softfloat: Support float_round_to_odd more places

Previously this was only supported for roundAndPackFloat64.

New support in round_canonical, round_to_int, float128_round_to_int,
roundAndPackFloat32, roundAndPackInt32, roundAndPackInt64,
roundAndPackUint64.  This does not include any of the floatx80 routines,
as we do not have users for that rounding mode there.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20190215170225.15537-1-richard.henderson@linaro.org>
Tested-by: David Hildenbrand <david@redhat.com>
[AJB: add missing break]
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
This commit is contained in:
Richard Henderson 2019-02-15 09:02:25 -08:00 committed by Alex Bennée
parent dc3f8a9dcf
commit 5d64abb32f
2 changed files with 94 additions and 14 deletions

View File

@ -696,6 +696,7 @@ static FloatParts sf_canonicalize(FloatParts part, const FloatFmt *parm,
static FloatParts round_canonical(FloatParts p, float_status *s,
const FloatFmt *parm)
{
const uint64_t frac_lsb = parm->frac_lsb;
const uint64_t frac_lsbm1 = parm->frac_lsbm1;
const uint64_t round_mask = parm->round_mask;
const uint64_t roundeven_mask = parm->roundeven_mask;
@ -731,6 +732,10 @@ static FloatParts round_canonical(FloatParts p, float_status *s,
inc = p.sign ? round_mask : 0;
overflow_norm = !p.sign;
break;
case float_round_to_odd:
overflow_norm = true;
inc = frac & frac_lsb ? 0 : round_mask;
break;
default:
g_assert_not_reached();
}
@ -778,9 +783,14 @@ static FloatParts round_canonical(FloatParts p, float_status *s,
shift64RightJamming(frac, 1 - exp, &frac);
if (frac & round_mask) {
/* Need to recompute round-to-even. */
if (s->float_rounding_mode == float_round_nearest_even) {
switch (s->float_rounding_mode) {
case float_round_nearest_even:
inc = ((frac & roundeven_mask) != frac_lsbm1
? frac_lsbm1 : 0);
break;
case float_round_to_odd:
inc = frac & frac_lsb ? 0 : round_mask;
break;
}
flags |= float_flag_inexact;
frac += inc;
@ -1988,6 +1998,9 @@ static FloatParts round_to_int(FloatParts a, int rmode,
case float_round_down:
one = a.sign;
break;
case float_round_to_odd:
one = true;
break;
default:
g_assert_not_reached();
}
@ -2021,6 +2034,9 @@ static FloatParts round_to_int(FloatParts a, int rmode,
case float_round_down:
inc = a.sign ? rnd_mask : 0;
break;
case float_round_to_odd:
inc = a.frac & frac_lsb ? 0 : rnd_mask;
break;
default:
g_assert_not_reached();
}
@ -3314,6 +3330,9 @@ static int32_t roundAndPackInt32(flag zSign, uint64_t absZ, float_status *status
case float_round_down:
roundIncrement = zSign ? 0x7f : 0;
break;
case float_round_to_odd:
roundIncrement = absZ & 0x80 ? 0 : 0x7f;
break;
default:
abort();
}
@ -3368,6 +3387,9 @@ static int64_t roundAndPackInt64(flag zSign, uint64_t absZ0, uint64_t absZ1,
case float_round_down:
increment = zSign && absZ1;
break;
case float_round_to_odd:
increment = !(absZ0 & 1) && absZ1;
break;
default:
abort();
}
@ -3424,6 +3446,9 @@ static int64_t roundAndPackUint64(flag zSign, uint64_t absZ0,
case float_round_down:
increment = zSign && absZ1;
break;
case float_round_to_odd:
increment = !(absZ0 & 1) && absZ1;
break;
default:
abort();
}
@ -3526,6 +3551,9 @@ static float32 roundAndPackFloat32(flag zSign, int zExp, uint32_t zSig,
case float_round_down:
roundIncrement = zSign ? 0x7f : 0;
break;
case float_round_to_odd:
roundIncrement = zSig & 0x80 ? 0 : 0x7f;
break;
default:
abort();
break;
@ -3536,8 +3564,10 @@ static float32 roundAndPackFloat32(flag zSign, int zExp, uint32_t zSig,
|| ( ( zExp == 0xFD )
&& ( (int32_t) ( zSig + roundIncrement ) < 0 ) )
) {
bool overflow_to_inf = roundingMode != float_round_to_odd &&
roundIncrement != 0;
float_raise(float_flag_overflow | float_flag_inexact, status);
return packFloat32( zSign, 0xFF, - ( roundIncrement == 0 ));
return packFloat32(zSign, 0xFF, -!overflow_to_inf);
}
if ( zExp < 0 ) {
if (status->flush_to_zero) {
@ -3555,6 +3585,13 @@ static float32 roundAndPackFloat32(flag zSign, int zExp, uint32_t zSig,
if (isTiny && roundBits) {
float_raise(float_flag_underflow, status);
}
if (roundingMode == float_round_to_odd) {
/*
* For round-to-odd case, the roundIncrement depends on
* zSig which just changed.
*/
roundIncrement = zSig & 0x80 ? 0 : 0x7f;
}
}
}
if (roundBits) {
@ -6987,6 +7024,15 @@ float128 float128_round_to_int(float128 a, float_status *status)
add128(z.high, z.low, 0, roundBitsMask, &z.high, &z.low);
}
break;
case float_round_to_odd:
/*
* Note that if lastBitMask == 0, the last bit is the lsb
* of high, and roundBitsMask == -1.
*/
if ((lastBitMask ? z.low & lastBitMask : z.high & 1) == 0) {
add128(z.high, z.low, 0, roundBitsMask, &z.high, &z.low);
}
break;
default:
abort();
}
@ -6998,7 +7044,7 @@ float128 float128_round_to_int(float128 a, float_status *status)
status->float_exception_flags |= float_flag_inexact;
aSign = extractFloat128Sign( a );
switch (status->float_rounding_mode) {
case float_round_nearest_even:
case float_round_nearest_even:
if ( ( aExp == 0x3FFE )
&& ( extractFloat128Frac0( a )
| extractFloat128Frac1( a ) )
@ -7011,14 +7057,17 @@ float128 float128_round_to_int(float128 a, float_status *status)
return packFloat128(aSign, 0x3FFF, 0, 0);
}
break;
case float_round_down:
case float_round_down:
return
aSign ? packFloat128( 1, 0x3FFF, 0, 0 )
: packFloat128( 0, 0, 0, 0 );
case float_round_up:
case float_round_up:
return
aSign ? packFloat128( 1, 0, 0, 0 )
: packFloat128( 0, 0x3FFF, 0, 0 );
case float_round_to_odd:
return packFloat128(aSign, 0x3FFF, 0, 0);
}
return packFloat128( aSign, 0, 0, 0 );
}
@ -7051,6 +7100,12 @@ float128 float128_round_to_int(float128 a, float_status *status)
z.high += roundBitsMask;
}
break;
case float_round_to_odd:
if ((z.high & lastBitMask) == 0) {
z.high |= (a.low != 0);
z.high += roundBitsMask;
}
break;
default:
abort();
}

