//===- HexagonIntrinsicsV4.td - V4 Instruction intrinsics --*- tablegen -*-===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
// This is populated based on the following specs:
// Hexagon V4 Architecture Extensions
// Application-Level Specification
// 80-V9418-12 Rev. A
// June 15, 2010

// Vector reduce multiply word by signed half (32x16)
//Rdd=vrmpyweh(Rss,Rtt)[:<<1]
def : T_PP_pat <M4_vrmpyeh_s0, int_hexagon_M4_vrmpyeh_s0>;
def : T_PP_pat <M4_vrmpyeh_s1, int_hexagon_M4_vrmpyeh_s1>;

//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
def : T_PP_pat <M4_vrmpyoh_s0, int_hexagon_M4_vrmpyoh_s0>;
def : T_PP_pat <M4_vrmpyoh_s1, int_hexagon_M4_vrmpyoh_s1>;

//Rdd+=vrmpyweh(Rss,Rtt)[:<<1]
def : T_PPP_pat <M4_vrmpyeh_acc_s0, int_hexagon_M4_vrmpyeh_acc_s0>;
def : T_PPP_pat <M4_vrmpyeh_acc_s1, int_hexagon_M4_vrmpyeh_acc_s1>;

//Rdd=vrmpywoh(Rss,Rtt)[:<<1]
def : T_PPP_pat <M4_vrmpyoh_acc_s0, int_hexagon_M4_vrmpyoh_acc_s0>;
def : T_PPP_pat <M4_vrmpyoh_acc_s1, int_hexagon_M4_vrmpyoh_acc_s1>;

// Vector multiply halfwords, signed by unsigned
// Rdd=vmpyhsu(Rs,Rt)[:<<1]:sat
def : T_RR_pat <M2_vmpy2su_s0, int_hexagon_M2_vmpy2su_s0>;
def : T_RR_pat <M2_vmpy2su_s1, int_hexagon_M2_vmpy2su_s1>;

// Rxx+=vmpyhsu(Rs,Rt)[:<<1]:sat
def : T_PRR_pat <M2_vmac2su_s0, int_hexagon_M2_vmac2su_s0>;
def : T_PRR_pat <M2_vmac2su_s1, int_hexagon_M2_vmac2su_s1>;

// Vector polynomial multiply halfwords
// Rdd=vpmpyh(Rs,Rt)
def : T_RR_pat <M4_vpmpyh, int_hexagon_M4_vpmpyh>;
// Rxx[^]=vpmpyh(Rs,Rt)
def : T_PRR_pat <M4_vpmpyh_acc, int_hexagon_M4_vpmpyh_acc>;

// Polynomial multiply words
// Rdd=pmpyw(Rs,Rt)
def : T_RR_pat <M4_pmpyw, int_hexagon_M4_pmpyw>;
// Rxx^=pmpyw(Rs,Rt)
def : T_PRR_pat <M4_pmpyw_acc, int_hexagon_M4_pmpyw_acc>;

//Rxx^=asr(Rss,Rt)
def : T_PPR_pat <S2_asr_r_p_xor, int_hexagon_S2_asr_r_p_xor>;
//Rxx^=asl(Rss,Rt)
def : T_PPR_pat <S2_asl_r_p_xor, int_hexagon_S2_asl_r_p_xor>;
//Rxx^=lsr(Rss,Rt)
def : T_PPR_pat <S2_lsr_r_p_xor, int_hexagon_S2_lsr_r_p_xor>;
//Rxx^=lsl(Rss,Rt)
def : T_PPR_pat <S2_lsl_r_p_xor, int_hexagon_S2_lsl_r_p_xor>;

