[ARM] Remove target-specific ITOFP/FPTOI nodes

[jmolloy]

Summary

Anton tried this 5 years ago but it was reverted due to extra VMOVs being emitted. This can be easily fixed with a liberal application of patterns - matching loads/stores and extractelts.

There appears now to be reasonably good LIT test coverage ensuring no extra VMOVs are inserted. I fell foul of a lot of them when developing this patch, so I'm reasonably confident this won't regress performance.

[rengolin]

Hi James,

Can't see anything wrong with that. Do we have tests for old ARM cores where the call is made?

cheers,
-renato

lib/Target/ARM/ARMISelLowering.h

enum NodeType {
       FMSTAT,       // ARM fmstat instruction.
 
       CMOV,         // ARM conditional move instructions.
 
       BCC_i64,
 
       RBIT,         // ARM bitreverse instruction
 
-      FTOSI,        // FP to sint within a FP register.
-      FTOUI,        // FP to uint within a FP register.
-      SITOF,        // sint to FP within a FP register.
-      UITOF,        // uint to FP within a FP register.
-
       SRL_FLAG,     // V,Flag = srl_flag X -> srl X, 1 + save carry out.
       SRA_FLAG,     // V,Flag = sra_flag X -> sra X, 1 + save carry out.
       RRX,          // V = RRX X, Flag     -> srl X, 1 + shift in carry flag.
 
       ADDC,         // Add with carry
       ADDE,         // Add using carry
       SUBC,         // Sub with carry
       SUBE,         // Sub using carry

lib/Target/ARM/ARMISelLowering.cpp

if (Subtarget->isFPOnlySP()) {
     setOperationAction(ISD::FLOG10,     MVT::f64, Expand);
     setOperationAction(ISD::FEXP,       MVT::f64, Expand);
     setOperationAction(ISD::FEXP2,      MVT::f64, Expand);
     setOperationAction(ISD::FCEIL,      MVT::f64, Expand);
     setOperationAction(ISD::FTRUNC,     MVT::f64, Expand);
     setOperationAction(ISD::FRINT,      MVT::f64, Expand);
     setOperationAction(ISD::FNEARBYINT, MVT::f64, Expand);
     setOperationAction(ISD::FFLOOR,     MVT::f64, Expand);
+    setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
+    setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
+    setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
+    setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
+    setOperationAction(ISD::FP_TO_SINT, MVT::f64, Custom);
+    setOperationAction(ISD::FP_TO_UINT, MVT::f64, Custom);
     setOperationAction(ISD::FP_ROUND,   MVT::f32, Custom);
     setOperationAction(ISD::FP_EXTEND,  MVT::f64, Custom);
   }
 
   computeRegisterProperties(Subtarget->getRegisterInfo());
 
   // ARM does not have floating-point extending loads.
   for (MVT VT : MVT::fp_valuetypes()) {
ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
 
   if (!Subtarget->hasVFP4()) {
     setOperationAction(ISD::FMA, MVT::f64, Expand);
     setOperationAction(ISD::FMA, MVT::f32, Expand);
   }
 
   // Various VFP goodness
   if (!TM.Options.UseSoftFloat && !Subtarget->isThumb1Only()) {
-    // int <-> fp are custom expanded into bit_convert + ARMISD ops.
-    if (Subtarget->hasVFP2()) {
-      setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
-      setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
-      setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
-      setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
-    }
-
     // FP-ARMv8 adds f64 <-> f16 conversion. Before that it should be expanded.
     if (!Subtarget->hasFPARMv8() || Subtarget->isFPOnlySP()) {
       setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
       setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
     }
 
     // fp16 is a special v7 extension that adds f16 <-> f32 conversions.
     if (!Subtarget->hasFP16()) {
const char *ARMTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case ARMISD::CMPFPw0:       return "ARMISD::CMPFPw0";
   case ARMISD::BCC_i64:       return "ARMISD::BCC_i64";
   case ARMISD::FMSTAT:        return "ARMISD::FMSTAT";
 
   case ARMISD::CMOV:          return "ARMISD::CMOV";
 
   case ARMISD::RBIT:          return "ARMISD::RBIT";
 
