Target/AMDGPU/AMDGPUInstructions.td

//===-- AMDGPUInstructions.td - Common instruction defs ---*- tablegen -*-===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains instruction defs that are common to all hw codegen
// targets.
//
//===----------------------------------------------------------------------===//

class AMDGPUInst <dag outs, dag ins, string asm = "",
  list<dag> pattern = []> : Instruction {
  field bit isRegisterLoad = 0;
  field bit isRegisterStore = 0;

  let Namespace = "AMDGPU";
  let OutOperandList = outs;
  let InOperandList = ins;
  let AsmString = asm;
  let Pattern = pattern;
  let Itinerary = NullALU;

  // SoftFail is a field the disassembler can use to provide a way for
  // instructions to not match without killing the whole decode process. It is
  // mainly used for ARM, but Tablegen expects this field to exist or it fails
  // to build the decode table.
  field bits<64> SoftFail = 0;

  let DecoderNamespace = Namespace;

  let TSFlags{63} = isRegisterLoad;
  let TSFlags{62} = isRegisterStore;
}

class AMDGPUShaderInst <dag outs, dag ins, string asm = "",
  list<dag> pattern = []> : AMDGPUInst<outs, ins, asm, pattern> {

  field bits<32> Inst = 0xffffffff;
}

def FP16Denormals : Predicate<"Subtarget->hasFP16Denormals()">;
def FP32Denormals : Predicate<"Subtarget->hasFP32Denormals()">;
def FP64Denormals : Predicate<"Subtarget->hasFP64Denormals()">;
def NoFP16Denormals : Predicate<"!Subtarget->hasFP16Denormals()">;
def NoFP32Denormals : Predicate<"!Subtarget->hasFP32Denormals()">;
def NoFP64Denormals : Predicate<"!Subtarget->hasFP64Denormals()">;
def UnsafeFPMath : Predicate<"TM.Options.UnsafeFPMath">;
def FMA : Predicate<"Subtarget->hasFMA()">;

def InstFlag : OperandWithDefaultOps <i32, (ops (i32 0))>;

def u16ImmTarget : AsmOperandClass {
  let Name = "U16Imm";
  let RenderMethod = "addImmOperands";
}

def s16ImmTarget : AsmOperandClass {
  let Name = "S16Imm";
  let RenderMethod = "addImmOperands";
}

let OperandType = "OPERAND_IMMEDIATE" in {

def u32imm : Operand<i32> {
  let PrintMethod = "printU32ImmOperand";
}

def u16imm : Operand<i16> {
  let PrintMethod = "printU16ImmOperand";
  let ParserMatchClass = u16ImmTarget;
}

def s16imm : Operand<i16> {
  let PrintMethod = "printU16ImmOperand";
  let ParserMatchClass = s16ImmTarget;
}

def u8imm : Operand<i8> {
  let PrintMethod = "printU8ImmOperand";
}

} // End OperandType = "OPERAND_IMMEDIATE"

//===--------------------------------------------------------------------===//
// Custom Operands
//===--------------------------------------------------------------------===//
def brtarget   : Operand<OtherVT>;

//===----------------------------------------------------------------------===//
// Misc. PatFrags
//===----------------------------------------------------------------------===//

class HasOneUseUnaryOp<SDPatternOperator op> : PatFrag<
  (ops node:$src0),
  (op $src0),
  [{ return N->hasOneUse(); }]
>;

class HasOneUseBinOp<SDPatternOperator op> : PatFrag<
  (ops node:$src0, node:$src1),
  (op $src0, $src1),
  [{ return N->hasOneUse(); }]
>;

class HasOneUseTernaryOp<SDPatternOperator op> : PatFrag<
  (ops node:$src0, node:$src1, node:$src2),
  (op $src0, $src1, $src2),
  [{ return N->hasOneUse(); }]
>;

def trunc_oneuse : HasOneUseUnaryOp<trunc>;

