diff options
Diffstat (limited to 'fpu/softfloat-specialize.h')
-rw-r--r-- | fpu/softfloat-specialize.h | 16 |
1 files changed, 8 insertions, 8 deletions
diff --git a/fpu/softfloat-specialize.h b/fpu/softfloat-specialize.h index d430f58a71..f9b6f0cbc6 100644 --- a/fpu/softfloat-specialize.h +++ b/fpu/softfloat-specialize.h @@ -68,7 +68,7 @@ typedef struct { | otherwise returns 0. *----------------------------------------------------------------------------*/ -flag float32_is_nan( float32 a ) +int float32_is_nan( float32 a ) { return ( 0xFF000000 < (bits32) ( a<<1 ) ); @@ -80,7 +80,7 @@ flag float32_is_nan( float32 a ) | NaN; otherwise returns 0. *----------------------------------------------------------------------------*/ -flag float32_is_signaling_nan( float32 a ) +int float32_is_signaling_nan( float32 a ) { return ( ( ( a>>22 ) & 0x1FF ) == 0x1FE ) && ( a & 0x003FFFFF ); @@ -161,7 +161,7 @@ static float32 propagateFloat32NaN( float32 a, float32 b STATUS_PARAM) | otherwise returns 0. *----------------------------------------------------------------------------*/ -flag float64_is_nan( float64 a ) +int float64_is_nan( float64 a ) { return ( LIT64( 0xFFE0000000000000 ) < (bits64) ( a<<1 ) ); @@ -173,7 +173,7 @@ flag float64_is_nan( float64 a ) | NaN; otherwise returns 0. *----------------------------------------------------------------------------*/ -flag float64_is_signaling_nan( float64 a ) +int float64_is_signaling_nan( float64 a ) { return @@ -264,7 +264,7 @@ static float64 propagateFloat64NaN( float64 a, float64 b STATUS_PARAM) | NaN; otherwise returns 0. *----------------------------------------------------------------------------*/ -flag floatx80_is_nan( floatx80 a ) +int floatx80_is_nan( floatx80 a ) { return ( ( a.high & 0x7FFF ) == 0x7FFF ) && (bits64) ( a.low<<1 ); @@ -276,7 +276,7 @@ flag floatx80_is_nan( floatx80 a ) | signaling NaN; otherwise returns 0. *----------------------------------------------------------------------------*/ -flag floatx80_is_signaling_nan( floatx80 a ) +int floatx80_is_signaling_nan( floatx80 a ) { bits64 aLow; @@ -371,7 +371,7 @@ static floatx80 propagateFloatx80NaN( floatx80 a, floatx80 b STATUS_PARAM) | otherwise returns 0. *----------------------------------------------------------------------------*/ -flag float128_is_nan( float128 a ) +int float128_is_nan( float128 a ) { return @@ -385,7 +385,7 @@ flag float128_is_nan( float128 a ) | signaling NaN; otherwise returns 0. *----------------------------------------------------------------------------*/ -flag float128_is_signaling_nan( float128 a ) +int float128_is_signaling_nan( float128 a ) { return |