// Multiply and use upper result
def : T_RR_pat <M2_mpysu_up, int_hexagon_M2_mpysu_up>;
def : T_RR_pat <M2_mpy_up_s1, int_hexagon_M2_mpy_up_s1>;
def : T_RR_pat <M2_hmmpyh_s1, int_hexagon_M2_hmmpyh_s1>;
def : T_RR_pat <M2_hmmpyl_s1, int_hexagon_M2_hmmpyl_s1>;
def : T_RR_pat <M2_mpy_up_s1_sat, int_hexagon_M2_mpy_up_s1_sat>;

def : T_PP_pat <A2_vaddub, int_hexagon_A2_vaddb_map>;
def : T_PP_pat <A2_vsubub, int_hexagon_A2_vsubb_map>;

// Vector reduce add unsigned halfwords
def : T_PP_pat <M2_vraddh, int_hexagon_M2_vraddh>;

def: T_P_pat<S2_brevp, int_hexagon_S2_brevp>;
def: T_P_pat<S2_ct0p,  int_hexagon_S2_ct0p>;
def: T_P_pat<S2_ct1p,  int_hexagon_S2_ct1p>;

def: T_Q_RR_pat<C4_nbitsset,  int_hexagon_C4_nbitsset>;
def: T_Q_RR_pat<C4_nbitsclr,  int_hexagon_C4_nbitsclr>;
def: T_Q_RI_pat<C4_nbitsclri, int_hexagon_C4_nbitsclri>;

def : T_Q_PI_pat<A4_vcmpbeqi,     int_hexagon_A4_vcmpbeqi>;
def : T_Q_PI_pat<A4_vcmpbgti,     int_hexagon_A4_vcmpbgti>;
def : T_Q_PI_pat<A4_vcmpbgtui,    int_hexagon_A4_vcmpbgtui>;
def : T_Q_PI_pat<A4_vcmpheqi,     int_hexagon_A4_vcmpheqi>;
def : T_Q_PI_pat<A4_vcmphgti,     int_hexagon_A4_vcmphgti>;
def : T_Q_PI_pat<A4_vcmphgtui,    int_hexagon_A4_vcmphgtui>;
def : T_Q_PI_pat<A4_vcmpweqi,     int_hexagon_A4_vcmpweqi>;
def : T_Q_PI_pat<A4_vcmpwgti,     int_hexagon_A4_vcmpwgti>;
def : T_Q_PI_pat<A4_vcmpwgtui,    int_hexagon_A4_vcmpwgtui>;
def : T_Q_PP_pat<A4_vcmpbeq_any,  int_hexagon_A4_vcmpbeq_any>;

def : T_Q_RR_pat<A4_cmpbeq,   int_hexagon_A4_cmpbeq>;
def : T_Q_RR_pat<A4_cmpbgt,   int_hexagon_A4_cmpbgt>;
def : T_Q_RR_pat<A4_cmpbgtu,  int_hexagon_A4_cmpbgtu>;
def : T_Q_RR_pat<A4_cmpheq,   int_hexagon_A4_cmpheq>;
def : T_Q_RR_pat<A4_cmphgt,   int_hexagon_A4_cmphgt>;
def : T_Q_RR_pat<A4_cmphgtu,  int_hexagon_A4_cmphgtu>;

def : T_Q_RI_pat<A4_cmpbeqi,  int_hexagon_A4_cmpbeqi>;
def : T_Q_RI_pat<A4_cmpbgti,  int_hexagon_A4_cmpbgti>;
def : T_Q_RI_pat<A4_cmpbgtui, int_hexagon_A4_cmpbgtui>;

def : T_Q_RI_pat<A4_cmpheqi,  int_hexagon_A4_cmpheqi>;
def : T_Q_RI_pat<A4_cmphgti,  int_hexagon_A4_cmphgti>;
def : T_Q_RI_pat<A4_cmphgtui, int_hexagon_A4_cmphgtui>;

def : T_Q_RP_pat<A4_boundscheck, int_hexagon_A4_boundscheck>;
def : T_Q_PR_pat<A4_tlbmatch,    int_hexagon_A4_tlbmatch>;

