X86InstrSSE.td

//====- X86InstrSSE.td - Describe the X86 Instruction Set --*- 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 X86 SSE instruction set, defining the instructions,
// and properties of the instructions which are needed for code generation,
// machine code emission, and analysis.
//
//===----------------------------------------------------------------------===//


//===----------------------------------------------------------------------===//
// SSE specific DAG Nodes.
//===----------------------------------------------------------------------===//

def SDTX86FPShiftOp : SDTypeProfile<1, 2, [ SDTCisSameAs<0, 1>,
                                            SDTCisFP<0>, SDTCisInt<2> ]>;
def SDTX86VFCMP : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<1, 2>,
                                       SDTCisFP<1>, SDTCisVT<3, i8>]>;

def X86fmin    : SDNode<"X86ISD::FMIN",      SDTFPBinOp>;
def X86fmax    : SDNode<"X86ISD::FMAX",      SDTFPBinOp>;
def X86fand    : SDNode<"X86ISD::FAND",      SDTFPBinOp,
                        [SDNPCommutative, SDNPAssociative]>;
def X86for     : SDNode<"X86ISD::FOR",       SDTFPBinOp,
                        [SDNPCommutative, SDNPAssociative]>;
def X86fxor    : SDNode<"X86ISD::FXOR",      SDTFPBinOp,
                        [SDNPCommutative, SDNPAssociative]>;
def X86frsqrt  : SDNode<"X86ISD::FRSQRT",    SDTFPUnaryOp>;
def X86frcp    : SDNode<"X86ISD::FRCP",      SDTFPUnaryOp>;
def X86fsrl    : SDNode<"X86ISD::FSRL",      SDTX86FPShiftOp>;
def X86comi    : SDNode<"X86ISD::COMI",      SDTX86CmpTest>;
def X86ucomi   : SDNode<"X86ISD::UCOMI",     SDTX86CmpTest>;
def X86pshufb  : SDNode<"X86ISD::PSHUFB", 
                 SDTypeProfile<1, 2, [SDTCisVT<0, v16i8>, SDTCisSameAs<0,1>,
                                      SDTCisSameAs<0,2>]>>;
def X86pextrb  : SDNode<"X86ISD::PEXTRB",
                 SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<2>]>>;
def X86pextrw  : SDNode<"X86ISD::PEXTRW",
                 SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<2>]>>;
def X86pinsrb  : SDNode<"X86ISD::PINSRB", 
                 SDTypeProfile<1, 3, [SDTCisVT<0, v16i8>, SDTCisSameAs<0,1>,
                                      SDTCisVT<2, i32>, SDTCisPtrTy<3>]>>;
def X86pinsrw  : SDNode<"X86ISD::PINSRW", 
                 SDTypeProfile<1, 3, [SDTCisVT<0, v8i16>, SDTCisSameAs<0,1>,
                                      SDTCisVT<2, i32>, SDTCisPtrTy<3>]>>;
def X86insrtps : SDNode<"X86ISD::INSERTPS", 
                 SDTypeProfile<1, 3, [SDTCisVT<0, v4f32>, SDTCisSameAs<0,1>,
                                      SDTCisVT<2, f32>, SDTCisPtrTy<3>]>>;
def X86vzmovl  : SDNode<"X86ISD::VZEXT_MOVL",
                 SDTypeProfile<1, 1, [SDTCisSameAs<0,1>]>>;
def X86vzload  : SDNode<"X86ISD::VZEXT_LOAD", SDTLoad,
                        [SDNPHasChain, SDNPMayLoad]>;
def X86vshl    : SDNode<"X86ISD::VSHL",      SDTIntShiftOp>;
def X86vshr    : SDNode<"X86ISD::VSRL",      SDTIntShiftOp>;
def X86cmpps   : SDNode<"X86ISD::CMPPS",     SDTX86VFCMP>;
def X86cmppd   : SDNode<"X86ISD::CMPPD",     SDTX86VFCMP>;
def X86pcmpeqb : SDNode<"X86ISD::PCMPEQB", SDTIntBinOp, [SDNPCommutative]>;
def X86pcmpeqw : SDNode<"X86ISD::PCMPEQW", SDTIntBinOp, [SDNPCommutative]>;
def X86pcmpeqd : SDNode<"X86ISD::PCMPEQD", SDTIntBinOp, [SDNPCommutative]>;
def X86pcmpeqq : SDNode<"X86ISD::PCMPEQQ", SDTIntBinOp, [SDNPCommutative]>;
def X86pcmpgtb : SDNode<"X86ISD::PCMPGTB", SDTIntBinOp>;
def X86pcmpgtw : SDNode<"X86ISD::PCMPGTW", SDTIntBinOp>;
def X86pcmpgtd : SDNode<"X86ISD::PCMPGTD", SDTIntBinOp>;
def X86pcmpgtq : SDNode<"X86ISD::PCMPGTQ", SDTIntBinOp>;

//===----------------------------------------------------------------------===//
// SSE Complex Patterns
//===----------------------------------------------------------------------===//

// These are 'extloads' from a scalar to the low element of a vector, zeroing
// the top elements.  These are used for the SSE 'ss' and 'sd' instruction
// forms.
def sse_load_f32 : ComplexPattern<v4f32, 5, "SelectScalarSSELoad", [],
                                  [SDNPHasChain, SDNPMayLoad]>;
def sse_load_f64 : ComplexPattern<v2f64, 5, "SelectScalarSSELoad", [],
                                  [SDNPHasChain, SDNPMayLoad]>;

def ssmem : Operand<v4f32> {
  let PrintMethod = "printf32mem";
  let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc, i32imm, i8imm);
}
def sdmem : Operand<v2f64> {
  let PrintMethod = "printf64mem";
  let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc, i32imm, i8imm);
}