-  case ARMISD::FTOSI:         return "ARMISD::FTOSI";
-  case ARMISD::FTOUI:         return "ARMISD::FTOUI";
-  case ARMISD::SITOF:         return "ARMISD::SITOF";
-  case ARMISD::UITOF:         return "ARMISD::UITOF";
-
   case ARMISD::SRL_FLAG:      return "ARMISD::SRL_FLAG";
   case ARMISD::SRA_FLAG:      return "ARMISD::SRA_FLAG";
   case ARMISD::RRX:           return "ARMISD::RRX";
 
   case ARMISD::ADDC:          return "ARMISD::ADDC";
   case ARMISD::ADDE:          return "ARMISD::ADDE";
   case ARMISD::SUBC:          return "ARMISD::SUBC";
   case ARMISD::SUBE:          return "ARMISD::SUBE";
static SDValue LowerVectorFP_TO_INT(SDValue Op, SelectionDAG &DAG) {
   Op = DAG.getNode(Op.getOpcode(), dl, MVT::v4i32, Op.getOperand(0));
   return DAG.getNode(ISD::TRUNCATE, dl, VT, Op);
 }
 
 SDValue ARMTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) const {
   EVT VT = Op.getValueType();
   if (VT.isVector())
     return LowerVectorFP_TO_INT(Op, DAG);
-
   if (Subtarget->isFPOnlySP() && Op.getOperand(0).getValueType() == MVT::f64) {
     RTLIB::Libcall LC;
     if (Op.getOpcode() == ISD::FP_TO_SINT)
       LC = RTLIB::getFPTOSINT(Op.getOperand(0).getValueType(),
                               Op.getValueType());
     else
       LC = RTLIB::getFPTOUINT(Op.getOperand(0).getValueType(),
                               Op.getValueType());
     return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1,
                        /*isSigned*/ false, SDLoc(Op)).first;
   }
 
-  SDLoc dl(Op);
-  unsigned Opc;
-
-  switch (Op.getOpcode()) {
-  default: llvm_unreachable("Invalid opcode!");
-  case ISD::FP_TO_SINT:
-    Opc = ARMISD::FTOSI;
-    break;
-  case ISD::FP_TO_UINT:
-    Opc = ARMISD::FTOUI;
-    break;
-  }
-  Op = DAG.getNode(Opc, dl, MVT::f32, Op.getOperand(0));
-  return DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);
+  return Op;
 }
 
 static SDValue LowerVectorINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
   EVT VT = Op.getValueType();
   SDLoc dl(Op);
 
   if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::i32) {
     if (VT.getVectorElementType() == MVT::f32)
static SDValue LowerVectorINT_TO_FP(SDValue Op, SelectionDAG &DAG) {
   Op = DAG.getNode(CastOpc, dl, MVT::v4i32, Op.getOperand(0));
   return DAG.getNode(Opc, dl, VT, Op);
 }
 
 SDValue ARMTargetLowering::LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const {
   EVT VT = Op.getValueType();
   if (VT.isVector())
     return LowerVectorINT_TO_FP(Op, DAG);
-
   if (Subtarget->isFPOnlySP() && Op.getValueType() == MVT::f64) {
     RTLIB::Libcall LC;
     if (Op.getOpcode() == ISD::SINT_TO_FP)
       LC = RTLIB::getSINTTOFP(Op.getOperand(0).getValueType(),
                               Op.getValueType());
     else
       LC = RTLIB::getUINTTOFP(Op.getOperand(0).getValueType(),
                               Op.getValueType());
     return makeLibCall(DAG, LC, Op.getValueType(), &Op.getOperand(0), 1,
                        /*isSigned*/ false, SDLoc(Op)).first;
   }
 
-  SDLoc dl(Op);
-  unsigned Opc;
-
-  switch (Op.getOpcode()) {
-  default: llvm_unreachable("Invalid opcode!");
-  case ISD::SINT_TO_FP:
-    Opc = ARMISD::SITOF;
-    break;
-  case ISD::UINT_TO_FP:
-    Opc = ARMISD::UITOF;
-    break;
-  }
-
-  Op = DAG.getNode(ISD::BITCAST, dl, MVT::f32, Op.getOperand(0));
-  return DAG.getNode(Opc, dl, VT, Op);
+  return Op;
 }
 
