/*
 *  Copyright(c) 2019-2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, see <http://www.gnu.org/licenses/>.
 */

/*
 * Floating-Point Instructions
 */

/*************************************/
/* Scalar FP                         */
/*************************************/
Q6INSN(F2_sfadd,"Rd32=sfadd(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Add",
{ RdV=fUNFLOAT(fFLOAT(RsV)+fFLOAT(RtV));})

Q6INSN(F2_sfsub,"Rd32=sfsub(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Subtract",
{ RdV=fUNFLOAT(fFLOAT(RsV)-fFLOAT(RtV));})

Q6INSN(F2_sfmpy,"Rd32=sfmpy(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Multiply",
{ RdV=fUNFLOAT(fSFMPY(fFLOAT(RsV),fFLOAT(RtV)));})

Q6INSN(F2_sffma,"Rx32+=sfmpy(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Fused Multiply Add",
{ RxV=fUNFLOAT(fFMAF(fFLOAT(RsV),fFLOAT(RtV),fFLOAT(RxV)));})

Q6INSN(F2_sffma_sc,"Rx32+=sfmpy(Rs32,Rt32,Pu4):scale",ATTRIBS(),
"Floating-Point Fused Multiply Add w/ Additional Scaling (2**Pu)",
{
    fHIDE(size4s_t tmp;)
    fCHECKSFNAN3(RxV,RxV,RsV,RtV);
    tmp=fUNFLOAT(fFMAFX(fFLOAT(RsV),fFLOAT(RtV),fFLOAT(RxV),PuV));
    if (!((fFLOAT(RxV) == 0.0) && fISZEROPROD(fFLOAT(RsV),fFLOAT(RtV)))) RxV = tmp;
})

Q6INSN(F2_sffms,"Rx32-=sfmpy(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Fused Multiply Add",
{ RxV=fUNFLOAT(fFMAF(-fFLOAT(RsV),fFLOAT(RtV),fFLOAT(RxV))); })

Q6INSN(F2_sffma_lib,"Rx32+=sfmpy(Rs32,Rt32):lib",ATTRIBS(),
"Floating-Point Fused Multiply Add for Library Routines",
{ fFPSETROUND_NEAREST(); fHIDE(int infinp; int infminusinf; size4s_t tmp;)
  infminusinf = ((isinf(fFLOAT(RxV))) &&
                 (fISINFPROD(fFLOAT(RsV),fFLOAT(RtV))) &&
                 (fGETBIT(31,RsV ^ RxV ^ RtV) != 0));
  infinp = (isinf(fFLOAT(RxV))) || (isinf(fFLOAT(RtV))) || (isinf(fFLOAT(RsV)));
  fCHECKSFNAN3(RxV,RxV,RsV,RtV);
  tmp=fUNFLOAT(fFMAF(fFLOAT(RsV),fFLOAT(RtV),fFLOAT(RxV)));
  if (!((fFLOAT(RxV) == 0.0) && fISZEROPROD(fFLOAT(RsV),fFLOAT(RtV)))) RxV = tmp;
  fFPCANCELFLAGS();
  if (isinf(fFLOAT(RxV)) && !infinp) RxV = RxV - 1;
  if (infminusinf) RxV = 0;
})

Q6INSN(F2_sffms_lib,"Rx32-=sfmpy(Rs32,Rt32):lib",ATTRIBS(),
"Floating-Point Fused Multiply Add for Library Routines",
{ fFPSETROUND_NEAREST(); fHIDE(int infinp; int infminusinf; size4s_t tmp;)
  infminusinf = ((isinf(fFLOAT(RxV))) &&
                 (fISINFPROD(fFLOAT(RsV),fFLOAT(RtV))) &&
                 (fGETBIT(31,RsV ^ RxV ^ RtV) == 0));
  infinp = (isinf(fFLOAT(RxV))) || (isinf(fFLOAT(RtV))) || (isinf(fFLOAT(RsV)));
  fCHECKSFNAN3(RxV,RxV,RsV,RtV);
  tmp=fUNFLOAT(fFMAF(-fFLOAT(RsV),fFLOAT(RtV),fFLOAT(RxV)));
  if (!((fFLOAT(RxV) == 0.0) && fISZEROPROD(fFLOAT(RsV),fFLOAT(RtV)))) RxV = tmp;
  fFPCANCELFLAGS();
  if (isinf(fFLOAT(RxV)) && !infinp) RxV = RxV - 1;
  if (infminusinf) RxV = 0;
})


Q6INSN(F2_sfcmpeq,"Pd4=sfcmp.eq(Rs32,Rt32)",ATTRIBS(),
"Floating Point Compare for Equal",
{PdV=f8BITSOF(fFLOAT(RsV)==fFLOAT(RtV));})