//===----------------------------------------------------------------------===//
// SSE pattern fragments
//===----------------------------------------------------------------------===//

def loadv4f32    : PatFrag<(ops node:$ptr), (v4f32 (load node:$ptr))>;
def loadv2f64    : PatFrag<(ops node:$ptr), (v2f64 (load node:$ptr))>;
def loadv4i32    : PatFrag<(ops node:$ptr), (v4i32 (load node:$ptr))>;
def loadv2i64    : PatFrag<(ops node:$ptr), (v2i64 (load node:$ptr))>;

// Like 'store', but always requires vector alignment.
def alignedstore : PatFrag<(ops node:$val, node:$ptr),
                           (store node:$val, node:$ptr), [{
  return cast<StoreSDNode>(N)->getAlignment() >= 16;
}]>;

// Like 'load', but always requires vector alignment.
def alignedload : PatFrag<(ops node:$ptr), (load node:$ptr), [{
  return cast<LoadSDNode>(N)->getAlignment() >= 16;
}]>;

def alignedloadfsf32 : PatFrag<(ops node:$ptr), (f32   (alignedload node:$ptr))>;
def alignedloadfsf64 : PatFrag<(ops node:$ptr), (f64   (alignedload node:$ptr))>;
def alignedloadv4f32 : PatFrag<(ops node:$ptr), (v4f32 (alignedload node:$ptr))>;
def alignedloadv2f64 : PatFrag<(ops node:$ptr), (v2f64 (alignedload node:$ptr))>;
def alignedloadv4i32 : PatFrag<(ops node:$ptr), (v4i32 (alignedload node:$ptr))>;
def alignedloadv2i64 : PatFrag<(ops node:$ptr), (v2i64 (alignedload node:$ptr))>;

// Like 'load', but uses special alignment checks suitable for use in
// memory operands in most SSE instructions, which are required to
// be naturally aligned on some targets but not on others.
// FIXME: Actually implement support for targets that don't require the
//        alignment. This probably wants a subtarget predicate.
def memop : PatFrag<(ops node:$ptr), (load node:$ptr), [{
  return cast<LoadSDNode>(N)->getAlignment() >= 16;
}]>;

def memopfsf32 : PatFrag<(ops node:$ptr), (f32   (memop node:$ptr))>;
def memopfsf64 : PatFrag<(ops node:$ptr), (f64   (memop node:$ptr))>;
def memopv4f32 : PatFrag<(ops node:$ptr), (v4f32 (memop node:$ptr))>;
def memopv2f64 : PatFrag<(ops node:$ptr), (v2f64 (memop node:$ptr))>;
def memopv4i32 : PatFrag<(ops node:$ptr), (v4i32 (memop node:$ptr))>;
def memopv2i64 : PatFrag<(ops node:$ptr), (v2i64 (memop node:$ptr))>;
def memopv16i8 : PatFrag<(ops node:$ptr), (v16i8 (memop node:$ptr))>;

// SSSE3 uses MMX registers for some instructions. They aren't aligned on a
// 16-byte boundary.
// FIXME: 8 byte alignment for mmx reads is not required
def memop64 : PatFrag<(ops node:$ptr), (load node:$ptr), [{
  return cast<LoadSDNode>(N)->getAlignment() >= 8;
}]>;

def memopv8i8  : PatFrag<(ops node:$ptr), (v8i8  (memop64 node:$ptr))>;
def memopv4i16 : PatFrag<(ops node:$ptr), (v4i16 (memop64 node:$ptr))>;
def memopv8i16 : PatFrag<(ops node:$ptr), (v8i16 (memop64 node:$ptr))>;
def memopv2i32 : PatFrag<(ops node:$ptr), (v2i32 (memop64 node:$ptr))>;

def bc_v4f32 : PatFrag<(ops node:$in), (v4f32 (bitconvert node:$in))>;
def bc_v2f64 : PatFrag<(ops node:$in), (v2f64 (bitconvert node:$in))>;
def bc_v16i8 : PatFrag<(ops node:$in), (v16i8 (bitconvert node:$in))>;
def bc_v8i16 : PatFrag<(ops node:$in), (v8i16 (bitconvert node:$in))>;
def bc_v4i32 : PatFrag<(ops node:$in), (v4i32 (bitconvert node:$in))>;
def bc_v2i64 : PatFrag<(ops node:$in), (v2i64 (bitconvert node:$in))>;

def vzmovl_v2i64 : PatFrag<(ops node:$src),
                           (bitconvert (v2i64 (X86vzmovl
                             (v2i64 (scalar_to_vector (loadi64 node:$src))))))>;
def vzmovl_v4i32 : PatFrag<(ops node:$src),
                           (bitconvert (v4i32 (X86vzmovl
                             (v4i32 (scalar_to_vector (loadi32 node:$src))))))>;

def vzload_v2i64 : PatFrag<(ops node:$src),
                           (bitconvert (v2i64 (X86vzload node:$src)))>;


def fp32imm0 : PatLeaf<(f32 fpimm), [{
  return N->isExactlyValue(+0.0);
}]>;

def PSxLDQ_imm  : SDNodeXForm<imm, [{
  // Transformation function: imm >> 3
  return getI32Imm(N->getZExtValue() >> 3);
}]>;

// SHUFFLE_get_shuf_imm xform function: convert vector_shuffle mask to PSHUF*,
// SHUFP* etc. imm.
def SHUFFLE_get_shuf_imm : SDNodeXForm<build_vector, [{
  return getI8Imm(X86::getShuffleSHUFImmediate(N));
}]>;

