: fcmp ole <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_max_red_float_le(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]
   %arrayidx = getelementptr inbounds [1024 x float]* @fA, i64 0, i64 %indvars.iv
   %0 = load float* %arrayidx, align 4
   %cmp3 = fcmp ole float %0, %max.red.08
   %max.red.0 = select i1 %cmp3, float %max.red.08, float %0
   %indvars.iv.next = add i64 %indvars.iv, 1
   %exitcond = icmp eq i64 %indvars.iv.next, 1024
   br i1 %exitcond, label %for.end, label %for.body
 
 for.end:
   ret float %max.red.0
 }
 
-; CHECK: @unordered_max_red
+; CHECK-LABEL: @unordered_max_red_float(
 ; CHECK: fcmp ugt <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @unordered_max_red_float(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]
for.end:
   ret float %max.red.0
 }
 
 ; CHECK-LABEL: @unordered_max_red_float_ge(
 ; CHECK: fcmp uge <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @unordered_max_red_float_ge(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]
   %arrayidx = getelementptr inbounds [1024 x float]* @fA, i64 0, i64 %indvars.iv
   %0 = load float* %arrayidx, align 4
   %cmp3 = fcmp uge float %0, %max.red.08
   %max.red.0 = select i1 %cmp3, float %0, float %max.red.08
   %indvars.iv.next = add i64 %indvars.iv, 1
   %exitcond = icmp eq i64 %indvars.iv.next, 1024
   br i1 %exitcond, label %for.end, label %for.body
 
 for.end:
   ret float %max.red.0
 }
 
-; CHECK: @inverted_unordered_max_red
+; CHECK-LABEL: @inverted_unordered_max_red_float(
 ; CHECK: fcmp ult <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_unordered_max_red_float(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]
for.end:
   ret float %max.red.0
 }
 
 ; CHECK-LABEL: @inverted_unordered_max_red_float_le(
 ; CHECK: fcmp ule <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp ogt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_unordered_max_red_float_le(float %max) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %max.red.08 = phi float [ %max, %entry ], [ %max.red.0, %for.body ]

 ; Minimum.
 
 ; Turn this into a min reduction in the presence of a no-nans-fp-math attribute.
 ; CHECK-LABEL: @min_red_float(
 ; CHECK: fcmp olt <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @min_red_float(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]
for.end:
   ret float %min.red.0
 }
 
 ; CHECK-LABEL: @min_red_float_le(
 ; CHECK: fcmp ole <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @min_red_float_le(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]
for.end:
   ret float %min.red.0
 }
 
 ; CHECK-LABEL: @inverted_min_red_float(
 ; CHECK: fcmp ogt <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_min_red_float(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]
for.end:
   ret float %min.red.0
 }
 
 ; CHECK-LABEL: @inverted_min_red_float_ge(
 ; CHECK: fcmp oge <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_min_red_float_ge(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]
   %arrayidx = getelementptr inbounds [1024 x float]* @fA, i64 0, i64 %indvars.iv
   %0 = load float* %arrayidx, align 4
   %cmp3 = fcmp oge float %0, %min.red.08
   %min.red.0 = select i1 %cmp3, float %min.red.08, float %0
   %indvars.iv.next = add i64 %indvars.iv, 1
   %exitcond = icmp eq i64 %indvars.iv.next, 1024
   br i1 %exitcond, label %for.end, label %for.body
 
 for.end:
   ret float %min.red.0
 }
 
-; CHECK: @unordered_min_red
+; CHECK-LABEL: @unordered_min_red_float(
 ; CHECK: fcmp ult <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @unordered_min_red_float(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]
for.end:
   ret float %min.red.0
 }
 
 ; CHECK-LABEL: @unordered_min_red_float_le(
 ; CHECK: fcmp ule <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @unordered_min_red_float_le(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]
   %arrayidx = getelementptr inbounds [1024 x float]* @fA, i64 0, i64 %indvars.iv
   %0 = load float* %arrayidx, align 4
   %cmp3 = fcmp ule float %0, %min.red.08
   %min.red.0 = select i1 %cmp3, float %0, float %min.red.08
   %indvars.iv.next = add i64 %indvars.iv, 1
   %exitcond = icmp eq i64 %indvars.iv.next, 1024
   br i1 %exitcond, label %for.end, label %for.body
 
 for.end:
   ret float %min.red.0
 }
 
-; CHECK: @inverted_unordered_min_red
+; CHECK-LABEL: @inverted_unordered_min_red_float(
 ; CHECK: fcmp ugt <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_unordered_min_red_float(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]
for.end:
   ret float %min.red.0
 }
 
 ; CHECK-LABEL: @inverted_unordered_min_red_float_ge(
 ; CHECK: fcmp uge <2 x float>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x float>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define float @inverted_unordered_min_red_float_ge(float %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi float [ %min, %entry ], [ %min.red.0, %for.body ]

 }
 
 ; Make sure we handle doubles, too.
 ; CHECK-LABEL: @min_red_double(
 ; CHECK: fcmp olt <2 x double>
 ; CHECK: select <2 x i1>
 ; CHECK: middle.block
 ; CHECK: fcmp olt <2 x double>
-; CHECK: select <2 x i1>
+; CHECK: select i1
 
 define double @min_red_double(double %min) #0 {
 entry:
   br label %for.body
 
 for.body:
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %min.red.08 = phi double [ %min, %entry ], [ %min.red.0, %for.body ]
for.body:
   %exitcond = icmp eq i64 %indvars.iv.next, 1024
   br i1 %exitcond, label %for.end, label %for.body
 
 for.end:
   ret float %max.red.0
 }
 
 
-attributes #0 = { "no-nans-fp-math"="true" } 
+attributes #0 = { "no-nans-fp-math"="true" }