long_double __addxf3(long_double A, long_double B)
{
#if __SIZEOF_LONG_DOUBLE__ == 12
return floatx80_add(A, B);
#else
return float128_add(A, B);
#endif
}
static void fpgen_rm_reg(m68000_base_device *m68k, UINT16 w2)
{
int ea = m68k->ir & 0x3f;
int rm = (w2 >> 14) & 0x1;
int src = (w2 >> 10) & 0x7;
int dst = (w2 >> 7) & 0x7;
int opmode = w2 & 0x7f;
floatx80 source;
// fmovecr #$f, fp0 f200 5c0f
if (rm)
{
switch (src)
{
case 0: // Long-Word Integer
{
INT32 d = READ_EA_32(m68k, ea);
source = int32_to_floatx80(d);
break;
}
case 1: // Single-precision Real
{
UINT32 d = READ_EA_32(m68k, ea);
source = float32_to_floatx80(d);
break;
}
case 2: // Extended-precision Real
{
source = READ_EA_FPE(m68k, ea);
break;
}
case 3: // Packed-decimal Real
{
source = READ_EA_PACK(m68k, ea);
break;
}
case 4: // Word Integer
{
INT16 d = READ_EA_16(m68k, ea);
source = int32_to_floatx80((INT32)d);
break;
}
case 5: // Double-precision Real
{
UINT64 d = READ_EA_64(m68k, ea);
source = float64_to_floatx80(d);
break;
}
case 6: // Byte Integer
{
INT8 d = READ_EA_8(m68k, ea);
source = int32_to_floatx80((INT32)d);
break;
}
case 7: // FMOVECR load from constant ROM
{
switch (w2 & 0x7f)
{
case 0x0: // Pi
source.high = 0x4000;
source.low = U64(0xc90fdaa22168c235);
break;
case 0xb: // log10(2)
source.high = 0x3ffd;
source.low = U64(0x9a209a84fbcff798);
break;
case 0xc: // e
source.high = 0x4000;
source.low = U64(0xadf85458a2bb4a9b);
break;
case 0xd: // log2(e)
source.high = 0x3fff;
source.low = U64(0xb8aa3b295c17f0bc);
break;
case 0xe: // log10(e)
source.high = 0x3ffd;
source.low = U64(0xde5bd8a937287195);
break;
case 0xf: // 0.0
source = int32_to_floatx80((INT32)0);
break;
case 0x30: // ln(2)
source.high = 0x3ffe;
source.low = U64(0xb17217f7d1cf79ac);
break;
case 0x31: // ln(10)
source.high = 0x4000;
source.low = U64(0x935d8dddaaa8ac17);
break;
case 0x32: // 1 (or 100? manuals are unclear, but 1 would make more sense)
source = int32_to_floatx80((INT32)1);
break;
case 0x33: // 10^1
source = int32_to_floatx80((INT32)10);
break;
case 0x34: // 10^2
source = int32_to_floatx80((INT32)10*10);
break;
case 0x35: // 10^4
source = int32_to_floatx80((INT32)1000*10);
break;
case 0x36: // 1.0e8
source = int32_to_floatx80((INT32)10000000*10);
break;
case 0x37: // 1.0e16 - can't get the right precision from INT32 so go "direct" with constants from h/w
source.high = 0x4034;
source.low = U64(0x8e1bc9bf04000000);
break;
case 0x38: // 1.0e32
source.high = 0x4069;
source.low = U64(0x9dc5ada82b70b59e);
break;
case 0x39: // 1.0e64
source.high = 0x40d3;
source.low = U64(0xc2781f49ffcfa6d5);
break;
case 0x3a: // 1.0e128
source.high = 0x41a8;
source.low = U64(0x93ba47c980e98ce0);
break;
case 0x3b: // 1.0e256
source.high = 0x4351;
source.low = U64(0xaa7eebfb9df9de8e);
break;
case 0x3c: // 1.0e512
source.high = 0x46a3;
source.low = U64(0xe319a0aea60e91c7);
break;
case 0x3d: // 1.0e1024
source.high = 0x4d48;
source.low = U64(0xc976758681750c17);
break;
case 0x3e: // 1.0e2048
source.high = 0x5a92;
source.low = U64(0x9e8b3b5dc53d5de5);
break;
case 0x3f: // 1.0e4096
source.high = 0x7525;
source.low = U64(0xc46052028a20979b);
