xref: /gem5/src/arch/alpha/isa/decoder.isa

// -*- mode:c++ -*-

// Copyright (c) 2003-2006 The Regents of The University of Michigan
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met: redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer;
// redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution;
// neither the name of the copyright holders nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// Authors: Steve Reinhardt

////////////////////////////////////////////////////////////////////
//
// The actual decoder specification
//

decode OPCODE default Unknown::unknown() {

    format LoadAddress {
        0x08: lda({{ Ra = Rb + disp; }});
        0x09: ldah({{ Ra = Rb + (disp << 16); }});
    }

    format LoadOrNop {
        0x0a: ldbu({{ Ra_uq = Mem_ub; }});
        0x0c: ldwu({{ Ra_uq = Mem_uw; }});
        0x0b: ldq_u({{ Ra = Mem_uq; }}, ea_code = {{ EA = (Rb + disp) & ~7; }});
        0x23: ldt({{ Fa = Mem_df; }});
        0x2a: ldl_l({{ Ra_sl = Mem_sl; }}, mem_flags = LLSC);
        0x2b: ldq_l({{ Ra_uq = Mem_uq; }}, mem_flags = LLSC);
    }

    format LoadOrPrefetch {
        0x28: ldl({{ Ra_sl = Mem_sl; }});
        0x29: ldq({{ Ra_uq = Mem_uq; }}, pf_flags = EVICT_NEXT);
        // IsFloating flag on lds gets the prefetch to disassemble
        // using f31 instead of r31... funcitonally it's unnecessary
        0x22: lds({{ Fa_uq = s_to_t(Mem_ul); }},
                  pf_flags = PF_EXCLUSIVE, inst_flags = IsFloating);
    }

    format Store {
        0x0e: stb({{ Mem_ub = Ra<7:0>; }});
        0x0d: stw({{ Mem_uw = Ra<15:0>; }});
        0x2c: stl({{ Mem_ul = Ra<31:0>; }});
        0x2d: stq({{ Mem_uq = Ra_uq; }});
        0x0f: stq_u({{ Mem_uq = Ra_uq; }}, {{ EA = (Rb + disp) & ~7; }});
        0x26: sts({{ Mem_ul = t_to_s(Fa_uq); }});
        0x27: stt({{ Mem_df = Fa; }});
    }

    format StoreCond {
        0x2e: stl_c({{ Mem_ul = Ra<31:0>; }},
                    {{
                        uint64_t tmp = write_result;
                        // see stq_c
                        Ra = (tmp == 0 || tmp == 1) ? tmp : Ra;
                        if (tmp == 1) {
                            xc->setStCondFailures(0);
                        }
                    }}, mem_flags = LLSC, inst_flags = IsStoreConditional);
        0x2f: stq_c({{ Mem_uq = Ra; }},
                    {{
                        uint64_t tmp = write_result;
                        // If the write operation returns 0 or 1, then
                        // this was a conventional store conditional,
                        // and the value indicates the success/failure
                        // of the operation.  If another value is
                        // returned, then this was a Turbolaser
                        // mailbox access, and we don't update the
                        // result register at all.
                        Ra = (tmp == 0 || tmp == 1) ? tmp : Ra;
                        if (tmp == 1) {
                            // clear failure counter... this is
                            // non-architectural and for debugging
                            // only.
                            xc->setStCondFailures(0);
                        }
                    }}, mem_flags = LLSC, inst_flags = IsStoreConditional);
    }

    format IntegerOperate {

        0x10: decode INTFUNC {  // integer arithmetic operations

            0x00: addl({{ Rc_sl = Ra_sl + Rb_or_imm_sl; }});
            0x40: addlv({{
                int32_t tmp  = Ra_sl + Rb_or_imm_sl;
                // signed overflow occurs when operands have same sign
                // and sign of result does not match.
                if (Ra_sl<31:> == Rb_or_imm_sl<31:> && tmp<31:> != Ra_sl<31:>)
                    fault = new IntegerOverflowFault;
                Rc_sl = tmp;
            }});
            0x02: s4addl({{ Rc_sl = (Ra_sl << 2) + Rb_or_imm_sl; }});
            0x12: s8addl({{ Rc_sl = (Ra_sl << 3) + Rb_or_imm_sl; }});

            0x20: addq({{ Rc = Ra + Rb_or_imm; }});
            0x60: addqv({{
                uint64_t tmp = Ra + Rb_or_imm;
                // signed overflow occurs when operands have same sign
                // and sign of result does not match.
                if (Ra<63:> == Rb_or_imm<63:> && tmp<63:> != Ra<63:>)
                    fault = new IntegerOverflowFault;
                Rc = tmp;
            }});
            0x22: s4addq({{ Rc = (Ra << 2) + Rb_or_imm; }});
            0x32: s8addq({{ Rc = (Ra << 3) + Rb_or_imm; }});

            0x09: subl({{ Rc_sl = Ra_sl - Rb_or_imm_sl; }});
            0x49: sublv({{
                int32_t tmp  = Ra_sl - Rb_or_imm_sl;
                // signed overflow detection is same as for add,
                // except we need to look at the *complemented*
                // sign bit of the subtrahend (Rb), i.e., if the initial
                // signs are the *same* then no overflow can occur
                if (Ra_sl<31:> != Rb_or_imm_sl<31:> && tmp<31:> != Ra_sl<31:>)
                    fault = new IntegerOverflowFault;
                Rc_sl = tmp;
            }});
            0x0b: s4subl({{ Rc_sl = (Ra_sl << 2) - Rb_or_imm_sl; }});
            0x1b: s8subl({{ Rc_sl = (Ra_sl << 3) - Rb_or_imm_sl; }});

