AuroraMiddleware/qt5base-lts
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

qfloat16: re-merge the F16C content into qfloat16.cpp

Browse Source 
The bugs we had noticed previously are believed to be fixed. MSVC will
do per-function updating of AVX content where necessary and the ICC
issue is no longer relevant.

Change-Id: I2bbf422288924c198645fffd16a9235f2d73cc19
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
This commit is contained in:
Thiago Macieira 2021-09-28 17:26:09 -07:00
parent cb0ea39895
commit 0df7831d01
3 changed files with 29 additions and 133 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
src/corelib/CMakeLists.txt
View File

@ -558,23 +558,6 @@ qt_internal_extend_target(Core CONDITION GCC AND (CMAKE_CXX_COMPILER_VERSION VER
# CMAKE_INSTALL_DATA_DIR = "$$[QT_HOST_DATA]/"
# CMAKE_INSTALL_DATA_DIR_IS_ABSOLUTE = "True"
#### Keys ignored in scope 20:.:global:global/global.pri:QT_CPU_FEATURES__ss_QT_ARCH___contains___f16c:
# f16c_cxx = "true"
#### Keys ignored in scope 22:.:global:global/global.pri:CLANG OR intel_icl OR ICC:
# f16c_cxx = "false"
#### Keys ignored in scope 24:.:global:global/global.pri:GCC AND f16c AND x86SimdAlways:
# f16c_cxx = "true"
#### Keys ignored in scope 26:.:global:global/global.pri:MSVC AND QT_CPU_FEATURES__ss_QT_ARCH___contains___avx:
# f16c_cxx = "true"
#### Keys ignored in scope 27:.:global:global/global.pri:else:
# f16c_cxx = "false"
# qt_internal_extend_target(Core CONDITION _ss_f16c_cxx ... # special case remove: Handled manually
# special case begin
# Remove this because it's handled manually
#qt_internal_extend_target(Core CONDITION NOT _x_-armcc_x_ AND NOT cross_compile AND NOT static AND (hurd_x_ OR linux_x_)
@ -1291,21 +1274,6 @@ qt_internal_extend_target(Core CONDITION WASM
kernel/qeventdispatcher_wasm.cpp kernel/qeventdispatcher_wasm_p.h
)
# special case begin
# needs refinement for the other compiler specific cases.
if(TEST_arch_${TEST_architecture_arch}_subarch_f16c)
# Equivalent to contains(QT_CPU_FEATURES.$$QT_ARCH, f16c)
qt_internal_extend_target(Core
DEFINES
QFLOAT16_INCLUDE_FAST
)
else()
qt_internal_add_simd_part(Core SIMD f16c
SOURCES
global/qfloat16_f16c.c
)
endif()
set_source_files_properties(
thread/qmutex_mac.cpp
thread/qmutex_unix.cpp

42
src/corelib/global/qfloat16.cpp
View File

@ -197,17 +197,36 @@ static inline bool hasFastF16()
return qCpuHasFeature(F16C) && qCpuHasFeature(AVX);
}
extern "C" {
#ifdef QFLOAT16_INCLUDE_FAST
# define f16cextern static
#else
# define f16cextern extern
#endif
QT_FUNCTION_TARGET(F16C)
void qFloatToFloat16_fast(quint16 *out, const float *in, qsizetype len) Q_DECL_NOEXCEPT
{
qsizetype i = 0;
int epilog_i;
for (; i < len - 7; i += 8)
_mm_storeu_si128((__m128i *)(out + i), _mm256_cvtps_ph(_mm256_loadu_ps(in + i), 0));
if (i < len - 3) {
_mm_storel_epi64((__m128i *)(out + i), _mm_cvtps_ph(_mm_loadu_ps(in + i), 0));
i += 4;
}
// Inlining "qfloat16::qfloat16(float f)":
for (epilog_i = 0; i < len && epilog_i < 3; ++i, ++epilog_i)
out[i] = _mm_extract_epi16(_mm_cvtps_ph(_mm_set_ss(in[i]), 0), 0);
}
f16cextern void qFloatToFloat16_fast(quint16 *out, const float *in, qsizetype len) noexcept;
f16cextern void qFloatFromFloat16_fast(float *out, const quint16 *in, qsizetype len) noexcept;
#undef f16cextern
QT_FUNCTION_TARGET(F16C)
void qFloatFromFloat16_fast(float *out, const quint16 *in, qsizetype len) Q_DECL_NOEXCEPT
{
qsizetype i = 0;
int epilog_i;
for (; i < len - 7; i += 8)
_mm256_storeu_ps(out + i, _mm256_cvtph_ps(_mm_loadu_si128((const __m128i *)(in + i))));
if (i < len - 3) {
_mm_storeu_ps(out + i, _mm_cvtph_ps(_mm_loadl_epi64((const __m128i *)(in + i))));
i += 4;
}
// Inlining "qfloat16::operator float()":
for (epilog_i = 0; i < len && epilog_i < 3; ++i, ++epilog_i)
out[i] = _mm_cvtss_f32(_mm_cvtph_ps(_mm_cvtsi32_si128(in[i])));
}
#elif defined(__ARM_FP16_FORMAT_IEEE) && defined(__ARM_NEON__) && (__ARM_FP & 2)
@ -292,6 +311,3 @@ Q_CORE_EXPORT void qFloatFromFloat16(float *out, const qfloat16 *in, qsizetype l
QT_END_NAMESPACE
#include "qfloat16tables.cpp"
#ifdef QFLOAT16_INCLUDE_FAST
# include "qfloat16_f16c.c"
#endif