break;
default:
fatalerror("fmove_rm_reg: unknown constant ROM offset %x at %08x\n", w2&0x7f, REG_PC(m68k)-4);
break;
}
// handle it right here, the usual opmode bits aren't valid in the FMOVECR case
REG_FP(m68k)[dst] = source;
m68k->remaining_cycles -= 4;
return;
}
default: fatalerror("fmove_rm_reg: invalid source specifier %x at %08X\n", src, REG_PC(m68k)-4);
}
}
else
{
source = REG_FP(m68k)[src];
}
switch (opmode)
{
case 0x00: // FMOVE
{
REG_FP(m68k)[dst] = source;
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 4;
break;
}
case 0x01: // FINT
{
INT32 temp;
temp = floatx80_to_int32(source);
REG_FP(m68k)[dst] = int32_to_floatx80(temp);
break;
}
case 0x03: // FINTRZ
{
INT32 temp;
temp = floatx80_to_int32_round_to_zero(source);
REG_FP(m68k)[dst] = int32_to_floatx80(temp);
break;
}
case 0x04: // FSQRT
{
REG_FP(m68k)[dst] = floatx80_sqrt(source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 109;
break;
}
case 0x06: // FLOGNP1
{
REG_FP(m68k)[dst] = floatx80_flognp1 (source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 594; // for MC68881
break;
}
case 0x0e: // FSIN
{
REG_FP(m68k)[dst] = source;
floatx80_fsin(REG_FP(m68k)[dst]);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 75;
break;
}
case 0x0f: // FTAN
{
REG_FP(m68k)[dst] = source;
floatx80_ftan(REG_FP(m68k)[dst]);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 75;
break;
}
case 0x14: // FLOGN
{
REG_FP(m68k)[dst] = floatx80_flogn (source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 548; // for MC68881
break;
}
case 0x15: // FLOG10
{
REG_FP(m68k)[dst] = floatx80_flog10 (source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 604; // for MC68881
break;
}
case 0x16: // FLOG2
{
REG_FP(m68k)[dst] = floatx80_flog2 (source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 604; // for MC68881
break;
}
case 0x18: // FABS
{
REG_FP(m68k)[dst] = source;
REG_FP(m68k)[dst].high &= 0x7fff;
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 3;
break;
}
case 0x1a: // FNEG
{
REG_FP(m68k)[dst] = source;
REG_FP(m68k)[dst].high ^= 0x8000;
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 3;
break;
}
case 0x1d: // FCOS
{
REG_FP(m68k)[dst] = source;
floatx80_fcos(REG_FP(m68k)[dst]);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 75;
break;
}
case 0x1e: // FGETEXP
{
INT16 temp2;
temp2 = source.high; // get the exponent
temp2 -= 0x3fff; // take off the bias
REG_FP(m68k)[dst] = double_to_fx80((double)temp2);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 6;
break;
}
case 0x20: // FDIV
{
REG_FP(m68k)[dst] = floatx80_div(REG_FP(m68k)[dst], source);
m68k->remaining_cycles -= 43;
break;
}
case 0x22: // FADD
{
REG_FP(m68k)[dst] = floatx80_add(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 9;
break;
}
case 0x23: // FMUL
{
REG_FP(m68k)[dst] = floatx80_mul(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 11;
break;
}
case 0x24: // FSGLDIV
{
float32 a = floatx80_to_float32( REG_FP(m68k)[dst] );
float32 b = floatx80_to_float32( source );
REG_FP(m68k)[dst] = float32_to_floatx80( float32_div(a, b) );
m68k->remaining_cycles -= 43; // // ? (value is from FDIV)