            0x29: subq({{ Rc = Ra - Rb_or_imm; }});
            0x69: subqv({{
                uint64_t tmp  = Ra - Rb_or_imm;
                // signed overflow detection is same as for add,
                // except we need to look at the *complemented*
                // sign bit of the subtrahend (Rb), i.e., if the initial
                // signs are the *same* then no overflow can occur
                if (Ra<63:> != Rb_or_imm<63:> && tmp<63:> != Ra<63:>)
                    fault = new IntegerOverflowFault;
                Rc = tmp;
            }});
            0x2b: s4subq({{ Rc = (Ra << 2) - Rb_or_imm; }});
            0x3b: s8subq({{ Rc = (Ra << 3) - Rb_or_imm; }});

            0x2d: cmpeq({{ Rc = (Ra == Rb_or_imm); }});
            0x6d: cmple({{ Rc = (Ra_sq <= Rb_or_imm_sq); }});
            0x4d: cmplt({{ Rc = (Ra_sq <  Rb_or_imm_sq); }});
            0x3d: cmpule({{ Rc = (Ra_uq <= Rb_or_imm_uq); }});
            0x1d: cmpult({{ Rc = (Ra_uq <  Rb_or_imm_uq); }});

            0x0f: cmpbge({{
                int hi = 7;
                int lo = 0;
                uint64_t tmp = 0;
                for (int i = 0; i < 8; ++i) {
                    tmp |= (Ra_uq<hi:lo> >= Rb_or_imm_uq<hi:lo>) << i;
                    hi += 8;
                    lo += 8;
                }
                Rc = tmp;
            }});
        }

        0x11: decode INTFUNC {  // integer logical operations

            0x00: and({{ Rc = Ra & Rb_or_imm; }});
            0x08: bic({{ Rc = Ra & ~Rb_or_imm; }});
            0x20: bis({{ Rc = Ra | Rb_or_imm; }});
            0x28: ornot({{ Rc = Ra | ~Rb_or_imm; }});
            0x40: xor({{ Rc = Ra ^ Rb_or_imm; }});
            0x48: eqv({{ Rc = Ra ^ ~Rb_or_imm; }});

            // conditional moves
            0x14: cmovlbs({{ Rc = ((Ra & 1) == 1) ? Rb_or_imm : Rc; }});
            0x16: cmovlbc({{ Rc = ((Ra & 1) == 0) ? Rb_or_imm : Rc; }});
            0x24: cmoveq({{ Rc = (Ra == 0) ? Rb_or_imm : Rc; }});
            0x26: cmovne({{ Rc = (Ra != 0) ? Rb_or_imm : Rc; }});
            0x44: cmovlt({{ Rc = (Ra_sq <  0) ? Rb_or_imm : Rc; }});
            0x46: cmovge({{ Rc = (Ra_sq >= 0) ? Rb_or_imm : Rc; }});
            0x64: cmovle({{ Rc = (Ra_sq <= 0) ? Rb_or_imm : Rc; }});
            0x66: cmovgt({{ Rc = (Ra_sq >  0) ? Rb_or_imm : Rc; }});

            // For AMASK, RA must be R31.
            0x61: decode RA {
                31: amask({{ Rc = Rb_or_imm & ~ULL(0x17); }});
            }

            // For IMPLVER, RA must be R31 and the B operand
            // must be the immediate value 1.
            0x6c: decode RA {
                31: decode IMM {
                    1: decode INTIMM {
                        // return EV5 for FullSystem and EV6 otherwise
                        1: implver({{ Rc = FullSystem ? 1 : 2 }});
                    }
                }
            }

            // The mysterious 11.25...
            0x25: WarnUnimpl::eleven25();
        }

        0x12: decode INTFUNC {
            0x39: sll({{ Rc = Ra << Rb_or_imm<5:0>; }});
            0x34: srl({{ Rc = Ra_uq >> Rb_or_imm<5:0>; }});
            0x3c: sra({{ Rc = Ra_sq >> Rb_or_imm<5:0>; }});

            0x02: mskbl({{ Rc = Ra & ~(mask( 8) << (Rb_or_imm<2:0> * 8)); }});
            0x12: mskwl({{ Rc = Ra & ~(mask(16) << (Rb_or_imm<2:0> * 8)); }});
            0x22: mskll({{ Rc = Ra & ~(mask(32) << (Rb_or_imm<2:0> * 8)); }});
            0x32: mskql({{ Rc = Ra & ~(mask(64) << (Rb_or_imm<2:0> * 8)); }});

            0x52: mskwh({{
                int bv = Rb_or_imm<2:0>;
                Rc =  bv ? (Ra & ~(mask(16) >> (64 - 8 * bv))) : Ra;
            }});
            0x62: msklh({{
                int bv = Rb_or_imm<2:0>;
                Rc =  bv ? (Ra & ~(mask(32) >> (64 - 8 * bv))) : Ra;
            }});
            0x72: mskqh({{
                int bv = Rb_or_imm<2:0>;
                Rc =  bv ? (Ra & ~(mask(64) >> (64 - 8 * bv))) : Ra;
            }});

