0b9b56e38e
currently supporting sparc64. After a `make depend all` there are three programs; testsoftfloat for testing against the SoftFloat in src/lib/libc/softfloat for reference purposes, testemufloat for testing the emulator source in src/lib/libc/sparc64/fpu and testfloat for testing with the installed libc. Support for other architectures can be added as needed. PR: 144900 Submitted by: Peter Jeremy
1184 lines
28 KiB
C
1184 lines
28 KiB
C
|
|
/*
|
|
===============================================================================
|
|
|
|
This C source file is part of TestFloat, Release 2a, a package of programs
|
|
for testing the correctness of floating-point arithmetic complying to the
|
|
IEC/IEEE Standard for Floating-Point.
|
|
|
|
Written by John R. Hauser. More information is available through the Web
|
|
page `http://HTTP.CS.Berkeley.EDU/~jhauser/arithmetic/TestFloat.html'.
|
|
|
|
THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort
|
|
has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
|
|
TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO
|
|
PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
|
|
AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
|
|
|
|
Derivative works are acceptable, even for commercial purposes, so long as
|
|
(1) they include prominent notice that the work is derivative, and (2) they
|
|
include prominent notice akin to these four paragraphs for those parts of
|
|
this code that are retained.
|
|
|
|
===============================================================================
|
|
*/
|
|
|
|
int8 slow_float_rounding_mode;
|
|
int8 slow_float_exception_flags;
|
|
int8 slow_float_detect_tininess;
|
|
|
|
typedef struct {
|
|
bits32 a0, a1;
|
|
} bits64X;
|
|
|
|
typedef struct {
|
|
flag isNaN;
|
|
flag isInf;
|
|
flag isZero;
|
|
flag sign;
|
|
int16 exp;
|
|
bits64X sig;
|
|
} floatX;
|
|
|
|
static const floatX floatXNaN = { TRUE, FALSE, FALSE, FALSE, 0, { 0, 0 } };
|
|
static const floatX floatXPositiveZero =
|
|
{ FALSE, FALSE, TRUE, FALSE, 0, { 0, 0 } };
|
|
static const floatX floatXNegativeZero =
|
|
{ FALSE, FALSE, TRUE, TRUE, 0, { 0, 0 } };
|
|
|
|
static bits64X shortShift64Left( bits64X a, int8 shiftCount )
|
|
{
|
|
int8 negShiftCount;
|
|
|
|
negShiftCount = ( - shiftCount & 31 );
|
|
a.a0 = ( a.a0<<shiftCount ) | ( a.a1>>negShiftCount );
|
|
a.a1 <<= shiftCount;
|
|
return a;
|
|
|
|
}
|
|
|
|
static bits64X shortShift64RightJamming( bits64X a, int8 shiftCount )
|
|
{
|
|
int8 negShiftCount;
|
|
bits32 extra;
|
|
|
|
negShiftCount = ( - shiftCount & 31 );
|
|
extra = a.a1<<negShiftCount;
|
|
a.a1 = ( a.a0<<negShiftCount ) | ( a.a1>>shiftCount ) | ( extra != 0 );
|
|
a.a0 >>= shiftCount;
|
|
return a;
|
|
|
|
}
|
|
|
|
static bits64X neg64( bits64X a )
|
|
{
|
|
|
|
if ( a.a1 == 0 ) {
|
|
a.a0 = - a.a0;
|
|
}
|
|
else {
|
|
a.a1 = - a.a1;
|
|
a.a0 = ~ a.a0;
|
|
}
|
|
return a;
|
|
|
|
}
|
|
|
|
static bits64X add64( bits64X a, bits64X b )
|
|
{
|
|
|
|
a.a1 += b.a1;
|
|
a.a0 += b.a0 + ( a.a1 < b.a1 );
|
|
return a;
|
|
|
|
}
|
|
|
|
static flag eq64( bits64X a, bits64X b )
|
|
{
|
|
|
|
return ( a.a0 == b.a0 ) && ( a.a1 == b.a1 );
|
|
|
|
}
|
|
|
|
static flag le64( bits64X a, bits64X b )
|
|
{
|
|
|
|
return ( a.a0 < b.a0 ) || ( ( a.a0 == b.a0 ) && ( a.a1 <= b.a1 ) );
|
|
|
|
}
|
|
|
|
static flag lt64( bits64X a, bits64X b )
|
|
{
|
|
|
|
return ( a.a0 < b.a0 ) || ( ( a.a0 == b.a0 ) && ( a.a1 < b.a1 ) );
|
|
|
|
}
|
|
|
|
static floatX roundFloatXTo24( flag isTiny, floatX zx )
|
|
{
|
|
|
|
if ( zx.sig.a1 ) {
|
|
slow_float_exception_flags |= float_flag_inexact;
|
|
if ( isTiny ) slow_float_exception_flags |= float_flag_underflow;
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
