[ARM][VecReduce] Force expand vector_reduce_fmin

Under MVE, we do not have any lowering for fminimum, which a
vector_reduce_fmin without NoNan will be expanded into. As with the
other recent patches, force this to expand in the pre-isel pass. Note
that Neon lowering would be OK because the scalar fminimum uses the
vector VMIN instruction, but is probably better to just rely on the
scalar operations, which is what is done here.

Also fixes what appears to be the reversal of INF vs -INF in the
vector_reduce_fmin widening code.

Author: davemgreen

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

@@ -4678,11 +4678,11 @@ SDValue DAGTypeLegalizer::WidenVecOp_VECREDUCE(SDNode *N) {
     break;
   case ISD::VECREDUCE_FMAX:
     NeutralElem = DAG.getConstantFP(
-        std::numeric_limits<double>::infinity(), dl, ElemVT);
+        -std::numeric_limits<double>::infinity(), dl, ElemVT);
     break;
   case ISD::VECREDUCE_FMIN:
     NeutralElem = DAG.getConstantFP(
-        -std::numeric_limits<double>::infinity(), dl, ElemVT);
+        std::numeric_limits<double>::infinity(), dl, ElemVT);
     break;
   }
 

llvm/lib/Target/ARM/ARMTargetTransformInfo.h

@@ -176,12 +176,15 @@ class ARMTTIImpl : public BasicTTIImplBase<ARMTTIImpl> {
       // We don't have legalization support for ordered FP reductions.
       if (!II->getFastMathFlags().allowReassoc())
         return true;
-      LLVM_FALLTHROUGH;
+      // Can't legalize reductions with soft floats.
+      return TLI->useSoftFloat() || !TLI->getSubtarget()->hasFPRegs();
 
     case Intrinsic::experimental_vector_reduce_fmin:
     case Intrinsic::experimental_vector_reduce_fmax:
-      // Can't legalize reductions with soft floats.
-      return TLI->useSoftFloat() || !TLI->getSubtarget()->hasFPRegs();
+      // Can't legalize reductions with soft floats, and NoNan will create
+      // fminimum which we do not know how to lower.
+      return TLI->useSoftFloat() || !TLI->getSubtarget()->hasFPRegs() ||
+             !II->getFastMathFlags().noNaNs();
 
     default:
       // Don't expand anything else, let legalization deal with it.

llvm/test/CodeGen/AArch64/vecreduce-fmax-legalization.ll

@@ -47,7 +47,7 @@ define fp128 @test_v1f128(<1 x fp128> %a) nounwind {
 define float @test_v3f32(<3 x float> %a) nounwind {
 ; CHECK-LABEL: test_v3f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #2139095040
+; CHECK-NEXT:    mov w8, #-8388608
 ; CHECK-NEXT:    fmov s1, w8
 ; CHECK-NEXT:    mov v0.s[3], v1.s[0]
 ; CHECK-NEXT:    fmaxnmv s0, v0.4s