let Properties = [SDNPCommutative, SDNPAssociative] in {
def smax_oneuse : HasOneUseBinOp<smax>;
def smin_oneuse : HasOneUseBinOp<smin>;
def umax_oneuse : HasOneUseBinOp<umax>;
def umin_oneuse : HasOneUseBinOp<umin>;
def fminnum_oneuse : HasOneUseBinOp<fminnum>;
def fmaxnum_oneuse : HasOneUseBinOp<fmaxnum>;
def and_oneuse : HasOneUseBinOp<and>;
def or_oneuse : HasOneUseBinOp<or>;
def xor_oneuse : HasOneUseBinOp<xor>;
} // Properties = [SDNPCommutative, SDNPAssociative]

def add_oneuse : HasOneUseBinOp<add>;
def sub_oneuse : HasOneUseBinOp<sub>;

def srl_oneuse : HasOneUseBinOp<srl>;
def shl_oneuse : HasOneUseBinOp<shl>;

def select_oneuse : HasOneUseTernaryOp<select>;

def srl_16 : PatFrag<
  (ops node:$src0), (srl_oneuse node:$src0, (i32 16))
>;


def hi_i16_elt : PatFrag<
  (ops node:$src0), (i16 (trunc (i32 (srl_16 node:$src0))))
>;


def hi_f16_elt : PatLeaf<
  (vt), [{
  if (N->getOpcode() != ISD::BITCAST)
    return false;
  SDValue Tmp = N->getOperand(0);

  if (Tmp.getOpcode() != ISD::SRL)
    return false;
    if (const auto *RHS = dyn_cast<ConstantSDNode>(Tmp.getOperand(1))
      return RHS->getZExtValue() == 16;
    return false;
}]>;

//===----------------------------------------------------------------------===//
// PatLeafs for floating-point comparisons
//===----------------------------------------------------------------------===//

def COND_OEQ : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETOEQ || N->get() == ISD::SETEQ;}]
>;

def COND_ONE : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETONE || N->get() == ISD::SETNE;}]
>;

def COND_OGT : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETOGT || N->get() == ISD::SETGT;}]
>;

def COND_OGE : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETOGE || N->get() == ISD::SETGE;}]
>;

def COND_OLT : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETOLT || N->get() == ISD::SETLT;}]
>;

def COND_OLE : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETOLE || N->get() == ISD::SETLE;}]
>;

def COND_O : PatLeaf <(cond), [{return N->get() == ISD::SETO;}]>;
def COND_UO : PatLeaf <(cond), [{return N->get() == ISD::SETUO;}]>;

//===----------------------------------------------------------------------===//
// PatLeafs for unsigned / unordered comparisons
//===----------------------------------------------------------------------===//

def COND_UEQ : PatLeaf <(cond), [{return N->get() == ISD::SETUEQ;}]>;
def COND_UNE : PatLeaf <(cond), [{return N->get() == ISD::SETUNE;}]>;
def COND_UGT : PatLeaf <(cond), [{return N->get() == ISD::SETUGT;}]>;
def COND_UGE : PatLeaf <(cond), [{return N->get() == ISD::SETUGE;}]>;
def COND_ULT : PatLeaf <(cond), [{return N->get() == ISD::SETULT;}]>;
def COND_ULE : PatLeaf <(cond), [{return N->get() == ISD::SETULE;}]>;

// XXX - For some reason R600 version is preferring to use unordered
// for setne?
def COND_UNE_NE : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETUNE || N->get() == ISD::SETNE;}]
>;

//===----------------------------------------------------------------------===//
// PatLeafs for signed comparisons
//===----------------------------------------------------------------------===//

def COND_SGT : PatLeaf <(cond), [{return N->get() == ISD::SETGT;}]>;
def COND_SGE : PatLeaf <(cond), [{return N->get() == ISD::SETGE;}]>;
def COND_SLT : PatLeaf <(cond), [{return N->get() == ISD::SETLT;}]>;
def COND_SLE : PatLeaf <(cond), [{return N->get() == ISD::SETLE;}]>;