if ( zx.sig.a1 < 0x80000000 ) goto noIncrement;
|
|
if ( ( zx.sig.a1 == 0x80000000 ) && ! ( zx.sig.a0 & 1 ) ) {
|
|
goto noIncrement;
|
|
}
|
|
break;
|
|
case float_round_to_zero:
|
|
goto noIncrement;
|
|
case float_round_down:
|
|
if ( ! zx.sign ) goto noIncrement;
|
|
break;
|
|
case float_round_up:
|
|
if ( zx.sign ) goto noIncrement;
|
|
break;
|
|
}
|
|
++zx.sig.a0;
|
|
if ( zx.sig.a0 == 0x01000000 ) {
|
|
zx.sig.a0 = 0x00800000;
|
|
++zx.exp;
|
|
}
|
|
}
|
|
noIncrement:
|
|
zx.sig.a1 = 0;
|
|
return zx;
|
|
|
|
}
|
|
|
|
static floatX roundFloatXTo53( flag isTiny, floatX zx )
|
|
{
|
|
int8 roundBits;
|
|
|
|
roundBits = zx.sig.a1 & 7;
|
|
zx.sig.a1 -= roundBits;
|
|
if ( roundBits ) {
|
|
slow_float_exception_flags |= float_flag_inexact;
|
|
if ( isTiny ) slow_float_exception_flags |= float_flag_underflow;
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
if ( roundBits < 4 ) goto noIncrement;
|
|
if ( ( roundBits == 4 ) && ! ( zx.sig.a1 & 8 ) ) goto noIncrement;
|
|
break;
|
|
case float_round_to_zero:
|
|
goto noIncrement;
|
|
case float_round_down:
|
|
if ( ! zx.sign ) goto noIncrement;
|
|
break;
|
|
case float_round_up:
|
|
if ( zx.sign ) goto noIncrement;
|
|
break;
|
|
}
|
|
zx.sig.a1 += 8;
|
|
zx.sig.a0 += ( zx.sig.a1 == 0 );
|
|
if ( zx.sig.a0 == 0x01000000 ) {
|
|
zx.sig.a0 = 0x00800000;
|
|
++zx.exp;
|
|
}
|
|
}
|
|
noIncrement:
|
|
return zx;
|
|
|
|
}
|
|
|
|
static floatX int32ToFloatX( int32 a )
|
|
{
|
|
floatX ax;
|
|
|
|
ax.isNaN = FALSE;
|
|
ax.isInf = FALSE;
|
|
ax.sign = ( a < 0 );
|
|
ax.sig.a1 = ax.sign ? - a : a;
|
|
ax.sig.a0 = 0;
|
|
if ( a == 0 ) {
|
|
ax.isZero = TRUE;
|
|
return ax;
|
|
}
|
|
ax.isZero = FALSE;
|
|
ax.sig = shortShift64Left( ax.sig, 23 );
|
|
ax.exp = 32;
|
|
while ( ax.sig.a0 < 0x00800000 ) {
|
|
ax.sig = shortShift64Left( ax.sig, 1 );
|
|
--ax.exp;
|
|
}
|
|
return ax;
|
|
|
|
}
|
|
|
|
static int32 floatXToInt32( floatX ax )
|
|
{
|
|
int8 savedExceptionFlags;
|
|
int16 shiftCount;
|
|
int32 z;
|
|
|
|
if ( ax.isInf || ax.isNaN ) {
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
return ( ax.isInf & ax.sign ) ? 0x80000000 : 0x7FFFFFFF;
|
|
}
|
|
if ( ax.isZero ) return 0;
|
|
savedExceptionFlags = slow_float_exception_flags;
|
|
shiftCount = 52 - ax.exp;
|
|
if ( 56 < shiftCount ) {
|
|
ax.sig.a1 = 1;
|
|
ax.sig.a0 = 0;
|
|
}
|
|
else {
|
|
while ( 0 < shiftCount ) {
|
|
ax.sig = shortShift64RightJamming( ax.sig, 1 );
|
|
--shiftCount;
|
|
}
|
|
}
|
|
ax = roundFloatXTo53( FALSE, ax );
|
|
ax.sig = shortShift64RightJamming( ax.sig, 3 );
|
|
z = ax.sig.a1;
|
|
if ( ax.sign ) z = - z;
|
|
if ( ( shiftCount < 0 )
|
|
|| ax.sig.a0
|
|
|| ( ( z != 0 ) && ( ( ax.sign ^ ( z < 0 ) ) != 0 ) )
|
|
) {
|
|
slow_float_exception_flags = savedExceptionFlags | float_flag_invalid;
|
|
return ax.sign ? 0x80000000 : 0x7FFFFFFF;
|
|
}
|
|
return z;
|
|
|
|
}
|
|
|
|
static floatX float32ToFloatX( float32 a )
|
|
{
|
|
int16 expField;
|
|
floatX ax;
|
|
|
|
ax.isNaN = FALSE;
|
|
ax.isInf = FALSE;
|
|
ax.isZero = FALSE;
|
|
ax.sign = ( ( a & 0x80000000 ) != 0 );
|
|
expField = ( a>>23 ) & 0xFF;
|
|
ax.sig.a1 = 0;
|
|
ax.sig.a0 = a & 0x007FFFFF;
|
|
if ( expField == 0 ) {
|
|
if ( ax.sig.a0 == 0 ) {
|
|
ax.isZero = TRUE;
|
|
}
|
|
else {
|
|
expField = 1 - 0x7F;
|
|
do {
|
|
ax.sig.a0 <<= 1;
|
|
--expField;
|
|
} while ( ax.sig.a0 < 0x00800000 );
|
|
ax.exp = expField;
|
|
}
|
|
}
|
|
else if ( expField == 0xFF ) {
|
|
if ( ax.sig.a0 == 0 ) {
|
|
ax.isInf = TRUE;
|
|
}
|
|
else {
|
|
ax.isNaN = TRUE;
|
|
}
|
|
}
|
|
else {
|
|
ax.sig.a0 |= 0x00800000;
|
|
ax.exp = expField - 0x7F;
|
|
}
|
|
return ax;
|
|
|
|
}
|
|
|
|
static float32 floatXToFloat32( floatX zx )
|
|
{
|
|
