configure: Add support for detecting AVX512 instructions

Tested on Linux with Clang 3.7, GCC 4.9, 5.1 and 6.0, ICC 16 beta; on OS
X with Clang-XCode 6.4, ICC 16 beta; on Windows with MSVC 2013 and ICC
15. MinGW is not tested.

GCC 4.9:     AVX512F AVX512ER AVX512CD AVX512PF
GCC 5 & 6:   AVX512F AVX512ER AVX512CD AVX512PF AVX512DQ AVX512BW AVX512VL AVX512IFMA AVX512VBMI
Clang 3.7:   AVX512F AVX512ER AVX512CD
Clang-XCode: <none>
ICC 15 & 16: AVX512F AVX512ER AVX512CD AVX512PF AVX512DQ AVX512BW AVX512VL
MSVC 2013:   <none>

Change-Id: Ib306f8f647014b399b87ffff13f1da1b161c31d7
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>

config.tests/common/avx512/avx512.cpp

@@ -0,0 +1,97 @@
+/****************************************************************************
+**
+** Copyright (C) 2012 Intel Corporation.
+** Contact: http://www.qt.io/licensing/
+**
+** This file is part of the configuration of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL21$
+** 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 http://www.qt.io/terms-conditions. For further
+** information use the contact form at http://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 2.1 or version 3 as published by the Free
+** Software Foundation and appearing in the file LICENSE.LGPLv21 and
+** LICENSE.LGPLv3 included in the packaging of this file. Please review the
+** following information to ensure the GNU Lesser General Public License
+** requirements will be met: https://www.gnu.org/licenses/lgpl.html and
+** http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
+**
+** As a special exception, The Qt Company gives you certain additional
+** rights. These rights are described in The Qt Company LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#include <immintrin.h>
+
+#ifndef AVX512WANT
+#  error ".pro file must define AVX512WANT macro to the AVX-512 feature to be tested"
+#endif
+
+// The following checks if __AVXx__ is defined, where x is the value in
+// AVX512WANT
+#define HAS2(x)             __AVX512 ## x ## __
+#define HAS(x)              HAS2(x)
+#if !HAS(AVX512WANT)
+#  error "Feature not supported"
+#endif
+
+int main(int, char**argv)
+{
+    /* AVX512 Foundation */
+    __m512i i;
+    __m512d d;
+    __m512 f;
+    __mmask16 m = ~1;
+    i = _mm512_maskz_loadu_epi32(0, argv);
+    d = _mm512_loadu_pd((double *)argv + 64);
+    f = _mm512_loadu_ps((float *)argv + 128);
+
+#ifdef __AVX512ER__
+    /* AVX512 Exponential and Reciprocal */
+    f =  _mm512_exp2a23_round_ps(f, 8);
+#endif
+#ifdef __AVX512CD__
+    /* AVX512 Conflict Detection */
+    i = _mm512_maskz_conflict_epi32(m, i);
+#endif
+#ifdef __AVX512PF__
+    /* AVX512 Prefetch */
+    _mm512_mask_prefetch_i64scatter_pd(argv, 0xf, i, 2, 2);
+#endif
+#ifdef __AVX512DQ__
+    /* AVX512 Doubleword and Quadword support */
+    m = _mm512_movepi32_mask(i);
+#endif
+#ifdef __AVX512BW__
+    /* AVX512 Byte and Word support */
+    i =  _mm512_mask_loadu_epi8(i, m, argv - 8);
+#endif
+#ifdef __AVX512VL__
+    /* AVX512 Vector Length */
+    __m256i i2 = _mm256_maskz_loadu_epi32(0, argv);
+    _mm256_mask_storeu_epi32(argv + 1, m, i2);
+#endif
+#ifdef __AVX512IFMA__
+    /* AVX512 Integer Fused Multiply-Add */
+    i = _mm512_madd52lo_epu64(i, i, i);
+#endif
+#ifdef __AVX512VBMI__
+    /* AVX512 Vector Byte Manipulation Instructions */
+    i = _mm512_permutexvar_epi8(i, i);
+#endif
+
+    _mm512_mask_storeu_epi64(argv, m, i);
+    _mm512_mask_storeu_ps(argv + 64, m, f);
+    _mm512_mask_storeu_pd(argv + 128, m, d);
+    return 0;
+}

config.tests/common/avx512/avx512.pro

@@ -0,0 +1,12 @@
+SOURCES = avx512.cpp
+CONFIG -= qt dylib release debug_and_release
+CONFIG += debug console
+
+isEmpty(AVX512): error("You must set the AVX512 variable!")
+
+varname = QMAKE_CFLAGS_AVX512$$AVX512
+value = $$eval($$varname)
+isEmpty($$varname): error("This compiler does not support AVX512")
+
+QMAKE_CXXFLAGS += $$value
+DEFINES += AVX512WANT=$$AVX512