 SDValue ARMTargetLowering::LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const {
   // Implement fcopysign with a fabs and a conditional fneg.
   SDValue Tmp0 = Op.getOperand(0);
   SDValue Tmp1 = Op.getOperand(1);
   SDLoc dl(Op);
   EVT VT = Op.getValueType();

lib/Target/ARM/ARMInstrFormats.td

 }
 // ARMV5MOPat - Same as ARMV5TEPat with UseMulOps.
 class ARMV5MOPat<dag pattern, dag result> : Pat<pattern, result> {
   list<Predicate> Predicates = [IsARM, HasV5TE, UseMulOps];
 }
 class ARMV6Pat<dag pattern, dag result> : Pat<pattern, result> {
   list<Predicate> Predicates = [IsARM, HasV6];
 }
-
+class VFPPat<dag pattern, dag result> : Pat<pattern, result> {
+  list<Predicate> Predicates = [HasVFP2];
+}
+class VFPNoNEONPat<dag pattern, dag result> : Pat<pattern, result> {
+  list<Predicate> Predicates = [HasVFP2, DontUseNEONForFP];
+}
 //===----------------------------------------------------------------------===//
 // Thumb Instruction Format Definitions.
 //
 
 class ThumbI<dag oops, dag iops, AddrMode am, int sz,
              InstrItinClass itin, string asm, string cstr, list<dag> pattern>
   : InstThumb<am, sz, IndexModeNone, ThumbFrm, GenericDomain, cstr, itin> {
   let OutOperandList = oops;

lib/Target/ARM/ARMInstrNEON.td

: NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
                  SPR:$acc, ssub_0),
                 (INSERT_SUBREG
                  (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
                  SPR:$a, ssub_0),
                 (INSERT_SUBREG
                  (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
                  SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
 
+class NVCVTIFPat<SDNode OpNode, NeonI Inst>
+  : NEONFPPat<(f32 (OpNode GPR:$a)),
+              (f32 (EXTRACT_SUBREG
+                     (v2f32 (Inst
+                       (INSERT_SUBREG
+                         (v2f32 (IMPLICIT_DEF)),
+                         (i32 (COPY_TO_REGCLASS GPR:$a, SPR)), ssub_0))),
+                     ssub_0))>;
+class NVCVTFIPat<SDNode OpNode, NeonI Inst>
+  : NEONFPPat<(i32 (OpNode SPR:$a)),
+              (i32 (EXTRACT_SUBREG
+                     (v2f32 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
+                                                 SPR:$a, ssub_0))),
+                     ssub_0))>;
+
 def : N3VSPat<fadd, VADDfd>;
 def : N3VSPat<fsub, VSUBfd>;
 def : N3VSPat<fmul, VMULfd>;
 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
       Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
       Requires<[HasNEON, UseNEONForFP, UseFPVMLx, DontUseFusedMAC]>;
 def : N3VSMulOpPat<fmul, fadd, VFMAfd>,
       Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
 def : N3VSMulOpPat<fmul, fsub, VFMSfd>,
       Requires<[HasVFP4, UseNEONForFP, UseFusedMAC]>;
 def : N2VSPat<fabs, VABSfd>;
 def : N2VSPat<fneg, VNEGfd>;
 def : N3VSPat<NEONfmax, VMAXfd>;
 def : N3VSPat<NEONfmin, VMINfd>;
-def : N2VSPat<arm_ftosi, VCVTf2sd>;
-def : N2VSPat<arm_ftoui, VCVTf2ud>;
-def : N2VSPat<arm_sitof, VCVTs2fd>;
-def : N2VSPat<arm_uitof, VCVTu2fd>;
+def : NVCVTFIPat<fp_to_sint, VCVTf2sd>;
+def : NVCVTFIPat<fp_to_uint, VCVTf2ud>;
+def : NVCVTIFPat<sint_to_fp, VCVTs2fd>;
+def : NVCVTIFPat<uint_to_fp, VCVTu2fd>;
+
+// NEON doesn't have any f64 conversions, so provide patterns to make
+// sure the VFP conversions match when extracting from a vector.
+def : VFPPat<(f64 (sint_to_fp (extractelt (v2i32 DPR:$src), imm:$lane))),
+             (VSITOD (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
+def : VFPPat<(f64 (sint_to_fp (extractelt (v4i32 QPR:$src), imm:$lane))),
+             (VSITOD (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane)))>;
+def : VFPPat<(f64 (uint_to_fp (extractelt (v2i32 DPR:$src), imm:$lane))),
+             (VUITOD (EXTRACT_SUBREG DPR:$src, (SSubReg_f32_reg imm:$lane)))>;
+def : VFPPat<(f64 (uint_to_fp (extractelt (v4i32 QPR:$src), imm:$lane))),
+             (VUITOD (EXTRACT_SUBREG QPR:$src, (SSubReg_f32_reg imm:$lane)))>;
+
 