//===----------------------------------------------------------------------===//
// PatLeafs for integer equality
//===----------------------------------------------------------------------===//

def COND_EQ : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETEQ || N->get() == ISD::SETUEQ;}]
>;

def COND_NE : PatLeaf <
  (cond),
  [{return N->get() == ISD::SETNE || N->get() == ISD::SETUNE;}]
>;

def COND_NULL : PatLeaf <
  (cond),
  [{(void)N; return false;}]
>;


//===----------------------------------------------------------------------===//
// Load/Store Pattern Fragments
//===----------------------------------------------------------------------===//

class Aligned8Bytes <dag ops, dag frag> : PatFrag <ops, frag, [{
  return cast<MemSDNode>(N)->getAlignment() % 8 == 0;
}]>;

class Aligned16Bytes <dag ops, dag frag> : PatFrag <ops, frag, [{
  return cast<MemSDNode>(N)->getAlignment() >= 16;
}]>;

class LoadFrag <SDPatternOperator op> : PatFrag<(ops node:$ptr), (op node:$ptr)>;

class StoreFrag<SDPatternOperator op> : PatFrag <
  (ops node:$value, node:$ptr), (op node:$value, node:$ptr)
>;

class StoreHi16<SDPatternOperator op> : PatFrag <
  (ops node:$value, node:$ptr), (op (srl node:$value, (i32 16)), node:$ptr)
>;

class PrivateAddress : CodePatPred<[{
  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.PRIVATE_ADDRESS;
}]>;

class ConstantAddress : CodePatPred<[{
  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.CONSTANT_ADDRESS;
}]>;

class LocalAddress : CodePatPred<[{
  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.LOCAL_ADDRESS;
}]>;

class GlobalAddress : CodePatPred<[{
  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.GLOBAL_ADDRESS;
}]>;

class GlobalLoadAddress : CodePatPred<[{
  auto AS = cast<MemSDNode>(N)->getAddressSpace();
  return AS == AMDGPUASI.GLOBAL_ADDRESS || AS == AMDGPUASI.CONSTANT_ADDRESS;
}]>;

class FlatLoadAddress : CodePatPred<[{
  const auto AS = cast<MemSDNode>(N)->getAddressSpace();
  return AS == AMDGPUASI.FLAT_ADDRESS ||
         AS == AMDGPUASI.GLOBAL_ADDRESS ||
         AS == AMDGPUASI.CONSTANT_ADDRESS;
}]>;

class FlatStoreAddress : CodePatPred<[{
  const auto AS = cast<MemSDNode>(N)->getAddressSpace();
  return AS == AMDGPUASI.FLAT_ADDRESS ||
         AS == AMDGPUASI.GLOBAL_ADDRESS;
}]>;

class AZExtLoadBase <SDPatternOperator ld_node>: PatFrag<(ops node:$ptr),
                                              (ld_node node:$ptr), [{
  LoadSDNode *L = cast<LoadSDNode>(N);
  return L->getExtensionType() == ISD::ZEXTLOAD ||
         L->getExtensionType() == ISD::EXTLOAD;
}]>;

def az_extload : AZExtLoadBase <unindexedload>;

def az_extloadi8 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i8;
}]>;

def az_extloadi16 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i16;
}]>;

def az_extloadi32 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i32;
}]>;

class PrivateLoad <SDPatternOperator op> : LoadFrag <op>, PrivateAddress;
class PrivateStore <SDPatternOperator op> : StoreFrag <op>, PrivateAddress;

class LocalLoad <SDPatternOperator op> : LoadFrag <op>, LocalAddress;
class LocalStore <SDPatternOperator op> : StoreFrag <op>, LocalAddress;

class GlobalLoad <SDPatternOperator op> : LoadFrag<op>, GlobalLoadAddress;
class GlobalStore <SDPatternOperator op> : StoreFrag<op>, GlobalAddress;

class FlatLoad <SDPatternOperator op> : LoadFrag <op>, FlatLoadAddress;
class FlatStore <SDPatternOperator op> : StoreFrag <op>, FlatStoreAddress;

class ConstantLoad <SDPatternOperator op> : LoadFrag <op>, ConstantAddress;