            0x06: extbl({{ Rc = (Ra_uq >> (Rb_or_imm<2:0> * 8))< 7:0>; }});
            0x16: extwl({{ Rc = (Ra_uq >> (Rb_or_imm<2:0> * 8))<15:0>; }});
            0x26: extll({{ Rc = (Ra_uq >> (Rb_or_imm<2:0> * 8))<31:0>; }});
            0x36: extql({{ Rc = (Ra_uq >> (Rb_or_imm<2:0> * 8)); }});

            0x5a: extwh({{
                Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<15:0>; }});
            0x6a: extlh({{
                Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<31:0>; }});
            0x7a: extqh({{
                Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>); }});

            0x0b: insbl({{ Rc = Ra< 7:0> << (Rb_or_imm<2:0> * 8); }});
            0x1b: inswl({{ Rc = Ra<15:0> << (Rb_or_imm<2:0> * 8); }});
            0x2b: insll({{ Rc = Ra<31:0> << (Rb_or_imm<2:0> * 8); }});
            0x3b: insql({{ Rc = Ra       << (Rb_or_imm<2:0> * 8); }});

            0x57: inswh({{
                int bv = Rb_or_imm<2:0>;
                Rc = bv ? (Ra_uq<15:0> >> (64 - 8 * bv)) : 0;
            }});
            0x67: inslh({{
                int bv = Rb_or_imm<2:0>;
                Rc = bv ? (Ra_uq<31:0> >> (64 - 8 * bv)) : 0;
            }});
            0x77: insqh({{
                int bv = Rb_or_imm<2:0>;
                Rc = bv ? (Ra_uq       >> (64 - 8 * bv)) : 0;
            }});

            0x30: zap({{
                uint64_t zapmask = 0;
                for (int i = 0; i < 8; ++i) {
                    if (Rb_or_imm<i:>)
                        zapmask |= (mask(8) << (i * 8));
                }
                Rc = Ra & ~zapmask;
            }});
            0x31: zapnot({{
                uint64_t zapmask = 0;
                for (int i = 0; i < 8; ++i) {
                    if (!Rb_or_imm<i:>)
                        zapmask |= (mask(8) << (i * 8));
                }
                Rc = Ra & ~zapmask;
            }});
        }

        0x13: decode INTFUNC {  // integer multiplies
            0x00: mull({{ Rc_sl = Ra_sl * Rb_or_imm_sl; }}, IntMultOp);
            0x20: mulq({{ Rc    = Ra    * Rb_or_imm;    }}, IntMultOp);
            0x30: umulh({{
                uint64_t hi, lo;
                mul128(Ra, Rb_or_imm, hi, lo);
                Rc = hi;
            }}, IntMultOp);
            0x40: mullv({{
                // 32-bit multiply with trap on overflow
                int64_t Rax = Ra_sl;    // sign extended version of Ra_sl
                int64_t Rbx = Rb_or_imm_sl;
                int64_t tmp = Rax * Rbx;
                // To avoid overflow, all the upper 32 bits must match
                // the sign bit of the lower 32.  We code this as
                // checking the upper 33 bits for all 0s or all 1s.
                uint64_t sign_bits = tmp<63:31>;
                if (sign_bits != 0 && sign_bits != mask(33))
                    fault = new IntegerOverflowFault;
                Rc_sl = tmp<31:0>;
            }}, IntMultOp);
            0x60: mulqv({{
                // 64-bit multiply with trap on overflow
                uint64_t hi, lo;
                mul128(Ra, Rb_or_imm, hi, lo);
                // all the upper 64 bits must match the sign bit of
                // the lower 64
                if (!((hi == 0 && lo<63:> == 0) ||
                      (hi == mask(64) && lo<63:> == 1)))
                    fault = new IntegerOverflowFault;
                Rc = lo;
            }}, IntMultOp);
        }

        0x1c: decode INTFUNC {
            0x00: decode RA { 31: sextb({{ Rc_sb = Rb_or_imm< 7:0>; }}); }
            0x01: decode RA { 31: sextw({{ Rc_sw = Rb_or_imm<15:0>; }}); }

            0x30: ctpop({{
                             uint64_t count = 0;
                             for (int i = 0; Rb<63:i>; ++i) {
                                 if (Rb<i:i> == 0x1)
                                     ++count;
                             }
                             Rc = count;
                           }}, IntAluOp);

            0x31: perr({{
                             uint64_t temp = 0;
                             int hi = 7;
                             int lo = 0;
                             for (int i = 0; i < 8; ++i) {
                                 uint8_t ra_ub = Ra_uq<hi:lo>;
                                 uint8_t rb_ub = Rb_uq<hi:lo>;
                                 temp += (ra_ub >= rb_ub) ? 
                                         (ra_ub - rb_ub) : (rb_ub - ra_ub);
                                 hi += 8;
                                 lo += 8;
                             }
                             Rc = temp;
                           }});

            0x32: ctlz({{
                             uint64_t count = 0;
                             uint64_t temp = Rb;
                             if (temp<63:32>) temp >>= 32; else count += 32;
                             if (temp<31:16>) temp >>= 16; else count += 16;
                             if (temp<15:8>) temp >>= 8; else count += 8;
                             if (temp<7:4>) temp >>= 4; else count += 4;
                             if (temp<3:2>) temp >>= 2; else count += 2;
                             if (temp<1:1>) temp >>= 1; else count += 1;
                             if ((temp<0:0>) != 0x1) count += 1;
                             Rc = count;
                           }}, IntAluOp);

