[ChangeLog][QtCore][Global] Added the QT_NO_FLOAT16_OPERATORS macro in
order to work around a Microsoft <= VS2017 compiler bug that is
exposed when using std::bitset along with any Qt header that includes
<qfloat16.h>.
This is fixed in MSVC 2019[1], but the workaround is needed for
earlier versions.
In this case, cl.exe fails with
C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.10.25017\include\bitset(270): error C2666: 'operator /': 10 overloads have similar conversions
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qsize.h(364): note: could be 'const QSizeF operator /(const QSizeF &,qreal)'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qsize.h(194): note: or 'const QSize operator /(const QSize &,qreal)'
c:\qt\5.12.0\msvc2017_64\include\qtcore\qmargins.h(427): note: or 'QMarginsF operator /(const QMarginsF &,qreal)'
c:\qt\5.12.0\msvc2017_64\include\qtcore\qmargins.h(213): note: or 'QMargins operator /(const QMargins &,qreal)'
c:\qt\5.12.0\msvc2017_64\include\qtcore\qmargins.h(207): note: or 'QMargins operator /(const QMargins &,int)'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(205): note: or 'double operator /(int,qfloat16) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(205): note: or 'double operator /(qfloat16,int) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(195): note: or 'float operator /(float,qfloat16) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(195): note: or 'float operator /(qfloat16,float) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(194): note: or 'double operator /(double,qfloat16) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(194): note: or 'double operator /(qfloat16,double) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(193): note: or 'long double operator /(long double,qfloat16) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(193): note: or 'long double operator /(qfloat16,long double) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qfloat16.h(176): note: or 'qfloat16 operator /(qfloat16,qfloat16) noexcept'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qpoint.h(402): note: or 'const QPointF operator /(const QPointF &,qreal)'
C:\Qt\5.12.0\msvc2017_64\include\QtCore/qpoint.h(206): note: or 'const QPoint operator /(const QPoint &,qreal)'
C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.10.25017\include\bitset(270): note: or 'built-in C++ operator/(::size_t, )'
C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.10.25017\include\bitset(270): note: while trying to match the argument list '(::size_t, )'
C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.10.25017\include\bitset(266): note: while compiling class template member function 'std::bitset<8> &std::bitset<8>::set(::size_t,bool)'
C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.10.25017\include\bitset(39): note: see reference to function template instantiation 'std::bitset<8> &std::bitset<8>::set(::size_t,bool)' being compiled
..\Qt5.12.0-C2666\main.cpp(7): note: see reference to class template instantiation 'std::bitset<8>' being compiled
Invoke this workaround by defining the macro QT_NO_FLOAT16_OPERATORS
in user code prior to the inclusion of Qt includes in a translation
unit.
Arithmetic operators from qfloat16 will then not be present in that
compilation unit.
[1] https://developercommunity.visualstudio.com/content/problem/406329/compiler-error-c2666-when-using-stdbitset-and-cust.html
Task-number: QTBUG-72073
Change-Id: I58f8400bf933ad781d4213731695e20e0c482166
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
This is going to be the most common scenario for GCC: all recent
versions allow compiling F16C code in the same source, thus generating
better code.
MSVC is excluded unless AVX is already turned on by the user in the
mkspec file, because it fails to use the VEX-prefixed instructions for
everything else, printing a warning that it knows it should have done
so.
ICC is excluded because it's known to generate invalid code when using
the F16C intrinsics unless F16C is turned explicitly on.
Clang is excluded because it runs into an internal error compiling this
code unless F16C is turned explicitly on.
Change-Id: I57a1bd6e0c194530b732fffd14f58de6a46f0de1
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
An over-correction snuck in at some point.
Change-Id: Ib67c4f6ecf8e6e244a0598c025a7cb7c15401070
Reviewed-by: Eirik Aavitsland <eirik.aavitsland@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Both ARM and x86 can convert fp16 much faster in bulk than one at a
time. This also enables hardware accelerated conversion on x86, when
F16C isn't unconditionally available at compile time.
This code is implemented in C to ensure that there's no leakage of
inline symbols from the .obj file that was compiled by Visual Studio
with AVX support. Unfortunately, simd.prf uses $(CXX) instead of $(CC)
for all its sources, which means the file gets interpreted as C++ by
g++, clang++ and icpc. Those compilers at least don't leak any symbols.
Done-with: Thiago Macieira <thiago.macieira@intel.com>
Change-Id: I9d26d99e83392861fb09564e0e8e8d76cd8483b3
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
C++ does not specify which kind of floating point implementation is
being used. The C Standard doesn't either, but it includes a normative
reference for implementations adoping it (ISO/IEC 9899:2011 Annex F).
There are a few existing checks in qfloat16.cpp; move them to qglobal.cpp
(next to the other, similar checks), and improve them by actually
checking that the radix used for floating point numbers is 2.
Change-Id: I704a3a8efeb51014b3be23fb236654d647a6f44f
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
GHS compiler is not fully compliant with iec559.
Therefore we need to replace is_iec559 assertion
with separate checks to build quint16.
Change-Id: I88c57e394b8d4e7899ee7d4a13cbfbac9436b2fc
Reviewed-by: Rolland Dudemaine <rolland@ghs.com>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This constitutes a fairly complete submission of an entirely new
floating point type which conforms to IEEE 754 as a 16-bit storage
class. Conversion between qfloat16 and float is currently performed
through a sequence of lookup tables. Global-level functions
qRound(), qRound64(), qFuzzyCompare(), qFuzzyIsNull(), and
qIsNull() each with a qfloat16 parameter have been included
for completeness.
[ChangeLog][QtCore] Added new qfloat16 class.
Change-Id: Ia52eb27846965c14f8140c00faf5ba33c9443976
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Keith Kyzivat
Allan Sandfeld Jensen
Lars Knoll
Thiago Macieira
Glen Mabey
Giuseppe D'Angelo
Kimmo Ollila