| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If -mavx2 is used, __AVX2__ is defined, which enables the F16C code
after commit 280e321e52fd4e86545f3f0d4bd4e047786a897e, but that was
wrong since we aren't allowed to use the F16C intrinsics with either
Clang or GCC (we can only do that with GCC 4.9 and Clang 4.8, and only
with an __attribute__ decoration).
With ICC and MSVC, we are allowed to use the intrinsics, but the
#include was missing.
[ChangeLog][QtCore] Fixed a compilation issue with qfloat16 if AVX2
support is enabled in the compiler. Since all processors that support
AVX2 also support F16C, for GCC and Clang it is recommended to either
add -mf16c to your build or to use the corresponding -march= switch.
Task-number: QTBUG-64529
Change-Id: I84e363d735b443cb9beefffd14b8ac1fd4baa978
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
|
|
|
|
|
| |
Change-Id: I65cd64dfc0ed357555e8b5276109303377a67e0e
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
The operator double() and operator long double() members of qfloat16
are causing cast ambiguities. This removes them, leaving only
operator float() which seems to be adequate.
Also, additional arithmetic operator tests were added which without
this removal fail to compile.
Change-Id: Id52a101b318fd754969b3de13c1e528d0aac2387
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
|
|
|
|
|
|
|
|
|
|
|
| |
That's before the return type or static, inline, constexpr or such
keywords (if any).
Perl Script:
s/^(\s+)(.*) Q_REQUIRED_RESULT(;)?(\s*\/\/.*)?$/\1Q_REQUIRED_RESULT \2\3\4/
Change-Id: I7814054a102a407d876ffffd14b6a16182f159e2
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
|
|
|
|
|
|
|
|
|
| |
_mm_cvtps_ph is usually defined as a macro:
qfloat16.h:122:37: error: use of old-style cast [-Werror=old-style-cast]
Change-Id: Icd0e0d4b27cb4e5eb892fffd14b516ec47826c0c
Reviewed-by: Ville Voutilainen <ville.voutilainen@qt.io>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Neither the Intel compiler nor Visual C++ have a dedicated switch to
enable F16C support, like GCC and Clang do. So we used the AVX switch
for that in commit 8241d51f7049d48912ce25fbd49ef4d22c58e340, as it was
the closest, lowest denominator. That was incorrect and insufficient.
The Intel compiler silently miscompiles the intrinsics with -xAVX,
making calls to out-of-line functions like _mm_cvtps_ph, which don't
exist. So we actually have to use AVX2 support to generate correct code.
That might be a problem later, since Ivy Bridge supports F16C but not
AVX2.
Visual C++ is able to generate F16C code with just -arch:AVX.
Either way, since there's no dedicated command-line switch, there's also
no dedicated preprocessor macro. We're using __AVX2__ for both
compilers, as that's a sufficient condition to indicate a processor that
supports F16C.
Change-Id: I27b55fdf514247549455fffd14b205b8d8b86da7
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
|
|
|
|
|
|
|
|
|
| |
MSVC, Apple's Clang and Clang prior to 3.9 do not recognize _cvtss_sh
and _cvtsh_ss. So expand the operation to use directly the packed
intrinsics.
Change-Id: I27b55fdf514247549455fffd14b2046fd638593d
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Clang implements the _cvtss_sh intrinsic by way of a macro, which
uses a C99 extension and that's not allowed in C++ mode:
float16.h:119:11: error: compound literals are a C99-specific feature [-Werror,-Wc99-extensions]
/usr/bin/../lib64/clang/3.9.1/include/f16cintrin.h:76:55: note: expanded from macro '_cvtss_sh'
Reported at https://bugs.llvm.org/show_bug.cgi?id=32491.
Change-Id: I27b55fdf514247549455fffd14b170df75dd4e1f
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
|
|
|
|
|
| |
Change-Id: Ie30fa13eb5f8eb44f0c9c21e9483c8aba40cec02
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
|
|
|
|
|
|
|
|
|
|
| |
Use F16C or ARM FP16 if available at compile time.
Configure check added because older clang compilers have F16C defines
and flags but not all the intrinsics.
Change-Id: I71f358b8fd003e70ab8fcf35097414591e485112
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
|
|
|
|
|
|
|
|
|
|
|
| |
We have observed that MSVC 2013 gets lost with the overloads that this
header adds, causing compilation bugs. This is believed to be a compiler
bug, but it's not something we can work around.
Task-number: QTBUG-58555
Change-Id: I536c32a88bff44dab37afffd14a1bad1d31dc16d
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
Reviewed-by: Glen Mabey <Glen.Mabey@swri.org>
|
|
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>
|