configure

@@ -716,6 +716,7 @@ CFG_SSE4_1=auto
 CFG_SSE4_2=auto
 CFG_AVX=auto
 CFG_AVX2=auto
+CFG_AVX512=auto
 CFG_REDUCE_RELOCATIONS=auto
 CFG_ACCESSIBILITY=auto
 CFG_ACCESSIBILITY_ATSPI_BRIDGE=no # will be enabled depending on dbus and accessibility being enabled
@@ -1640,6 +1641,13 @@ while [ "$#" -gt 0 ]; do
             UNKNOWN_OPT=yes
         fi
         ;;
+    avx512)
+        if [  "$VAL" = "no" ]; then
+            CFG_AVX512=""
+        else
+            UNKNOWN_OPT=yes
+        fi
+        ;;
     mips_dsp)
         if [ "$VAL" = "no" ]; then
             CFG_MIPS_DSP="$VAL"
@@ -2463,6 +2471,7 @@ Configure options:
     -no-sse4.2 ......... Do not compile with use of SSE4.2 instructions.
     -no-avx ............ Do not compile with use of AVX instructions.
     -no-avx2 ........... Do not compile with use of AVX2 instructions.
+    -no-avx512 ......... Do not compile with use of AVX512 instructions.
     -no-mips_dsp ....... Do not compile with use of MIPS DSP instructions.
     -no-mips_dspr2 ..... Do not compile with use of MIPS DSP rev2 instructions.
 
@@ -4498,6 +4507,31 @@ if [ "${CFG_AVX2}" = "auto" ]; then
     fi
 fi
 
+# detect avx512 support
+if [ "${CFG_AVX512}" == "auto" ]; then
+    # First, test for AVX-512 Foundation
+    if compileTest common/avx512 "avx512f" AVX512=F; then
+        # Test for the sub-features
+        CFG_AVX512=f
+        CFG_AVX512_UPPER=AVX512F
+        for feature in er cd pf dq bw vl ifma vbmi; do
+            if [ -n "BASH_VERSION" ] && [ "${BASH_VERSION%%.*}" -gt 3 ]; then
+                upper=${feature^^*}
+            elif [ -n "$ZSH_VERSION" ]; then
+                upper=${(U)feature}
+            else
+                upper=`echo $feature | tr a-z A-Z`
+            fi
+            if compileTest common/avx512 "avx512$feature" AVX512=$upper; then
+                CFG_AVX512="$CFG_AVX512 $feature"
+                CFG_AVX512_UPPER="$CFG_AVX512_UPPER AVX512$upper"
+            fi
+        done
+    else
+        CFG_AVX512=
+    fi
+fi
+
 # check Neon support
 if [ "$CFG_NEON" = "auto" ]; then
     # no compile test, just check what the compiler has
@@ -6246,6 +6280,9 @@ fi
 [ "$CFG_SSE4_2" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG sse4_2"
 [ "$CFG_AVX" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG avx"
 [ "$CFG_AVX2" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG avx2"
+for feature in $CFG_AVX512; do
+    QMAKE_CONFIG="$QMAKE_CONFIG avx512$feature"
+done
 [ "$CFG_NEON" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG neon"
 if [ "$CFG_ARCH" = "mips" ]; then
     [ "$CFG_MIPS_DSP" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG mips_dsp"
@@ -6715,6 +6752,9 @@ for SUBARCH in SSE2 SSE3 SSSE3 SSE4_1 SSE4_2 AVX AVX2 \
             ;;
     esac
 done
+for feature in $CFG_AVX512_UPPER; do
+    echo "#define QT_COMPILER_SUPPORTS_$feature 1"  >>"$outpath/src/corelib/global/qconfig.h.new"
+done
 
 echo "" >>"$outpath/src/corelib/global/qconfig.h.new"
 