def : T_RRR_pat <M4_mpyrr_addr,    int_hexagon_M4_mpyrr_addr>;
def : T_IRR_pat <M4_mpyrr_addi,    int_hexagon_M4_mpyrr_addi>;
def : T_IRI_pat <M4_mpyri_addi,    int_hexagon_M4_mpyri_addi>;
def : T_RIR_pat <M4_mpyri_addr_u2, int_hexagon_M4_mpyri_addr_u2>;
def : T_RRI_pat <M4_mpyri_addr,    int_hexagon_M4_mpyri_addr>;
def : T_RRR_pat <M4_mac_up_s1_sat, int_hexagon_M4_mac_up_s1_sat>;
def : T_RRR_pat <M4_nac_up_s1_sat, int_hexagon_M4_nac_up_s1_sat>;

// Complex multiply 32x16
def : T_PR_pat <M4_cmpyi_wh, int_hexagon_M4_cmpyi_wh>;
def : T_PR_pat <M4_cmpyr_wh, int_hexagon_M4_cmpyr_wh>;

def : T_PR_pat <M4_cmpyi_whc, int_hexagon_M4_cmpyi_whc>;
def : T_PR_pat <M4_cmpyr_whc, int_hexagon_M4_cmpyr_whc>;

def : T_PP_pat<A4_andnp, int_hexagon_A4_andnp>;
def : T_PP_pat<A4_ornp,  int_hexagon_A4_ornp>;

// Complex add/sub halfwords/words
def : T_PP_pat <S4_vxaddsubw, int_hexagon_S4_vxaddsubw>;
def : T_PP_pat <S4_vxsubaddw, int_hexagon_S4_vxsubaddw>;
def : T_PP_pat <S4_vxaddsubh, int_hexagon_S4_vxaddsubh>;
def : T_PP_pat <S4_vxsubaddh, int_hexagon_S4_vxsubaddh>;

def : T_PP_pat <S4_vxaddsubhr, int_hexagon_S4_vxaddsubhr>;
def : T_PP_pat <S4_vxsubaddhr, int_hexagon_S4_vxsubaddhr>;

// Extract bitfield
def : T_PP_pat  <S4_extractp_rp, int_hexagon_S4_extractp_rp>;
def : T_RP_pat  <S4_extract_rp, int_hexagon_S4_extract_rp>;
def : T_PII_pat <S4_extractp, int_hexagon_S4_extractp>;
def : T_RII_pat <S4_extract, int_hexagon_S4_extract>;

// Vector conditional negate
// Rdd=vcnegh(Rss,Rt)
def : T_PR_pat <S2_vcnegh, int_hexagon_S2_vcnegh>;

// Shift an immediate left by register amount
def : T_IR_pat<S4_lsli, int_hexagon_S4_lsli>;

// Vector reduce maximum halfwords
def : T_PPR_pat <A4_vrmaxh, int_hexagon_A4_vrmaxh>;
def : T_PPR_pat <A4_vrmaxuh, int_hexagon_A4_vrmaxuh>;

// Vector reduce maximum words
def : T_PPR_pat <A4_vrmaxw, int_hexagon_A4_vrmaxw>;
def : T_PPR_pat <A4_vrmaxuw, int_hexagon_A4_vrmaxuw>;

// Vector reduce minimum halfwords
def : T_PPR_pat <A4_vrminh, int_hexagon_A4_vrminh>;
def : T_PPR_pat <A4_vrminuh, int_hexagon_A4_vrminuh>;

// Vector reduce minimum words
def : T_PPR_pat <A4_vrminw, int_hexagon_A4_vrminw>;
def : T_PPR_pat <A4_vrminuw, int_hexagon_A4_vrminuw>;

// Rotate and reduce bytes
def : Pat <(int_hexagon_S4_vrcrotate DoubleRegs:$src1, IntRegs:$src2,
                                     u2_0ImmPred:$src3),
           (S4_vrcrotate DoubleRegs:$src1, IntRegs:$src2, u2_0ImmPred:$src3)>;