def load_private : PrivateLoad <load>;
def az_extloadi8_private : PrivateLoad <az_extloadi8>;
def sextloadi8_private : PrivateLoad <sextloadi8>;
def az_extloadi16_private : PrivateLoad <az_extloadi16>;
def sextloadi16_private : PrivateLoad <sextloadi16>;

def store_private : PrivateStore <store>;
def truncstorei8_private : PrivateStore<truncstorei8>;
def truncstorei16_private : PrivateStore <truncstorei16>;
def store_hi16_private : StoreHi16 <truncstorei16>, PrivateAddress;
def truncstorei8_hi16_private : StoreHi16<truncstorei8>, PrivateAddress;


def load_global : GlobalLoad <load>;
def sextloadi8_global : GlobalLoad <sextloadi8>;
def az_extloadi8_global : GlobalLoad <az_extloadi8>;
def sextloadi16_global : GlobalLoad <sextloadi16>;
def az_extloadi16_global : GlobalLoad <az_extloadi16>;
def atomic_load_global : GlobalLoad<atomic_load>;

def store_global : GlobalStore <store>;
def truncstorei8_global : GlobalStore <truncstorei8>;
def truncstorei16_global : GlobalStore <truncstorei16>;
def store_atomic_global : GlobalStore<atomic_store>;
def truncstorei8_hi16_global : StoreHi16 <truncstorei8>, GlobalAddress;
def truncstorei16_hi16_global : StoreHi16 <truncstorei16>, GlobalAddress;

def load_local : LocalLoad <load>;
def az_extloadi8_local : LocalLoad <az_extloadi8>;
def sextloadi8_local : LocalLoad <sextloadi8>;
def az_extloadi16_local : LocalLoad <az_extloadi16>;
def sextloadi16_local : LocalLoad <sextloadi16>;
def atomic_load_32_local : LocalLoad<atomic_load_32>;
def atomic_load_64_local : LocalLoad<atomic_load_64>;

def store_local : LocalStore <store>;
def truncstorei8_local : LocalStore <truncstorei8>;
def truncstorei16_local : LocalStore <truncstorei16>;
def store_local_hi16 : StoreHi16 <truncstorei16>, LocalAddress;
def truncstorei8_local_hi16 : StoreHi16<truncstorei8>, LocalAddress;
def atomic_store_local : LocalStore <atomic_store>;

def load_align8_local : Aligned8Bytes <
  (ops node:$ptr), (load_local node:$ptr)
>;

def load_align16_local : Aligned16Bytes <
  (ops node:$ptr), (load_local node:$ptr)
>;

def store_align8_local : Aligned8Bytes <
  (ops node:$val, node:$ptr), (store_local node:$val, node:$ptr)
>;

def store_align16_local : Aligned16Bytes <
  (ops node:$val, node:$ptr), (store_local node:$val, node:$ptr)
>;

def load_flat          : FlatLoad <load>;
def az_extloadi8_flat  : FlatLoad <az_extloadi8>;
def sextloadi8_flat    : FlatLoad <sextloadi8>;
def az_extloadi16_flat : FlatLoad <az_extloadi16>;
def sextloadi16_flat   : FlatLoad <sextloadi16>;
def atomic_load_flat   : FlatLoad<atomic_load>;

def store_flat         : FlatStore <store>;
def truncstorei8_flat  : FlatStore <truncstorei8>;
def truncstorei16_flat : FlatStore <truncstorei16>;
def atomic_store_flat  : FlatStore <atomic_store>;
def truncstorei8_hi16_flat  : StoreHi16<truncstorei8>, FlatStoreAddress;
def truncstorei16_hi16_flat : StoreHi16<truncstorei16>, FlatStoreAddress;


def constant_load : ConstantLoad<load>;
def sextloadi8_constant : ConstantLoad <sextloadi8>;
def az_extloadi8_constant : ConstantLoad <az_extloadi8>;
def sextloadi16_constant : ConstantLoad <sextloadi16>;
def az_extloadi16_constant : ConstantLoad <az_extloadi16>;