 // Prefer VMOVDRR for i32 -> f32 bitcasts, it can write all DPR registers.
 def : Pat<(f32 (bitconvert GPR:$a)),
           (EXTRACT_SUBREG (VMOVDRR GPR:$a, GPR:$a), ssub_0)>,
         Requires<[HasNEON, DontUseVMOVSR]>;
 
 //===----------------------------------------------------------------------===//
 // Non-Instruction Patterns

lib/Target/ARM/ARMInstrVFP.td

 //===-- ARMInstrVFP.td - VFP support for ARM ---------------*- tablegen -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // This file describes the ARM VFP instruction set.
 //
 //===----------------------------------------------------------------------===//
 
-def SDT_FTOI    : SDTypeProfile<1, 1, [SDTCisVT<0, f32>, SDTCisFP<1>]>;
-def SDT_ITOF    : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisVT<1, f32>]>;
 def SDT_CMPFP0  : SDTypeProfile<0, 1, [SDTCisFP<0>]>;
 def SDT_VMOVDRR : SDTypeProfile<1, 2, [SDTCisVT<0, f64>, SDTCisVT<1, i32>,
                                        SDTCisSameAs<1, 2>]>;
 
-def arm_ftoui  : SDNode<"ARMISD::FTOUI",   SDT_FTOI>;
-def arm_ftosi  : SDNode<"ARMISD::FTOSI",   SDT_FTOI>;
-def arm_sitof  : SDNode<"ARMISD::SITOF",   SDT_ITOF>;
-def arm_uitof  : SDNode<"ARMISD::UITOF",   SDT_ITOF>;
 def arm_fmstat : SDNode<"ARMISD::FMSTAT",  SDTNone, [SDNPInGlue, SDNPOutGlue]>;
 def arm_cmpfp  : SDNode<"ARMISD::CMPFP",   SDT_ARMCmp, [SDNPOutGlue]>;
 def arm_cmpfp0 : SDNode<"ARMISD::CMPFPw0", SDT_CMPFP0, [SDNPOutGlue]>;
 def arm_fmdrr  : SDNode<"ARMISD::VMOVDRR", SDT_VMOVDRR>;
 
 
 //===----------------------------------------------------------------------===//
 // Operand Definitions.
def : Pat<(f64 (f16_to_fp GPR:$a)),
           (VCVTBHD (COPY_TO_REGCLASS GPR:$a, SPR))>;
 
 multiclass vcvt_inst<string opc, bits<2> rm,
                      SDPatternOperator node = null_frag> {
   let PostEncoderMethod = "", DecoderNamespace = "VFPV8" in {
     def SS : ASuInp<0b11101, 0b11, 0b1100, 0b11, 0,
                     (outs SPR:$Sd), (ins SPR:$Sm),
                     NoItinerary, !strconcat("vcvt", opc, ".s32.f32\t$Sd, $Sm"),
-                    [(set SPR:$Sd, (arm_ftosi (node SPR:$Sm)))]>,
+                    []>,
                     Requires<[HasFPARMv8]> {
       let Inst{17-16} = rm;
     }
 