break;
}
case 0x25: // FREM
{
REG_FP(m68k)[dst] = floatx80_rem(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 43; // guess
break;
}
case 0x27: // FSGLMUL
{
float32 a = floatx80_to_float32( REG_FP(m68k)[dst] );
float32 b = floatx80_to_float32( source );
REG_FP(m68k)[dst] = float32_to_floatx80( float32_mul(a, b) );
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 11; // ? (value is from FMUL)
break;
}
case 0x28: // FSUB
{
REG_FP(m68k)[dst] = floatx80_sub(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 9;
break;
}
case 0x38: // FCMP
{
floatx80 res;
res = floatx80_sub(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, res);
m68k->remaining_cycles -= 7;
break;
}
case 0x3a: // FTST
{
floatx80 res;
res = source;
SET_CONDITION_CODES(m68k, res);
m68k->remaining_cycles -= 7;
break;
}
default: fatalerror("fpgen_rm_reg: unimplemented opmode %02X at %08X\n", opmode, REG_PPC(m68k));
}
}
unsigned int ExtendedCPDO(const unsigned int opcode)
{
FPA11 *fpa11 = GET_FPA11();
floatx80 rFm, rFn;
unsigned int Fd, Fm, Fn, nRc = 1;
//printk("ExtendedCPDO(0x%08x)\n",opcode);
Fm = getFm(opcode);
if (CONSTANT_FM(opcode))
{
rFm = getExtendedConstant(Fm);
}
else
{
switch (fpa11->fType[Fm])
{
case typeSingle:
rFm = float32_to_floatx80(fpa11->fpreg[Fm].fSingle, &fpa11->fp_status);
break;
case typeDouble:
rFm = float64_to_floatx80(fpa11->fpreg[Fm].fDouble, &fpa11->fp_status);
break;
case typeExtended:
rFm = fpa11->fpreg[Fm].fExtended;
break;
default: return 0;
}
}
if (!MONADIC_INSTRUCTION(opcode))
{
Fn = getFn(opcode);
switch (fpa11->fType[Fn])
{
case typeSingle:
rFn = float32_to_floatx80(fpa11->fpreg[Fn].fSingle, &fpa11->fp_status);
break;
case typeDouble:
rFn = float64_to_floatx80(fpa11->fpreg[Fn].fDouble, &fpa11->fp_status);
break;
case typeExtended:
rFn = fpa11->fpreg[Fn].fExtended;
break;
default: return 0;
}
}
Fd = getFd(opcode);
switch (opcode & MASK_ARITHMETIC_OPCODE)
{
/* dyadic opcodes */
case ADF_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_add(rFn,rFm, &fpa11->fp_status);
break;
case MUF_CODE:
case FML_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_mul(rFn,rFm, &fpa11->fp_status);
break;
case SUF_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_sub(rFn,rFm, &fpa11->fp_status);
break;
case RSF_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_sub(rFm,rFn, &fpa11->fp_status);
break;
case DVF_CODE:
case FDV_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_div(rFn,rFm, &fpa11->fp_status);
break;
case RDF_CODE:
case FRD_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_div(rFm,rFn, &fpa11->fp_status);
break;
#if 0
case POW_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_pow(rFn,rFm);
break;
case RPW_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_pow(rFm,rFn);
break;
#endif
case RMF_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_rem(rFn,rFm, &fpa11->fp_status);
break;
#if 0
case POL_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_pol(rFn,rFm);
break;
#endif
/* monadic opcodes */
case MVF_CODE:
fpa11->fpreg[Fd].fExtended = rFm;
break;
case MNF_CODE:
rFm.high ^= 0x8000;
fpa11->fpreg[Fd].fExtended = rFm;