@@ -7167,6 +7207,7 @@ if [ "$CFG_ARCH" = "i386" -o "$CFG_ARCH" = "x86_64" ]; then
     echo "    SSE2/SSE3/SSSE3 ...... ${CFG_SSE2}/${CFG_SSE3}/${CFG_SSSE3}"
     echo "    SSE4.1/SSE4.2 ........ ${CFG_SSE4_1}/${CFG_SSE4_2}"
     echo "    AVX/AVX2 ............. ${CFG_AVX}/${CFG_AVX2}"
+    echo "    AVX512 ............... ${CFG_AVX512_UPPER:-<none>}"
 elif [ "$CFG_ARCH" = "arm" ]; then
     echo "    Neon ................. ${CFG_NEON}"
 elif [ "$CFG_ARCH" = "mips" ]; then

mkspecs/common/gcc-base.conf

@@ -86,6 +86,15 @@ QMAKE_CFLAGS_SSE4_1    += -msse4.1
 QMAKE_CFLAGS_SSE4_2    += -msse4.2
 QMAKE_CFLAGS_AVX       += -mavx
 QMAKE_CFLAGS_AVX2      += -mavx2
+QMAKE_CFLAGS_AVX512F   += -mavx512f
+QMAKE_CFLAGS_AVX512ER  += -mavx512er
+QMAKE_CFLAGS_AVX512CD  += -mavx512cd
+QMAKE_CFLAGS_AVX512PF  += -mavx512pf
+QMAKE_CFLAGS_AVX512DQ  += -mavx512dq
+QMAKE_CFLAGS_AVX512BW  += -mavx512bw
+QMAKE_CFLAGS_AVX512VL  += -mavx512vl
+QMAKE_CFLAGS_AVX512IFMA += -mavx512ifma
+QMAKE_CFLAGS_AVX512VBMI += -mavx512vbmi
 QMAKE_CFLAGS_NEON      += -mfpu=neon
 
 # Wrapper tools that understand .o/.a files with GIMPLE instead of machine code

mkspecs/features/simd.prf

@@ -95,6 +95,47 @@ addSimdCompiler(sse4_1)
 addSimdCompiler(sse4_2)
 addSimdCompiler(avx)
 addSimdCompiler(avx2)
+addSimdCompiler(avx512f)
+addSimdCompiler(avx512cd)
+addSimdCompiler(avx512er)
+addSimdCompiler(avx512pf)
+addSimdCompiler(avx512dq)
+addSimdCompiler(avx512bw)
+addSimdCompiler(avx512vl)
+addSimdCompiler(avx512ifma)
+addSimdCompiler(avx512vbmi)
 addSimdCompiler(neon)
 addSimdCompiler(mips_dsp)
 addSimdCompiler(mips_dspr2)
+
+# Follow the Intel compiler's lead and define profiles of AVX512 instructions
+defineTest(addAvx512Profile) {
+    name = $$1
+    dependencies = $$2
+    upname = $$upper($$name)
+    varname = QMAKE_CFLAGS_$$upname
+
+    cpu_features_missing =
+    cflags = $$QMAKE_CFLAGS_AVX512F
+    for(part, dependencies) {
+        !CONFIG($$part): return()           # Profile isn't supported by the compiler
+
+        uppart = $$upper($$part)
+        cflags *= $$eval(QMAKE_CFLAGS_$${uppart})
+        !contains(QT_CPU_FEATURES, $$uppart): cpu_features_missing += $$uppart
+    }
+
+    CONFIG += $$name
+    isEmpty(cpu_features_missing): QT_CPU_FEATURES += $$name
+    $$varname = $$cflags
+
+    export(QT_CPU_FEATURES)
+    export(CONFIG)
+    export($$varname)
+    addSimdCompiler($$name)
+}
+addAvx512Profile(avx512common, avx512cd)
+addAvx512Profile(avx512mic, avx512cd avx512er avx512pf)
+addAvx512Profile(avx512core, avx512cd avx512bw avx512dq avx512vl)
+addAvx512Profile(avx512ifmavl, avx512ifma avx512vl)
+addAvx512Profile(avx512vbmivl, avx512vbmi avx512vl)

mkspecs/linux-icc/qmake.conf

@@ -35,6 +35,13 @@ QMAKE_CFLAGS_SSE4_1    += -xSSE4.1
 QMAKE_CFLAGS_SSE4_2    += -xSSE4.2
 QMAKE_CFLAGS_AVX       += -xAVX
 QMAKE_CFLAGS_AVX2      += -xCORE-AVX2
+QMAKE_CFLAGS_AVX512F   += -xCOMMON-AVX512
+QMAKE_CFLAGS_AVX512CD  += -xCOMMON-AVX512
+QMAKE_CFLAGS_AVX512ER  += -xMIC-AVX512
+QMAKE_CFLAGS_AVX512PF  += -xMIC-AVX512
+QMAKE_CFLAGS_AVX512DQ  += -xCORE-AVX512
+QMAKE_CFLAGS_AVX512BW  += -xCORE-AVX512
+QMAKE_CFLAGS_AVX512VL  += -xCORE-AVX512
 