class local_binary_atomic_op<SDNode atomic_op> :
  PatFrag<(ops node:$ptr, node:$value),
    (atomic_op node:$ptr, node:$value), [{
  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.LOCAL_ADDRESS;
}]>;

def atomic_swap_local : local_binary_atomic_op<atomic_swap>;
def atomic_load_add_local : local_binary_atomic_op<atomic_load_add>;
def atomic_load_sub_local : local_binary_atomic_op<atomic_load_sub>;
def atomic_load_and_local : local_binary_atomic_op<atomic_load_and>;
def atomic_load_or_local : local_binary_atomic_op<atomic_load_or>;
def atomic_load_xor_local : local_binary_atomic_op<atomic_load_xor>;
def atomic_load_nand_local : local_binary_atomic_op<atomic_load_nand>;
def atomic_load_min_local : local_binary_atomic_op<atomic_load_min>;
def atomic_load_max_local : local_binary_atomic_op<atomic_load_max>;
def atomic_load_umin_local : local_binary_atomic_op<atomic_load_umin>;
def atomic_load_umax_local : local_binary_atomic_op<atomic_load_umax>;

def mskor_global : PatFrag<(ops node:$val, node:$ptr),
                            (AMDGPUstore_mskor node:$val, node:$ptr), [{
  return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.GLOBAL_ADDRESS;
}]>;

class AtomicCmpSwapLocal <SDNode cmp_swap_node> : PatFrag<
    (ops node:$ptr, node:$cmp, node:$swap),
    (cmp_swap_node node:$ptr, node:$cmp, node:$swap), [{
      AtomicSDNode *AN = cast<AtomicSDNode>(N);
      return AN->getAddressSpace() == AMDGPUASI.LOCAL_ADDRESS;
}]>;

def atomic_cmp_swap_local : AtomicCmpSwapLocal <atomic_cmp_swap>;

multiclass global_binary_atomic_op<SDNode atomic_op> {
  def "" : PatFrag<
        (ops node:$ptr, node:$value),
        (atomic_op node:$ptr, node:$value),
        [{return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.GLOBAL_ADDRESS;}]>;

  def _noret : PatFrag<
        (ops node:$ptr, node:$value),
        (atomic_op node:$ptr, node:$value),
        [{return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.GLOBAL_ADDRESS && (SDValue(N, 0).use_empty());}]>;

  def _ret : PatFrag<
        (ops node:$ptr, node:$value),
        (atomic_op node:$ptr, node:$value),
        [{return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.GLOBAL_ADDRESS && (!SDValue(N, 0).use_empty());}]>;
}

defm atomic_swap_global : global_binary_atomic_op<atomic_swap>;
defm atomic_add_global : global_binary_atomic_op<atomic_load_add>;
defm atomic_and_global : global_binary_atomic_op<atomic_load_and>;
defm atomic_max_global : global_binary_atomic_op<atomic_load_max>;
defm atomic_min_global : global_binary_atomic_op<atomic_load_min>;
defm atomic_or_global : global_binary_atomic_op<atomic_load_or>;
defm atomic_sub_global : global_binary_atomic_op<atomic_load_sub>;
defm atomic_umax_global : global_binary_atomic_op<atomic_load_umax>;
defm atomic_umin_global : global_binary_atomic_op<atomic_load_umin>;
defm atomic_xor_global : global_binary_atomic_op<atomic_load_xor>;

// Legacy.
def AMDGPUatomic_cmp_swap_global : PatFrag<
  (ops node:$ptr, node:$value),
  (AMDGPUatomic_cmp_swap node:$ptr, node:$value)>, GlobalAddress;

def atomic_cmp_swap_global : PatFrag<
  (ops node:$ptr, node:$cmp, node:$value),
  (atomic_cmp_swap node:$ptr, node:$cmp, node:$value)>, GlobalAddress;


def atomic_cmp_swap_global_noret : PatFrag<
  (ops node:$ptr, node:$cmp, node:$value),
  (atomic_cmp_swap node:$ptr, node:$cmp, node:$value),
  [{return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.GLOBAL_ADDRESS && (SDValue(N, 0).use_empty());}]>;

def atomic_cmp_swap_global_ret : PatFrag<
  (ops node:$ptr, node:$cmp, node:$value),
  (atomic_cmp_swap node:$ptr, node:$cmp, node:$value),
  [{return cast<MemSDNode>(N)->getAddressSpace() == AMDGPUASI.GLOBAL_ADDRESS && (!SDValue(N, 0).use_empty());}]>;