     def US : ASuInp<0b11101, 0b11, 0b1100, 0b01, 0,
                     (outs SPR:$Sd), (ins SPR:$Sm),
                     NoItinerary, !strconcat("vcvt", opc, ".u32.f32\t$Sd, $Sm"),
-                    [(set SPR:$Sd, (arm_ftoui (node SPR:$Sm)))]>,
+                    []>,
                     Requires<[HasFPARMv8]> {
       let Inst{17-16} = rm;
     }
 
     def SD : ASuInp<0b11101, 0b11, 0b1100, 0b11, 0,
                     (outs SPR:$Sd), (ins DPR:$Dm),
                     NoItinerary, !strconcat("vcvt", opc, ".s32.f64\t$Sd, $Dm"),
-                    [(set SPR:$Sd, (arm_ftosi (f64 (node (f64 DPR:$Dm)))))]>,
+                    []>,
                     Requires<[HasFPARMv8, HasDPVFP]> {
       bits<5> Dm;
 
       let Inst{17-16} = rm;
 
       // Encode instruction operands
       let Inst{3-0} = Dm{3-0};
       let Inst{5}   = Dm{4};
       let Inst{8} = 1;
     }
 
     def UD : ASuInp<0b11101, 0b11, 0b1100, 0b01, 0,
                     (outs SPR:$Sd), (ins DPR:$Dm),
                     NoItinerary, !strconcat("vcvt", opc, ".u32.f64\t$Sd, $Dm"),
-                    [(set SPR:$Sd, (arm_ftoui (f64 (node (f64 DPR:$Dm)))))]>,
+                    []>,
                     Requires<[HasFPARMv8, HasDPVFP]> {
       bits<5> Dm;
 
       let Inst{17-16} = rm;
 
       // Encode instruction operands
       let Inst{3-0}  = Dm{3-0};
       let Inst{5}    = Dm{4};
       let Inst{8} = 1;
     }
   }
+
+  let Predicates = [HasFPARMv8] in {
+    def : Pat<(i32 (fp_to_sint (node SPR:$a))),
+              (COPY_TO_REGCLASS
+                (!cast<Instruction>(NAME#"SS") SPR:$a),
+                GPR)>;
+    def : Pat<(i32 (fp_to_uint (node SPR:$a))),
+              (COPY_TO_REGCLASS
+                (!cast<Instruction>(NAME#"US") SPR:$a),
+                GPR)>;
+  }
+  let Predicates = [HasFPARMv8, HasDPVFP] in {
+    def : Pat<(i32 (fp_to_sint (node (f64 DPR:$a)))),
+              (COPY_TO_REGCLASS
+                (!cast<Instruction>(NAME#"SD") DPR:$a),
+                GPR)>;
+    def : Pat<(i32 (fp_to_uint (node (f64 DPR:$a)))),
+              (COPY_TO_REGCLASS
+                (!cast<Instruction>(NAME#"UD") DPR:$a),
+                GPR)>;
+  }
 }
 
 defm VCVTA : vcvt_inst<"a", 0b00, frnd>;
 defm VCVTN : vcvt_inst<"n", 0b01>;
 defm VCVTP : vcvt_inst<"p", 0b10, fceil>;
 defm VCVTM : vcvt_inst<"m", 0b11, ffloor>;
 
 def VNEGD  : ADuI<0b11101, 0b11, 0b0001, 0b01, 0,
class AVConv1InSs_Encode<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3,
   let Inst{5}     = Sm{0};
   let Inst{15-12} = Sd{4-1};
   let Inst{22}    = Sd{0};
 }
 
 def VSITOD : AVConv1IDs_Encode<0b11101, 0b11, 0b1000, 0b1011,
                                (outs DPR:$Dd), (ins SPR:$Sm),
                                IIC_fpCVTID, "vcvt", ".f64.s32\t$Dd, $Sm",
-                               [(set DPR:$Dd, (f64 (arm_sitof SPR:$Sm)))]> {
+                               []> {
   let Inst{7} = 1; // s32
 }
 
