[Peter Dimov]
lexical_cast does a good job of optimizing the from_chars and to_chars
cases when it detects it can do so, but that by definition can't be better
than the programmer just calling the primitive operations from_chars
and to_chars directly, as they are (a) locale-independent, (b) non-
allocating, (c) non-throwing.
I wondered how lexical_cast's "good job" (limited by the algorithms at the time, not its fault!) compared to charconv, so I dusted off the performance comparison I wrote for CppCon 2019 and added lexical_cast.
MSVC's charconv is approximately 14x (for float) and 25x (for double) faster - that's times, not percent. (Thanks to Ryu by Ulf Adams.) This is with avoiding dynamic memory allocation in lexical_cast by converting into a small std::array. My absolute numbers (in a Ryzen 5950X VM):
452.3 ns | Boost float
1003.0 ns | Boost double
32.1 ns | STL float plain shortest
39.9 ns | STL double plain shortest
Here's my full test case, built with VS 2022 17.6 Preview 7 x64 (note that x64 vs. x86 makes a HUGE difference to charconv). I didn't bother to compare the from_chars scenario (as MSVC doesn't use a highly optimized algorithm there; investigating Eisel-Lemire is on our todo list).
C:\Temp>type charconv.cpp
// cl /std:c++17 /EHsc /nologo /W4 /MT /O2 /Iboost_1_82_0 charconv.cpp && charconv
#ifndef _MSC_VER
#define AVOID_CHARCONV
#define AVOID_SPRINTF_S
#endif // _MSC_VER
#ifndef AVOID_CHARCONV
#include <charconv>
#endif // AVOID_CHARCONV
#include <array>
#include <chrono>
#include <exception>
#include <random>
#include
#include
#include
#include
#include
#include
#include <vector>
#include
using namespace std;
using namespace std::chrono;
void verify(const bool b) {
if (!b) {
puts("FAIL");
exit(EXIT_FAILURE);
}
}
enum class RoundTrip { Sci, Fix, Gen, Hex, Lossy };
constexpr size_t N = 2'000'000; // how many floating-point values to test
constexpr size_t K = 5; // how many times to repeat the test, for cleaner timing
constexpr size_t BufSize = 2'000; // more than enough
unsigned int global_dummy = 0;
template <typename Floating>
void test_lexical_cast(const char* const str, const vector<Floating>& vec) {
const auto start = steady_clock::now();
for (size_t k = 0; k < K; ++k) {
for (const auto& elem : vec) {
const auto ret = boost::lexical_cast>(elem); // "-1.2345678901234567e-100" plus null term
global_dummy += static_cast<unsigned int>(ret[0]);
}
}
const auto finish = steady_clock::now();
printf("%6.1f ns | %s\n", duration{finish - start}.count() / (N * K), str);
}
template <typename Floating>
int sprintf_wrapper(char (&buf)[BufSize], const char* const fmt, const Floating elem) {
#ifdef AVOID_SPRINTF_S
return sprintf(buf, fmt, elem);
#else // AVOID_SPRINTF_S
return sprintf_s(buf, BufSize, fmt, elem);
#endif // AVOID_SPRINTF_S
}
template
void test_sprintf(const char* const str, const vector<Floating>& vec, const char* const fmt) {
char buf[BufSize];
const auto start = steady_clock::now();
for (size_t k = 0; k < K; ++k) {
for (const auto& elem : vec) {
const int ret = sprintf_wrapper(buf, fmt, elem);
global_dummy += static_cast<unsigned int>(ret);
global_dummy += static_cast<unsigned int>(buf[0]);
}
}
const auto finish = steady_clock::now();
printf("%6.1f ns | %s\n", duration{finish - start}.count() / (N * K), str);
for (const auto& elem : vec) {