// SHUFFLE_get_pshufhw_imm xform function: convert vector_shuffle mask to 
// PSHUFHW imm.
def SHUFFLE_get_pshufhw_imm : SDNodeXForm<build_vector, [{
  return getI8Imm(X86::getShufflePSHUFHWImmediate(N));
}]>;

// SHUFFLE_get_pshuflw_imm xform function: convert vector_shuffle mask to 
// PSHUFLW imm.
def SHUFFLE_get_pshuflw_imm : SDNodeXForm<build_vector, [{
  return getI8Imm(X86::getShufflePSHUFLWImmediate(N));
}]>;

def SSE_splat_mask : PatLeaf<(build_vector), [{
  return X86::isSplatMask(N);
}], SHUFFLE_get_shuf_imm>;

def SSE_splat_lo_mask : PatLeaf<(build_vector), [{
  return X86::isSplatLoMask(N);
}]>;

def MOVDDUP_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVDDUPMask(N);
}]>;

def MOVHLPS_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVHLPSMask(N);
}]>;

def MOVHLPS_v_undef_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVHLPS_v_undef_Mask(N);
}]>;

def MOVHP_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVHPMask(N);
}]>;

def MOVLP_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVLPMask(N);
}]>;

def MOVL_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVLMask(N);
}]>;

def MOVSHDUP_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVSHDUPMask(N);
}]>;

def MOVSLDUP_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isMOVSLDUPMask(N);
}]>;

def UNPCKL_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isUNPCKLMask(N);
}]>;

def UNPCKH_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isUNPCKHMask(N);
}]>;

def UNPCKL_v_undef_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isUNPCKL_v_undef_Mask(N);
}]>;

def UNPCKH_v_undef_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isUNPCKH_v_undef_Mask(N);
}]>;

def PSHUFD_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isPSHUFDMask(N);
}], SHUFFLE_get_shuf_imm>;

def PSHUFHW_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isPSHUFHWMask(N);
}], SHUFFLE_get_pshufhw_imm>;

def PSHUFLW_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isPSHUFLWMask(N);
}], SHUFFLE_get_pshuflw_imm>;

def SHUFP_unary_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isPSHUFDMask(N);
}], SHUFFLE_get_shuf_imm>;

def SHUFP_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isSHUFPMask(N);
}], SHUFFLE_get_shuf_imm>;

def PSHUFD_binary_shuffle_mask : PatLeaf<(build_vector), [{
  return X86::isSHUFPMask(N);
}], SHUFFLE_get_shuf_imm>;


//===----------------------------------------------------------------------===//
// SSE scalar FP Instructions
//===----------------------------------------------------------------------===//

// CMOV* - Used to implement the SSE SELECT DAG operation.  Expanded by the
// scheduler into a branch sequence.
// These are expanded by the scheduler.
let Uses = [EFLAGS], usesCustomDAGSchedInserter = 1 in {
  def CMOV_FR32 : I<0, Pseudo,
                    (outs FR32:$dst), (ins FR32:$t, FR32:$f, i8imm:$cond),
                    "#CMOV_FR32 PSEUDO!",
                    [(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond,
                                                  EFLAGS))]>;
  def CMOV_FR64 : I<0, Pseudo,
                    (outs FR64:$dst), (ins FR64:$t, FR64:$f, i8imm:$cond),
                    "#CMOV_FR64 PSEUDO!",
                    [(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond,
                                                  EFLAGS))]>;
  def CMOV_V4F32 : I<0, Pseudo,
                    (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
                    "#CMOV_V4F32 PSEUDO!",
                    [(set VR128:$dst,
                      (v4f32 (X86cmov VR128:$t, VR128:$f, imm:$cond,
                                          EFLAGS)))]>;
  def CMOV_V2F64 : I<0, Pseudo,
                    (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
                    "#CMOV_V2F64 PSEUDO!",
                    [(set VR128:$dst,
                      (v2f64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
                                          EFLAGS)))]>;
  def CMOV_V2I64 : I<0, Pseudo,
                    (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
                    "#CMOV_V2I64 PSEUDO!",
                    [(set VR128:$dst,
                      (v2i64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
                                          EFLAGS)))]>;
}

//===----------------------------------------------------------------------===//
// SSE1 Instructions
//===----------------------------------------------------------------------===//

// Move Instructions
let neverHasSideEffects = 1 in
def MOVSSrr : SSI<0x10, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
                  "movss\t{$src, $dst|$dst, $src}", []>;
let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
def MOVSSrm : SSI<0x10, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
                  "movss\t{$src, $dst|$dst, $src}",
                  [(set FR32:$dst, (loadf32 addr:$src))]>;
def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
                  "movss\t{$src, $dst|$dst, $src}",
                  [(store FR32:$src, addr:$dst)]>;

// Conversion instructions
def CVTTSS2SIrr : SSI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
                      "cvttss2si\t{$src, $dst|$dst, $src}",
                      [(set GR32:$dst, (fp_to_sint FR32:$src))]>;
def CVTTSS2SIrm : SSI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
                      "cvttss2si\t{$src, $dst|$dst, $src}",
                      [(set GR32:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
def CVTSI2SSrr  : SSI<0x2A, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
                      "cvtsi2ss\t{$src, $dst|$dst, $src}",
                      [(set FR32:$dst, (sint_to_fp GR32:$src))]>;
def CVTSI2SSrm  : SSI<0x2A, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
                      "cvtsi2ss\t{$src, $dst|$dst, $src}",
                      [(set FR32:$dst, (sint_to_fp (loadi32 addr:$src)))]>;

