darling-JavaScriptCore/runtime/PureNaN.h
2017-08-12 09:48:01 -07:00

96 lines
4.6 KiB
C++

/*
* Copyright (C) 2014 Apple Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#pragma once
#include <wtf/Assertions.h>
#include <wtf/StdLibExtras.h>
namespace JSC {
// NaN (not-a-number) double values are central to how JavaScriptCore encodes JavaScript
// values (JSValues). All values, including integers and non-numeric values, are always
// encoded using the IEEE 754 binary double format. Non-double values are encoded using
// a NaN with the sign bit set. The 51-bit payload is then used for encoding the actual
// value - be it an integer or a pointer to an object, or something else. But we only
// make use of the low 49 bits and the top 15 bits being all set to 1 is the indicator
// that a value is not a double. Top 15 bits being set to 1 also indicate a signed
// signaling NaN with some additional NaN payload bits.
//
// Our use of NaN encoding means that we have to be careful with how we use NaNs for
// ordinary doubles. For example, it would be wrong to ever use a NaN that has the top
// 15 bits set, as that would look like a non-double value to JSC.
//
// We can trust that on all of the hardware/OS combinations that we care about,
// NaN-producing math operations never produce a NaN that looks like a tagged value. But
// if we're ever in a situation where we worry about it, we can use purifyNaN() to get a
// NaN that doesn't look like a tagged non-double value. The JavaScript language doesn't
// distinguish between different flavors of NaN and there is no way to detect what kind
// of NaN you have - hence so long as all double NaNs are purified then our tagging
// scheme remains sound.
//
// It's worth noting that there are cases, like sin(), that will almost produce a NaN
// that breaks us. sin(-inf) returns 0xfff8000000000000. This doesn't break us because
// not all of the top 15 bits are set. But it's very close. Hence our assumptions about
// NaN are just about the most aggressive assumptions we could possibly make without
// having to call purifyNaN() in surprising places.
//
// For naming purposes, we say that a NaN is "pure" if it is safe to tag, in the sense
// that doing so would result in a tagged value that would pass the "are you a double"
// test. We say that a NaN is "impure" if attempting to tag it would result in a value
// that would look like something other than a double.
// Returns some kind of pure NaN.
inline double pureNaN()
{
// Be sure that we return exactly the kind of NaN that is safe. We engineer the bits
// ourselves to ensure that it's !isImpureNaN(). FWIW, this is what
// numeric_limits<double>::quiet_NaN() returns on Mac/X86_64. But AFAICT there is
// no guarantee that quiet_NaN would return a pureNaN on all platforms. For example,
// the docs appear to imply that quiet_NaN could even return a double with the
// signaling bit set on hardware that doesn't do signaling. That would probably
// never happen, but it's healthy to be paranoid.
return bitwise_cast<double>(0x7ff8000000000000ll);
}
#define PNaN (pureNaN())
inline bool isImpureNaN(double value)
{
// Tests if the double value would break JSVALUE64 encoding, which is the most
// aggressive kind of encoding that we currently use.
return bitwise_cast<uint64_t>(value) >= 0xfffe000000000000llu;
}
// If the given value is NaN then return a NaN that is known to be pure.
inline double purifyNaN(double value)
{
if (value != value)
return PNaN;
return value;
}
} // namespace JSC