 QMAKE_CXX               = icpc
 QMAKE_CXXFLAGS          = $$QMAKE_CFLAGS

mkspecs/macx-icc/qmake.conf

@@ -33,6 +33,13 @@ QMAKE_CFLAGS_SSE4_1    += -msse4.1
 QMAKE_CFLAGS_SSE4_2    += -msse4.2
 QMAKE_CFLAGS_AVX       += -mavx
 QMAKE_CFLAGS_AVX2      += -march=core-avx2
+QMAKE_CFLAGS_AVX512F   += -xCOMMON-AVX512
+QMAKE_CFLAGS_AVX512CD  += -xCOMMON-AVX512
+QMAKE_CFLAGS_AVX512ER  += -xMIC-AVX512
+QMAKE_CFLAGS_AVX512PF  += -xMIC-AVX512
+QMAKE_CFLAGS_AVX512DQ  += -xCORE-AVX512
+QMAKE_CFLAGS_AVX512BW  += -xCORE-AVX512
+QMAKE_CFLAGS_AVX512VL  += -xCORE-AVX512
 
 QMAKE_OBJECTIVE_CC      = clang
 QMAKE_OBJECTIVE_CFLAGS          = -pipe

mkspecs/win32-icc/qmake.conf

@@ -32,6 +32,13 @@ QMAKE_CFLAGS_SSE4_1     = -QxSSE4.1
 QMAKE_CFLAGS_SSE4_2     = -QxSSE4.2
 QMAKE_CFLAGS_AVX        = -QxAVX
 QMAKE_CFLAGS_AVX2       = -QxCORE-AVX2
+QMAKE_CFLAGS_AVX512F   += -QxCOMMON-AVX512
+QMAKE_CFLAGS_AVX512CD  += -QxCOMMON-AVX512
+QMAKE_CFLAGS_AVX512ER  += -QxMIC-AVX512
+QMAKE_CFLAGS_AVX512PF  += -QxMIC-AVX512
+QMAKE_CFLAGS_AVX512DQ  += -QxCORE-AVX512
+QMAKE_CFLAGS_AVX512BW  += -QxCORE-AVX512
+QMAKE_CFLAGS_AVX512VL  += -QxCORE-AVX512
 
 QMAKE_CXX               = $$QMAKE_CC
 QMAKE_CXXFLAGS          = $$QMAKE_CFLAGS /Zc:forScope

src/corelib/tools/qsimd_p.h

@@ -73,6 +73,7 @@
  *  SSE4_2   | x86  | I & C | I & C    | I only |
  *  AVX      | x86  | I & C | I & C    | I & C  |
  *  AVX2     | x86  | I & C | I & C    | I only |
+ *  AVX512xx | x86  | I & C | I & C    | I only |
  * I = intrinsics; C = code generation
  *
  * Code can use the following constructs to determine compiler support & status:

tools/configure/configureapp.cpp

@@ -160,6 +160,7 @@ Configure::Configure(int& argc, char** argv)
     dictionary[ "SSE4_2" ]          = "auto";
     dictionary[ "AVX" ]             = "auto";
     dictionary[ "AVX2" ]            = "auto";
+    dictionary[ "AVX512" ]          = "auto";
     dictionary[ "SYNCQT" ]          = "auto";
     dictionary[ "CE_CRT" ]          = "no";
     dictionary[ "CETEST" ]          = "auto";
@@ -885,6 +886,10 @@ void Configure::parseCmdLine()
             dictionary[ "AVX2" ] = "no";
         else if (configCmdLine.at(i) == "-avx2")
             dictionary[ "AVX2" ] = "yes";
+        else if (configCmdLine.at(i) == "-no-avx512")
+            dictionary[ "AVX512" ] = "";
+        else if (configCmdLine.at(i) == "-avx512")
+            dictionary[ "AVX512" ] = "auto";
 