Q6INSN(F2_sfcmpgt,"Pd4=sfcmp.gt(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Compare for Greater Than",
{PdV=f8BITSOF(fFLOAT(RsV)>fFLOAT(RtV));})

/* cmpge is not the same as !cmpgt(swapops) in IEEE */

Q6INSN(F2_sfcmpge,"Pd4=sfcmp.ge(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Compare for Greater Than / Equal To",
{PdV=f8BITSOF(fFLOAT(RsV)>=fFLOAT(RtV));})

/* Everyone seems to have this... */

Q6INSN(F2_sfcmpuo,"Pd4=sfcmp.uo(Rs32,Rt32)",ATTRIBS(),
"Floating-Point Compare for Unordered",
{PdV=f8BITSOF(isunordered(fFLOAT(RsV),fFLOAT(RtV)));})


Q6INSN(F2_sfmax,"Rd32=sfmax(Rs32,Rt32)",ATTRIBS(),
"Maximum of Floating-Point values",
{ RdV = fUNFLOAT(fSF_MAX(fFLOAT(RsV),fFLOAT(RtV))); })

Q6INSN(F2_sfmin,"Rd32=sfmin(Rs32,Rt32)",ATTRIBS(),
"Minimum of Floating-Point values",
{ RdV = fUNFLOAT(fSF_MIN(fFLOAT(RsV),fFLOAT(RtV))); })


Q6INSN(F2_sfclass,"Pd4=sfclass(Rs32,#u5)",ATTRIBS(),
"Classify Floating-Point Value",
{
    fHIDE(int class;)
    PdV = 0;
    class = fpclassify(fFLOAT(RsV));
    /* Is the value zero? */
    if (fGETBIT(0,uiV) && (class == FP_ZERO)) PdV = 0xff;
    if (fGETBIT(1,uiV) && (class == FP_NORMAL)) PdV = 0xff;
    if (fGETBIT(2,uiV) && (class == FP_SUBNORMAL)) PdV = 0xff;
    if (fGETBIT(3,uiV) && (class == FP_INFINITE)) PdV = 0xff;
    if (fGETBIT(4,uiV) && (class == FP_NAN)) PdV = 0xff;
    fFPCANCELFLAGS();
})

/* Range: +/- (1.0 .. 1+(63/64)) * 2**(-6 .. +9) */
/* More immediate bits should probably be used for more precision? */

Q6INSN(F2_sfimm_p,"Rd32=sfmake(#u10):pos",ATTRIBS(),
"Make Floating Point Value",
{
    RdV = (127 - 6) << 23;
    RdV += uiV << 17;
})

Q6INSN(F2_sfimm_n,"Rd32=sfmake(#u10):neg",ATTRIBS(),
"Make Floating Point Value",
{
    RdV = (127 - 6) << 23;
    RdV += (uiV << 17);
    RdV |= (1 << 31);
})


Q6INSN(F2_sfrecipa,"Rd32,Pe4=sfrecipa(Rs32,Rt32)",ATTRIBS(),
"Reciprocal Approximation for Division",
{
    fHIDE(int idx;)
    fHIDE(int adjust;)
    fHIDE(int mant;)
    fHIDE(int exp;)
    if (fSF_RECIP_COMMON(RsV,RtV,RdV,adjust)) {
        PeV = adjust;
        idx = (RtV >> 16) & 0x7f;
        mant = (fSF_RECIP_LOOKUP(idx) << 15) | 1;
        exp = fSF_BIAS() - (fSF_GETEXP(RtV) - fSF_BIAS()) - 1;
        RdV = fMAKESF(fGETBIT(31,RtV),exp,mant);
    }
})

Q6INSN(F2_sffixupn,"Rd32=sffixupn(Rs32,Rt32)",ATTRIBS(),
"Fix Up Numerator",
{
    fHIDE(int adjust;)
    fSF_RECIP_COMMON(RsV,RtV,RdV,adjust);
    RdV = RsV;
})

Q6INSN(F2_sffixupd,"Rd32=sffixupd(Rs32,Rt32)",ATTRIBS(),
"Fix Up Denominator",
{
    fHIDE(int adjust;)
    fSF_RECIP_COMMON(RsV,RtV,RdV,adjust);
    RdV = RtV;
})

Q6INSN(F2_sfinvsqrta,"Rd32,Pe4=sfinvsqrta(Rs32)",ATTRIBS(),
"Reciprocal Square Root Approximation",
{
    fHIDE(int idx;)
    fHIDE(int adjust;)
    fHIDE(int mant;)
    fHIDE(int exp;)
    if (fSF_INVSQRT_COMMON(RsV,RdV,adjust)) {
        PeV = adjust;
        idx = (RsV >> 17) & 0x7f;
        mant = (fSF_INVSQRT_LOOKUP(idx) << 15);
        exp = fSF_BIAS() - ((fSF_GETEXP(RsV) - fSF_BIAS()) >> 1) - 1;
        RdV = fMAKESF(fGETBIT(31,RsV),exp,mant);
    }
})