floatX savedZ;
|
|
flag isTiny;
|
|
int16 expField;
|
|
float32 z;
|
|
|
|
if ( zx.isZero ) return zx.sign ? 0x80000000 : 0;
|
|
if ( zx.isInf ) return zx.sign ? 0xFF800000 : 0x7F800000;
|
|
if ( zx.isNaN ) return 0xFFFFFFFF;
|
|
while ( 0x01000000 <= zx.sig.a0 ) {
|
|
zx.sig = shortShift64RightJamming( zx.sig, 1 );
|
|
++zx.exp;
|
|
}
|
|
while ( zx.sig.a0 < 0x00800000 ) {
|
|
zx.sig = shortShift64Left( zx.sig, 1 );
|
|
--zx.exp;
|
|
}
|
|
savedZ = zx;
|
|
isTiny =
|
|
( slow_float_detect_tininess == float_tininess_before_rounding )
|
|
&& ( zx.exp + 0x7F <= 0 );
|
|
zx = roundFloatXTo24( isTiny, zx );
|
|
expField = zx.exp + 0x7F;
|
|
if ( 0xFF <= expField ) {
|
|
slow_float_exception_flags |=
|
|
float_flag_overflow | float_flag_inexact;
|
|
if ( zx.sign ) {
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
case float_round_down:
|
|
z = 0xFF800000;
|
|
break;
|
|
case float_round_to_zero:
|
|
case float_round_up:
|
|
z = 0xFF7FFFFF;
|
|
break;
|
|
}
|
|
}
|
|
else {
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
case float_round_up:
|
|
z = 0x7F800000;
|
|
break;
|
|
case float_round_to_zero:
|
|
case float_round_down:
|
|
z = 0x7F7FFFFF;
|
|
break;
|
|
}
|
|
}
|
|
return z;
|
|
}
|
|
if ( expField <= 0 ) {
|
|
isTiny = TRUE;
|
|
zx = savedZ;
|
|
expField = zx.exp + 0x7F;
|
|
if ( expField < -27 ) {
|
|
zx.sig.a1 = ( zx.sig.a0 != 0 ) || ( zx.sig.a1 != 0 );
|
|
zx.sig.a0 = 0;
|
|
}
|
|
else {
|
|
while ( expField <= 0 ) {
|
|
zx.sig = shortShift64RightJamming( zx.sig, 1 );
|
|
++expField;
|
|
}
|
|
}
|
|
zx = roundFloatXTo24( isTiny, zx );
|
|
expField = ( 0x00800000 <= zx.sig.a0 ) ? 1 : 0;
|
|
}
|
|
z = expField;
|
|
z <<= 23;
|
|
if ( zx.sign ) z |= 0x80000000;
|
|
z |= zx.sig.a0 & 0x007FFFFF;
|
|
return z;
|
|
|
|
}
|
|
|
|
static floatX float64ToFloatX( float64 a )
|
|
{
|
|
int16 expField;
|
|
floatX ax;
|
|
|
|
ax.isNaN = FALSE;
|
|
ax.isInf = FALSE;
|
|
ax.isZero = FALSE;
|
|
#ifdef BITS64
|
|
ax.sign = ( ( a & LIT64( 0x8000000000000000 ) ) != 0 );
|
|
expField = ( a>>52 ) & 0x7FF;
|
|
ax.sig.a1 = a;
|
|
ax.sig.a0 = ( a>>32 ) & 0x000FFFFF;
|
|
#else
|
|
ax.sign = ( ( a.high & 0x80000000 ) != 0 );
|
|
expField = ( a.high>>( 52 - 32 ) ) & 0x7FF;
|
|
ax.sig.a1 = a.low;
|
|
ax.sig.a0 = a.high & 0x000FFFFF;
|
|
#endif
|
|
if ( expField == 0 ) {
|
|
if ( ( ax.sig.a0 == 0 ) && ( ax.sig.a1 == 0 ) ) {
|
|
ax.isZero = TRUE;
|
|
}
|
|
else {
|
|
expField = 1 - 0x3FF;
|
|
do {
|
|
ax.sig = shortShift64Left( ax.sig, 1 );
|
|
--expField;
|
|
} while ( ax.sig.a0 < 0x00100000 );
|
|
ax.exp = expField;
|
|
}
|
|
}
|
|
else if ( expField == 0x7FF ) {
|
|
if ( ( ax.sig.a0 == 0 ) && ( ax.sig.a1 == 0 ) ) {
|
|
ax.isInf = TRUE;
|
|
}
|
|
else {
|
|
ax.isNaN = TRUE;
|
|
}
|
|
}
|
|
else {
|
|
ax.exp = expField - 0x3FF;
|
|
ax.sig.a0 |= 0x00100000;
|
|
}
|
|
ax.sig = shortShift64Left( ax.sig, 3 );
|
|
return ax;
|
|
|
|
}
|
|
|
|
static float64 floatXToFloat64( floatX zx )
|
|
{
|
|
floatX savedZ;
|
|
flag isTiny;
|
|
int16 expField;
|
|
float64 z;
|
|
|
|
#ifdef BITS64
|
|
if ( zx.isZero ) return zx.sign ? LIT64( 0x8000000000000000 ) : 0;
|
|
if ( zx.isInf ) {
|
|
return
|
|
zx.sign ? LIT64( 0xFFF0000000000000 )
|
|
: LIT64( 0x7FF0000000000000 );
|
|
}
|
|
if ( zx.isNaN ) return LIT64( 0xFFFFFFFFFFFFFFFF );
|
|
#else
|
|
if ( zx.isZero ) {
|
|
z.low = 0;
|
|
z.high = zx.sign ? 0x80000000 : 0;
|
|
return z;
|
|
}
|
|
if ( zx.isInf ) {
|
|
z.low = 0;
|
|
z.high = zx.sign ? 0xFFF00000 : 0x7FF00000;
|
|
return z;
|
|
}
|
|
if ( zx.isNaN ) {
|
|
z.high = z.low = 0xFFFFFFFF;
|
|
return z;
|
|
}
|
|
#endif
|
|
while ( 0x01000000 <= zx.sig.a0 ) {
|
|
zx.sig = shortShift64RightJamming( zx.sig, 1 );
|
|
++zx.exp;
|
|
}
|
|
while ( zx.sig.a0 < 0x00800000 ) {