         else if (configCmdLine.at(i) == "-no-ssl") {
             dictionary[ "SSL"] = "no";
@@ -1699,6 +1704,7 @@ void Configure::applySpecSpecifics()
         dictionary[ "SSE4_2" ]              = "no";
         dictionary[ "AVX" ]                 = "no";
         dictionary[ "AVX2" ]                = "no";
+        dictionary[ "AVX512" ]              = "no";
         dictionary[ "CE_CRT" ]              = "yes";
         dictionary[ "LARGE_FILE" ]          = "no";
         dictionary[ "ANGLE" ]               = "no";
@@ -2014,6 +2020,8 @@ bool Configure::displayHelp()
         desc("AVX", "yes",      "-avx",                 "Compile with use of AVX instructions.");
         desc("AVX2", "no",      "-no-avx2",             "Do not compile with use of AVX2 instructions.");
         desc("AVX2", "yes",     "-avx2",                "Compile with use of AVX2 instructions.\n");
+        desc("AVX512", "no",    "-no-avx512",           "Do not compile with use of AVX512 instructions.");
+        desc("AVX512", "yes",   "-avx512",              "Compile with use of AVX512 instructions.\n");
         desc("SSL", "no",        "-no-ssl",             "Do not compile support for SSL.");
         desc("SSL", "yes",       "-ssl",                "Enable run-time SSL support.");
         desc("OPENSSL", "no",    "-no-openssl",         "Do not compile support for OpenSSL.");
@@ -2201,6 +2209,24 @@ bool Configure::checkAngleAvailability(QString *errorMessage /* = 0 */) const
     return true;
 }
 
+QString Configure::checkAvx512Availability()
+{
+    static const char avx512features[][5] = { "cd", "er", "pf", "bw", "dq", "vl", "ifma", "vbmi" };
+
+    // try AVX512 Foundation. No Foundation, nothing else works.
+    if (!tryCompileProject("common/avx512", "AVX512=F"))
+        return QString();
+
+    QString available = "avx512f";
+    for (int i = 0; i < sizeof(avx512features)/sizeof(avx512features[0]); ++i) {
+        if (tryCompileProject("common/avx512", QStringLiteral("AVX512=%0").arg(avx512features[i]).toUpper())) {
+            available += " avx512";
+            available += avx512features[i];
+        }
+    }
+    return available;
+}
+
 /*!
     Checks the system for the availability of a feature.
     Returns true if the feature is available, else false.
@@ -2476,6 +2502,8 @@ void Configure::autoDetection()
         dictionary["AVX"] = checkAvailability("AVX") ? "yes" : "no";
     if (dictionary["AVX2"] == "auto")
         dictionary["AVX2"] = checkAvailability("AVX2") ? "yes" : "no";
+    if (dictionary["AVX512"] == "auto")
+        dictionary["AVX512"] = checkAvx512Availability();
     if (dictionary["NEON"] == "auto")
         dictionary["NEON"] = checkAvailability("NEON") ? "yes" : "no";
     if (dictionary["SSL"] == "auto") {
@@ -3194,6 +3222,8 @@ void Configure::generateCachefile()
             moduleStream << " avx";
         if (dictionary[ "AVX2" ] == "yes")
             moduleStream << " avx2";
+        if (!dictionary[ "AVX512" ].isEmpty())
+            moduleStream << ' ' << dictionary[ "AVX512" ];
         if (dictionary[ "NEON" ] == "yes")
             moduleStream << " neon";
         if (dictionary[ "LARGE_FILE" ] == "yes")
@@ -3659,6 +3689,8 @@ void Configure::generateConfigfiles()
             tmpStream << "#define QT_COMPILER_SUPPORTS_AVX 1" << endl;
         if (dictionary[ "AVX2" ] == "yes")
             tmpStream << "#define QT_COMPILER_SUPPORTS_AVX2 1" << endl;
+        foreach (const QString &avx512feature, dictionary[ "AVX512" ].split(' ', QString::SkipEmptyParts))
+            tmpStream << "#define QT_COMPILER_SUPPRTS_" << avx512feature.toUpper() << " 1" << endl;
 
         if (dictionary["QREAL"] != "double") {
             tmpStream << "#define QT_COORD_TYPE " << dictionary["QREAL"] << endl;
@@ -3831,6 +3863,8 @@ void Configure::displayConfig()
     sout << "SSE4.2 support.............." << dictionary[ "SSE4_2" ] << endl;
     sout << "AVX support................." << dictionary[ "AVX" ] << endl;
     sout << "AVX2 support................" << dictionary[ "AVX2" ] << endl;
+    sout << "AVX512 support.............."
+         << (dictionary[ "AVX512" ].isEmpty() ? QString("<none>") : dictionary[ "AVX512" ].toUpper()) << endl;
     sout << "NEON support................" << dictionary[ "NEON" ] << endl;
     sout << "OpenGL support.............." << dictionary[ "OPENGL" ] << endl;
     sout << "Large File support.........." << dictionary[ "LARGE_FILE" ] << endl;

tools/configure/configureapp.h

@@ -92,6 +92,7 @@ public:
 
 private:
     bool checkAngleAvailability(QString *errorMessage = 0) const;
+    QString checkAvx512Availability();
 
     // Our variable dictionaries
     QMap<QString,QString> dictionary;