verify(sprintf_wrapper(buf, fmt, elem) != -1);
if constexpr (RT == RoundTrip::Lossy) {
// skip lossy conversions
} else if constexpr (is_same_v) {
verify(strtof(buf, nullptr) == elem);
} else {
verify(strtod(buf, nullptr) == elem);
}
}
}
#ifndef AVOID_CHARCONV
constexpr chars_format chars_format_from_RoundTrip(const RoundTrip rt) {
switch (rt) {
case RoundTrip::Sci:
return chars_format::scientific;
case RoundTrip::Fix:
return chars_format::fixed;
case RoundTrip::Gen:
return chars_format::general;
case RoundTrip::Hex:
return chars_format::hex;
case RoundTrip::Lossy:
default:
puts("FAIL");
exit(EXIT_FAILURE);
}
}
template
void test_to_chars(const char* const str, const vector<Floating>& vec, const Args&... args) {
char buf[BufSize];
const auto start = steady_clock::now();
for (size_t k = 0; k < K; ++k) {
for (const auto& elem : vec) {
const auto result = to_chars(buf, buf + BufSize, elem, args...);
global_dummy += static_cast<unsigned int>(result.ptr - buf);
global_dummy += static_cast<unsigned int>(buf[0]);
}
}
const auto finish = steady_clock::now();
printf("%6.1f ns | %s\n", duration{finish - start}.count() / (N * K), str);
for (const auto& elem : vec) {
const auto result = to_chars(buf, buf + BufSize, elem, args...);
verify(result.ec == errc{});
if constexpr (RT == RoundTrip::Lossy) {
// skip lossy conversions
} else {
Floating round_trip;
const auto from_result = from_chars(buf, result.ptr, round_trip, chars_format_from_RoundTrip(RT));
verify(from_result.ec == errc{});
verify(from_result.ptr == result.ptr);
verify(round_trip == elem);
}
}
}
#endif // AVOID_CHARCONV
template
vector<char> prepare_strings(const vector<Floating>& vec) {
vector<char> output;
char buf[BufSize];
for (const auto& elem : vec) {
int ret;
if constexpr (RT == RoundTrip::Sci) {
if constexpr (is_same_v) {
ret = sprintf_wrapper(buf, "%.8e", elem);
} else {
ret = sprintf_wrapper(buf, "%.16e", elem);
}
} else {
static_assert(RT == RoundTrip::Hex);
if constexpr (is_same_v) {
ret = sprintf_wrapper(buf, "%.6a", elem);
} else {
ret = sprintf_wrapper(buf, "%.13a", elem);
}
}
verify(ret != -1);
output.insert(output.end(), buf, buf + ret + 1); // include null terminator
}
return output;
}
template <typename Floating>
void test_strtox(const char* const str, const vector<Floating>& original, const vector<char>& strings) {
vector<Floating> round_trip(N);
const auto start = steady_clock::now();
for (size_t k = 0; k < K; ++k) {
const char* ptr = strings.data();
char* endptr = nullptr;
for (size_t n = 0; n < N; ++n) {
if constexpr (is_same_v) {
round_trip[n] = strtof(ptr, &endptr);
} else {
round_trip[n] = strtod(ptr, &endptr);
}
ptr = endptr + 1; // advance past null terminator
}
}
const auto finish = steady_clock::now();
printf("%6.1f ns | %s\n", duration{finish - start}.count() / (N * K), str);
verify(round_trip == original);
}
#ifndef AVOID_CHARCONV
vector<char> erase_0x(const vector<char>& strings) {
vector<char> output;
output.reserve(strings.size() - 2 * N);
for (auto i = strings.begin(); i != strings.end();) {
if (*i == '-') {
output.push_back('-');
i += 3; // advance past "-0x";
} else {
i += 2; // advance past "0x";
}
for (;;) {
const char c = *i++;
output.push_back(c);
if (c == '\0') {
break;
}
}
}
return output;
}
template
void test_from_chars(const char* const str, const vector<Floating>& original, const vector<char>& strings) {