|
|
zx.sig = shortShift64Left( zx.sig, 1 );
|
|
--zx.exp;
|
|
}
|
|
savedZ = zx;
|
|
isTiny =
|
|
( slow_float_detect_tininess == float_tininess_before_rounding )
|
|
&& ( zx.exp + 0x3FF <= 0 );
|
|
zx = roundFloatXTo53( isTiny, zx );
|
|
expField = zx.exp + 0x3FF;
|
|
if ( 0x7FF <= expField ) {
|
|
slow_float_exception_flags |=
|
|
float_flag_overflow | float_flag_inexact;
|
|
#ifdef BITS64
|
|
if ( zx.sign ) {
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
case float_round_down:
|
|
z = LIT64( 0xFFF0000000000000 );
|
|
break;
|
|
case float_round_to_zero:
|
|
case float_round_up:
|
|
z = LIT64( 0xFFEFFFFFFFFFFFFF );
|
|
break;
|
|
}
|
|
}
|
|
else {
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
case float_round_up:
|
|
z = LIT64( 0x7FF0000000000000 );
|
|
break;
|
|
case float_round_to_zero:
|
|
case float_round_down:
|
|
z = LIT64( 0x7FEFFFFFFFFFFFFF );
|
|
break;
|
|
}
|
|
}
|
|
#else
|
|
if ( zx.sign ) {
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
case float_round_down:
|
|
z.low = 0;
|
|
z.high = 0xFFF00000;
|
|
break;
|
|
case float_round_to_zero:
|
|
case float_round_up:
|
|
z.low = 0xFFFFFFFF;
|
|
z.high = 0xFFEFFFFF;
|
|
break;
|
|
}
|
|
}
|
|
else {
|
|
switch ( slow_float_rounding_mode ) {
|
|
case float_round_nearest_even:
|
|
case float_round_up:
|
|
z.low = 0;
|
|
z.high = 0x7FF00000;
|
|
break;
|
|
case float_round_to_zero:
|
|
case float_round_down:
|
|
z.low = 0xFFFFFFFF;
|
|
z.high = 0x7FEFFFFF;
|
|
break;
|
|
}
|
|
}
|
|
#endif
|
|
return z;
|
|
}
|
|
if ( expField <= 0 ) {
|
|
isTiny = TRUE;
|
|
zx = savedZ;
|
|
expField = zx.exp + 0x3FF;
|
|
if ( expField < -56 ) {
|
|
zx.sig.a1 = ( zx.sig.a0 != 0 ) || ( zx.sig.a1 != 0 );
|
|
zx.sig.a0 = 0;
|
|
}
|
|
else {
|
|
while ( expField <= 0 ) {
|
|
zx.sig = shortShift64RightJamming( zx.sig, 1 );
|
|
++expField;
|
|
}
|
|
}
|
|
zx = roundFloatXTo53( isTiny, zx );
|
|
expField = ( 0x00800000 <= zx.sig.a0 ) ? 1 : 0;
|
|
}
|
|
zx.sig = shortShift64RightJamming( zx.sig, 3 );
|
|
#ifdef BITS64
|
|
z = expField;
|
|
z <<= 52;
|
|
if ( zx.sign ) z |= LIT64( 0x8000000000000000 );
|
|
z |= ( ( (bits64) ( zx.sig.a0 & 0x000FFFFF ) )<<32 ) | zx.sig.a1;
|
|
#else
|
|
z.low = zx.sig.a1;
|
|
z.high = expField;
|
|
z.high <<= 52 - 32;
|
|
if ( zx.sign ) z.high |= 0x80000000;
|
|
z.high |= zx.sig.a0 & 0x000FFFFF;
|
|
#endif
|
|
return z;
|
|
|
|
}
|
|
|
|
static floatX floatXInvalid( void )
|
|
{
|
|
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
return floatXNaN;
|
|
|
|
}
|
|
|
|
static floatX floatXRoundToInt( floatX ax )
|
|
{
|
|
int16 shiftCount, i;
|
|
|
|
if ( ax.isNaN || ax.isInf ) return ax;
|
|
shiftCount = 52 - ax.exp;
|
|
if ( shiftCount <= 0 ) return ax;
|
|
if ( 55 < shiftCount ) {
|
|
ax.exp = 52;
|
|
ax.sig.a1 = ! ax.isZero;
|
|
ax.sig.a0 = 0;
|
|
}
|
|
else {
|
|
while ( 0 < shiftCount ) {
|
|
ax.sig = shortShift64RightJamming( ax.sig, 1 );
|
|
++ax.exp;
|
|
--shiftCount;
|
|
}
|
|
}
|
|
ax = roundFloatXTo53( FALSE, ax );
|
|
if ( ( ax.sig.a0 == 0 ) && ( ax.sig.a1 == 0 ) ) ax.isZero = TRUE;
|
|
return ax;
|
|
|
|
}
|
|
|
|
static floatX floatXAdd( floatX ax, floatX bx )
|
|
{
|
|
int16 expDiff;
|
|
floatX zx;
|
|
|
|
if ( ax.isNaN ) return ax;
|
|
if ( bx.isNaN ) return bx;
|
|
if ( ax.isInf && bx.isInf ) {
|
|
if ( ax.sign == bx.sign ) return ax;
|
|
return floatXInvalid();
|
|
}
|
|
if ( ax.isInf ) return ax;
|
|
if ( bx.isInf ) return bx;
|
|
if ( ax.isZero && bx.isZero ) {
|
|
if ( ax.sign == bx.sign ) return ax;
|
|
goto completeCancellation;
|
|
}
|
|
if ( ( ax.sign != bx.sign )
|
|
&& ( ax.exp == bx.exp )
|
|
&& eq64( ax.sig, bx.sig )
|
|
) {
|
|
completeCancellation:
|
|
return
|
|
( slow_float_rounding_mode == float_round_down ) ?