Q6INSN(F2_sffixupr,"Rd32=sffixupr(Rs32)",ATTRIBS(),
"Fix Up Radicand",
{
    fHIDE(int adjust;)
    fSF_INVSQRT_COMMON(RsV,RdV,adjust);
    RdV = RsV;
})

/*************************************/
/* Scalar DP                         */
/*************************************/
Q6INSN(F2_dfadd,"Rdd32=dfadd(Rss32,Rtt32)",ATTRIBS(),
"Floating-Point Add",
{ RddV=fUNDOUBLE(fDOUBLE(RssV)+fDOUBLE(RttV));})

Q6INSN(F2_dfsub,"Rdd32=dfsub(Rss32,Rtt32)",ATTRIBS(),
"Floating-Point Subtract",
{ RddV=fUNDOUBLE(fDOUBLE(RssV)-fDOUBLE(RttV));})

Q6INSN(F2_dfmax,"Rdd32=dfmax(Rss32,Rtt32)",ATTRIBS(),
"Maximum of Floating-Point values",
{ RddV = fUNDOUBLE(fDF_MAX(fDOUBLE(RssV),fDOUBLE(RttV))); })

Q6INSN(F2_dfmin,"Rdd32=dfmin(Rss32,Rtt32)",ATTRIBS(),
"Minimum of Floating-Point values",
{ RddV = fUNDOUBLE(fDF_MIN(fDOUBLE(RssV),fDOUBLE(RttV))); })

Q6INSN(F2_dfmpyfix,"Rdd32=dfmpyfix(Rss32,Rtt32)",ATTRIBS(),
"Fix Up Multiplicand for Multiplication",
{
    if (fDF_ISDENORM(RssV) && fDF_ISBIG(RttV) && fDF_ISNORMAL(RttV)) RddV = fUNDOUBLE(fDOUBLE(RssV) * 0x1.0p52);
    else if (fDF_ISDENORM(RttV) && fDF_ISBIG(RssV) && fDF_ISNORMAL(RssV)) RddV = fUNDOUBLE(fDOUBLE(RssV) * 0x1.0p-52);
    else RddV = RssV;
})

Q6INSN(F2_dfmpyll,"Rdd32=dfmpyll(Rss32,Rtt32)",ATTRIBS(),
"Multiply low*low and shift off low 32 bits into sticky (in MSB)",
{
    fHIDE(size8u_t prod;)
    prod = fMPY32UU(fGETUWORD(0,RssV),fGETUWORD(0,RttV));
    RddV = (prod >> 32) << 1;
    if (fGETUWORD(0,prod) != 0) fSETBIT(0,RddV,1);
})

Q6INSN(F2_dfmpylh,"Rxx32+=dfmpylh(Rss32,Rtt32)",ATTRIBS(),
"Multiply low*high and accumulate",
{
    RxxV += (fGETUWORD(0,RssV) * (0x00100000 | fZXTN(20,64,fGETUWORD(1,RttV)))) << 1;
})

Q6INSN(F2_dfmpyhh,"Rxx32+=dfmpyhh(Rss32,Rtt32)",ATTRIBS(),
"Multiply high*high and accumulate with L*H value",
{
    RxxV = fUNDOUBLE(fDF_MPY_HH(fDOUBLE(RssV),fDOUBLE(RttV),RxxV));
})


Q6INSN(F2_dfcmpeq,"Pd4=dfcmp.eq(Rss32,Rtt32)",ATTRIBS(),
"Floating Point Compare for Equal",
{PdV=f8BITSOF(fDOUBLE(RssV)==fDOUBLE(RttV));})

Q6INSN(F2_dfcmpgt,"Pd4=dfcmp.gt(Rss32,Rtt32)",ATTRIBS(),
"Floating-Point Compare for Greater Than",
{PdV=f8BITSOF(fDOUBLE(RssV)>fDOUBLE(RttV));})


/* cmpge is not the same as !cmpgt(swapops) in IEEE */

Q6INSN(F2_dfcmpge,"Pd4=dfcmp.ge(Rss32,Rtt32)",ATTRIBS(),
"Floating-Point Compare for Greater Than / Equal To",
{PdV=f8BITSOF(fDOUBLE(RssV)>=fDOUBLE(RttV));})

