[libc] Simplify decimalStringToFloat and hexadecimalStringToFloat and improve their performance.
Combine two loops in decimalStringToFloat and hexadecimalStringToFloat that extract the digits and re-arrange them a little bit. This slightly improves the performance of strtof and strtod:
Running libc_str_to_float_comparison_test parse-number-fxx-test_data/data/* on my machine (Ryzen 1700)
- with glibc: ~1.92 seconds
- with current implementation: ~1.78 seconds
- with this change: ~1.67 seconds
Differential Revision: https://reviews.llvm.org/D113681
GitOrigin-RevId: 499ca806dcfb5f5c842fb240e150140fb15011ed
diff --git a/src/__support/str_to_float.h b/src/__support/str_to_float.h
index a768889..3f4a050 100644
--- a/src/__support/str_to_float.h
+++ b/src/__support/str_to_float.h
@@ -467,14 +467,15 @@
// This is the number of leading zeroes a properly normalized float of type T
// should have.
- constexpr uint32_t NORMALIZED_LEADING_ZEROES =
+ constexpr int32_t NORMALIZED_LEADING_ZEROES =
(sizeof(BitsType) * 8) - fputil::FloatProperties<T>::mantissaWidth - 1;
constexpr BitsType OVERFLOWED_MANTISSA =
BitsType(1) << (fputil::FloatProperties<T>::mantissaWidth + 1);
// Normalization
int32_t amountToShift =
- NORMALIZED_LEADING_ZEROES - leadingZeroes<BitsType>(mantissa);
+ NORMALIZED_LEADING_ZEROES -
+ static_cast<int32_t>(leadingZeroes<BitsType>(mantissa));
if (amountToShift < 0) {
mantissa <<= -amountToShift;
} else {
@@ -569,59 +570,45 @@
// The goal for the first step of parsing is to convert the number in src to
// the format mantissa * (base ^ exponent)
- // The first loop fills the mantissa with as many digits as it can hold
+ // The loop fills the mantissa with as many digits as it can hold
const BitsType BITSTYPE_MAX_DIV_BY_BASE =
__llvm_libc::cpp::NumericLimits<BitsType>::max() / BASE;
- while ((isdigit(*src) || *src == DECIMAL_POINT) &&
- mantissa < BITSTYPE_MAX_DIV_BY_BASE) {
+ while (true) {
+ if (isdigit(*src)) {
+ uint32_t digit = *src - '0';
+ seenDigit = true;
+
+ if (mantissa < BITSTYPE_MAX_DIV_BY_BASE) {
+ mantissa = (mantissa * BASE) + digit;
+ if (afterDecimal) {
+ --exponent;
+ }
+ } else {
+ if (digit > 0)
+ truncated = true;
+ if (!afterDecimal)
+ ++exponent;
+ }
+
+ ++src;
+ continue;
+ }
if (*src == DECIMAL_POINT) {
if (afterDecimal) {
break; // this means that *src points to a second decimal point, ending
// the number.
- } else {
- afterDecimal = true;
- ++src;
- continue;
}
+ afterDecimal = true;
+ ++src;
+ continue;
}
- uint32_t digit = *src - '0';
-
- mantissa = (mantissa * BASE) + digit;
- seenDigit = true;
- if (afterDecimal) {
- --exponent;
- }
-
- ++src;
+ // The character is neither a digit nor a decimal point.
+ break;
}
if (!seenDigit)
return false;
- // The second loop is to run through the remaining digits after we've filled
- // the mantissa.
- while (isdigit(*src) || *src == DECIMAL_POINT) {
- if (*src == DECIMAL_POINT) {
- if (afterDecimal) {
- break; // this means that *src points to a second decimal point, ending
- // the number.
- } else {
- afterDecimal = true;
- ++src;
- continue;
- }
- }
- uint32_t digit = *src - '0';
-
- if (digit > 0)
- truncated = true;
-
- if (!afterDecimal)
- ++exponent;
-
- ++src;
- }
-
if ((*src | 32) == EXPONENT_MARKER) {
if (*(src + 1) == '+' || *(src + 1) == '-' || isdigit(*(src + 1))) {
++src;
@@ -673,62 +660,47 @@
// The goal for the first step of parsing is to convert the number in src to
// the format mantissa * (base ^ exponent)
- // The first loop fills the mantissa with as many digits as it can hold
+ // The loop fills the mantissa with as many digits as it can hold
const BitsType BITSTYPE_MAX_DIV_BY_BASE =
__llvm_libc::cpp::NumericLimits<BitsType>::max() / BASE;
- while ((isalnum(*src) || *src == DECIMAL_POINT) &&
- mantissa < BITSTYPE_MAX_DIV_BY_BASE) {
+ while (true) {
+ if (isalnum(*src)) {
+ uint32_t digit = b36_char_to_int(*src);
+ if (digit >= BASE) {
+ seenDigit = false;
+ break;
+ }
+ seenDigit = true;
+
+ if (mantissa < BITSTYPE_MAX_DIV_BY_BASE) {
+ mantissa = (mantissa * BASE) + digit;
+ if (afterDecimal)
+ --exponent;
+ } else {
+ if (digit > 0)
+ truncated = true;
+ if (!afterDecimal)
+ ++exponent;
+ }
+ ++src;
+ continue;
+ }
if (*src == DECIMAL_POINT) {
if (afterDecimal) {
break; // this means that *src points to a second decimal point, ending
// the number.
- } else {
- afterDecimal = true;
- ++src;
- continue;
}
+ afterDecimal = true;
+ ++src;
+ continue;
}
- uint32_t digit = b36_char_to_int(*src);
- if (digit >= BASE)
- break;
-
- mantissa = (mantissa * BASE) + digit;
- seenDigit = true;
- if (afterDecimal)
- --exponent;
-
- ++src;
+ // The character is neither a hexadecimal digit nor a decimal point.
+ break;
}
if (!seenDigit)
return false;
- // The second loop is to run through the remaining digits after we've filled
- // the mantissa.
- while (isalnum(*src) || *src == DECIMAL_POINT) {
- if (*src == DECIMAL_POINT) {
- if (afterDecimal) {
- break; // this means that *src points to a second decimal point, ending
- // the number.
- } else {
- afterDecimal = true;
- ++src;
- continue;
- }
- }
- uint32_t digit = b36_char_to_int(*src);
- if (digit >= BASE)
- break;
-
- if (digit > 0)
- truncated = true;
-
- if (!afterDecimal)
- ++exponent;
-
- ++src;
- }
-
// Convert the exponent from having a base of 16 to having a base of 2.
exponent *= 4;