|
|
floatXNegativeZero
|
|
: floatXPositiveZero;
|
|
}
|
|
if ( ax.isZero ) return bx;
|
|
if ( bx.isZero ) return ax;
|
|
expDiff = ax.exp - bx.exp;
|
|
if ( expDiff < 0 ) {
|
|
zx = ax;
|
|
zx.exp = bx.exp;
|
|
if ( expDiff < -56 ) {
|
|
zx.sig.a1 = 1;
|
|
zx.sig.a0 = 0;
|
|
}
|
|
else {
|
|
while ( expDiff < 0 ) {
|
|
zx.sig = shortShift64RightJamming( zx.sig, 1 );
|
|
++expDiff;
|
|
}
|
|
}
|
|
if ( ax.sign != bx.sign ) zx.sig = neg64( zx.sig );
|
|
zx.sign = bx.sign;
|
|
zx.sig = add64( zx.sig, bx.sig );
|
|
}
|
|
else {
|
|
zx = bx;
|
|
zx.exp = ax.exp;
|
|
if ( 56 < expDiff ) {
|
|
zx.sig.a1 = 1;
|
|
zx.sig.a0 = 0;
|
|
}
|
|
else {
|
|
while ( 0 < expDiff ) {
|
|
zx.sig = shortShift64RightJamming( zx.sig, 1 );
|
|
--expDiff;
|
|
}
|
|
}
|
|
if ( ax.sign != bx.sign ) zx.sig = neg64( zx.sig );
|
|
zx.sign = ax.sign;
|
|
zx.sig = add64( zx.sig, ax.sig );
|
|
}
|
|
if ( zx.sig.a0 & 0x80000000 ) {
|
|
zx.sig = neg64( zx.sig );
|
|
zx.sign = ! zx.sign;
|
|
}
|
|
return zx;
|
|
|
|
}
|
|
|
|
static floatX floatXMul( floatX ax, floatX bx )
|
|
{
|
|
int8 bitNum;
|
|
floatX zx;
|
|
|
|
if ( ax.isNaN ) return ax;
|
|
if ( bx.isNaN ) return bx;
|
|
if ( ax.isInf ) {
|
|
if ( bx.isZero ) return floatXInvalid();
|
|
if ( bx.sign ) ax.sign = ! ax.sign;
|
|
return ax;
|
|
}
|
|
if ( bx.isInf ) {
|
|
if ( ax.isZero ) return floatXInvalid();
|
|
if ( ax.sign ) bx.sign = ! bx.sign;
|
|
return bx;
|
|
}
|
|
zx = ax;
|
|
zx.sign ^= bx.sign;
|
|
if ( ax.isZero || bx.isZero ) {
|
|
return zx.sign ? floatXNegativeZero : floatXPositiveZero;
|
|
}
|
|
zx.exp += bx.exp + 1;
|
|
zx.sig.a1 = 0;
|
|
zx.sig.a0 = 0;
|
|
for ( bitNum = 0; bitNum < 55; ++bitNum ) {
|
|
if ( bx.sig.a1 & 2 ) zx.sig = add64( zx.sig, ax.sig );
|
|
bx.sig = shortShift64RightJamming( bx.sig, 1 );
|
|
zx.sig = shortShift64RightJamming( zx.sig, 1 );
|
|
}
|
|
return zx;
|
|
|
|
}
|
|
|
|
static floatX floatXDiv( floatX ax, floatX bx )
|
|
{
|
|
bits64X negBSig;
|
|
int8 bitNum;
|
|
floatX zx;
|
|
|
|
if ( ax.isNaN ) return ax;
|
|
if ( bx.isNaN ) return bx;
|
|
if ( ax.isInf ) {
|
|
if ( bx.isInf ) return floatXInvalid();
|
|
if ( bx.sign ) ax.sign = ! ax.sign;
|
|
return ax;
|
|
}
|
|
if ( bx.isZero ) {
|
|
if ( ax.isZero ) return floatXInvalid();
|
|
slow_float_exception_flags |= float_flag_divbyzero;
|
|
if ( ax.sign ) bx.sign = ! bx.sign;
|
|
bx.isZero = FALSE;
|
|
bx.isInf = TRUE;
|
|
return bx;
|
|
}
|
|
zx = ax;
|
|
zx.sign ^= bx.sign;
|
|
if ( ax.isZero || bx.isInf ) {
|
|
return zx.sign ? floatXNegativeZero : floatXPositiveZero;
|
|
}
|
|
zx.exp -= bx.exp + 1;
|
|
zx.sig.a1 = 0;
|
|
zx.sig.a0 = 0;
|
|
negBSig = neg64( bx.sig );
|
|
for ( bitNum = 0; bitNum < 56; ++bitNum ) {
|
|
if ( le64( bx.sig, ax.sig ) ) {
|
|
zx.sig.a1 |= 1;
|
|
ax.sig = add64( ax.sig, negBSig );
|
|
}
|
|
ax.sig = shortShift64Left( ax.sig, 1 );
|
|
zx.sig = shortShift64Left( zx.sig, 1 );
|
|
}
|
|
if ( ax.sig.a0 || ax.sig.a1 ) zx.sig.a1 |= 1;
|
|
return zx;
|
|
|
|
}
|
|
|
|
static floatX floatXRem( floatX ax, floatX bx )
|
|
{
|
|
bits64X negBSig;
|
|
flag lastQuotientBit;
|
|
bits64X savedASig;
|
|
|
|
if ( ax.isNaN ) return ax;
|
|
if ( bx.isNaN ) return bx;
|
|
if ( ax.isInf || bx.isZero ) return floatXInvalid();
|
|
if ( ax.isZero || bx.isInf ) return ax;
|
|
--bx.exp;
|
|
if ( ax.exp < bx.exp ) return ax;
|
|
bx.sig = shortShift64Left( bx.sig, 1 );