/* Everyone seems to have this... */

Q6INSN(F2_dfcmpuo,"Pd4=dfcmp.uo(Rss32,Rtt32)",ATTRIBS(),
"Floating-Point Compare for Unordered",
{PdV=f8BITSOF(isunordered(fDOUBLE(RssV),fDOUBLE(RttV)));})


Q6INSN(F2_dfclass,"Pd4=dfclass(Rss32,#u5)",ATTRIBS(),
"Classify Floating-Point Value",
{
    fHIDE(int class;)
    PdV = 0;
    class = fpclassify(fDOUBLE(RssV));
    /* Is the value zero? */
    if (fGETBIT(0,uiV) && (class == FP_ZERO)) PdV = 0xff;
    if (fGETBIT(1,uiV) && (class == FP_NORMAL)) PdV = 0xff;
    if (fGETBIT(2,uiV) && (class == FP_SUBNORMAL)) PdV = 0xff;
    if (fGETBIT(3,uiV) && (class == FP_INFINITE)) PdV = 0xff;
    if (fGETBIT(4,uiV) && (class == FP_NAN)) PdV = 0xff;
    fFPCANCELFLAGS();
})


/* Range: +/- (1.0 .. 1+(63/64)) * 2**(-6 .. +9) */
/* More immediate bits should probably be used for more precision? */

Q6INSN(F2_dfimm_p,"Rdd32=dfmake(#u10):pos",ATTRIBS(),
"Make Floating Point Value",
{
    RddV = (1023ULL - 6) << 52;
    RddV += (fHIDE((size8u_t))uiV) << 46;
})

Q6INSN(F2_dfimm_n,"Rdd32=dfmake(#u10):neg",ATTRIBS(),
"Make Floating Point Value",
{
    RddV = (1023ULL - 6) << 52;
    RddV += (fHIDE((size8u_t))uiV) << 46;
    RddV |= ((1ULL) << 63);
})


/* CONVERSION */

#define CONVERT(TAG,DEST,DESTV,SRC,SRCV,OUTCAST,OUTTYPE,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH) \
    Q6INSN(F2_conv_##TAG##MODETAG,#DEST"=convert_"#TAG"("#SRC")"#MODESYN,ATTRIBS(), \
    "Floating point format conversion", \
    { MODEBEH DESTV = OUTCAST(conv_##INTYPE##_to_##OUTTYPE(INCAST(SRCV))); })

CONVERT(sf2df,Rdd32,RddV,Rs32,RsV,fUNDOUBLE,df,fFLOAT,sf,,,)
CONVERT(df2sf,Rd32,RdV,Rss32,RssV,fUNFLOAT,sf,fDOUBLE,df,,,)

#define ALLINTDST(TAGSTART,SRC,SRCV,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH) \
CONVERT(TAGSTART##uw,Rd32,RdV,SRC,SRCV,fCAST4u,4u,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH) \
CONVERT(TAGSTART##w,Rd32,RdV,SRC,SRCV,fCAST4s,4s,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH) \
CONVERT(TAGSTART##ud,Rdd32,RddV,SRC,SRCV,fCAST8u,8u,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH) \
CONVERT(TAGSTART##d,Rdd32,RddV,SRC,SRCV,fCAST8s,8s,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH)

#define ALLFPDST(TAGSTART,SRC,SRCV,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH) \
CONVERT(TAGSTART##sf,Rd32,RdV,SRC,SRCV,fUNFLOAT,sf,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH) \
CONVERT(TAGSTART##df,Rdd32,RddV,SRC,SRCV,fUNDOUBLE,df,INCAST,INTYPE,MODETAG,MODESYN,MODEBEH)

#define ALLINTSRC(GEN,MODETAG,MODESYN,MODEBEH) \
GEN(uw##2,Rs32,RsV,fCAST4u,4u,MODETAG,MODESYN,MODEBEH) \
GEN(w##2,Rs32,RsV,fCAST4s,4s,MODETAG,MODESYN,MODEBEH) \
GEN(ud##2,Rss32,RssV,fCAST8u,8u,MODETAG,MODESYN,MODEBEH) \
GEN(d##2,Rss32,RssV,fCAST8s,8s,MODETAG,MODESYN,MODEBEH)

#define ALLFPSRC(GEN,MODETAG,MODESYN,MODEBEH) \
GEN(sf##2,Rs32,RsV,fFLOAT,sf,MODETAG,MODESYN,MODEBEH) \
GEN(df##2,Rss32,RssV,fDOUBLE,df,MODETAG,MODESYN,MODEBEH)

ALLINTSRC(ALLFPDST,,,)
ALLFPSRC(ALLINTDST,,,)
ALLFPSRC(ALLINTDST,_chop,:chop,fFPSETROUND_CHOP();)