// Match intrinsics which expect XMM operand(s).
def Int_CVTSS2SIrr : SSI<0x2D, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
                         "cvtss2si\t{$src, $dst|$dst, $src}",
                         [(set GR32:$dst, (int_x86_sse_cvtss2si VR128:$src))]>;
def Int_CVTSS2SIrm : SSI<0x2D, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
                         "cvtss2si\t{$src, $dst|$dst, $src}",
                         [(set GR32:$dst, (int_x86_sse_cvtss2si
                                           (load addr:$src)))]>;

// Match intrinisics which expect MM and XMM operand(s).
def Int_CVTPS2PIrr : PSI<0x2D, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
                         "cvtps2pi\t{$src, $dst|$dst, $src}",
                         [(set VR64:$dst, (int_x86_sse_cvtps2pi VR128:$src))]>;
def Int_CVTPS2PIrm : PSI<0x2D, MRMSrcMem, (outs VR64:$dst), (ins f64mem:$src),
                         "cvtps2pi\t{$src, $dst|$dst, $src}",
                         [(set VR64:$dst, (int_x86_sse_cvtps2pi 
                                           (load addr:$src)))]>;
def Int_CVTTPS2PIrr: PSI<0x2C, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
                         "cvttps2pi\t{$src, $dst|$dst, $src}",
                         [(set VR64:$dst, (int_x86_sse_cvttps2pi VR128:$src))]>;
def Int_CVTTPS2PIrm: PSI<0x2C, MRMSrcMem, (outs VR64:$dst), (ins f64mem:$src),
                         "cvttps2pi\t{$src, $dst|$dst, $src}",
                         [(set VR64:$dst, (int_x86_sse_cvttps2pi 
                                           (load addr:$src)))]>;
let Constraints = "$src1 = $dst" in {
  def Int_CVTPI2PSrr : PSI<0x2A, MRMSrcReg, 
                           (outs VR128:$dst), (ins VR128:$src1, VR64:$src2),
                        "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
                        [(set VR128:$dst, (int_x86_sse_cvtpi2ps VR128:$src1,
                                           VR64:$src2))]>;
  def Int_CVTPI2PSrm : PSI<0x2A, MRMSrcMem, 
                           (outs VR128:$dst), (ins VR128:$src1, i64mem:$src2),
                        "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
                        [(set VR128:$dst, (int_x86_sse_cvtpi2ps VR128:$src1, 
                                            (load addr:$src2)))]>;
}

// Aliases for intrinsics
def Int_CVTTSS2SIrr : SSI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
                          "cvttss2si\t{$src, $dst|$dst, $src}",
                          [(set GR32:$dst,
                            (int_x86_sse_cvttss2si VR128:$src))]>;
def Int_CVTTSS2SIrm : SSI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
                          "cvttss2si\t{$src, $dst|$dst, $src}",
                          [(set GR32:$dst,
                            (int_x86_sse_cvttss2si(load addr:$src)))]>;

let Constraints = "$src1 = $dst" in {
  def Int_CVTSI2SSrr : SSI<0x2A, MRMSrcReg,
                           (outs VR128:$dst), (ins VR128:$src1, GR32:$src2),
                           "cvtsi2ss\t{$src2, $dst|$dst, $src2}",
                           [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
                                              GR32:$src2))]>;
  def Int_CVTSI2SSrm : SSI<0x2A, MRMSrcMem,
                           (outs VR128:$dst), (ins VR128:$src1, i32mem:$src2),
                           "cvtsi2ss\t{$src2, $dst|$dst, $src2}",
                           [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
                                              (loadi32 addr:$src2)))]>;
}

// Comparison instructions
let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
  def CMPSSrr : SSIi8<0xC2, MRMSrcReg, 
                    (outs FR32:$dst), (ins FR32:$src1, FR32:$src, SSECC:$cc),
                    "cmp${cc}ss\t{$src, $dst|$dst, $src}", []>;
let mayLoad = 1 in
  def CMPSSrm : SSIi8<0xC2, MRMSrcMem, 
                    (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, SSECC:$cc),
                    "cmp${cc}ss\t{$src, $dst|$dst, $src}", []>;
}

let Defs = [EFLAGS] in {
def UCOMISSrr: PSI<0x2E, MRMSrcReg, (outs), (ins FR32:$src1, FR32:$src2),
                   "ucomiss\t{$src2, $src1|$src1, $src2}",
                   [(X86cmp FR32:$src1, FR32:$src2), (implicit EFLAGS)]>;
def UCOMISSrm: PSI<0x2E, MRMSrcMem, (outs), (ins FR32:$src1, f32mem:$src2),
                   "ucomiss\t{$src2, $src1|$src1, $src2}",
                   [(X86cmp FR32:$src1, (loadf32 addr:$src2)),
                    (implicit EFLAGS)]>;
} // Defs = [EFLAGS]

// Aliases to match intrinsics which expect XMM operand(s).
let Constraints = "$src1 = $dst" in {
  def Int_CMPSSrr : SSIi8<0xC2, MRMSrcReg, 
                        (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
                        "cmp${cc}ss\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
                                           VR128:$src, imm:$cc))]>;
  def Int_CMPSSrm : SSIi8<0xC2, MRMSrcMem, 
                        (outs VR128:$dst), (ins VR128:$src1, f32mem:$src, SSECC:$cc),
                        "cmp${cc}ss\t{$src, $dst|$dst, $src}",
                        [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
                                           (load addr:$src), imm:$cc))]>;
}

let Defs = [EFLAGS] in {
def Int_UCOMISSrr: PSI<0x2E, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
                       "ucomiss\t{$src2, $src1|$src1, $src2}",
                       [(X86ucomi (v4f32 VR128:$src1), VR128:$src2),
                        (implicit EFLAGS)]>;
def Int_UCOMISSrm: PSI<0x2E, MRMSrcMem, (outs),(ins VR128:$src1, f128mem:$src2),
                       "ucomiss\t{$src2, $src1|$src1, $src2}",
                       [(X86ucomi (v4f32 VR128:$src1), (load addr:$src2)),
                        (implicit EFLAGS)]>;

def Int_COMISSrr: PSI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
                      "comiss\t{$src2, $src1|$src1, $src2}",
                      [(X86comi (v4f32 VR128:$src1), VR128:$src2),
                       (implicit EFLAGS)]>;
def Int_COMISSrm: PSI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
                      "comiss\t{$src2, $src1|$src1, $src2}",
                      [(X86comi (v4f32 VR128:$src1), (load addr:$src2)),
                       (implicit EFLAGS)]>;
} // Defs = [EFLAGS]

// Aliases of packed SSE1 instructions for scalar use. These all have names that
// start with 'Fs'.

