feature: add runtime cpu check (#343) (#344)

Check at startup for required cpu instructions (SSE2, SSSE3, POPCOUNT)
variar · Jun 16, 2021 · 4657f4a · 4657f4a
1 parent 9bde202
commit 4657f4a
Show file tree

Hide file tree

Showing 6 changed files with 223 additions and 4 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -104,8 +104,9 @@ if(MSVC)
     ENDIF (CMAKE_C_FLAGS MATCHES "/W[0-4]")
 else()
     ucm_add_flags(CXX "-fno-sized-deallocation")
+    ucm_add_flags(C CXX "-mmmx -msse -msse2 -msse3 -mssse3 -mpopcnt")
     if (KLOGG_GENERIC_CPU)
-        ucm_add_flags(C CXX "-march=core2 -mtune=generic")
+        ucm_add_flags(C CXX "-march=x86-64 -mtune=generic")
     else()
         ucm_add_flags(C CXX "-march=native -mtune=generic")
     endif()

diff --git a/src/app/crashhandler.cpp b/src/app/crashhandler.cpp
@@ -35,6 +35,8 @@
 #include <QTimer>
 #include <QUrlQuery>
 #include <QVBoxLayout>
+#include <cstdint>
+#include <cstdlib>
 #include <string_view>
 
 #ifdef KLOGG_USE_MIMALLOC
@@ -44,6 +46,7 @@
 #include "client/crash_report_database.h"
 #include "sentry.h"
 
+#include "cpu_info.h"
 #include "issuereporter.h"
 #include "klogg_version.h"
 #include "log.h"
@@ -251,14 +254,16 @@ CrashHandler::CrashHandler()
 
     addExtra( "memory", physicalMemory() );
 
+    addExtra( "cpuInstructions", static_cast<unsigned>( supportedCpuInstructions() ) );
+
     memoryUsageTimer_ = std::make_unique<QTimer>();
     QObject::connect( memoryUsageTimer_.get(), &QTimer::timeout, [ addExtra ]() {
         const auto vmUsed = usedMemory();
         addExtra( "vm_used", vmUsed );
 
 #ifdef KLOGG_USE_MIMALLOC
-        size_t elapsedMsecs, userMsecs, systemMsecs, currentRss, peakRss, currentCommit,
-            peakCommit, pageFaults;
+        size_t elapsedMsecs, userMsecs, systemMsecs, currentRss, peakRss, currentCommit, peakCommit,
+            pageFaults;
 
         mi_process_info( &elapsedMsecs, &userMsecs, &systemMsecs, &currentRss, &peakRss,
                          &currentCommit, &peakCommit, &pageFaults );

diff --git a/src/app/main.cpp b/src/app/main.cpp
@@ -37,6 +37,7 @@
  */
 
 #include <QtGlobal>
+#include <QMessageBox>
 
 #ifdef Q_OS_WIN
 #define WIN32_LEAN_AND_MEAN
@@ -53,6 +54,7 @@
 #endif
 
 #include "configuration.h"
+#include "cpu_info.h"
 #include "log.h"
 #include "mainwindow.h"
 #include "styles.h"
@@ -90,7 +92,7 @@ void setApplicationAttributes( bool enableQtHdpi, int scaleFactorRounding )
         QGuiApplication::setHighDpiScaleFactorRoundingPolicy(
             static_cast<Qt::HighDpiScaleFactorRoundingPolicy>( scaleFactorRounding ) );
 #else
-    Q_UNUSED(scaleFactorRounding);
+        Q_UNUSED( scaleFactorRounding );
 #endif
     }
     else {
@@ -106,6 +108,15 @@ void setApplicationAttributes( bool enableQtHdpi, int scaleFactorRounding )
 
 int main( int argc, char* argv[] )
 {
+    auto requiredInstructuins = CpuInstructions::SSE2;
+    requiredInstructuins |= CpuInstructions::SSSE3;
+    requiredInstructuins |= CpuInstructions::POPCNT;
+    if ( !hasRequiredInstructions( supportedCpuInstructions(), requiredInstructuins ) ) {
+        QMessageBox::critical( nullptr, "Klogg", "Current CPU is not supported",
+                               QMessageBox::Close );
+        exit( EXIT_FAILURE );
+    }
+
 #ifdef KLOGG_USE_MIMALLOC
     mi_stats_reset();
 #endif

