[MIPS] Use custom bitcast lowering to avoid excessive instructions

On Mips32r2 bitcast can be expanded to two sw instructions and an ldc1
when using bitcast i64 to double or an sdc1 and two lw instructions when
using bitcast double to i64. By introducing custom lowering that uses
mtc1/mthc1 we can avoid excessive instructions.

Patch by Mirko Brkusanin.

Differential Revision: https://reviews.llvm.org/D61069

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@359171 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/Mips/MipsISelLowering.cpp

@@ -1203,8 +1203,9 @@ MipsTargetLowering::LowerOperationWrapper(SDNode *N,
                                           SelectionDAG &DAG) const {
   SDValue Res = LowerOperation(SDValue(N, 0), DAG);
 
-  for (unsigned I = 0, E = Res->getNumValues(); I != E; ++I)
-    Results.push_back(Res.getValue(I));
+  if (Res)
+    for (unsigned I = 0, E = Res->getNumValues(); I != E; ++I)
+      Results.push_back(Res.getValue(I));
 }
 
 void

lib/Target/Mips/MipsSEISelLowering.cpp

@@ -213,6 +213,11 @@ MipsSETargetLowering::MipsSETargetLowering(const MipsTargetMachine &TM,
   setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
   setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
 
+  if (Subtarget.hasMips32r2() && !Subtarget.useSoftFloat() &&
+      !Subtarget.hasMips64()) {
+    setOperationAction(ISD::BITCAST, MVT::i64, Custom);
+  }
+
   if (NoDPLoadStore) {
     setOperationAction(ISD::LOAD, MVT::f64, Custom);
     setOperationAction(ISD::STORE, MVT::f64, Custom);
@@ -462,6 +467,7 @@ SDValue MipsSETargetLowering::LowerOperation(SDValue Op,
   case ISD::BUILD_VECTOR:       return lowerBUILD_VECTOR(Op, DAG);
   case ISD::VECTOR_SHUFFLE:     return lowerVECTOR_SHUFFLE(Op, DAG);
   case ISD::SELECT:             return lowerSELECT(Op, DAG);
+  case ISD::BITCAST:            return lowerBITCAST(Op, DAG);
   }
 
   return MipsTargetLowering::LowerOperation(Op, DAG);
@@ -1220,6 +1226,36 @@ SDValue MipsSETargetLowering::lowerSTORE(SDValue Op, SelectionDAG &DAG) const {
                       Nd.getMemOperand()->getFlags(), Nd.getAAInfo());
 }
 
+SDValue MipsSETargetLowering::lowerBITCAST(SDValue Op,
+                                           SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  MVT Src = Op.getOperand(0).getValueType().getSimpleVT();
+  MVT Dest = Op.getValueType().getSimpleVT();
+
+  // Bitcast i64 to double.
+  if (Src == MVT::i64 && Dest == MVT::f64) {
+    SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
+                             Op.getOperand(0), DAG.getIntPtrConstant(0, DL));
+    SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32,
+                             Op.getOperand(0), DAG.getIntPtrConstant(1, DL));
+    return DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
+  }
+
+  // Bitcast double to i64.
+  if (Src == MVT::f64 && Dest == MVT::i64) {
+    SDValue Lo =
+        DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Op.getOperand(0),
+                    DAG.getConstant(0, DL, MVT::i32));
+    SDValue Hi =
+        DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32, Op.getOperand(0),
+                    DAG.getConstant(1, DL, MVT::i32));
+    return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi);
+  }
+
+  // Skip other cases of bitcast and use default lowering.
+  return SDValue();
+}
+
 SDValue MipsSETargetLowering::lowerMulDiv(SDValue Op, unsigned NewOpc,
                                           bool HasLo, bool HasHi,
                                           SelectionDAG &DAG) const {

lib/Target/Mips/MipsSEISelLowering.h

@@ -72,6 +72,7 @@ class TargetRegisterClass;
 
     SDValue lowerLOAD(SDValue Op, SelectionDAG &DAG) const;
     SDValue lowerSTORE(SDValue Op, SelectionDAG &DAG) const;
+    SDValue lowerBITCAST(SDValue Op, SelectionDAG &DAG) const;
 
     SDValue lowerMulDiv(SDValue Op, unsigned NewOpc, bool HasLo, bool HasHi,
                         SelectionDAG &DAG) const;

test/CodeGen/Mips/llvm-ir/bitcast.ll

@@ -1,13 +1,13 @@
 ; RUN: llc -march=mips -mcpu=mips32r2                         -asm-show-inst \
-; RUN: -mno-ldc1-sdc1 < %s | FileCheck %s --check-prefix=MIPS32R2
+; RUN: < %s | FileCheck %s --check-prefix=MIPS32R2
 ; RUN: llc -march=mips -mcpu=mips32r2 -mattr=+fp64            -asm-show-inst \
-; RUN: -mno-ldc1-sdc1 < %s | FileCheck %s --check-prefix=MIPS32FP64
+; RUN: < %s | FileCheck %s --check-prefix=MIPS32FP64
 ; RUN: llc -march=mips -mcpu=mips32r3 -mattr=+micromips       -asm-show-inst \
-; RUN: -mno-ldc1-sdc1 < %s | FileCheck %s --check-prefix=MM
+; RUN: < %s | FileCheck %s --check-prefix=MM
 ; RUN: llc -march=mips -mcpu=mips32r3 -mattr=+micromips,+fp64 -asm-show-inst \
-; RUN: -mno-ldc1-sdc1 < %s | FileCheck %s --check-prefix=MMFP64
+; RUN: < %s | FileCheck %s --check-prefix=MMFP64
 ; RUN: llc -march=mips -mcpu=mips32r6 -mattr=+micromips       -asm-show-inst \
-; RUN: -mno-ldc1-sdc1 < %s | FileCheck %s --check-prefix=MMR6
+; RUN: < %s | FileCheck %s --check-prefix=MMR6
 
 define double @mthc1(i64 %a) {
 ; MIPS32R2:   mthc1   {{.*}}                 # <MCInst #{{[0-9]+}} MTHC1_D32