            0x33: cttz({{
                             uint64_t count = 0;
                             uint64_t temp = Rb;
                             if (!(temp<31:0>)) { temp >>= 32; count += 32; }
                             if (!(temp<15:0>)) { temp >>= 16; count += 16; }
                             if (!(temp<7:0>)) { temp >>= 8; count += 8; }
                             if (!(temp<3:0>)) { temp >>= 4; count += 4; }
                             if (!(temp<1:0>)) { temp >>= 2; count += 2; }
                             if (!(temp<0:0> & ULL(0x1))) { 
                                 temp >>= 1; count += 1; 
                             }
                             if (!(temp<0:0> & ULL(0x1))) count += 1;
                             Rc = count;
                           }}, IntAluOp);


            0x34: unpkbw({{ 
                             Rc = (Rb_uq<7:0>
                                   | (Rb_uq<15:8> << 16)
                                   | (Rb_uq<23:16> << 32)
                                   | (Rb_uq<31:24> << 48));
                           }}, IntAluOp);

            0x35: unpkbl({{
                             Rc = (Rb_uq<7:0> | (Rb_uq<15:8> << 32));
                           }}, IntAluOp);

            0x36: pkwb({{
                             Rc = (Rb_uq<7:0>
                                   | (Rb_uq<23:16> << 8)
                                   | (Rb_uq<39:32> << 16)
                                   | (Rb_uq<55:48> << 24));
                           }}, IntAluOp);

            0x37: pklb({{
                             Rc = (Rb_uq<7:0> | (Rb_uq<39:32> << 8));
                           }}, IntAluOp);

            0x38: minsb8({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 56;
                             for (int i = 7; i >= 0; --i) {
                                 int8_t ra_sb = Ra_uq<hi:lo>;
                                 int8_t rb_sb = Rb_uq<hi:lo>;
                                 temp = ((temp << 8) 
                                         | ((ra_sb < rb_sb) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 8;
                                 lo -= 8;
                             }
                             Rc = temp;
                          }});

            0x39: minsw4({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 48;
                             for (int i = 3; i >= 0; --i) {
                                 int16_t ra_sw = Ra_uq<hi:lo>;
                                 int16_t rb_sw = Rb_uq<hi:lo>;
                                 temp = ((temp << 16) 
                                         | ((ra_sw < rb_sw) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 16;
                                 lo -= 16;
                             }
                             Rc = temp;
                          }});

            0x3a: minub8({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 56;
                             for (int i = 7; i >= 0; --i) {
                                 uint8_t ra_ub = Ra_uq<hi:lo>;
                                 uint8_t rb_ub = Rb_uq<hi:lo>;
                                 temp = ((temp << 8) 
                                         | ((ra_ub < rb_ub) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 8;
                                 lo -= 8;
                             }
                             Rc = temp;
                          }});

            0x3b: minuw4({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 48;
                             for (int i = 3; i >= 0; --i) {
                                 uint16_t ra_sw = Ra_uq<hi:lo>;
                                 uint16_t rb_sw = Rb_uq<hi:lo>;
                                 temp = ((temp << 16) 
                                         | ((ra_sw < rb_sw) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 16;
                                 lo -= 16;
                             }
                             Rc = temp;
                          }});

            0x3c: maxub8({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 56;
                             for (int i = 7; i >= 0; --i) {
                                 uint8_t ra_ub = Ra_uq<hi:lo>;
                                 uint8_t rb_ub = Rb_uq<hi:lo>;
                                 temp = ((temp << 8) 
                                         | ((ra_ub > rb_ub) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 8;
                                 lo -= 8;
                             }
                             Rc = temp;
                          }});

            0x3d: maxuw4({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 48;
                             for (int i = 3; i >= 0; --i) {
                                 uint16_t ra_uw = Ra_uq<hi:lo>;
                                 uint16_t rb_uw = Rb_uq<hi:lo>;
                                 temp = ((temp << 16) 
                                         | ((ra_uw > rb_uw) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 16;
                                 lo -= 16;
                             }
                             Rc = temp;
                          }});

            0x3e: maxsb8({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 56;
                             for (int i = 7; i >= 0; --i) {
                                 int8_t ra_sb = Ra_uq<hi:lo>;
                                 int8_t rb_sb = Rb_uq<hi:lo>;
                                 temp = ((temp << 8) 
                                         | ((ra_sb > rb_sb) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 8;
                                 lo -= 8;
                             }
                             Rc = temp;
                          }});

            0x3f: maxsw4({{
                             uint64_t temp = 0;
                             int hi = 63;
                             int lo = 48;
                             for (int i = 3; i >= 0; --i) {
                                 int16_t ra_sw = Ra_uq<hi:lo>;
                                 int16_t rb_sw = Rb_uq<hi:lo>;
                                 temp = ((temp << 16) 
                                         | ((ra_sw > rb_sw) ? Ra_uq<hi:lo>
                                                          : Rb_uq<hi:lo>));
                                 hi -= 16;
                                 lo -= 16;
                             }
                             Rc = temp;
                          }});

            format BasicOperateWithNopCheck {
                0x70: decode RB {
                    31: ftoit({{ Rc = Fa_uq; }}, FloatCvtOp);
                }
                0x78: decode RB {
                    31: ftois({{ Rc_sl = t_to_s(Fa_uq); }},
                              FloatCvtOp);
                }
            }
        }
    }