+let Predicates=[HasVFP2, HasDPVFP] in {
+  def : VFPPat<(f64 (sint_to_fp GPR:$a)),
+               (VSITOD (COPY_TO_REGCLASS GPR:$a, SPR))>;
+
+  def : VFPPat<(f64 (sint_to_fp (i32 (load addrmode5:$a)))),
+               (VSITOD (VLDRS addrmode5:$a))>;
+}
+
 def VSITOS : AVConv1InSs_Encode<0b11101, 0b11, 0b1000, 0b1010,
                                 (outs SPR:$Sd),(ins SPR:$Sm),
                                 IIC_fpCVTIS, "vcvt", ".f32.s32\t$Sd, $Sm",
-                                [(set SPR:$Sd, (arm_sitof SPR:$Sm))]> {
+                                []> {
   let Inst{7} = 1; // s32
 
   // Some single precision VFP instructions may be executed on both NEON and
   // VFP pipelines on A8.
   let D = VFPNeonA8Domain;
 }
 
+def : VFPNoNEONPat<(f32 (sint_to_fp GPR:$a)),
+                   (VSITOS (COPY_TO_REGCLASS GPR:$a, SPR))>;
+
+def : VFPNoNEONPat<(f32 (sint_to_fp (i32 (load addrmode5:$a)))),
+                   (VSITOS (VLDRS addrmode5:$a))>;
+
 def VUITOD : AVConv1IDs_Encode<0b11101, 0b11, 0b1000, 0b1011,
                                (outs DPR:$Dd), (ins SPR:$Sm),
                                IIC_fpCVTID, "vcvt", ".f64.u32\t$Dd, $Sm",
-                               [(set DPR:$Dd, (f64 (arm_uitof SPR:$Sm)))]> {
+                               []> {
   let Inst{7} = 0; // u32
 }
 
+let Predicates=[HasVFP2, HasDPVFP] in {
+  def : VFPPat<(f64 (uint_to_fp GPR:$a)),
+               (VUITOD (COPY_TO_REGCLASS GPR:$a, SPR))>;
+
+  def : VFPPat<(f64 (uint_to_fp (i32 (load addrmode5:$a)))),
+               (VUITOD (VLDRS addrmode5:$a))>;
+}
+
 def VUITOS : AVConv1InSs_Encode<0b11101, 0b11, 0b1000, 0b1010,
                                 (outs SPR:$Sd), (ins SPR:$Sm),
                                 IIC_fpCVTIS, "vcvt", ".f32.u32\t$Sd, $Sm",
-                                [(set SPR:$Sd, (arm_uitof SPR:$Sm))]> {
+                                []> {
   let Inst{7} = 0; // u32
 
   // Some single precision VFP instructions may be executed on both NEON and
   // VFP pipelines on A8.
   let D = VFPNeonA8Domain;
 }
 
+def : VFPNoNEONPat<(f32 (uint_to_fp GPR:$a)),
+                   (VUITOS (COPY_TO_REGCLASS GPR:$a, SPR))>;
+
+def : VFPNoNEONPat<(f32 (uint_to_fp (i32 (load addrmode5:$a)))),
+                   (VUITOS (VLDRS addrmode5:$a))>;
+
 // FP -> Int:
 
 class AVConv1IsD_Encode<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3,
                         bits<4> opcod4, dag oops, dag iops,
                         InstrItinClass itin, string opc, string asm,
                         list<dag> pattern>
   : AVConv1I<opcod1, opcod2, opcod3, opcod4, oops, iops, itin, opc, asm,
              pattern> {
class AVConv1InsS_Encode<bits<5> opcod1, bits<2> opcod2, bits<4> opcod3,
   let Inst{15-12} = Sd{4-1};
   let Inst{22}    = Sd{0};
 }
 
 // Always set Z bit in the instruction, i.e. "round towards zero" variants.
 def VTOSIZD : AVConv1IsD_Encode<0b11101, 0b11, 0b1101, 0b1011,
                                 (outs SPR:$Sd), (ins DPR:$Dm),
                                 IIC_fpCVTDI, "vcvt", ".s32.f64\t$Sd, $Dm",
-                                [(set SPR:$Sd, (arm_ftosi (f64 DPR:$Dm)))]> {
+                                []> {
   let Inst{7} = 1; // Z bit
 }
 