|
|
negBSig = neg64( bx.sig );
|
|
while ( bx.exp < ax.exp ) {
|
|
if ( le64( bx.sig, ax.sig ) ) ax.sig = add64( ax.sig, negBSig );
|
|
ax.sig = shortShift64Left( ax.sig, 1 );
|
|
--ax.exp;
|
|
}
|
|
lastQuotientBit = le64( bx.sig, ax.sig );
|
|
if ( lastQuotientBit ) ax.sig = add64( ax.sig, negBSig );
|
|
savedASig = ax.sig;
|
|
ax.sig = neg64( add64( ax.sig, negBSig ) );
|
|
if ( lt64( ax.sig, savedASig ) ) {
|
|
ax.sign = ! ax.sign;
|
|
}
|
|
else if ( lt64( savedASig, ax.sig ) ) {
|
|
ax.sig = savedASig;
|
|
}
|
|
else {
|
|
if ( lastQuotientBit ) {
|
|
ax.sign = ! ax.sign;
|
|
}
|
|
else {
|
|
ax.sig = savedASig;
|
|
}
|
|
}
|
|
if ( ( ax.sig.a0 == 0 ) && ( ax.sig.a1 == 0 ) ) ax.isZero = TRUE;
|
|
return ax;
|
|
|
|
}
|
|
|
|
static floatX floatXSqrt( floatX ax )
|
|
{
|
|
int8 bitNum;
|
|
bits64X bitSig, savedASig;
|
|
floatX zx;
|
|
|
|
if ( ax.isNaN || ax.isZero ) return ax;
|
|
if ( ax.sign ) return floatXInvalid();
|
|
if ( ax.isInf ) return ax;
|
|
zx = ax;
|
|
zx.exp >>= 1;
|
|
if ( ( ax.exp & 1 ) == 0 ) ax.sig = shortShift64RightJamming( ax.sig, 1 );
|
|
zx.sig.a1 = 0;
|
|
zx.sig.a0 = 0;
|
|
bitSig.a1 = 0;
|
|
bitSig.a0 = 0x00800000;
|
|
for ( bitNum = 0; bitNum < 56; ++bitNum ) {
|
|
savedASig = ax.sig;
|
|
ax.sig = add64( ax.sig, neg64( zx.sig ) );
|
|
ax.sig = shortShift64Left( ax.sig, 1 );
|
|
ax.sig = add64( ax.sig, neg64( bitSig ) );
|
|
if ( ax.sig.a0 & 0x80000000 ) {
|
|
ax.sig = shortShift64Left( savedASig, 1 );
|
|
}
|
|
else {
|
|
zx.sig.a1 |= bitSig.a1;
|
|
zx.sig.a0 |= bitSig.a0;
|
|
}
|
|
bitSig = shortShift64RightJamming( bitSig, 1 );
|
|
}
|
|
if ( ax.sig.a0 || ax.sig.a1 ) zx.sig.a1 |= 1;
|
|
return zx;
|
|
|
|
}
|
|
|
|
static flag floatXEq( floatX ax, floatX bx )
|
|
{
|
|
|
|
if ( ax.isNaN || bx.isNaN ) return FALSE;
|
|
if ( ax.isZero && bx.isZero ) return TRUE;
|
|
if ( ax.sign != bx.sign ) return FALSE;
|
|
if ( ax.isInf || bx.isInf ) return ax.isInf && bx.isInf;
|
|
return ( ax.exp == bx.exp ) && eq64( ax.sig, bx.sig );
|
|
|
|
}
|
|
|
|
static flag floatXLe( floatX ax, floatX bx )
|
|
{
|
|
|
|
if ( ax.isNaN || bx.isNaN ) return FALSE;
|
|
if ( ax.isZero && bx.isZero ) return TRUE;
|
|
if ( ax.sign != bx.sign ) return ax.sign;
|
|
if ( ax.sign ) {
|
|
if ( ax.isInf || bx.isZero ) return TRUE;
|
|
if ( bx.isInf || ax.isZero ) return FALSE;
|
|
if ( bx.exp < ax.exp ) return TRUE;
|
|
if ( ax.exp < bx.exp ) return FALSE;
|
|
return le64( bx.sig, ax.sig );
|
|
}
|
|
else {
|
|
if ( bx.isInf || ax.isZero ) return TRUE;
|
|
if ( ax.isInf || bx.isZero ) return FALSE;
|
|
if ( ax.exp < bx.exp ) return TRUE;
|
|
if ( bx.exp < ax.exp ) return FALSE;
|
|
return le64( ax.sig, bx.sig );
|
|
}
|
|
|
|
}
|
|
|
|
static flag floatXLt( floatX ax, floatX bx )
|
|
{
|
|
|
|
if ( ax.isNaN || bx.isNaN ) return FALSE;
|
|
if ( ax.isZero && bx.isZero ) return FALSE;
|
|
if ( ax.sign != bx.sign ) return ax.sign;
|
|
if ( ax.isInf && bx.isInf ) return FALSE;
|
|
if ( ax.sign ) {
|
|
if ( ax.isInf || bx.isZero ) return TRUE;
|
|
if ( bx.isInf || ax.isZero ) return FALSE;
|
|
if ( bx.exp < ax.exp ) return TRUE;
|
|
if ( ax.exp < bx.exp ) return FALSE;
|
|
return lt64( bx.sig, ax.sig );
|
|
}
|
|
else {
|
|
if ( bx.isInf || ax.isZero ) return TRUE;
|
|
if ( ax.isInf || bx.isZero ) return FALSE;
|
|
if ( ax.exp < bx.exp ) return TRUE;
|
|