//===----------------------------------------------------------------------===//
// Misc Pattern Fragments
//===----------------------------------------------------------------------===//

class Constants {
int TWO_PI = 0x40c90fdb;
int PI = 0x40490fdb;
int TWO_PI_INV = 0x3e22f983;
int FP_UINT_MAX_PLUS_1 = 0x4f800000;    // 1 << 32 in floating point encoding
int FP16_ONE = 0x3C00;
int V2FP16_ONE = 0x3C003C00;
int FP32_ONE = 0x3f800000;
int FP32_NEG_ONE = 0xbf800000;
int FP64_ONE = 0x3ff0000000000000;
int FP64_NEG_ONE = 0xbff0000000000000;
}
def CONST : Constants;

def FP_ZERO : PatLeaf <
  (fpimm),
  [{return N->getValueAPF().isZero();}]
>;

def FP_ONE : PatLeaf <
  (fpimm),
  [{return N->isExactlyValue(1.0);}]
>;

def FP_HALF : PatLeaf <
  (fpimm),
  [{return N->isExactlyValue(0.5);}]
>;

/* Generic helper patterns for intrinsics */
/* -------------------------------------- */

class POW_Common <AMDGPUInst log_ieee, AMDGPUInst exp_ieee, AMDGPUInst mul>
  : AMDGPUPat <
  (fpow f32:$src0, f32:$src1),
  (exp_ieee (mul f32:$src1, (log_ieee f32:$src0)))
>;

/* Other helper patterns */
/* --------------------- */

/* Extract element pattern */
class Extract_Element <ValueType sub_type, ValueType vec_type, int sub_idx,
                       SubRegIndex sub_reg>
  : AMDGPUPat<
  (sub_type (extractelt vec_type:$src, sub_idx)),
  (EXTRACT_SUBREG $src, sub_reg)
> {
  let SubtargetPredicate = TruePredicate;
}

/* Insert element pattern */
class Insert_Element <ValueType elem_type, ValueType vec_type,
                      int sub_idx, SubRegIndex sub_reg>
  : AMDGPUPat <
  (insertelt vec_type:$vec, elem_type:$elem, sub_idx),
  (INSERT_SUBREG $vec, $elem, sub_reg)
> {
  let SubtargetPredicate = TruePredicate;
}

// XXX: Convert to new syntax and use COPY_TO_REG, once the DFAPacketizer
// can handle COPY instructions.
// bitconvert pattern
class BitConvert <ValueType dt, ValueType st, RegisterClass rc> : AMDGPUPat <
  (dt (bitconvert (st rc:$src0))),
  (dt rc:$src0)
>;

// XXX: Convert to new syntax and use COPY_TO_REG, once the DFAPacketizer
// can handle COPY instructions.
class DwordAddrPat<ValueType vt, RegisterClass rc> : AMDGPUPat <
  (vt (AMDGPUdwordaddr (vt rc:$addr))),
  (vt rc:$addr)
>;

// BFI_INT patterns

multiclass BFIPatterns <Instruction BFI_INT,
                        Instruction LoadImm32,
                        RegisterClass RC64> {
  // Definition from ISA doc:
  // (y & x) | (z & ~x)
  def : AMDGPUPat <
    (or (and i32:$y, i32:$x), (and i32:$z, (not i32:$x))),
    (BFI_INT $x, $y, $z)
  >;

  // 64-bit version
  def : AMDGPUPat <
    (or (and i64:$y, i64:$x), (and i64:$z, (not i64:$x))),
    (REG_SEQUENCE RC64,
      (BFI_INT (i32 (EXTRACT_SUBREG $x, sub0)),
               (i32 (EXTRACT_SUBREG $y, sub0)),
               (i32 (EXTRACT_SUBREG $z, sub0))), sub0,
      (BFI_INT (i32 (EXTRACT_SUBREG $x, sub1)),
               (i32 (EXTRACT_SUBREG $y, sub1)),
               (i32 (EXTRACT_SUBREG $z, sub1))), sub1)
  >;