// Rotate and reduce bytes with accumulation
// Rxx+=vrcrotate(Rss,Rt,#u2)
def : Pat <(int_hexagon_S4_vrcrotate_acc DoubleRegs:$src1, DoubleRegs:$src2,
                                         IntRegs:$src3, u2_0ImmPred:$src4),
           (S4_vrcrotate_acc DoubleRegs:$src1, DoubleRegs:$src2,
                             IntRegs:$src3, u2_0ImmPred:$src4)>;

// Vector conditional negate
def : T_PPR_pat<S2_vrcnegh, int_hexagon_S2_vrcnegh>;

// Logical xor with xor accumulation
def : T_PPP_pat<M4_xor_xacc, int_hexagon_M4_xor_xacc>;

// ALU64 - Vector min/max byte
def : T_PP_pat <A2_vminb, int_hexagon_A2_vminb>;
def : T_PP_pat <A2_vmaxb, int_hexagon_A2_vmaxb>;

// Shift and add/sub/and/or
def : T_IRI_pat <S4_andi_asl_ri, int_hexagon_S4_andi_asl_ri>;
def : T_IRI_pat <S4_ori_asl_ri,  int_hexagon_S4_ori_asl_ri>;
def : T_IRI_pat <S4_addi_asl_ri, int_hexagon_S4_addi_asl_ri>;
def : T_IRI_pat <S4_subi_asl_ri, int_hexagon_S4_subi_asl_ri>;
def : T_IRI_pat <S4_andi_lsr_ri, int_hexagon_S4_andi_lsr_ri>;
def : T_IRI_pat <S4_ori_lsr_ri,  int_hexagon_S4_ori_lsr_ri>;
def : T_IRI_pat <S4_addi_lsr_ri, int_hexagon_S4_addi_lsr_ri>;
def : T_IRI_pat <S4_subi_lsr_ri, int_hexagon_S4_subi_lsr_ri>;

// Split bitfield
def : T_RI_pat <A4_bitspliti, int_hexagon_A4_bitspliti>;
def : T_RR_pat <A4_bitsplit,  int_hexagon_A4_bitsplit>;

def: T_RR_pat<S4_parity,      int_hexagon_S4_parity>;

def: T_Q_RI_pat<S4_ntstbit_i, int_hexagon_S4_ntstbit_i>;
def: T_Q_RR_pat<S4_ntstbit_r, int_hexagon_S4_ntstbit_r>;

def: T_RI_pat<S4_clbaddi,     int_hexagon_S4_clbaddi>;
def: T_PI_pat<S4_clbpaddi,    int_hexagon_S4_clbpaddi>;
def: T_P_pat <S4_clbpnorm,    int_hexagon_S4_clbpnorm>;

//*******************************************************************
//            ALU32/ALU
//*******************************************************************

// ALU32 / ALU / Logical Operations.
def: T_RR_pat<A4_andn, int_hexagon_A4_andn>;
def: T_RR_pat<A4_orn,  int_hexagon_A4_orn>;

//*******************************************************************
//            ALU32/PERM
//*******************************************************************

// Combine Words Into Doublewords.
def: T_RI_pat<A4_combineri, int_hexagon_A4_combineri, s32_0ImmPred>;
def: T_IR_pat<A4_combineir, int_hexagon_A4_combineir, s32_0ImmPred>;

//*******************************************************************
//           ALU32/PRED
//*******************************************************************

// Compare
def : T_Q_RI_pat<C4_cmpneqi, int_hexagon_C4_cmpneqi, s32_0ImmPred>;
def : T_Q_RI_pat<C4_cmpltei, int_hexagon_C4_cmpltei, s32_0ImmPred>;
def : T_Q_RI_pat<C4_cmplteui, int_hexagon_C4_cmplteui, u32_0ImmPred>;