    // Conditional branches.
    format CondBranch {
        0x39: beq({{ cond = (Ra == 0); }});
        0x3d: bne({{ cond = (Ra != 0); }});
        0x3e: bge({{ cond = (Ra_sq >= 0); }});
        0x3f: bgt({{ cond = (Ra_sq >  0); }});
        0x3b: ble({{ cond = (Ra_sq <= 0); }});
        0x3a: blt({{ cond = (Ra_sq < 0); }});
        0x38: blbc({{ cond = ((Ra & 1) == 0); }});
        0x3c: blbs({{ cond = ((Ra & 1) == 1); }});

        0x31: fbeq({{ cond = (Fa == 0); }});
        0x35: fbne({{ cond = (Fa != 0); }});
        0x36: fbge({{ cond = (Fa >= 0); }});
        0x37: fbgt({{ cond = (Fa >  0); }});
        0x33: fble({{ cond = (Fa <= 0); }});
        0x32: fblt({{ cond = (Fa < 0); }});
    }

    // unconditional branches
    format UncondBranch {
        0x30: br();
        0x34: bsr(IsCall);
    }

    // indirect branches
    0x1a: decode JMPFUNC {
        format Jump {
            0: jmp();
            1: jsr(IsCall);
            2: ret(IsReturn);
            3: jsr_coroutine(IsCall, IsReturn);
        }
    }

    // Square root and integer-to-FP moves
    0x14: decode FP_SHORTFUNC {
        // Integer to FP register moves must have RB == 31
        0x4: decode RB {
            31: decode FP_FULLFUNC {
                format BasicOperateWithNopCheck {
                    0x004: itofs({{ Fc_uq = s_to_t(Ra_ul); }}, FloatCvtOp);
                    0x024: itoft({{ Fc_uq = Ra_uq; }}, FloatCvtOp);
                    0x014: FailUnimpl::itoff(); // VAX-format conversion
                }
            }
        }

        // Square root instructions must have FA == 31
        0xb: decode FA {
            31: decode FP_TYPEFUNC {
                format FloatingPointOperate {
#if SS_COMPATIBLE_FP
                    0x0b: sqrts({{
                        if (Fb < 0.0)
                            fault = new ArithmeticFault;
                        Fc = sqrt(Fb);
                    }}, FloatSqrtOp);
#else
                    0x0b: sqrts({{
                        if (Fb_sf < 0.0)
                            fault = new ArithmeticFault;
                        Fc_sf = sqrt(Fb_sf);
                    }}, FloatSqrtOp);
#endif
                    0x2b: sqrtt({{
                        if (Fb < 0.0)
                            fault = new ArithmeticFault;
                        Fc = sqrt(Fb);
                    }}, FloatSqrtOp);
                }
            }
        }

        // VAX-format sqrtf and sqrtg are not implemented
        0xa: FailUnimpl::sqrtfg();
    }

    // IEEE floating point
    0x16: decode FP_SHORTFUNC_TOP2 {
        // The top two bits of the short function code break this
        // space into four groups: binary ops, compares, reserved, and
        // conversions.  See Table 4-12 of AHB.  There are different
        // special cases in these different groups, so we decode on
        // these top two bits first just to select a decode strategy.
        // Most of these instructions may have various trapping and
        // rounding mode flags set; these are decoded in the
        // FloatingPointDecode template used by the
        // FloatingPointOperate format.

        // add/sub/mul/div: just decode on the short function code
        // and source type.  All valid trapping and rounding modes apply.
        0: decode FP_TRAPMODE {
            // check for valid trapping modes here
            0,1,5,7: decode FP_TYPEFUNC {
                   format FloatingPointOperate {
#if SS_COMPATIBLE_FP
                       0x00: adds({{ Fc = Fa + Fb; }});
                       0x01: subs({{ Fc = Fa - Fb; }});
                       0x02: muls({{ Fc = Fa * Fb; }}, FloatMultOp);
                       0x03: divs({{ Fc = Fa / Fb; }}, FloatDivOp);
#else
                       0x00: adds({{ Fc_sf = Fa_sf + Fb_sf; }});
                       0x01: subs({{ Fc_sf = Fa_sf - Fb_sf; }});
                       0x02: muls({{ Fc_sf = Fa_sf * Fb_sf; }}, FloatMultOp);
                       0x03: divs({{ Fc_sf = Fa_sf / Fb_sf; }}, FloatDivOp);
#endif

                       0x20: addt({{ Fc = Fa + Fb; }});
                       0x21: subt({{ Fc = Fa - Fb; }});
                       0x22: mult({{ Fc = Fa * Fb; }}, FloatMultOp);
                       0x23: divt({{ Fc = Fa / Fb; }}, FloatDivOp);
                   }
             }
        }

        // Floating-point compare instructions must have the default
        // rounding mode, and may use the default trapping mode or
        // /SU.  Both trapping modes are treated the same by M5; the
        // only difference on the real hardware (as far a I can tell)
        // is that without /SU you'd get an imprecise trap if you
        // tried to compare a NaN with something else (instead of an
        // "unordered" result).
        1: decode FP_FULLFUNC {
            format BasicOperateWithNopCheck {
                0x0a5, 0x5a5: cmpteq({{ Fc = (Fa == Fb) ? 2.0 : 0.0; }},
                                     FloatCmpOp);
                0x0a7, 0x5a7: cmptle({{ Fc = (Fa <= Fb) ? 2.0 : 0.0; }},
                                     FloatCmpOp);
                0x0a6, 0x5a6: cmptlt({{ Fc = (Fa <  Fb) ? 2.0 : 0.0; }},
                                     FloatCmpOp);
                0x0a4, 0x5a4: cmptun({{ // unordered
                    Fc = (!(Fa < Fb) && !(Fa == Fb) && !(Fa > Fb)) ? 2.0 : 0.0;
                }}, FloatCmpOp);
            }
        }