// Alias instructions that map fld0 to pxor for sse.
let isReMaterializable = 1, isAsCheapAsAMove = 1 in
def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins),
                 "pxor\t$dst, $dst", [(set FR32:$dst, fp32imm0)]>,
               Requires<[HasSSE1]>, TB, OpSize;

// Alias instruction to do FR32 reg-to-reg copy using movaps. Upper bits are
// disregarded.
let neverHasSideEffects = 1 in 
def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
                     "movaps\t{$src, $dst|$dst, $src}", []>;

// Alias instruction to load FR32 from f128mem using movaps. Upper bits are
// disregarded.
let canFoldAsLoad = 1 in
def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
                     "movaps\t{$src, $dst|$dst, $src}",
                     [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;

// Alias bitwise logical operations using SSE logical ops on packed FP values.
let Constraints = "$src1 = $dst" in {
let isCommutable = 1 in {
  def FsANDPSrr : PSI<0x54, MRMSrcReg, (outs FR32:$dst),
                                       (ins FR32:$src1, FR32:$src2),
                      "andps\t{$src2, $dst|$dst, $src2}",
                      [(set FR32:$dst, (X86fand FR32:$src1, FR32:$src2))]>;
  def FsORPSrr  : PSI<0x56, MRMSrcReg, (outs FR32:$dst),
                                       (ins FR32:$src1, FR32:$src2),
                      "orps\t{$src2, $dst|$dst, $src2}",
                      [(set FR32:$dst, (X86for FR32:$src1, FR32:$src2))]>;
  def FsXORPSrr : PSI<0x57, MRMSrcReg, (outs FR32:$dst),
                                       (ins FR32:$src1, FR32:$src2),
                      "xorps\t{$src2, $dst|$dst, $src2}",
                      [(set FR32:$dst, (X86fxor FR32:$src1, FR32:$src2))]>;
}

def FsANDPSrm : PSI<0x54, MRMSrcMem, (outs FR32:$dst),
                                     (ins FR32:$src1, f128mem:$src2),
                    "andps\t{$src2, $dst|$dst, $src2}",
                    [(set FR32:$dst, (X86fand FR32:$src1,
                                      (memopfsf32 addr:$src2)))]>;
def FsORPSrm  : PSI<0x56, MRMSrcMem, (outs FR32:$dst),
                                     (ins FR32:$src1, f128mem:$src2),
                    "orps\t{$src2, $dst|$dst, $src2}",
                    [(set FR32:$dst, (X86for FR32:$src1,
                                      (memopfsf32 addr:$src2)))]>;
def FsXORPSrm : PSI<0x57, MRMSrcMem, (outs FR32:$dst),
                                     (ins FR32:$src1, f128mem:$src2),
                    "xorps\t{$src2, $dst|$dst, $src2}",
                    [(set FR32:$dst, (X86fxor FR32:$src1,
                                      (memopfsf32 addr:$src2)))]>;

let neverHasSideEffects = 1 in {
def FsANDNPSrr : PSI<0x55, MRMSrcReg,
                     (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
                     "andnps\t{$src2, $dst|$dst, $src2}", []>;
let mayLoad = 1 in
def FsANDNPSrm : PSI<0x55, MRMSrcMem,
                     (outs FR32:$dst), (ins FR32:$src1, f128mem:$src2),
                     "andnps\t{$src2, $dst|$dst, $src2}", []>;
}
}

/// basic_sse1_fp_binop_rm - SSE1 binops come in both scalar and vector forms.
///
/// In addition, we also have a special variant of the scalar form here to
/// represent the associated intrinsic operation.  This form is unlike the
/// plain scalar form, in that it takes an entire vector (instead of a scalar)
/// and leaves the top elements unmodified (therefore these cannot be commuted).
///
/// These three forms can each be reg+reg or reg+mem, so there are a total of
/// six "instructions".
///
let Constraints = "$src1 = $dst" in {
multiclass basic_sse1_fp_binop_rm<bits<8> opc, string OpcodeStr,
                                  SDNode OpNode, Intrinsic F32Int,
                                  bit Commutable = 0> {
  // Scalar operation, reg+reg.
  def SSrr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
                 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                 [(set FR32:$dst, (OpNode FR32:$src1, FR32:$src2))]> {
    let isCommutable = Commutable;
  }

  // Scalar operation, reg+mem.
  def SSrm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
                                 (ins FR32:$src1, f32mem:$src2),
                 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                 [(set FR32:$dst, (OpNode FR32:$src1, (load addr:$src2)))]>;
                 
  // Vector operation, reg+reg.
  def PSrr : PSI<opc, MRMSrcReg, (outs VR128:$dst),
                                 (ins VR128:$src1, VR128:$src2),
               !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
               [(set VR128:$dst, (v4f32 (OpNode VR128:$src1, VR128:$src2)))]> {
    let isCommutable = Commutable;
  }

