diff options
author | Thiago Macieira <thiago.macieira@intel.com> | 2021-09-28 17:26:09 -0700 |
---|---|---|
committer | Thiago Macieira <thiago.macieira@intel.com> | 2021-10-13 12:49:49 -0700 |
commit | 0df7831d01b03910ae74a432c2a07a29d99fc315 (patch) | |
tree | 70348515fd1812c3712b9815f661a42069e805c1 /src/corelib/global | |
parent | cb0ea3989598a83b2326f63e7da1669018a965e5 (diff) |
qfloat16: re-merge the F16C content into qfloat16.cpp
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>
Diffstat (limited to 'src/corelib/global')
-rw-r--r-- | src/corelib/global/qfloat16.cpp | 42 | ||||
-rw-r--r-- | src/corelib/global/qfloat16_f16c.c | 88 |
2 files changed, 29 insertions, 101 deletions
diff --git a/src/corelib/global/qfloat16.cpp b/src/corelib/global/qfloat16.cpp index 9ef197b3a1..834d82e0ea 100644 --- a/src/corelib/global/qfloat16.cpp +++ b/src/corelib/global/qfloat16.cpp @@ -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 - -f16cextern void qFloatToFloat16_fast(quint16 *out, const float *in, qsizetype len) noexcept; -f16cextern void qFloatFromFloat16_fast(float *out, const quint16 *in, qsizetype len) noexcept; +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); +} -#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 diff --git a/src/corelib/global/qfloat16_f16c.c b/src/corelib/global/qfloat16_f16c.c deleted file mode 100644 index d60a021bdb..0000000000 --- a/src/corelib/global/qfloat16_f16c.c +++ /dev/null @@ -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) |