        // The FP-to-integer and integer-to-FP conversion insts
        // require that FA be 31.
        3: decode FA {
            31: decode FP_TYPEFUNC {
                format FloatingPointOperate {
                    0x2f: decode FP_ROUNDMODE {
                        format FPFixedRounding {
                            // "chopped" i.e. round toward zero
                            0: cvttq({{ Fc_sq = (int64_t)trunc(Fb); }},
                                     Chopped);
                            // round to minus infinity
                            1: cvttq({{ Fc_sq = (int64_t)floor(Fb); }},
                                     MinusInfinity);
                        }
                      default: cvttq({{ Fc_sq = (int64_t)nearbyint(Fb); }});
                    }

                    // The cvtts opcode is overloaded to be cvtst if the trap
                    // mode is 2 or 6 (which are not valid otherwise)
                    0x2c: decode FP_FULLFUNC {
                        format BasicOperateWithNopCheck {
                            // trap on denorm version "cvtst/s" is
                            // simulated same as cvtst
                            0x2ac, 0x6ac: cvtst({{ Fc = Fb_sf; }});
                        }
                      default: cvtts({{ Fc_sf = Fb; }});
                    }

                    // The trapping mode for integer-to-FP conversions
                    // must be /SUI or nothing; /U and /SU are not
                    // allowed.  The full set of rounding modes are
                    // supported though.
                    0x3c: decode FP_TRAPMODE {
                        0,7: cvtqs({{ Fc_sf = Fb_sq; }});
                    }
                    0x3e: decode FP_TRAPMODE {
                        0,7: cvtqt({{ Fc    = Fb_sq; }});
                    }
                }
            }
        }
    }

    // misc FP operate
    0x17: decode FP_FULLFUNC {
        format BasicOperateWithNopCheck {
            0x010: cvtlq({{
                Fc_sl = (Fb_uq<63:62> << 30) | Fb_uq<58:29>;
            }});
            0x030: cvtql({{
                Fc_uq = (Fb_uq<31:30> << 62) | (Fb_uq<29:0> << 29);
            }});

            // We treat the precise & imprecise trapping versions of
            // cvtql identically.
            0x130, 0x530: cvtqlv({{
                // To avoid overflow, all the upper 32 bits must match
                // the sign bit of the lower 32.  We code this as
                // checking the upper 33 bits for all 0s or all 1s.
                uint64_t sign_bits = Fb_uq<63:31>;
                if (sign_bits != 0 && sign_bits != mask(33))
                    fault = new IntegerOverflowFault;
                Fc_uq = (Fb_uq<31:30> << 62) | (Fb_uq<29:0> << 29);
            }});

            0x020: cpys({{  // copy sign
                Fc_uq = (Fa_uq<63:> << 63) | Fb_uq<62:0>;
            }});
            0x021: cpysn({{ // copy sign negated
                Fc_uq = (~Fa_uq<63:> << 63) | Fb_uq<62:0>;
            }});
            0x022: cpyse({{ // copy sign and exponent
                Fc_uq = (Fa_uq<63:52> << 52) | Fb_uq<51:0>;
            }});

            0x02a: fcmoveq({{ Fc = (Fa == 0) ? Fb : Fc; }});
            0x02b: fcmovne({{ Fc = (Fa != 0) ? Fb : Fc; }});
            0x02c: fcmovlt({{ Fc = (Fa <  0) ? Fb : Fc; }});
            0x02d: fcmovge({{ Fc = (Fa >= 0) ? Fb : Fc; }});
            0x02e: fcmovle({{ Fc = (Fa <= 0) ? Fb : Fc; }});
            0x02f: fcmovgt({{ Fc = (Fa >  0) ? Fb : Fc; }});

            0x024: mt_fpcr({{ FPCR = Fa_uq; }}, IsIprAccess);
            0x025: mf_fpcr({{ Fa_uq = FPCR; }}, IsIprAccess);
        }
    }

    // miscellaneous mem-format ops
    0x18: decode MEMFUNC {
        format WarnUnimpl {
            0x8000: fetch();
            0xa000: fetch_m();
            0xe800: ecb();
        }

        format MiscPrefetch {
            0xf800: wh64({{ EA = Rb & ~ULL(63); }},
                         {{ ; }},
                         mem_flags = PREFETCH);
        }

        format BasicOperate {
            0xc000: rpcc({{
                /* Rb is a fake dependency so here is a fun way to get
                 * the parser to understand that.
                 */
                uint64_t unused_var M5_VAR_USED = Rb;
                Ra = FullSystem ? xc->readMiscReg(IPR_CC) : curTick();
            }}, IsUnverifiable);

            // All of the barrier instructions below do nothing in
            // their execute() methods (hence the empty code blocks).
            // All of their functionality is hard-coded in the
            // pipeline based on the flags IsSerializing,
            // IsMemBarrier, and IsWriteBarrier.  In the current
            // detailed CPU model, the execute() function only gets
            // called at fetch, so there's no way to generate pipeline
            // behavior at any other stage.  Once we go to an
            // exec-in-exec CPU model we should be able to get rid of
            // these flags and implement this behavior via the
            // execute() methods.

