mirror of
https://codeberg.org/ziglang/zig.git
synced 2026-04-28 11:27:09 +03:00
Marc Tiehuis
2085a4af56
The previous float-parsing method was lacking in a lot of areas. This commit introduces a state-of-the art implementation that is both accurate and fast to std. Code is derived from working repo https://github.com/tiehuis/zig-parsefloat. This includes more test-cases and performance numbers that are present in this commit. * Accuracy The primary testing regime has been using test-data found at https://github.com/tiehuis/parse-number-fxx-test-data. This is a fork of upstream with support for f128 test-cases added. This data has been verified against other independent implementations and represents accurate round-to-even IEEE-754 floating point semantics. * Performance Compared to the existing parseFloat implementation there is ~5-10x performance improvement using the above corpus. (f128 parsing excluded in below measurements). ** Old $ time ./test_all_fxx_data 3520298/5296694 succeeded (1776396 fail) ________________________________________________________ Executed in 28.68 secs fish external usr time 28.48 secs 0.00 micros 28.48 secs sys time 0.08 secs 694.00 micros 0.08 secs ** This Implementation $ time ./test_all_fxx_data 5296693/5296694 succeeded (1 fail) ________________________________________________________ Executed in 4.54 secs fish external usr time 4.37 secs 515.00 micros 4.37 secs sys time 0.10 secs 171.00 micros 0.10 secs Further performance numbers can be seen using the https://github.com/tiehuis/simple_fastfloat_benchmark/ repository, which compares against some other well-known string-to-float conversion functions. A breakdown can be found here: https://github.com/tiehuis/zig-parsefloat/blob/0d9f020f1a37ca88bf889703b397c1c41779f090/PERFORMANCE.md#commit-b15406a0d2e18b50a4b62fceb5a6a3bb60ca5706 In summary, we are within 20% of the C++ reference implementation and have about ~600-700MB/s throughput on a Intel I5-6500 3.5Ghz. * F128 Support Finally, f128 is now completely supported with full accuracy. This does use a slower path which is possible to improve in future. * Behavioural Changes There are a few behavioural changes to note. - `parseHexFloat` is now redundant and these are now supported directly in `parseFloat`. - We implement round-to-even in all parsing routines. This is as specified by IEEE-754. Previous code used different rounding mechanisms (standard was round-to-zero, hex-parsing looked to use round-up) so there may be subtle differences. Closes #2207. Fixes #11169.
92 lines
2.8 KiB
Zig
92 lines
2.8 KiB
Zig
const std = @import("std");
|
|
|
|
/// A custom N-bit floating point type, representing `f * 2^e`.
|
|
/// e is biased, so it be directly shifted into the exponent bits.
|
|
/// Negative exponent indicates an invalid result.
|
|
pub fn BiasedFp(comptime T: type) type {
|
|
const MantissaT = mantissaType(T);
|
|
|
|
return struct {
|
|
const Self = @This();
|
|
|
|
/// The significant digits.
|
|
f: MantissaT,
|
|
/// The biased, binary exponent.
|
|
e: i32,
|
|
|
|
pub fn zero() Self {
|
|
return .{ .f = 0, .e = 0 };
|
|
}
|
|
|
|
pub fn zeroPow2(e: i32) Self {
|
|
return .{ .f = 0, .e = e };
|
|
}
|
|
|
|
pub fn inf(comptime FloatT: type) Self {
|
|
return .{ .f = 0, .e = (1 << std.math.floatExponentBits(FloatT)) - 1 };
|
|
}
|
|
|
|
pub fn eql(self: Self, other: Self) bool {
|
|
return self.f == other.f and self.e == other.e;
|
|
}
|
|
|
|
pub fn toFloat(self: Self, comptime FloatT: type, negative: bool) FloatT {
|
|
var word = self.f;
|
|
word |= @intCast(MantissaT, self.e) << std.math.floatMantissaBits(FloatT);
|
|
var f = floatFromUnsigned(FloatT, MantissaT, word);
|
|
if (negative) f = -f;
|
|
return f;
|
|
}
|
|
};
|
|
}
|
|
|
|
pub fn floatFromUnsigned(comptime T: type, comptime MantissaT: type, v: MantissaT) T {
|
|
return switch (T) {
|
|
f16 => @bitCast(f16, @truncate(u16, v)),
|
|
f32 => @bitCast(f32, @truncate(u32, v)),
|
|
f64 => @bitCast(f64, @truncate(u64, v)),
|
|
f128 => @bitCast(f128, v),
|
|
else => unreachable,
|
|
};
|
|
}
|
|
|
|
/// Represents a parsed floating point value as its components.
|
|
pub fn Number(comptime T: type) type {
|
|
return struct {
|
|
exponent: i64,
|
|
mantissa: mantissaType(T),
|
|
negative: bool,
|
|
/// More than max_mantissa digits were found during parse
|
|
many_digits: bool,
|
|
/// The number was a hex-float (e.g. 0x1.234p567)
|
|
hex: bool,
|
|
};
|
|
}
|
|
|
|
/// Determine if 8 bytes are all decimal digits.
|
|
/// This does not care about the order in which the bytes were loaded.
|
|
pub fn isEightDigits(v: u64) bool {
|
|
const a = v +% 0x4646_4646_4646_4646;
|
|
const b = v -% 0x3030_3030_3030_3030;
|
|
return ((a | b) & 0x8080_8080_8080_8080) == 0;
|
|
}
|
|
|
|
pub fn isDigit(c: u8, comptime base: u8) bool {
|
|
std.debug.assert(base == 10 or base == 16);
|
|
|
|
return if (base == 10)
|
|
'0' <= c and c <= '9'
|
|
else
|
|
'0' <= c and c <= '9' or 'a' <= c and c <= 'f' or 'A' <= c and c <= 'F';
|
|
}
|
|
|
|
/// Returns the underlying storage type used for the mantissa of floating-point type.
|
|
/// The output unsigned type must have at least as many bits as the input floating-point type.
|
|
pub fn mantissaType(comptime T: type) type {
|
|
return switch (T) {
|
|
f16, f32, f64 => u64,
|
|
f128 => u128,
|
|
else => unreachable,
|
|
};
|
|
}
|