  // SHA-256 Ch function
  // z ^ (x & (y ^ z))
  def : AMDGPUPat <
    (xor i32:$z, (and i32:$x, (xor i32:$y, i32:$z))),
    (BFI_INT $x, $y, $z)
  >;

  // 64-bit version
  def : AMDGPUPat <
    (xor i64:$z, (and i64:$x, (xor i64:$y, i64:$z))),
    (REG_SEQUENCE RC64,
      (BFI_INT (i32 (EXTRACT_SUBREG $x, sub0)),
               (i32 (EXTRACT_SUBREG $y, sub0)),
               (i32 (EXTRACT_SUBREG $z, sub0))), sub0,
      (BFI_INT (i32 (EXTRACT_SUBREG $x, sub1)),
               (i32 (EXTRACT_SUBREG $y, sub1)),
               (i32 (EXTRACT_SUBREG $z, sub1))), sub1)
  >;

  def : AMDGPUPat <
    (fcopysign f32:$src0, f32:$src1),
    (BFI_INT (LoadImm32 (i32 0x7fffffff)), $src0, $src1)
  >;

  def : AMDGPUPat <
    (f32 (fcopysign f32:$src0, f64:$src1)),
    (BFI_INT (LoadImm32 (i32 0x7fffffff)), $src0,
             (i32 (EXTRACT_SUBREG $src1, sub1)))
  >;

  def : AMDGPUPat <
    (f64 (fcopysign f64:$src0, f64:$src1)),
    (REG_SEQUENCE RC64,
      (i32 (EXTRACT_SUBREG $src0, sub0)), sub0,
      (BFI_INT (LoadImm32 (i32 0x7fffffff)),
               (i32 (EXTRACT_SUBREG $src0, sub1)),
               (i32 (EXTRACT_SUBREG $src1, sub1))), sub1)
  >;

  def : AMDGPUPat <
    (f64 (fcopysign f64:$src0, f32:$src1)),
    (REG_SEQUENCE RC64,
      (i32 (EXTRACT_SUBREG $src0, sub0)), sub0,
      (BFI_INT (LoadImm32 (i32 0x7fffffff)),
               (i32 (EXTRACT_SUBREG $src0, sub1)),
               $src1), sub1)
  >;
}

// SHA-256 Ma patterns

// ((x & z) | (y & (x | z))) -> BFI_INT (XOR x, y), z, y
multiclass SHA256MaPattern <Instruction BFI_INT, Instruction XOR, RegisterClass RC64> {
  def : AMDGPUPat <
    (or (and i32:$x, i32:$z), (and i32:$y, (or i32:$x, i32:$z))),
    (BFI_INT (XOR i32:$x, i32:$y), i32:$z, i32:$y)
  >;

  def : AMDGPUPat <
    (or (and i64:$x, i64:$z), (and i64:$y, (or i64:$x, i64:$z))),
    (REG_SEQUENCE RC64,
      (BFI_INT (XOR (i32 (EXTRACT_SUBREG $x, sub0)),
                    (i32 (EXTRACT_SUBREG $y, sub0))),
               (i32 (EXTRACT_SUBREG $z, sub0)),
               (i32 (EXTRACT_SUBREG $y, sub0))), sub0,
      (BFI_INT (XOR (i32 (EXTRACT_SUBREG $x, sub1)),
                    (i32 (EXTRACT_SUBREG $y, sub1))),
               (i32 (EXTRACT_SUBREG $z, sub1)),
               (i32 (EXTRACT_SUBREG $y, sub1))), sub1)
  >;
}

