diff options
Diffstat (limited to 'asm/float.c')
-rw-r--r-- | asm/float.c | 12 |
1 files changed, 6 insertions, 6 deletions
diff --git a/asm/float.c b/asm/float.c index 54af63cb..6d27dc9a 100644 --- a/asm/float.c +++ b/asm/float.c @@ -236,7 +236,7 @@ static bool ieee_flconvert(const char *string, fp_limb *mant, *!float-toolong [on] too many digits in floating-point number *! warns about too many digits in floating-point numbers. */ - nasm_warnf(WARN_FLOAT_TOOLONG|ERR_PASS2, + nasm_warn(WARN_FLOAT_TOOLONG|ERR_PASS2, "floating-point constant significand contains " "more than %i digits", MANT_DIGITS); warned = true; @@ -685,7 +685,7 @@ static int to_packed_bcd(const char *str, const char *p, if (c >= '0' && c <= '9') { if (tv < 0) { if (n == 9) - nasm_warnf(ERR_PASS2, "packed BCD truncated to 18 digits"); + nasm_warn(ERR_PASS2, "packed BCD truncated to 18 digits"); tv = c-'0'; } else { if (n < 9) @@ -802,7 +802,7 @@ static int to_float(const char *str, int s, uint8_t *result, if (exponent >= 2 - expmax && exponent <= expmax) { type = FL_NORMAL; } else if (exponent > 0) { - nasm_warnf(WARN_FLOAT_OVERFLOW|ERR_PASS2, + nasm_warn(WARN_FLOAT_OVERFLOW|ERR_PASS2, "overflow in floating-point constant"); type = FL_INFINITY; } else { @@ -841,7 +841,7 @@ static int to_float(const char *str, int s, uint8_t *result, *! warns about floating point underflow (a nonzero *! constant rounded to zero.) */ - nasm_warnf(WARN_FLOAT_UNDERFLOW|ERR_PASS2, + nasm_warn(WARN_FLOAT_UNDERFLOW|ERR_PASS2, "underflow in floating-point constant"); goto zero; } else { @@ -849,7 +849,7 @@ static int to_float(const char *str, int s, uint8_t *result, *!float-denorm [off] floating point denormal *! warns about denormal floating point constants. */ - nasm_warnf(WARN_FLOAT_DENORM|ERR_PASS2, + nasm_warn(WARN_FLOAT_DENORM|ERR_PASS2, "denormal floating-point constant"); } } @@ -869,7 +869,7 @@ static int to_float(const char *str, int s, uint8_t *result, *!float-overflow [on] floating point overflow *! warns about floating point underflow. */ - nasm_warnf(WARN_FLOAT_OVERFLOW|ERR_PASS2, + nasm_warn(WARN_FLOAT_OVERFLOW|ERR_PASS2, "overflow in floating-point constant"); type = FL_INFINITY; goto overflow; |