88
src/corelib/global/qfloat16_f16c.c
View File

@ -1,88 +0,0 @@
/****************************************************************************
**
** Copyright (C) 2017 The Qt Company Ltd.
** Contact: https://www.qt.io/licensing/
**
** This file is part of the QtCore module of the Qt Toolkit.
**
** $QT_BEGIN_LICENSE:LGPL$
** Commercial License Usage
** Licensees holding valid commercial Qt licenses may use this file in
** accordance with the commercial license agreement provided with the
** Software or, alternatively, in accordance with the terms contained in
** a written agreement between you and The Qt Company. For licensing terms
** and conditions see https://www.qt.io/terms-conditions. For further
** information use the contact form at https://www.qt.io/contact-us.
**
** GNU Lesser General Public License Usage
** Alternatively, this file may be used under the terms of the GNU Lesser
** General Public License version 3 as published by the Free Software
** Foundation and appearing in the file LICENSE.LGPL3 included in the
** packaging of this file. Please review the following information to
** ensure the GNU Lesser General Public License version 3 requirements
** will be met: https://www.gnu.org/licenses/lgpl-3.0.html.
**
** GNU General Public License Usage
** Alternatively, this file may be used under the terms of the GNU
** General Public License version 2.0 or (at your option) the GNU General
** Public license version 3 or any later version approved by the KDE Free
** Qt Foundation. The licenses are as published by the Free Software
** Foundation and appearing in the file LICENSE.GPL2 and LICENSE.GPL3
** included in the packaging of this file. Please review the following
** information to ensure the GNU General Public License requirements will
** be met: https://www.gnu.org/licenses/gpl-2.0.html and
** https://www.gnu.org/licenses/gpl-3.0.html.
**
** $QT_END_LICENSE$
**
****************************************************************************/
#include "private/qsimd_p.h"
// The x86 F16C instructions operate on AVX registers, so AVX support is
// required.
#if QT_COMPILER_SUPPORTS_HERE(AVX)
#ifdef __cplusplus
QT_BEGIN_NAMESPACE
extern "C" {
#endif
QT_FUNCTION_TARGET(F16C)
void qFloatToFloat16_fast(quint16 *out, const float *in, qsizetype len) Q_DECL_NOEXCEPT
{
qsizetype i = 0;
int epilog_i;
for (; i < len - 7; i += 8)
_mm_storeu_si128((__m128i *)(out + i), _mm256_cvtps_ph(_mm256_loadu_ps(in + i), 0));
if (i < len - 3) {
_mm_storel_epi64((__m128i *)(out + i), _mm_cvtps_ph(_mm_loadu_ps(in + i), 0));
i += 4;
}
// Inlining "qfloat16::qfloat16(float f)":
for (epilog_i = 0; i < len && epilog_i < 3; ++i, ++epilog_i)
out[i] = _mm_extract_epi16(_mm_cvtps_ph(_mm_set_ss(in[i]), 0), 0);
}
QT_FUNCTION_TARGET(F16C)
void qFloatFromFloat16_fast(float *out, const quint16 *in, qsizetype len) Q_DECL_NOEXCEPT
{
qsizetype i = 0;
int epilog_i;
for (; i < len - 7; i += 8)
_mm256_storeu_ps(out + i, _mm256_cvtph_ps(_mm_loadu_si128((const __m128i *)(in + i))));
if (i < len - 3) {
_mm_storeu_ps(out + i, _mm_cvtph_ps(_mm_loadl_epi64((const __m128i *)(in + i))));
i += 4;
}
// Inlining "qfloat16::operator float()":
for (epilog_i = 0; i < len && epilog_i < 3; ++i, ++epilog_i)
out[i] = _mm_cvtss_f32(_mm_cvtph_ps(_mm_cvtsi32_si128(in[i])));
}
#ifdef __cplusplus
} // extern "C"
QT_END_NAMESPACE
#endif
#endif // QT_COMPILER_SUPPORTS_HERE(AVX)