if ( bx.exp < ax.exp ) return FALSE;
|
|
return lt64( ax.sig, bx.sig );
|
|
}
|
|
|
|
}
|
|
|
|
float32 slow_int32_to_float32( int32 a )
|
|
{
|
|
|
|
return floatXToFloat32( int32ToFloatX( a ) );
|
|
|
|
}
|
|
|
|
float64 slow_int32_to_float64( int32 a )
|
|
{
|
|
|
|
return floatXToFloat64( int32ToFloatX( a ) );
|
|
|
|
}
|
|
|
|
int32 slow_float32_to_int32( float32 a )
|
|
{
|
|
|
|
return floatXToInt32( float32ToFloatX( a ) );
|
|
|
|
}
|
|
|
|
int32 slow_float32_to_int32_round_to_zero( float32 a )
|
|
{
|
|
int8 savedRoundingMode;
|
|
int32 z;
|
|
|
|
savedRoundingMode = slow_float_rounding_mode;
|
|
slow_float_rounding_mode = float_round_to_zero;
|
|
z = floatXToInt32( float32ToFloatX( a ) );
|
|
slow_float_rounding_mode = savedRoundingMode;
|
|
return z;
|
|
|
|
}
|
|
|
|
float64 slow_float32_to_float64( float32 a )
|
|
{
|
|
|
|
return floatXToFloat64( float32ToFloatX( a ) );
|
|
|
|
}
|
|
|
|
float32 slow_float32_round_to_int( float32 a )
|
|
{
|
|
|
|
return floatXToFloat32( floatXRoundToInt( float32ToFloatX( a ) ) );
|
|
|
|
}
|
|
|
|
float32 slow_float32_add( float32 a, float32 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat32(
|
|
floatXAdd( float32ToFloatX( a ), float32ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float32 slow_float32_sub( float32 a, float32 b )
|
|
{
|
|
|
|
b ^= 0x80000000;
|
|
return
|
|
floatXToFloat32(
|
|
floatXAdd( float32ToFloatX( a ), float32ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float32 slow_float32_mul( float32 a, float32 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat32(
|
|
floatXMul( float32ToFloatX( a ), float32ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float32 slow_float32_div( float32 a, float32 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat32(
|
|
floatXDiv( float32ToFloatX( a ), float32ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float32 slow_float32_rem( float32 a, float32 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat32(
|
|
floatXRem( float32ToFloatX( a ), float32ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float32 slow_float32_sqrt( float32 a )
|
|
{
|
|
|
|
return floatXToFloat32( floatXSqrt( float32ToFloatX( a ) ) );
|
|
|
|
}
|
|
|
|
flag slow_float32_eq( float32 a, float32 b )
|
|
{
|
|
|
|
return floatXEq( float32ToFloatX( a ), float32ToFloatX( b ) );
|
|
|
|
}
|
|
|
|
flag slow_float32_le( float32 a, float32 b )
|
|
{
|
|
floatX ax, bx;
|
|
|
|
ax = float32ToFloatX( a );
|
|
bx = float32ToFloatX( b );
|
|
if ( ax.isNaN || bx.isNaN ) {
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
}
|
|
return floatXLe( ax, bx );
|
|
|
|
}
|
|
|
|
flag slow_float32_lt( float32 a, float32 b )
|
|
{
|
|
floatX ax, bx;
|
|
|
|
ax = float32ToFloatX( a );
|
|
bx = float32ToFloatX( b );
|
|
if ( ax.isNaN || bx.isNaN ) {
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
}
|
|
return floatXLt( ax, bx );
|
|
|
|
}
|
|
|
|
flag slow_float32_eq_signaling( float32 a, float32 b )
|
|
{
|
|
floatX ax, bx;
|
|
|
|
ax = float32ToFloatX( a );
|
|
bx = float32ToFloatX( b );
|
|
if ( ax.isNaN || bx.isNaN ) {
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
}
|
|
return floatXEq( ax, bx );
|
|
|
|
}
|
|
|
|
flag slow_float32_le_quiet( float32 a, float32 b )
|
|
{
|
|
|
|
return floatXLe( float32ToFloatX( a ), float32ToFloatX( b ) );
|
|
|
|
}
|
|
|
|