const char* const last = strings.data() + strings.size();
vector<Floating> round_trip(N);
const auto start = steady_clock::now();
for (size_t k = 0; k < K; ++k) {
const char* first = strings.data();
for (size_t n = 0; n < N; ++n) {
const auto from_result = from_chars(first, last, round_trip[n], chars_format_from_RoundTrip(RT));
first = from_result.ptr + 1; // advance past null terminator
}
}
const auto finish = steady_clock::now();
printf("%6.1f ns | %s\n", duration{finish - start}.count() / (N * K), str);
verify(round_trip == original);
}
#endif // AVOID_CHARCONV
void test_all() {
#if defined(__clang__) && defined(_M_IX86)
const char* const toolset = "Clang/LLVM x86 + MSVC STL";
#elif defined(__clang__) && defined(_M_X64)
const char* const toolset = "Clang/LLVM x64 + MSVC STL";
#elif !defined(__clang__) && defined(_M_IX86)
const char* const toolset = "C1XX/C2 x86 + MSVC STL";
#elif !defined(__clang__) && defined(_M_X64)
const char* const toolset = "C1XX/C2 x64 + MSVC STL";
#else
const char* const toolset = "Unknown Toolset";
#endif
puts(toolset);
vector<float> vec_flt;
vector<double> vec_dbl;
{
mt19937_64 mt64;
vec_flt.reserve(N);
while (vec_flt.size() < N) {
const uint32_t val = static_cast(mt64());
constexpr uint32_t inf_nan = 0x7F800000U;
if ((val & inf_nan) == inf_nan) {
continue; // skip INF/NAN
}
float flt;
static_assert(sizeof(flt) == sizeof(val));
memcpy(&flt, &val, sizeof(flt));
vec_flt.push_back(flt);
}
vec_dbl.reserve(N);
while (vec_dbl.size() < N) {
const uint64_t val = mt64();
constexpr uint64_t inf_nan = 0x7FF0000000000000ULL;
if ((val & inf_nan) == inf_nan) {
continue; // skip INF/NAN
}
double dbl;
static_assert(sizeof(dbl) == sizeof(val));
memcpy(&dbl, &val, sizeof(dbl));
vec_dbl.push_back(dbl);
}
}
test_lexical_cast("Boost float", vec_flt);
test_lexical_cast("Boost double", vec_dbl);
test_sprintfRoundTrip::Sci("CRT float scientific 8", vec_flt, "%.8e");
test_sprintfRoundTrip::Sci("CRT double scientific 16", vec_dbl, "%.16e");
test_sprintfRoundTrip::Lossy("CRT float fixed 6 (lossy)", vec_flt, "%f");
test_sprintfRoundTrip::Lossy("CRT double fixed 6 (lossy)", vec_dbl, "%f");
test_sprintfRoundTrip::Gen("CRT float general 9", vec_flt, "%.9g");
test_sprintfRoundTrip::Gen("CRT double general 17", vec_dbl, "%.17g");
test_sprintfRoundTrip::Hex("CRT float hex 6", vec_flt, "%.6a");
test_sprintfRoundTrip::Hex("CRT double hex 13", vec_dbl, "%.13a");
#ifndef AVOID_CHARCONV
test_to_charsRoundTrip::Gen("STL float plain shortest", vec_flt);
test_to_charsRoundTrip::Gen("STL double plain shortest", vec_dbl);
test_to_charsRoundTrip::Sci("STL float scientific shortest", vec_flt, chars_format::scientific);
test_to_charsRoundTrip::Sci("STL double scientific shortest", vec_dbl, chars_format::scientific);
test_to_charsRoundTrip::Fix("STL float fixed shortest", vec_flt, chars_format::fixed);
test_to_charsRoundTrip::Fix("STL double fixed shortest", vec_dbl, chars_format::fixed);
test_to_charsRoundTrip::Gen("STL float general shortest", vec_flt, chars_format::general);
test_to_charsRoundTrip::Gen("STL double general shortest", vec_dbl, chars_format::general);
test_to_charsRoundTrip::Hex("STL float hex shortest", vec_flt, chars_format::hex);
test_to_charsRoundTrip::Hex("STL double hex shortest", vec_dbl, chars_format::hex);
