summaryrefslogtreecommitdiff
path: root/float.c
diff options
context:
space:
mode:
authorH. Peter Anvin <hpa@zytor.com>2007-09-18 17:50:34 -0700
committerH. Peter Anvin <hpa@zytor.com>2007-09-18 17:50:34 -0700
commite31747e95bba75c7e27d0a76f0e385c6d12351e2 (patch)
treeea6bc9c4c7a3cfc8378a2b91250778e14188d351 /float.c
parentcfbe7c3cc2dbdfe1268e2d0a19fc59b52cbcfcc5 (diff)
downloadnasm-e31747e95bba75c7e27d0a76f0e385c6d12351e2.tar.gz
nasm-e31747e95bba75c7e27d0a76f0e385c6d12351e2.tar.bz2
nasm-e31747e95bba75c7e27d0a76f0e385c6d12351e2.zip
Unify all standard IEEE floating-point formats; add 128-bit
Unify all the standard IEEE formats into one function, add support for IEEE standard 128-bit floating point numbers. The 80-bit format is still special since it explicitly represents the integer portion.
Diffstat (limited to 'float.c')
-rw-r--r--float.c206
1 files changed, 64 insertions, 142 deletions
diff --git a/float.c b/float.c
index afa84d2..a6ad393 100644
--- a/float.c
+++ b/float.c
@@ -18,8 +18,8 @@
#define TRUE 1
#define FALSE 0
-#define MANT_WORDS 6 /* 64 bits + 32 for accuracy == 96 */
-#define MANT_DIGITS 28 /* 29 digits don't fit in 96 bits */
+#define MANT_WORDS 10 /* 112 bits + 48 for accuracy == 160 */
+#define MANT_DIGITS 49 /* 50 digits don't fit in 160 bits */
/*
* guaranteed top bit of from is set
@@ -47,9 +47,8 @@ static int ieee_multiply(uint16_t *to, uint16_t *from)
temp[i] &= 0xFFFF;
}
if (temp[0] & 0x8000) {
- for (i = 0; i < MANT_WORDS; i++)
- to[i] = temp[i] & 0xFFFF;
- return 0;
+ memcpy(to, temp, 2*MANT_WORDS);
+ return 0;
} else {
for (i = 0; i < MANT_WORDS; i++)
to[i] = (temp[i] << 1) + !!(temp[i + 1] & 0x8000);
@@ -213,75 +212,33 @@ static int ieee_round(uint16_t *mant, int i)
#define put(a,b) ( (*(a)=(b)), ((a)[1]=(b)>>8) )
-/* 64-bit format with 52-bit mantissa and 11-bit exponent */
-static int to_double(char *str, int32_t sign, uint8_t *result,
- efunc error)
-{
- uint16_t mant[MANT_WORDS];
- int32_t exponent;
+/* Produce standard IEEE formats, with implicit "1" bit; this makes
+ the following assumptions:
- sign = (sign < 0 ? 0x8000L : 0L);
+ - the sign bit is the MSB, followed by the exponent.
+ - the sign bit plus exponent fit in 16 bits.
+ - the exponent bias is 2^(n-1)-1 for an n-bit exponent */
- ieee_flconvert(str, mant, &exponent, error);
- if (mant[0] & 0x8000) {
- /*
- * Non-zero.
- */
- exponent--;
- if (exponent >= -1022 && exponent <= 1024) {
- /*
- * Normalised.
- */
- exponent += 1023;
- ieee_shr(mant, 11);
- ieee_round(mant, 4);
- if (mant[0] & 0x20) /* did we scale up by one? */
- ieee_shr(mant, 1), exponent++;
- mant[0] &= 0xF; /* remove leading one */
- put(result + 6, (exponent << 4) | mant[0] | sign);
- put(result + 4, mant[1]);
- put(result + 2, mant[2]);
- put(result + 0, mant[3]);
- } else if (exponent < -1022 && exponent >= -1074) {
- /*
- * Denormal.
- */
- int shift = -(exponent + 1011);
- int sh = shift % 16, wds = shift / 16;
- ieee_shr(mant, sh);
- if (ieee_round(mant, 4 - wds)
- || (sh > 0 && (mant[0] & (0x8000 >> (sh - 1))))) {
- ieee_shr(mant, 1);
- if (sh == 0)
- mant[0] |= 0x8000;
- exponent++;
- }
- put(result + 6, (wds == 0 ? mant[0] : 0) | sign);
- put(result + 4, (wds <= 1 ? mant[1 - wds] : 0));
- put(result + 2, (wds <= 2 ? mant[2 - wds] : 0));
- put(result + 0, (wds <= 3 ? mant[3 - wds] : 0));
- } else {
- if (exponent > 0) {
- error(ERR_NONFATAL, "overflow in floating-point constant");
- return 0;
- } else
- memset(result, 0, 8);
- }
- } else {
- /*
- * Zero.
- */
- memset(result, 0, 8);
- }
- return 1; /* success */
-}
+struct ieee_format {
+ int words;
+ int mantissa; /* Bits in the mantissa */
+ int exponent; /* Bits in the exponent */
+};
+
+static const struct ieee_format ieee_16 = { 1, 10, 5 };
+static const struct ieee_format ieee_32 = { 2, 23, 8 };
+static const struct ieee_format ieee_64 = { 4, 52, 11 };
+static const struct ieee_format ieee_128 = { 8, 112, 15 };
-/* 32-bit format with 23-bit mantissa and 8-bit exponent */
+/* Produce all the standard IEEE formats: 16, 32, 64, and 128 bits */
static int to_float(char *str, int32_t sign, uint8_t *result,
- efunc error)
+ const struct ieee_format *fmt, efunc error)
{
- uint16_t mant[MANT_WORDS];
+ uint16_t mant[MANT_WORDS], *mp;
int32_t exponent;
+ int32_t expmax = 1 << (fmt->exponent-1);
+ uint16_t implicit_one = 0x8000 >> fmt->exponent;
+ int i;
sign = (sign < 0 ? 0x8000L : 0L);
@@ -291,101 +248,64 @@ static int to_float(char *str, int32_t sign, uint8_t *result,
* Non-zero.