break;
case ABS_CODE:
rFm.high &= 0x7fff;
fpa11->fpreg[Fd].fExtended = rFm;
break;
case RND_CODE:
case URD_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_round_to_int(rFm, &fpa11->fp_status);
break;
case SQT_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_sqrt(rFm, &fpa11->fp_status);
break;
#if 0
case LOG_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_log(rFm);
break;
case LGN_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_ln(rFm);
break;
case EXP_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_exp(rFm);
break;
case SIN_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_sin(rFm);
break;
case COS_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_cos(rFm);
break;
case TAN_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_tan(rFm);
break;
case ASN_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_arcsin(rFm);
break;
case ACS_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_arccos(rFm);
break;
case ATN_CODE:
fpa11->fpreg[Fd].fExtended = floatx80_arctan(rFm);
break;
#endif
case NRM_CODE:
break;
default:
{
nRc = 0;
}
}
if (0 != nRc) fpa11->fType[Fd] = typeExtended;
return nRc;
}
static void fpgen_rm_reg(m68ki_cpu_core *m68k, UINT16 w2)
{
int ea = m68k->ir & 0x3f;
int rm = (w2 >> 14) & 0x1;
int src = (w2 >> 10) & 0x7;
int dst = (w2 >> 7) & 0x7;
int opmode = w2 & 0x7f;
floatx80 source;
// fmovecr #$f, fp0 f200 5c0f
if (rm)
{
switch (src)
{
case 0: // Long-Word Integer
{
INT32 d = READ_EA_32(m68k, ea);
source = int32_to_floatx80(d);
break;
}
case 1: // Single-precision Real
{
UINT32 d = READ_EA_32(m68k, ea);
source = float32_to_floatx80(d);
break;
}
case 2: // Extended-precision Real
{
source = READ_EA_FPE(m68k, ea);
break;
}
case 3: // Packed-decimal Real
{
source = READ_EA_PACK(m68k, ea);
break;
}
case 4: // Word Integer
{
INT16 d = READ_EA_16(m68k, ea);
source = int32_to_floatx80((INT32)d);
break;
}
case 5: // Double-precision Real
{
UINT64 d = READ_EA_64(m68k, ea);
source = float64_to_floatx80(d);
break;
}
case 6: // Byte Integer
{
INT8 d = READ_EA_8(m68k, ea);
source = int32_to_floatx80((INT32)d);
break;
}
case 7: // FMOVECR load from constant ROM
{
switch (w2 & 0x7f)
{
case 0x0: // Pi
source.high = 0x4000;
source.low = U64(0xc90fdaa22168c235);
break;
case 0xb: // log10(2)
source.high = 0x3ffd;
source.low = U64(0x9a209a84fbcff798);
break;
case 0xc: // e
source.high = 0x4000;
source.low = U64(0xadf85458a2bb4a9b);
break;
case 0xd: // log2(e)
source.high = 0x3fff;
source.low = U64(0xb8aa3b295c17f0bc);
break;
case 0xe: // log10(e)
source.high = 0x3ffd;
source.low = U64(0xde5bd8a937287195);
break;
case 0xf: // 0.0
source = int32_to_floatx80((INT32)0);
break;
case 0x30: // ln(2)
source.high = 0x3ffe;
source.low = U64(0xb17217f7d1cf79ac);
break;
case 0x31: // ln(10)
source.high = 0x4000;
source.low = U64(0x935d8dddaaa8ac17);
break;
case 0x32: // 1 (or 100? manuals are unclear, but 1 would make more sense)
source = int32_to_floatx80((INT32)1);
break;
case 0x33: // 10^1
source = int32_to_floatx80((INT32)10);
break;
case 0x34: // 10^2
source = int32_to_floatx80((INT32)10*10);
break;
default:
fatalerror("fmove_rm_reg: unknown constant ROM offset %x at %08x\n", w2&0x7f, REG_PC(m68k)-4);
break;
}
// handle it right here, the usual opmode bits aren't valid in the FMOVECR case
REG_FP(m68k)[dst] = source;
m68k->remaining_cycles -= 4;
return;
}
default: fatalerror("fmove_rm_reg: invalid source specifier %x at %08X\n", src, REG_PC(m68k)-4);
}
}
else
{
source = REG_FP(m68k)[src];
}
switch (opmode)
{
case 0x00: // FMOVE
{
REG_FP(m68k)[dst] = source;
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 4;
break;
}
case 0x01: // FINT
{
INT32 temp;
temp = floatx80_to_int32(source);
REG_FP(m68k)[dst] = int32_to_floatx80(temp);
break;
}
case 0x03: // FINTRZ
{
INT32 temp;
temp = floatx80_to_int32_round_to_zero(source);
REG_FP(m68k)[dst] = int32_to_floatx80(temp);
break;
}
case 0x04: // FSQRT
{
REG_FP(m68k)[dst] = floatx80_sqrt(source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 109;
break;
}
case 0x0e: // FSIN
{
REG_FP(m68k)[dst] = source;
floatx80_fsin(REG_FP(m68k)[dst]);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 75;
break;
}
case 0x0f: // FTAN
{
REG_FP(m68k)[dst] = source;
floatx80_ftan(REG_FP(m68k)[dst]);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 75;
break;
}
case 0x18: // FABS
{
REG_FP(m68k)[dst] = source;
REG_FP(m68k)[dst].high &= 0x7fff;
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 3;
break;
}
case 0x1a: // FNEG
{
REG_FP(m68k)[dst] = source;
REG_FP(m68k)[dst].high ^= 0x8000;
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 3;
break;
}
case 0x1d: // FCOS
{
REG_FP(m68k)[dst] = source;
floatx80_fcos(REG_FP(m68k)[dst]);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 75;
break;
}
case 0x1e: // FGETEXP
{
INT16 temp2;
temp2 = source.high; // get the exponent
temp2 -= 0x3fff; // take off the bias
REG_FP(m68k)[dst] = double_to_fx80((double)temp2);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 6;
}
case 0x20: // FDIV
{
REG_FP(m68k)[dst] = floatx80_div(REG_FP(m68k)[dst], source);
m68k->remaining_cycles -= 43;
break;
}
case 0x22: // FADD
{
REG_FP(m68k)[dst] = floatx80_add(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 9;
break;
}
case 0x23: // FMUL
{
REG_FP(m68k)[dst] = floatx80_mul(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 11;
break;
}
case 0x24: // FSGLDIV
{
REG_FP(m68k)[dst] = floatx80_div(REG_FP(m68k)[dst], source);
m68k->remaining_cycles -= 43; // // ? (value is from FDIV)
break;
}
case 0x25: // FREM
{
REG_FP(m68k)[dst] = floatx80_rem(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 43; // guess
break;
}
case 0x27: // FSGLMUL
{
REG_FP(m68k)[dst] = floatx80_mul(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 11; // ? (value is from FMUL)
break;
}
case 0x28: // FSUB
{
REG_FP(m68k)[dst] = floatx80_sub(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, REG_FP(m68k)[dst]);
m68k->remaining_cycles -= 9;
break;
}
case 0x38: // FCMP
{
floatx80 res;
res = floatx80_sub(REG_FP(m68k)[dst], source);
SET_CONDITION_CODES(m68k, res);
m68k->remaining_cycles -= 7;
break;
}
case 0x3a: // FTST
{
floatx80 res;
res = source;
SET_CONDITION_CODES(m68k, res);
m68k->remaining_cycles -= 7;