  // Vector operation, reg+mem.
  def PSrm : PSI<opc, MRMSrcMem, (outs VR128:$dst),
                                 (ins VR128:$src1, f128mem:$src2),
                 !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
             [(set VR128:$dst, (OpNode VR128:$src1, (memopv4f32 addr:$src2)))]>;

  // Intrinsic operation, reg+reg.
  def SSrr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
                                     (ins VR128:$src1, VR128:$src2),
                     !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                     [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2))]>;

  // Intrinsic operation, reg+mem.
  def SSrm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
                                     (ins VR128:$src1, ssmem:$src2),
                     !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                     [(set VR128:$dst, (F32Int VR128:$src1,
                                               sse_load_f32:$src2))]>;
}
}

// Arithmetic instructions
defm ADD : basic_sse1_fp_binop_rm<0x58, "add", fadd, int_x86_sse_add_ss, 1>;
defm MUL : basic_sse1_fp_binop_rm<0x59, "mul", fmul, int_x86_sse_mul_ss, 1>;
defm SUB : basic_sse1_fp_binop_rm<0x5C, "sub", fsub, int_x86_sse_sub_ss>;
defm DIV : basic_sse1_fp_binop_rm<0x5E, "div", fdiv, int_x86_sse_div_ss>;

/// sse1_fp_binop_rm - Other SSE1 binops
///
/// This multiclass is like basic_sse1_fp_binop_rm, with the addition of
/// instructions for a full-vector intrinsic form.  Operations that map
/// onto C operators don't use this form since they just use the plain
/// vector form instead of having a separate vector intrinsic form.
///
/// This provides a total of eight "instructions".
///
let Constraints = "$src1 = $dst" in {
multiclass sse1_fp_binop_rm<bits<8> opc, string OpcodeStr,
                            SDNode OpNode,
                            Intrinsic F32Int,
                            Intrinsic V4F32Int,
                            bit Commutable = 0> {

  // Scalar operation, reg+reg.
  def SSrr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
                 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                 [(set FR32:$dst, (OpNode FR32:$src1, FR32:$src2))]> {
    let isCommutable = Commutable;
  }

  // Scalar operation, reg+mem.
  def SSrm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
                                 (ins FR32:$src1, f32mem:$src2),
                 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                 [(set FR32:$dst, (OpNode FR32:$src1, (load addr:$src2)))]>;
                 
  // Vector operation, reg+reg.
  def PSrr : PSI<opc, MRMSrcReg, (outs VR128:$dst),
                                 (ins VR128:$src1, VR128:$src2),
               !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
               [(set VR128:$dst, (v4f32 (OpNode VR128:$src1, VR128:$src2)))]> {
    let isCommutable = Commutable;
  }

  // Vector operation, reg+mem.
  def PSrm : PSI<opc, MRMSrcMem, (outs VR128:$dst),
                                 (ins VR128:$src1, f128mem:$src2),
                 !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
             [(set VR128:$dst, (OpNode VR128:$src1, (memopv4f32 addr:$src2)))]>;

  // Intrinsic operation, reg+reg.
  def SSrr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
                                     (ins VR128:$src1, VR128:$src2),
                     !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                     [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2))]> {
    let isCommutable = Commutable;
  }

  // Intrinsic operation, reg+mem.
  def SSrm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
                                     (ins VR128:$src1, ssmem:$src2),
                     !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
                     [(set VR128:$dst, (F32Int VR128:$src1,
                                               sse_load_f32:$src2))]>;

  // Vector intrinsic operation, reg+reg.
  def PSrr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst),
                                     (ins VR128:$src1, VR128:$src2),
                     !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
                     [(set VR128:$dst, (V4F32Int VR128:$src1, VR128:$src2))]> {
    let isCommutable = Commutable;
  }

  // Vector intrinsic operation, reg+mem.
  def PSrm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst),
                                     (ins VR128:$src1, f128mem:$src2),
                     !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
           [(set VR128:$dst, (V4F32Int VR128:$src1, (memopv4f32 addr:$src2)))]>;
}
}

defm MAX : sse1_fp_binop_rm<0x5F, "max", X86fmax,
                            int_x86_sse_max_ss, int_x86_sse_max_ps>;
defm MIN : sse1_fp_binop_rm<0x5D, "min", X86fmin,
                            int_x86_sse_min_ss, int_x86_sse_min_ps>;

//===----------------------------------------------------------------------===//
// SSE packed FP Instructions

// Move Instructions
let neverHasSideEffects = 1 in 
def MOVAPSrr : PSI<0x28, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                   "movaps\t{$src, $dst|$dst, $src}", []>;
let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
def MOVAPSrm : PSI<0x28, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                   "movaps\t{$src, $dst|$dst, $src}",
                   [(set VR128:$dst, (alignedloadv4f32 addr:$src))]>;

def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                   "movaps\t{$src, $dst|$dst, $src}",
                   [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;

let neverHasSideEffects = 1 in
def MOVUPSrr : PSI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                   "movups\t{$src, $dst|$dst, $src}", []>;
let canFoldAsLoad = 1 in
def MOVUPSrm : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                   "movups\t{$src, $dst|$dst, $src}",
                   [(set VR128:$dst, (loadv4f32 addr:$src))]>;
def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                   "movups\t{$src, $dst|$dst, $src}",
                   [(store (v4f32 VR128:$src), addr:$dst)]>;