*/
exponent--;
- if (exponent >= -126 && exponent <= 128) {
+ if (exponent >= 2-expmax && exponent <= expmax) {
/*
* Normalised.
*/
- exponent += 127;
- ieee_shr(mant, 8);
- ieee_round(mant, 2);
- if (mant[0] & 0x100) /* did we scale up by one? */
- ieee_shr(mant, 1), exponent++;
- mant[0] &= 0x7F; /* remove leading one */
- put(result + 2, (exponent << 7) | mant[0] | sign);
- put(result + 0, mant[1]);
- } else if (exponent < -126 && exponent >= -149) {
+ exponent += expmax;
+ ieee_shr(mant, fmt->exponent);
+ ieee_round(mant, fmt->words);
+ /* did we scale up by one? */
+ if (mant[0] & (implicit_one << 1)) {
+ ieee_shr(mant, 1);
+ exponent++;
+ }
+
+ mant[0] &= (implicit_one-1); /* remove leading one */
+ mant[0] |= exponent << (15 - fmt->exponent);
+ } else if (exponent < 2-expmax && exponent >= 2-expmax-fmt->mantissa) {
/*
* Denormal.
*/
- int shift = -(exponent + 118);
+ int shift = -(exponent + expmax-2-fmt->exponent);
int sh = shift % 16, wds = shift / 16;
ieee_shr(mant, sh);
- if (ieee_round(mant, 2 - wds)
+ if (ieee_round(mant, fmt->words - wds)
|| (sh > 0 && (mant[0] & (0x8000 >> (sh - 1))))) {
ieee_shr(mant, 1);
if (sh == 0)
mant[0] |= 0x8000;
exponent++;
}
- put(result + 2, (wds == 0 ? mant[0] : 0) | sign);
- put(result + 0, (wds <= 1 ? mant[1 - wds] : 0));
+
+ if (wds) {
+ for (i = fmt->words-1; i >= wds; i--)
+ mant[i] = mant[i-wds];
+ for (; i >= 0; i--)
+ mant[i] = 0;
+ }
} else {
if (exponent > 0) {
error(ERR_NONFATAL, "overflow in floating-point constant");
return 0;
- } else
- memset(result, 0, 4);
+ } else {
+ memset(mant, 0, 2*fmt->words);
+ }
}
} else {
- memset(result, 0, 4);
+ /* Zero */
+ memset(mant, 0, 2*fmt->words);
}
- return 1;
-}
-/* 16-bit format with 10-bit mantissa and 5-bit exponent.
- Defined in IEEE 754r. Used in SSE5. See the AMD SSE5 manual, AMD
- document number 43479. */
-static int to_float16(char *str, int32_t sign, uint8_t *result,
- efunc error)
-{
- uint16_t mant[MANT_WORDS];
- int32_t exponent;
-
- sign = (sign < 0 ? 0x8000L : 0L);
+ mant[0] |= sign;
- ieee_flconvert(str, mant, &exponent, error);
- if (mant[0] & 0x8000) {
- /*
- * Non-zero.
- */
- exponent--;
- if (exponent >= -14 && exponent <= 16) {
- /*
- * Normalised.
- */
- exponent += 15;
- ieee_shr(mant, 5);
- ieee_round(mant, 1);
- if (mant[0] & 0x800) /* did we scale up by one? */
- ieee_shr(mant, 1), exponent++;
- mant[0] &= 0x3FF; /* remove leading one */
- put(result + 0, (exponent << 7) | mant[0] | sign);
- } else if (exponent < -14 && exponent >= -24) {
- /*
- * Denormal.
- */
- int shift = -(exponent + 8);
- int sh = shift % 16, wds = shift / 16;
- ieee_shr(mant, sh);
- if (ieee_round(mant, 1 - wds)
- || (sh > 0 && (mant[0] & (0x8000 >> (sh - 1))))) {
- ieee_shr(mant, 1);
- if (sh == 0)
- mant[0] |= 0x8000;
- exponent++;
- }
- put(result + 0, (wds == 0 ? mant[0] : 0) | sign);
- } else {
- if (exponent > 0) {
- error(ERR_NONFATAL, "overflow in floating-point constant");
- return 0;
- } else
- memset(result, 0, 2);
- }
- } else {
- memset(result, 0, 2);
+ for (mp = &mant[fmt->words], i = 0; i < fmt->words; i++) {
+ uint16_t m = *--mp;
+ put(result, m);
+ result += 2;
}
- return 1;
+
+ return 1; /* success */
}
/* 80-bit format with 64-bit mantissa *including an explicit integer 1*
@@ -456,13 +376,15 @@ int float_const(char *number, int32_t sign, uint8_t *result, int bytes,
{
switch (bytes) {
case 2:
- return to_float16(number, sign, result, error);
+ return to_float(number, sign, result, &ieee_16, error);
case 4:
- return to_float(number, sign, result, error);
+ return to_float(number, sign, result, &ieee_32, error);
case 8:
- return to_double(number, sign, result, error);
+ return to_float(number, sign, result, &ieee_64, error);
case 10:
return to_ldoub(number, sign, result, error);
+ case 16:
+ return to_float(number, sign, result, &ieee_128, error);
default:
error(ERR_PANIC, "strange value %d passed to float_const", bytes);
return 0;