test_to_charsRoundTrip::Sci("STL float scientific 8", vec_flt, chars_format::scientific, 8);
test_to_charsRoundTrip::Sci("STL double scientific 16", vec_dbl, chars_format::scientific, 16);
test_to_charsRoundTrip::Lossy("STL float fixed 6 (lossy)", vec_flt, chars_format::fixed, 6);
test_to_charsRoundTrip::Lossy("STL double fixed 6 (lossy)", vec_dbl, chars_format::fixed, 6);
test_to_charsRoundTrip::Gen("STL float general 9", vec_flt, chars_format::general, 9);
test_to_charsRoundTrip::Gen("STL double general 17", vec_dbl, chars_format::general, 17);
test_to_charsRoundTrip::Hex("STL float hex 6", vec_flt, chars_format::hex, 6);
test_to_charsRoundTrip::Hex("STL double hex 13", vec_dbl, chars_format::hex, 13);
#endif // AVOID_CHARCONV
puts("----------");
const vector<char> strings_sci_flt = prepare_stringsRoundTrip::Sci(vec_flt);
const vector<char> strings_sci_dbl = prepare_stringsRoundTrip::Sci(vec_dbl);
const vector<char> strings_hex_flt = prepare_stringsRoundTrip::Hex(vec_flt);
const vector<char> strings_hex_dbl = prepare_stringsRoundTrip::Hex(vec_dbl);
test_strtox("CRT strtof float scientific", vec_flt, strings_sci_flt);
test_strtox("CRT strtod double scientific", vec_dbl, strings_sci_dbl);
test_strtox("CRT strtof float hex", vec_flt, strings_hex_flt);
test_strtox("CRT strtod double hex", vec_dbl, strings_hex_dbl);
#ifndef AVOID_CHARCONV
test_from_charsRoundTrip::Sci("STL from_chars float scientific", vec_flt, strings_sci_flt);
test_from_charsRoundTrip::Sci("STL from_chars double scientific", vec_dbl, strings_sci_dbl);
test_from_charsRoundTrip::Hex("STL from_chars float hex", vec_flt, erase_0x(strings_hex_flt));
test_from_charsRoundTrip::Hex("STL from_chars double hex", vec_dbl, erase_0x(strings_hex_dbl));
#endif // AVOID_CHARCONV
printf("global_dummy: %u\n", global_dummy);
}
int main() {
try {
test_all();
} catch (const exception& e) {
printf("Exception: %s\n", e.what());
} catch (...) {
printf("Unknown exception.\n");
}
}
C:\Temp>cl /std:c++17 /EHsc /nologo /W4 /MT /O2 /Iboost_1_82_0 charconv.cpp && charconv
charconv.cpp
C1XX/C2 x64 + MSVC STL
452.3 ns | Boost float
1003.0 ns | Boost double
222.7 ns | CRT float scientific 8
381.3 ns | CRT double scientific 16
231.3 ns | CRT float fixed 6 (lossy)
818.3 ns | CRT double fixed 6 (lossy)
260.1 ns | CRT float general 9
807.6 ns | CRT double general 17
96.0 ns | CRT float hex 6
117.4 ns | CRT double hex 13
32.1 ns | STL float plain shortest
39.9 ns | STL double plain shortest
31.2 ns | STL float scientific shortest
39.7 ns | STL double scientific shortest
37.6 ns | STL float fixed shortest
111.7 ns | STL double fixed shortest
31.8 ns | STL float general shortest
39.9 ns | STL double general shortest
14.9 ns | STL float hex shortest
18.3 ns | STL double hex shortest
48.9 ns | STL float scientific 8
57.4 ns | STL double scientific 16
35.2 ns | STL float fixed 6 (lossy)
93.2 ns | STL double fixed 6 (lossy)
59.4 ns | STL float general 9
70.4 ns | STL double general 17
16.2 ns | STL float hex 6
19.1 ns | STL double hex 13
----------
103.7 ns | CRT strtof float scientific
191.2 ns | CRT strtod double scientific
56.6 ns | CRT strtof float hex
86.0 ns | CRT strtod double hex
77.2 ns | STL from_chars float scientific
154.3 ns | STL from_chars double scientific
26.2 ns | STL from_chars float hex
30.1 ns | STL from_chars double hex
global_dummy: 3715056096
Hope this helps,
STL
