diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/corelib/tools/qsimd_p.h | 43 | ||||
| -rw-r--r-- | src/gui/painting/painting.pri | 2 | ||||
| -rw-r--r-- | src/gui/painting/qdrawhelper.cpp | 2 |
3 files changed, 44 insertions, 3 deletions
diff --git a/src/corelib/tools/qsimd_p.h b/src/corelib/tools/qsimd_p.h index af262ec88f..9f1321df94 100644 --- a/src/corelib/tools/qsimd_p.h +++ b/src/corelib/tools/qsimd_p.h @@ -1,7 +1,7 @@ /**************************************************************************** ** ** Copyright (C) 2016 The Qt Company Ltd. -** Copyright (C) 2016 Intel Corporation. +** Copyright (C) 2018 Intel Corporation. ** Contact: https://www.qt.io/licensing/ ** ** This file is part of the QtCore module of the Qt Toolkit. @@ -232,8 +232,49 @@ # define __RDRND__ 1 # endif +# if defined(__BMI__) && !defined(__BMI2__) && defined(Q_CC_INTEL) +// BMI2 instructions: +// All processors that support BMI support BMI2 (and AVX2) +// (but neither MSVC nor the Intel compiler define this macro) +# define __BMI2__ 1 +# endif + # include "qsimd_x86_p.h" +// Haswell sub-architecture +// +// The Intel Core 4th generation was codenamed "Haswell" and introduced AVX2, +// BMI1, BMI2, FMA, LZCNT, MOVBE, which makes it a good divider for a +// sub-target for us. The first AMD processor with AVX2 support (Zen) has the +// same features. +// +// macOS's fat binaries support the "x86_64h" sub-architecture and the GNU libc +// ELF loader also supports a "haswell/" subdir (e.g., /usr/lib/haswell). +# define QT_FUNCTION_TARGET_STRING_ARCH_HASWELL "arch=haswell" +# if defined(__AVX2__) && defined(__BMI__) && defined(__BMI2__) && defined(__F16C__) && \ + defined(__FMA__) && defined(__LZCNT__) && defined(__RDRND__) +# define __haswell__ 1 +# endif + +// This constant does not include all CPU features found in a Haswell, only +// those that we'd have optimized code for. +// Note: must use Q_CONSTEXPR here, as this file may be compiled in C mode. +QT_BEGIN_NAMESPACE +static const quint64 CpuFeatureArchHaswell = 0 + | CpuFeatureSSE2 + | CpuFeatureSSE3 + | CpuFeatureSSSE3 + | CpuFeatureSSE4_1 + | CpuFeatureSSE4_2 + | CpuFeatureFMA + | CpuFeaturePOPCNT + | CpuFeatureAVX + | CpuFeatureF16C + | CpuFeatureAVX2 + | CpuFeatureBMI + | CpuFeatureBMI2; +QT_END_NAMESPACE + #endif /* Q_PROCESSOR_X86 */ // Clang compiler fix, see http://lists.llvm.org/pipermail/cfe-commits/Week-of-Mon-20160222/151168.html diff --git a/src/gui/painting/painting.pri b/src/gui/painting/painting.pri index 749cc221ff..c3585a4647 100644 --- a/src/gui/painting/painting.pri +++ b/src/gui/painting/painting.pri @@ -127,7 +127,7 @@ SSE2_SOURCES += painting/qdrawhelper_sse2.cpp SSSE3_SOURCES += painting/qdrawhelper_ssse3.cpp SSE4_1_SOURCES += painting/qdrawhelper_sse4.cpp \ painting/qimagescale_sse4.cpp -AVX2_SOURCES += painting/qdrawhelper_avx2.cpp +ARCH_HASWELL_SOURCES += painting/qdrawhelper_avx2.cpp NEON_SOURCES += painting/qdrawhelper_neon.cpp painting/qimagescale_neon.cpp NEON_HEADERS += painting/qdrawhelper_neon_p.h diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp index 5892d59a64..235bba7206 100644 --- a/src/gui/painting/qdrawhelper.cpp +++ b/src/gui/painting/qdrawhelper.cpp @@ -5772,7 +5772,7 @@ static void qInitDrawhelperFunctions() #endif #if defined(QT_COMPILER_SUPPORTS_AVX2) - if (qCpuHasFeature(AVX2)) { + if (qCpuHasFeature(ArchHaswell)) { extern void qt_blend_rgb32_on_rgb32_avx2(uchar *destPixels, int dbpl, const uchar *srcPixels, int sbpl, int w, int h, int const_alpha); |