+let Predicates=[HasVFP2, HasDPVFP] in {
+  def : VFPPat<(i32 (fp_to_sint (f64 DPR:$a))),
+               (COPY_TO_REGCLASS (VTOSIZD DPR:$a), GPR)>;
+
+  def : VFPPat<(store (i32 (fp_to_sint (f64 DPR:$a))), addrmode5:$ptr),
+               (VSTRS (VTOSIZD DPR:$a), addrmode5:$ptr)>;
+}
+
 def VTOSIZS : AVConv1InsS_Encode<0b11101, 0b11, 0b1101, 0b1010,
                                  (outs SPR:$Sd), (ins SPR:$Sm),
                                  IIC_fpCVTSI, "vcvt", ".s32.f32\t$Sd, $Sm",
-                                 [(set SPR:$Sd, (arm_ftosi SPR:$Sm))]> {
+                                 []> {
   let Inst{7} = 1; // Z bit
 
   // Some single precision VFP instructions may be executed on both NEON and
   // VFP pipelines on A8.
   let D = VFPNeonA8Domain;
 }
 
+def : VFPNoNEONPat<(i32 (fp_to_sint SPR:$a)),
+                   (COPY_TO_REGCLASS (VTOSIZS SPR:$a), GPR)>;
+
+def : VFPNoNEONPat<(store (i32 (fp_to_sint (f32 SPR:$a))), addrmode5:$ptr),
+                   (VSTRS (VTOSIZS SPR:$a), addrmode5:$ptr)>;
+
 def VTOUIZD : AVConv1IsD_Encode<0b11101, 0b11, 0b1100, 0b1011,
                                (outs SPR:$Sd), (ins DPR:$Dm),
                                IIC_fpCVTDI, "vcvt", ".u32.f64\t$Sd, $Dm",
-                               [(set SPR:$Sd, (arm_ftoui (f64 DPR:$Dm)))]> {
+                               []> {
   let Inst{7} = 1; // Z bit
 }
 
+let Predicates=[HasVFP2, HasDPVFP] in {
+  def : VFPPat<(i32 (fp_to_uint (f64 DPR:$a))),
+               (COPY_TO_REGCLASS (VTOUIZD DPR:$a), GPR)>;
+
+  def : VFPPat<(store (i32 (fp_to_uint (f64 DPR:$a))), addrmode5:$ptr),
+               (VSTRS (VTOUIZD DPR:$a), addrmode5:$ptr)>;
+}
+
 def VTOUIZS : AVConv1InsS_Encode<0b11101, 0b11, 0b1100, 0b1010,
                                  (outs SPR:$Sd), (ins SPR:$Sm),
                                  IIC_fpCVTSI, "vcvt", ".u32.f32\t$Sd, $Sm",
-                                 [(set SPR:$Sd, (arm_ftoui SPR:$Sm))]> {
+                                 []> {
   let Inst{7} = 1; // Z bit
 
   // Some single precision VFP instructions may be executed on both NEON and
   // VFP pipelines on A8.
   let D = VFPNeonA8Domain;
 }
 
+def : VFPNoNEONPat<(i32 (fp_to_uint SPR:$a)),
+                   (COPY_TO_REGCLASS (VTOUIZS SPR:$a), GPR)>;
+
+def : VFPNoNEONPat<(store (i32 (fp_to_uint (f32 SPR:$a))), addrmode5:$ptr),
+                  (VSTRS (VTOUIZS SPR:$a), addrmode5:$ptr)>;
+
 // And the Z bit '0' variants, i.e. use the rounding mode specified by FPSCR.
 let Uses = [FPSCR] in {
 // FIXME: Verify encoding after integrated assembler is working.
 def VTOSIRD : AVConv1IsD_Encode<0b11101, 0b11, 0b1101, 0b1011,
                                 (outs SPR:$Sd), (ins DPR:$Dm),
                                 IIC_fpCVTDI, "vcvtr", ".s32.f64\t$Sd, $Dm",
                                 [(set SPR:$Sd, (int_arm_vcvtr (f64 DPR:$Dm)))]>{
   let Inst{7} = 0; // Z bit