break;
}
default: fatalerror("fpgen_rm_reg: unimplemented opmode %02X at %08X\n", opmode, REG_PPC(m68k));
}
}
floatx80 fyl2xp1(floatx80 a, floatx80 b)
{
INT32 aExp, bExp;
UINT64 aSig, bSig, zSig0, zSig1, zSig2;
int aSign, bSign;
aSig = extractFloatx80Frac(a);
aExp = extractFloatx80Exp(a);
aSign = extractFloatx80Sign(a);
bSig = extractFloatx80Frac(b);
bExp = extractFloatx80Exp(b);
bSign = extractFloatx80Sign(b);
int zSign = aSign ^ bSign;
if (aExp == 0x7FFF) {
if ((UINT64) (aSig<<1)
|| ((bExp == 0x7FFF) && (UINT64) (bSig<<1)))
{
return propagateFloatx80NaN(a, b);
}
if (aSign)
{
invalid:
float_raise(float_flag_invalid);
return floatx80_default_nan;
}
else {
if (bExp == 0) {
if (bSig == 0) goto invalid;
float_raise(float_flag_denormal);
}
return packFloatx80(bSign, 0x7FFF, U64(0x8000000000000000));
}
}
if (bExp == 0x7FFF)
{
if ((UINT64) (bSig<<1))
return propagateFloatx80NaN(a, b);
if (aExp == 0) {
if (aSig == 0) goto invalid;
float_raise(float_flag_denormal);
}
return packFloatx80(zSign, 0x7FFF, U64(0x8000000000000000));
}
if (aExp == 0) {
if (aSig == 0) {
if (bSig && (bExp == 0)) float_raise(float_flag_denormal);
return packFloatx80(zSign, 0, 0);
}
float_raise(float_flag_denormal);
normalizeFloatx80Subnormal(aSig, &aExp, &aSig);
}
if (bExp == 0) {
if (bSig == 0) return packFloatx80(zSign, 0, 0);
float_raise(float_flag_denormal);
normalizeFloatx80Subnormal(bSig, &bExp, &bSig);
}
float_raise(float_flag_inexact);
if (aSign && aExp >= 0x3FFF)
return a;
if (aExp >= 0x3FFC) // big argument
{
return fyl2x(floatx80_add(a, floatx80_one), b);
}
// handle tiny argument
if (aExp < EXP_BIAS-70)
{
// first order approximation, return (a*b)/ln(2)
INT32 zExp = aExp + FLOAT_LN2INV_EXP - 0x3FFE;
mul128By64To192(FLOAT_LN2INV_HI, FLOAT_LN2INV_LO, aSig, &zSig0, &zSig1, &zSig2);
if (0 < (INT64) zSig0) {
shortShift128Left(zSig0, zSig1, 1, &zSig0, &zSig1);
--zExp;
}
zExp = zExp + bExp - 0x3FFE;
mul128By64To192(zSig0, zSig1, bSig, &zSig0, &zSig1, &zSig2);
if (0 < (INT64) zSig0) {
shortShift128Left(zSig0, zSig1, 1, &zSig0, &zSig1);
--zExp;
}
return
roundAndPackFloatx80(80, aSign ^ bSign, zExp, zSig0, zSig1);
}
/* ******************************** */
/* using float128 for approximation */
/* ******************************** */
shift128Right(aSig<<1, 0, 16, &zSig0, &zSig1);
float128 x = packFloat128(aSign, aExp, zSig0, zSig1);
x = poly_l2p1(x);
return floatx80_mul(b, float128_to_floatx80(x));
}
floatx80 fpatan(floatx80 a, floatx80 b, float_status_t &status)
{
// handle unsupported extended double-precision floating encodings
if (floatx80_is_unsupported(a) || floatx80_is_unsupported(b)) {
float_raise(status, float_flag_invalid);
return floatx80_default_nan;
}
Bit64u aSig = extractFloatx80Frac(a);
Bit32s aExp = extractFloatx80Exp(a);
int aSign = extractFloatx80Sign(a);
Bit64u bSig = extractFloatx80Frac(b);
Bit32s bExp = extractFloatx80Exp(b);
int bSign = extractFloatx80Sign(b);
int zSign = aSign ^ bSign;
if (bExp == 0x7FFF)
{
if ((Bit64u) (bSig<<1))
return propagateFloatx80NaN(a, b, status);
if (aExp == 0x7FFF) {
if ((Bit64u) (aSig<<1))
return propagateFloatx80NaN(a, b, status);