            // trapb is just a barrier on integer traps, where excb is
            // a barrier on integer and FP traps.  "EXCB is thus a
            // superset of TRAPB." (Alpha ARM, Sec 4.11.4) We treat
            // them the same though.
            0x0000: trapb({{ }}, IsSerializing, IsSerializeBefore, No_OpClass);
            0x0400: excb({{ }}, IsSerializing, IsSerializeBefore, No_OpClass);
            0x4000: mb({{ }}, IsMemBarrier, MemReadOp);
            0x4400: wmb({{ }}, IsWriteBarrier, MemWriteOp);
        }

        0xe000: decode FullSystem {
            0: FailUnimpl::rc_se();
            default: BasicOperate::rc({{
                Ra = IntrFlag;
                IntrFlag = 0;
            }}, IsNonSpeculative, IsUnverifiable);
        }
        0xf000: decode FullSystem {
            0: FailUnimpl::rs_se();
            default: BasicOperate::rs({{
                Ra = IntrFlag;
                IntrFlag = 1;
            }}, IsNonSpeculative, IsUnverifiable);
        }
    }

#if FULL_SYSTEM
    0x00: CallPal::call_pal({{
        if (!palValid ||
            (palPriv
             && xc->readMiscReg(IPR_ICM) != mode_kernel)) {
            // invalid pal function code, or attempt to do privileged
            // PAL call in non-kernel mode
            fault = new UnimplementedOpcodeFault;
        } else {
            // check to see if simulator wants to do something special
            // on this PAL call (including maybe suppress it)
            bool dopal = xc->simPalCheck(palFunc);

            if (dopal) {
                xc->setMiscReg(IPR_EXC_ADDR, NPC);
                NPC = xc->readMiscReg(IPR_PAL_BASE) + palOffset;
            }
        }
    }}, IsNonSpeculative);
#else
    0x00: decode PALFUNC {
        format EmulatedCallPal {
            0x00: halt ({{
                exitSimLoop("halt instruction encountered");
            }}, IsNonSpeculative);
            0x83: callsys({{
                xc->syscall(R0);
            }}, IsSerializeAfter, IsNonSpeculative, IsSyscall);
            // Read uniq reg into ABI return value register (r0)
            0x9e: rduniq({{ R0 = Runiq; }}, IsIprAccess);
            // Write uniq reg with value from ABI arg register (r16)
            0x9f: wruniq({{ Runiq = R16; }}, IsIprAccess);
        }
    }
#endif

#if FULL_SYSTEM
    0x1b: decode PALMODE {
        0: OpcdecFault::hw_st_quad();
        1: decode HW_LDST_QUAD {
            format HwLoad {
                0: hw_ld({{ EA = (Rb + disp) & ~3; }}, {{ Ra = Mem_ul; }},
                         L, IsSerializing, IsSerializeBefore);
                1: hw_ld({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem_uq; }},
                         Q, IsSerializing, IsSerializeBefore);
            }
        }
    }

    0x1f: decode PALMODE {
        0: OpcdecFault::hw_st_cond();
        format HwStore {
            1: decode HW_LDST_COND {
                0: decode HW_LDST_QUAD {
                    0: hw_st({{ EA = (Rb + disp) & ~3; }},
                {{ Mem_ul = Ra<31:0>; }}, L, IsSerializing, IsSerializeBefore);
                    1: hw_st({{ EA = (Rb + disp) & ~7; }},
                {{ Mem_uq = Ra_uq; }}, Q, IsSerializing, IsSerializeBefore);
                }

                1: FailUnimpl::hw_st_cond();
            }
        }
    }

    0x19: decode PALMODE {
        0: OpcdecFault::hw_mfpr();
        format HwMoveIPR {
            1: hw_mfpr({{
                int miscRegIndex = (ipr_index < MaxInternalProcRegs) ?
                        IprToMiscRegIndex[ipr_index] : -1;
                if(miscRegIndex < 0 || !IprIsReadable(miscRegIndex) ||
                    miscRegIndex >= NumInternalProcRegs)
                        fault = new UnimplementedOpcodeFault;
                else
                    Ra = xc->readMiscReg(miscRegIndex);
            }}, IsIprAccess);
        }
    }

    0x1d: decode PALMODE {
        0: OpcdecFault::hw_mtpr();
        format HwMoveIPR {
            1: hw_mtpr({{
                int miscRegIndex = (ipr_index < MaxInternalProcRegs) ?
                        IprToMiscRegIndex[ipr_index] : -1;
                if(miscRegIndex < 0 || !IprIsWritable(miscRegIndex) ||
                    miscRegIndex >= NumInternalProcRegs)
                        fault = new UnimplementedOpcodeFault;
                else
                    xc->setMiscReg(miscRegIndex, Ra);
                if (traceData) { traceData->setData(Ra); }
            }}, IsIprAccess);
        }
    }

  0x1e: decode PALMODE {
      0: OpcdecFault::hw_rei();
        format BasicOperate {
          1: hw_rei({{ xc->hwrei(); }}, IsSerializing, IsSerializeBefore);
        }
    }