// Compare To General Register.
def: T_Q_RR_pat<C4_cmpneq,  int_hexagon_C4_cmpneq>;
def: T_Q_RR_pat<C4_cmplte,  int_hexagon_C4_cmplte>;
def: T_Q_RR_pat<C4_cmplteu, int_hexagon_C4_cmplteu>;

def: T_RR_pat<A4_rcmpeq,  int_hexagon_A4_rcmpeq>;
def: T_RR_pat<A4_rcmpneq, int_hexagon_A4_rcmpneq>;

def: T_RI_pat<A4_rcmpeqi,  int_hexagon_A4_rcmpeqi>;
def: T_RI_pat<A4_rcmpneqi, int_hexagon_A4_rcmpneqi>;

//*******************************************************************
//           CR
//*******************************************************************

// CR / Logical Operations On Predicates.
def: T_Q_QQQ_pat<C4_and_and,  int_hexagon_C4_and_and>;
def: T_Q_QQQ_pat<C4_and_andn, int_hexagon_C4_and_andn>;
def: T_Q_QQQ_pat<C4_and_or,   int_hexagon_C4_and_or>;
def: T_Q_QQQ_pat<C4_and_orn,  int_hexagon_C4_and_orn>;
def: T_Q_QQQ_pat<C4_or_and,   int_hexagon_C4_or_and>;
def: T_Q_QQQ_pat<C4_or_andn,  int_hexagon_C4_or_andn>;
def: T_Q_QQQ_pat<C4_or_or,    int_hexagon_C4_or_or>;
def: T_Q_QQQ_pat<C4_or_orn,   int_hexagon_C4_or_orn>;

//*******************************************************************
//           XTYPE/ALU
//*******************************************************************

// Add And Accumulate.

def : T_RRI_pat <S4_addaddi, int_hexagon_S4_addaddi>;
def : T_RIR_pat <S4_subaddi, int_hexagon_S4_subaddi>;


// XTYPE / ALU / Logical-logical Words.
def : T_RRR_pat <M4_or_xor,   int_hexagon_M4_or_xor>;
def : T_RRR_pat <M4_and_xor,  int_hexagon_M4_and_xor>;
def : T_RRR_pat <M4_or_and,   int_hexagon_M4_or_and>;
def : T_RRR_pat <M4_and_and,  int_hexagon_M4_and_and>;
def : T_RRR_pat <M4_xor_and,  int_hexagon_M4_xor_and>;
def : T_RRR_pat <M4_or_or,    int_hexagon_M4_or_or>;
def : T_RRR_pat <M4_and_or,   int_hexagon_M4_and_or>;
def : T_RRR_pat <M4_xor_or,   int_hexagon_M4_xor_or>;
def : T_RRR_pat <M4_or_andn,  int_hexagon_M4_or_andn>;
def : T_RRR_pat <M4_and_andn, int_hexagon_M4_and_andn>;
def : T_RRR_pat <M4_xor_andn, int_hexagon_M4_xor_andn>;

def : T_RRI_pat <S4_or_andi, int_hexagon_S4_or_andi>;
def : T_RRI_pat <S4_or_andix,  int_hexagon_S4_or_andix>;
def : T_RRI_pat <S4_or_ori, int_hexagon_S4_or_ori>;

// Modulo wrap.
def : T_RR_pat <A4_modwrapu, int_hexagon_A4_modwrapu>;

// Arithmetic/Convergent round
// Rd=[cround|round](Rs,Rt)[:sat]
// Rd=[cround|round](Rs,#u5)[:sat]
def : T_RI_pat <A4_cround_ri, int_hexagon_A4_cround_ri>;
def : T_RR_pat <A4_cround_rr, int_hexagon_A4_cround_rr>;

def : T_RI_pat <A4_round_ri, int_hexagon_A4_round_ri>;
def : T_RR_pat <A4_round_rr, int_hexagon_A4_round_rr>;

def : T_RI_pat <A4_round_ri_sat, int_hexagon_A4_round_ri_sat>;
def : T_RR_pat <A4_round_rr_sat, int_hexagon_A4_round_rr_sat>;

def : T_P_pat <A2_roundsat, int_hexagon_A2_roundsat>;