if (aSign) { /* return 3PI/4 */
return roundAndPackFloatx80(80, bSign,
FLOATX80_3PI4_EXP, FLOAT_3PI4_HI, FLOAT_3PI4_LO, status);
}
else { /* return PI/4 */
return roundAndPackFloatx80(80, bSign,
FLOATX80_PI4_EXP, FLOAT_PI_HI, FLOAT_PI_LO, status);
}
}
if (aSig && (aExp == 0))
float_raise(status, float_flag_denormal);
/* return PI/2 */
return roundAndPackFloatx80(80, bSign, FLOATX80_PI2_EXP, FLOAT_PI_HI, FLOAT_PI_LO, status);
}
if (aExp == 0x7FFF)
{
if ((Bit64u) (aSig<<1))
return propagateFloatx80NaN(a, b, status);
if (bSig && (bExp == 0))
float_raise(status, float_flag_denormal);
return_PI_or_ZERO:
if (aSign) { /* return PI */
return roundAndPackFloatx80(80, bSign, FLOATX80_PI_EXP, FLOAT_PI_HI, FLOAT_PI_LO, status);
} else { /* return 0 */
return packFloatx80(bSign, 0, 0);
}
}
if (bExp == 0)
{
if (bSig == 0) {
if (aSig && (aExp == 0)) float_raise(status, float_flag_denormal);
goto return_PI_or_ZERO;
}
float_raise(status, float_flag_denormal);
normalizeFloatx80Subnormal(bSig, &bExp, &bSig);
}
if (aExp == 0)
{
if (aSig == 0) /* return PI/2 */
return roundAndPackFloatx80(80, bSign, FLOATX80_PI2_EXP, FLOAT_PI_HI, FLOAT_PI_LO, status);
float_raise(status, float_flag_denormal);
normalizeFloatx80Subnormal(aSig, &aExp, &aSig);
}
float_raise(status, float_flag_inexact);
/* |a| = |b| ==> return PI/4 */
if (aSig == bSig && aExp == bExp)
return roundAndPackFloatx80(80, bSign, FLOATX80_PI4_EXP, FLOAT_PI_HI, FLOAT_PI_LO, status);
/* ******************************** */
/* using float128 for approximation */
/* ******************************** */
float128 a128 = normalizeRoundAndPackFloat128(0, aExp-0x10, aSig, 0, status);
float128 b128 = normalizeRoundAndPackFloat128(0, bExp-0x10, bSig, 0, status);
float128 x;
int swap = 0, add_pi6 = 0, add_pi4 = 0;
if (aExp > bExp || (aExp == bExp && aSig > bSig))
{
x = float128_div(b128, a128, status);
}
else {
x = float128_div(a128, b128, status);
swap = 1;
}
Bit32s xExp = extractFloat128Exp(x);
if (xExp <= EXP_BIAS-40)
goto approximation_completed;
if (x.hi >= BX_CONST64(0x3ffe800000000000)) // 3/4 < x < 1
{
/*
arctan(x) = arctan((x-1)/(x+1)) + pi/4
*/
float128 t1 = float128_sub(x, float128_one, status);
float128 t2 = float128_add(x, float128_one, status);
x = float128_div(t1, t2, status);
add_pi4 = 1;
}
else
{
/* argument correction */
if (xExp >= 0x3FFD) // 1/4 < x < 3/4
{
/*
arctan(x) = arctan((x*sqrt(3)-1)/(x+sqrt(3))) + pi/6
*/
float128 t1 = float128_mul(x, float128_sqrt3, status);
float128 t2 = float128_add(x, float128_sqrt3, status);
x = float128_sub(t1, float128_one, status);
x = float128_div(x, t2, status);
add_pi6 = 1;
}
}
x = poly_atan(x, status);
if (add_pi6) x = float128_add(x, float128_pi6, status);
if (add_pi4) x = float128_add(x, float128_pi4, status);
approximation_completed:
if (swap) x = float128_sub(float128_pi2, x, status);
floatx80 result = float128_to_floatx80(x, status);
if (zSign) floatx80_chs(result);
int rSign = extractFloatx80Sign(result);
if (!bSign && rSign)
return floatx80_add(result, floatx80_pi, status);
if (bSign && !rSign)
return floatx80_sub(result, floatx80_pi, status);
return result;
}