flag slow_float32_lt_quiet( float32 a, float32 b )
|
|
{
|
|
|
|
return floatXLt( float32ToFloatX( a ), float32ToFloatX( b ) );
|
|
|
|
}
|
|
|
|
int32 slow_float64_to_int32( float64 a )
|
|
{
|
|
|
|
return floatXToInt32( float64ToFloatX( a ) );
|
|
|
|
}
|
|
|
|
int32 slow_float64_to_int32_round_to_zero( float64 a )
|
|
{
|
|
int8 savedRoundingMode;
|
|
int32 z;
|
|
|
|
savedRoundingMode = slow_float_rounding_mode;
|
|
slow_float_rounding_mode = float_round_to_zero;
|
|
z = floatXToInt32( float64ToFloatX( a ) );
|
|
slow_float_rounding_mode = savedRoundingMode;
|
|
return z;
|
|
|
|
}
|
|
|
|
float32 slow_float64_to_float32( float64 a )
|
|
{
|
|
|
|
return floatXToFloat32( float64ToFloatX( a ) );
|
|
|
|
}
|
|
|
|
float64 slow_float64_round_to_int( float64 a )
|
|
{
|
|
|
|
return floatXToFloat64( floatXRoundToInt( float64ToFloatX( a ) ) );
|
|
|
|
}
|
|
|
|
float64 slow_float64_add( float64 a, float64 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat64(
|
|
floatXAdd( float64ToFloatX( a ), float64ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float64 slow_float64_sub( float64 a, float64 b )
|
|
{
|
|
|
|
#ifdef BITS64
|
|
b ^= LIT64( 0x8000000000000000 );
|
|
#else
|
|
b.high ^= 0x80000000;
|
|
#endif
|
|
return
|
|
floatXToFloat64(
|
|
floatXAdd( float64ToFloatX( a ), float64ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float64 slow_float64_mul( float64 a, float64 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat64(
|
|
floatXMul( float64ToFloatX( a ), float64ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float64 slow_float64_div( float64 a, float64 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat64(
|
|
floatXDiv( float64ToFloatX( a ), float64ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float64 slow_float64_rem( float64 a, float64 b )
|
|
{
|
|
|
|
return
|
|
floatXToFloat64(
|
|
floatXRem( float64ToFloatX( a ), float64ToFloatX( b ) ) );
|
|
|
|
}
|
|
|
|
float64 slow_float64_sqrt( float64 a )
|
|
{
|
|
|
|
return floatXToFloat64( floatXSqrt( float64ToFloatX( a ) ) );
|
|
|
|
}
|
|
|
|
flag slow_float64_eq( float64 a, float64 b )
|
|
{
|
|
|
|
return floatXEq( float64ToFloatX( a ), float64ToFloatX( b ) );
|
|
|
|
}
|
|
|
|
flag slow_float64_le( float64 a, float64 b )
|
|
{
|
|
floatX ax, bx;
|
|
|
|
ax = float64ToFloatX( a );
|
|
bx = float64ToFloatX( b );
|
|
if ( ax.isNaN || bx.isNaN ) {
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
}
|
|
return floatXLe( ax, bx );
|
|
|
|
}
|
|
|
|
flag slow_float64_lt( float64 a, float64 b )
|
|
{
|
|
floatX ax, bx;
|
|
|
|
ax = float64ToFloatX( a );
|
|
bx = float64ToFloatX( b );
|
|
if ( ax.isNaN || bx.isNaN ) {
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
}
|
|
return floatXLt( ax, bx );
|
|
|
|
}
|
|
|
|
flag slow_float64_eq_signaling( float64 a, float64 b )
|
|
{
|
|
floatX ax, bx;
|
|
|
|
ax = float64ToFloatX( a );
|
|
bx = float64ToFloatX( b );
|
|
if ( ax.isNaN || bx.isNaN ) {
|
|
slow_float_exception_flags |= float_flag_invalid;
|
|
}
|
|
return floatXEq( ax, bx );
|
|
|
|
}
|
|
|
|
flag slow_float64_le_quiet( float64 a, float64 b )
|
|
{
|
|
|
|
return floatXLe( float64ToFloatX( a ), float64ToFloatX( b ) );
|
|
|
|
}
|
|
|
|
flag slow_float64_lt_quiet( float64 a, float64 b )
|
|
{
|
|
|
|
return floatXLt( float64ToFloatX( a ), float64ToFloatX( b ) );
|
|
|
|
}
|
|
|