// Intrinsic forms of MOVUPS load and store
let canFoldAsLoad = 1 in
def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                       "movups\t{$src, $dst|$dst, $src}",
                       [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
                       "movups\t{$src, $dst|$dst, $src}",
                       [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;

let Constraints = "$src1 = $dst" in {
  let AddedComplexity = 20 in {
    def MOVLPSrm : PSI<0x12, MRMSrcMem,
                       (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
                       "movlps\t{$src2, $dst|$dst, $src2}",
       [(set VR128:$dst, 
             (v4f32 (vector_shuffle VR128:$src1,
                     (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2)))),
                     MOVLP_shuffle_mask)))]>;
    def MOVHPSrm : PSI<0x16, MRMSrcMem,
                       (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
                       "movhps\t{$src2, $dst|$dst, $src2}",
       [(set VR128:$dst, 
             (v4f32 (vector_shuffle VR128:$src1,
                     (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2)))),
                     MOVHP_shuffle_mask)))]>;
  } // AddedComplexity
} // Constraints = "$src1 = $dst"


def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                   "movlps\t{$src, $dst|$dst, $src}",
                   [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
                                 (iPTR 0))), addr:$dst)]>;

// v2f64 extract element 1 is always custom lowered to unpack high to low
// and extract element 0 so the non-store version isn't too horrible.
def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
                   "movhps\t{$src, $dst|$dst, $src}",
                   [(store (f64 (vector_extract
                                 (v2f64 (vector_shuffle
                                         (bc_v2f64 (v4f32 VR128:$src)), (undef),
                                         UNPCKH_shuffle_mask)), (iPTR 0))),
                     addr:$dst)]>;

let Constraints = "$src1 = $dst" in {
let AddedComplexity = 20 in {
def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                    "movlhps\t{$src2, $dst|$dst, $src2}",
                    [(set VR128:$dst,
                      (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
                              MOVHP_shuffle_mask)))]>;

def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                    "movhlps\t{$src2, $dst|$dst, $src2}",
                    [(set VR128:$dst,
                      (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
                              MOVHLPS_shuffle_mask)))]>;
} // AddedComplexity
} // Constraints = "$src1 = $dst"

let AddedComplexity = 20 in
def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef), MOVDDUP_shuffle_mask)),
          (MOVLHPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>;




// Arithmetic

/// sse1_fp_unop_rm - SSE1 unops come in both scalar and vector forms.
///
/// In addition, we also have a special variant of the scalar form here to
/// represent the associated intrinsic operation.  This form is unlike the
/// plain scalar form, in that it takes an entire vector (instead of a
/// scalar) and leaves the top elements undefined.
///
/// And, we have a special variant form for a full-vector intrinsic form.
///
/// These four forms can each have a reg or a mem operand, so there are a
/// total of eight "instructions".
///
multiclass sse1_fp_unop_rm<bits<8> opc, string OpcodeStr,
                           SDNode OpNode,
                           Intrinsic F32Int,
                           Intrinsic V4F32Int,
                           bit Commutable = 0> {
  // Scalar operation, reg.
  def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
                !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
                [(set FR32:$dst, (OpNode FR32:$src))]> {
    let isCommutable = Commutable;
  }

  // Scalar operation, mem.
  def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
                !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
                [(set FR32:$dst, (OpNode (load addr:$src)))]>;
                 
  // Vector operation, reg.
  def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
              !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
              [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]> {
    let isCommutable = Commutable;
  }

  // Vector operation, mem.
  def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
                [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;

  // Intrinsic operation, reg.
  def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                    !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
                    [(set VR128:$dst, (F32Int VR128:$src))]> {
    let isCommutable = Commutable;
  }

  // Intrinsic operation, mem.
  def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
                    !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
                    [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;

  // Vector intrinsic operation, reg
  def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
                    !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
                    [(set VR128:$dst, (V4F32Int VR128:$src))]> {
    let isCommutable = Commutable;
  }

  // Vector intrinsic operation, mem
  def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
                    !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
                    [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
}

// Square root.
defm SQRT  : sse1_fp_unop_rm<0x51, "sqrt",  fsqrt,
                             int_x86_sse_sqrt_ss, int_x86_sse_sqrt_ps>;

// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
defm RSQRT : sse1_fp_unop_rm<0x52, "rsqrt", X86frsqrt,
                             int_x86_sse_rsqrt_ss, int_x86_sse_rsqrt_ps>;
defm RCP   : sse1_fp_unop_rm<0x53, "rcp",   X86frcp,
                             int_x86_sse_rcp_ss, int_x86_sse_rcp_ps>;

// Logical
let Constraints = "$src1 = $dst" in {
  let isCommutable = 1 in {
    def ANDPSrr : PSI<0x54, MRMSrcReg,
                      (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                      "andps\t{$src2, $dst|$dst, $src2}",
                      [(set VR128:$dst, (v2i64
                                         (and VR128:$src1, VR128:$src2)))]>;
    def ORPSrr  : PSI<0x56, MRMSrcReg,
                      (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                      "orps\t{$src2, $dst|$dst, $src2}",
                      [(set VR128:$dst, (v2i64
                                         (or VR128:$src1, VR128:$src2)))]>;
    def XORPSrr : PSI<0x57, MRMSrcReg,
                      (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                      "xorps\t{$src2, $dst|$dst, $src2}",
                      [(set VR128:$dst, (v2i64
                                         (xor VR128:$src1, VR128:$src2)))]>;
  }