#endif

    format BasicOperate {
        // M5 special opcodes use the reserved 0x01 opcode space
        0x01: decode M5FUNC {
            0x00: arm({{
                PseudoInst::arm(xc->tcBase());
            }}, IsNonSpeculative);
            0x01: quiesce({{
                PseudoInst::quiesce(xc->tcBase());
            }}, IsNonSpeculative, IsQuiesce);
            0x02: quiesceNs({{
                PseudoInst::quiesceNs(xc->tcBase(), R16);
            }}, IsNonSpeculative, IsQuiesce);
            0x03: quiesceCycles({{
                PseudoInst::quiesceCycles(xc->tcBase(), R16);
            }}, IsNonSpeculative, IsQuiesce, IsUnverifiable);
            0x04: quiesceTime({{
                R0 = PseudoInst::quiesceTime(xc->tcBase());
            }}, IsNonSpeculative, IsUnverifiable);
            0x07: rpns({{
                R0 = PseudoInst::rpns(xc->tcBase());
            }}, IsNonSpeculative, IsUnverifiable);
            0x09: wakeCPU({{
                PseudoInst::wakeCPU(xc->tcBase(), R16);
            }}, IsNonSpeculative, IsUnverifiable);
            0x10: deprecated_ivlb({{
                warn_once("Obsolete M5 ivlb instruction encountered.\n");
            }});
            0x11: deprecated_ivle({{
                warn_once("Obsolete M5 ivlb instruction encountered.\n");
            }});
            0x20: deprecated_exit ({{
                warn_once("deprecated M5 exit instruction encountered.\n");
                PseudoInst::m5exit(xc->tcBase(), 0);
            }}, No_OpClass, IsNonSpeculative);
            0x21: m5exit({{
                PseudoInst::m5exit(xc->tcBase(), R16);
            }}, No_OpClass, IsNonSpeculative);
            0x31: loadsymbol({{
                PseudoInst::loadsymbol(xc->tcBase());
            }}, No_OpClass, IsNonSpeculative);
            0x30: initparam({{
                Ra = PseudoInst::initParam(xc->tcBase());
            }});
            0x40: resetstats({{
                PseudoInst::resetstats(xc->tcBase(), R16, R17);
            }}, IsNonSpeculative);
            0x41: dumpstats({{
                PseudoInst::dumpstats(xc->tcBase(), R16, R17);
            }}, IsNonSpeculative);
            0x42: dumpresetstats({{
                PseudoInst::dumpresetstats(xc->tcBase(), R16, R17);
            }}, IsNonSpeculative);
            0x43: m5checkpoint({{
                PseudoInst::m5checkpoint(xc->tcBase(), R16, R17);
            }}, IsNonSpeculative);
            0x50: m5readfile({{
                R0 = PseudoInst::readfile(xc->tcBase(), R16, R17, R18);
            }}, IsNonSpeculative);
            0x51: m5break({{
                PseudoInst::debugbreak(xc->tcBase());
            }}, IsNonSpeculative);
            0x52: m5switchcpu({{
                PseudoInst::switchcpu(xc->tcBase());
            }}, IsNonSpeculative);
            0x53: m5addsymbol({{
                PseudoInst::addsymbol(xc->tcBase(), R16, R17);
            }}, IsNonSpeculative);
            0x54: m5panic({{
                panic("M5 panic instruction called at pc = %#x.", PC);
            }}, IsNonSpeculative);
#define  CPANN(lbl) CPA::cpa()->lbl(xc->tcBase())
            0x55: decode RA {
                0x00: m5a_old({{
                    panic("Deprecated M5 annotate instruction executed "
                          "at pc = %#x\n", PC);
                }}, IsNonSpeculative);
                0x01: m5a_bsm({{
                    CPANN(swSmBegin);
                }}, IsNonSpeculative);
                0x02: m5a_esm({{
                    CPANN(swSmEnd);
                }}, IsNonSpeculative);
                0x03: m5a_begin({{
                    CPANN(swExplictBegin);
                }}, IsNonSpeculative);
                0x04: m5a_end({{
                    CPANN(swEnd);
                }}, IsNonSpeculative);
                0x06: m5a_q({{
                    CPANN(swQ);
                }}, IsNonSpeculative);
                0x07: m5a_dq({{
                    CPANN(swDq);
                }}, IsNonSpeculative);
                0x08: m5a_wf({{
                    CPANN(swWf);
                }}, IsNonSpeculative);
                0x09: m5a_we({{
                    CPANN(swWe);
                }}, IsNonSpeculative);
                0x0C: m5a_sq({{
                    CPANN(swSq);
                }}, IsNonSpeculative);
                0x0D: m5a_aq({{
                    CPANN(swAq);
                }}, IsNonSpeculative);
                0x0E: m5a_pq({{
                    CPANN(swPq);
                }}, IsNonSpeculative);
                0x0F: m5a_l({{
                    CPANN(swLink);
                }}, IsNonSpeculative);
                0x10: m5a_identify({{
                    CPANN(swIdentify);
                }}, IsNonSpeculative);
                0x11: m5a_getid({{
                    R0 = CPANN(swGetId);
                }}, IsNonSpeculative);
                0x13: m5a_scl({{
                    CPANN(swSyscallLink);
                }}, IsNonSpeculative);
                0x14: m5a_rq({{
                    CPANN(swRq);
                }}, IsNonSpeculative);
            } // M5 Annotate Operations
#undef CPANN
            0x56: m5reserved2({{
                warn("M5 reserved opcode ignored");
            }}, IsNonSpeculative);
            0x57: m5reserved3({{
                warn("M5 reserved opcode ignored");
            }}, IsNonSpeculative);
            0x58: m5reserved4({{
                warn("M5 reserved opcode ignored");
            }}, IsNonSpeculative);
            0x59: m5reserved5({{
                warn("M5 reserved opcode ignored");
            }}, IsNonSpeculative);
        }
    }
}