// Bitfield extract patterns

def IMMZeroBasedBitfieldMask : PatLeaf <(imm), [{
  return isMask_32(N->getZExtValue());
}]>;

def IMMPopCount : SDNodeXForm<imm, [{
  return CurDAG->getTargetConstant(countPopulation(N->getZExtValue()), SDLoc(N),
                                   MVT::i32);
}]>;

multiclass BFEPattern <Instruction UBFE, Instruction SBFE, Instruction MOV> {
  def : AMDGPUPat <
    (i32 (and (i32 (srl i32:$src, i32:$rshift)), IMMZeroBasedBitfieldMask:$mask)),
    (UBFE $src, $rshift, (MOV (i32 (IMMPopCount $mask))))
  >;

  // x & ((1 << y) - 1)
  def : AMDGPUPat <
    (and i32:$src, (add_oneuse (shl_oneuse 1, i32:$width), -1)),
    (UBFE $src, (i32 0), $width)
  >;

  // x & ~(-1 << y)
  def : AMDGPUPat <
    (and i32:$src, (xor_oneuse (shl_oneuse -1, i32:$width), -1)),
    (UBFE $src, (i32 0), $width)
  >;

  // x & (-1 >> (bitwidth - y))
  def : AMDGPUPat <
    (and i32:$src, (srl_oneuse -1, (sub 32, i32:$width))),
    (UBFE $src, (i32 0), $width)
  >;

  // x << (bitwidth - y) >> (bitwidth - y)
  def : AMDGPUPat <
    (srl (shl_oneuse i32:$src, (sub 32, i32:$width)), (sub 32, i32:$width)),
    (UBFE $src, (i32 0), $width)
  >;

  def : AMDGPUPat <
    (sra (shl_oneuse i32:$src, (sub 32, i32:$width)), (sub 32, i32:$width)),
    (SBFE $src, (i32 0), $width)
  >;
}

// rotr pattern
class ROTRPattern <Instruction BIT_ALIGN> : AMDGPUPat <
  (rotr i32:$src0, i32:$src1),
  (BIT_ALIGN $src0, $src0, $src1)
>;

// This matches 16 permutations of
// max(min(x, y), min(max(x, y), z))
class IntMed3Pat<Instruction med3Inst,
                 SDPatternOperator max,
                 SDPatternOperator max_oneuse,
                 SDPatternOperator min_oneuse,
                 ValueType vt = i32> : AMDGPUPat<
  (max (min_oneuse vt:$src0, vt:$src1),
       (min_oneuse (max_oneuse vt:$src0, vt:$src1), vt:$src2)),
  (med3Inst $src0, $src1, $src2)
>;

// Special conversion patterns

def cvt_rpi_i32_f32 : PatFrag <
  (ops node:$src),
  (fp_to_sint (ffloor (fadd $src, FP_HALF))),
  [{ (void) N; return TM.Options.NoNaNsFPMath; }]
>;

def cvt_flr_i32_f32 : PatFrag <
  (ops node:$src),
  (fp_to_sint (ffloor $src)),
  [{ (void)N; return TM.Options.NoNaNsFPMath; }]
>;

class IMad24Pat<Instruction Inst, bit HasClamp = 0> : AMDGPUPat <
  (add (AMDGPUmul_i24 i32:$src0, i32:$src1), i32:$src2),
  !if(HasClamp, (Inst $src0, $src1, $src2, (i1 0)),
                (Inst $src0, $src1, $src2))
>;

class UMad24Pat<Instruction Inst, bit HasClamp = 0> : AMDGPUPat <
  (add (AMDGPUmul_u24 i32:$src0, i32:$src1), i32:$src2),
  !if(HasClamp, (Inst $src0, $src1, $src2, (i1 0)),
                (Inst $src0, $src1, $src2))
>;

class RcpPat<Instruction RcpInst, ValueType vt> : AMDGPUPat <
  (fdiv FP_ONE, vt:$src),
  (RcpInst $src)
>;

class RsqPat<Instruction RsqInst, ValueType vt> : AMDGPUPat <
  (AMDGPUrcp (fsqrt vt:$src)),
  (RsqInst $src)
>;

include "R600Instructions.td"
include "R700Instructions.td"
include "EvergreenInstructions.td"
include "CaymanInstructions.td"

include "SIInstrInfo.td"