  def ANDPSrm : PSI<0x54, MRMSrcMem,
                    (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
                    "andps\t{$src2, $dst|$dst, $src2}",
                    [(set VR128:$dst, (and (bc_v2i64 (v4f32 VR128:$src1)),
                                       (memopv2i64 addr:$src2)))]>;
  def ORPSrm  : PSI<0x56, MRMSrcMem,
                    (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
                    "orps\t{$src2, $dst|$dst, $src2}",
                    [(set VR128:$dst, (or (bc_v2i64 (v4f32 VR128:$src1)),
                                       (memopv2i64 addr:$src2)))]>;
  def XORPSrm : PSI<0x57, MRMSrcMem,
                    (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
                    "xorps\t{$src2, $dst|$dst, $src2}",
                    [(set VR128:$dst, (xor (bc_v2i64 (v4f32 VR128:$src1)),
                                       (memopv2i64 addr:$src2)))]>;
  def ANDNPSrr : PSI<0x55, MRMSrcReg,
                     (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                     "andnps\t{$src2, $dst|$dst, $src2}",
                     [(set VR128:$dst,
                       (v2i64 (and (xor VR128:$src1,
                                    (bc_v2i64 (v4i32 immAllOnesV))),
                               VR128:$src2)))]>;
  def ANDNPSrm : PSI<0x55, MRMSrcMem,
                     (outs VR128:$dst), (ins VR128:$src1,f128mem:$src2),
                     "andnps\t{$src2, $dst|$dst, $src2}",
                     [(set VR128:$dst,
                       (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
                                    (bc_v2i64 (v4i32 immAllOnesV))),
                               (memopv2i64 addr:$src2))))]>;
}

let Constraints = "$src1 = $dst" in {
  def CMPPSrri : PSIi8<0xC2, MRMSrcReg, 
                    (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
                    "cmp${cc}ps\t{$src, $dst|$dst, $src}",
                    [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
                                                        VR128:$src, imm:$cc))]>;
  def CMPPSrmi : PSIi8<0xC2, MRMSrcMem, 
                  (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, SSECC:$cc),
                  "cmp${cc}ps\t{$src, $dst|$dst, $src}",
                  [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
                                            (memop addr:$src), imm:$cc))]>;
}
def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
          (CMPPSrri VR128:$src1, VR128:$src2, imm:$cc)>;
def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
          (CMPPSrmi VR128:$src1, addr:$src2, imm:$cc)>;

// Shuffle and unpack instructions
let Constraints = "$src1 = $dst" in {
  let isConvertibleToThreeAddress = 1 in // Convert to pshufd
    def SHUFPSrri : PSIi8<0xC6, MRMSrcReg, 
                          (outs VR128:$dst), (ins VR128:$src1,
                           VR128:$src2, i32i8imm:$src3),
                          "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
                          [(set VR128:$dst,
                            (v4f32 (vector_shuffle
                                    VR128:$src1, VR128:$src2,
                                    SHUFP_shuffle_mask:$src3)))]>;
  def SHUFPSrmi : PSIi8<0xC6, MRMSrcMem, 
                        (outs VR128:$dst), (ins VR128:$src1,
                         f128mem:$src2, i32i8imm:$src3),
                        "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
                        [(set VR128:$dst,
                          (v4f32 (vector_shuffle
                                  VR128:$src1, (memopv4f32 addr:$src2),
                                  SHUFP_shuffle_mask:$src3)))]>;

  let AddedComplexity = 10 in {
    def UNPCKHPSrr : PSI<0x15, MRMSrcReg, 
                         (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                         "unpckhps\t{$src2, $dst|$dst, $src2}",
                         [(set VR128:$dst,
                           (v4f32 (vector_shuffle
                                   VR128:$src1, VR128:$src2,
                                   UNPCKH_shuffle_mask)))]>;
    def UNPCKHPSrm : PSI<0x15, MRMSrcMem, 
                         (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
                         "unpckhps\t{$src2, $dst|$dst, $src2}",
                         [(set VR128:$dst,
                           (v4f32 (vector_shuffle
                                   VR128:$src1, (memopv4f32 addr:$src2),
                                   UNPCKH_shuffle_mask)))]>;

    def UNPCKLPSrr : PSI<0x14, MRMSrcReg, 
                         (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
                         "unpcklps\t{$src2, $dst|$dst, $src2}",
                         [(set VR128:$dst,
                           (v4f32 (vector_shuffle
                                   VR128:$src1, VR128:$src2,
                                   UNPCKL_shuffle_mask)))]>;
    def UNPCKLPSrm : PSI<0x14, MRMSrcMem, 
                         (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
                         "unpcklps\t{$src2, $dst|$dst, $src2}",
                         [(set VR128:$dst,
                           (v4f32 (vector_shuffle
                                   VR128:$src1, (memopv4f32 addr:$src2),
                                   UNPCKL_shuffle_mask)))]>;
  } // AddedComplexity
} // Constraints = "$src1 = $dst"

// Mask creation
def MOVMSKPSrr : PSI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
                     "movmskps\t{$src, $dst|$dst, $src}",
                     [(set GR32:$dst, (int_x86_sse_movmsk_ps VR128:$src))]>;
def MOVMSKPDrr : PSI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
                     "movmskpd\t{$src, $dst|$dst, $src}",
                     [(set GR32:$dst, (int_x86_sse2_movmsk_pd VR128:$src))]>;

// Prefetch intrinsic.
def PREFETCHT0   : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
    "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
def PREFETCHT1   : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
    "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
def PREFETCHT2   : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
    "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
def PREFETCHNTA  : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
    "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;

// Non-temporal stores
def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
                    "movntps\t{$src, $dst|$dst, $src}",
                    [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;

// Load, store, and memory fence
def SFENCE : PSI<0xAE, MRM7m, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>;

// MXCSR register
def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
                  "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
                  "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;

// Alias instructions that map zero vector to pxor / xorp* for sse.
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-zeros value if folding it would be beneficial.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1 in
def V_SET0 : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins),
                 "xorps\t$dst, $dst",
                 [(set VR128:$dst, (v4i32 immAllZerosV))]>;

let Predicates = [HasSSE1] in {
  def : Pat<(v2i64 immAllZerosV), (V_SET0)>;