View File

@ -125,17 +125,42 @@ static void not_implemented(void)
static bool blacklisted(unsigned op, int rmode)
{
/* odd has only been implemented for a few 128-bit ops */
/* odd has not been implemented for any 80-bit ops */
if (rmode == softfloat_round_odd) {
switch (op) {
case F128_ADD:
case F128_SUB:
case F128_MUL:
case F128_DIV:
case F128_TO_F64:
case F128_SQRT:
return false;
default:
case EXTF80_TO_UI32:
case EXTF80_TO_UI64:
case EXTF80_TO_I32:
case EXTF80_TO_I64:
case EXTF80_TO_UI32_R_MINMAG:
case EXTF80_TO_UI64_R_MINMAG:
case EXTF80_TO_I32_R_MINMAG:
case EXTF80_TO_I64_R_MINMAG:
case EXTF80_TO_F16:
case EXTF80_TO_F32:
case EXTF80_TO_F64:
case EXTF80_TO_F128:
case EXTF80_ROUNDTOINT:
case EXTF80_ADD:
case EXTF80_SUB:
case EXTF80_MUL:
case EXTF80_DIV:
case EXTF80_REM:
case EXTF80_SQRT:
case EXTF80_EQ:
case EXTF80_LE:
case EXTF80_LT:
case EXTF80_EQ_SIGNALING:
case EXTF80_LE_QUIET:
case EXTF80_LT_QUIET:
case UI32_TO_EXTF80:
case UI64_TO_EXTF80:
case I32_TO_EXTF80:
case I64_TO_EXTF80:
case F16_TO_EXTF80:
case F32_TO_EXTF80:
case F64_TO_EXTF80:
case F128_TO_EXTF80:
return true;
}
}