diff --git a/src/utils/CMakeLists.txt b/src/utils/CMakeLists.txt
@@ -6,6 +6,7 @@ add_library(utils STATIC
         ${CMAKE_CURRENT_SOURCE_DIR}/src/klogg_version.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/src/readablesize.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/src/memory_info.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/src/cpu_info.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/src/issuereporter.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/include/atomicflag.h
         ${CMAKE_CURRENT_SOURCE_DIR}/include/configuration.h
@@ -23,6 +24,7 @@ add_library(utils STATIC
         ${CMAKE_CURRENT_SOURCE_DIR}/include/readablesize.h
         ${CMAKE_CURRENT_SOURCE_DIR}/include/synchronization.h
         ${CMAKE_CURRENT_SOURCE_DIR}/include/memory_info.h
+        ${CMAKE_CURRENT_SOURCE_DIR}/include/cpu_info.h
         ${CMAKE_CURRENT_SOURCE_DIR}/include/overload_visitor.h
         ${CMAKE_CURRENT_SOURCE_DIR}/include/resourcewrapper.h
         ${CMAKE_CURRENT_SOURCE_DIR}/include/styles.h

diff --git a/src/utils/include/cpu_info.h b/src/utils/include/cpu_info.h
@@ -0,0 +1,61 @@
+/*
+ * Copyright (C) 2021 Anton Filimonov and other contributors
+ *
+ * This file is part of klogg.
+ *
+ * klogg is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * klogg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with klogg.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef KLOGG_CPU_INFO_H
+#define KLOGG_CPU_INFO_H
+
+enum class CpuInstructions : unsigned {
+    /// Indicates that no SIMD instructions are supported
+    NONE = 0,
+    SSE2 = 1 << 1,
+    SSE3 = 1 << 2,
+    SSSE3 = 1 << 3,
+    SSE41 = 1 << 4,
+    POPCNT = 1 << 5,
+    AVX = 1 << 6,
+    AVX2 = 1 << 7,
+};
+
+inline CpuInstructions& operator|=( CpuInstructions& x, const CpuInstructions& y )
+{
+    x = static_cast<CpuInstructions>( static_cast<unsigned>( x ) | static_cast<unsigned>( y ) );
+    return x;
+}
+
+inline CpuInstructions operator&( const CpuInstructions& x, const CpuInstructions& y )
+{
+    return static_cast<CpuInstructions>( static_cast<unsigned>( x ) & static_cast<unsigned>( y ) );
+}
+
+inline CpuInstructions operator~( const CpuInstructions& x )
+{
+    return static_cast<CpuInstructions>( ~static_cast<unsigned>( x ) );
+}
+
+inline bool hasRequiredInstructions( CpuInstructions current, CpuInstructions required )
+{
+    if ( ( ~current & required ) == CpuInstructions::NONE ) {
+        return true;
+    }
+    return false;
+}
+
+CpuInstructions supportedCpuInstructions();
+
+#endif
diff --git a/src/utils/src/cpu_info.cpp b/src/utils/src/cpu_info.cpp
@@ -0,0 +1,139 @@
+/*
+ * Copyright (C) 2021 Anton Filimonov and other contributors
+ *
+ * This file is part of klogg.
+ *
+ * klogg is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * klogg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with klogg.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "cpu_info.h"
+#include <QtGlobal>
+
+#if defined( Q_OS_WIN )
+
+#include <array>
+#include <bitset>
+#include <vector>
+
+#include <intrin.h>
+
+CpuInstructions supportedCpuInstructions()
+{
+    CpuInstructions cpuInstructions = CpuInstructions::NONE;
+
+    std::array<int, 4> cpui;
+
+    // Calling __cpuid with 0x0 as the function_id argument
+    // gets the number of the highest valid function ID.
+    __cpuid( cpui.data(), 0 );
+
+    const auto nIds = cpui[ 0 ];
+    std::vector<std::array<int, 4>> data;
+
+    for ( int i = 0; i <= nIds; ++i ) {
+        __cpuidex( cpui.data(), i, 0 );
+        data.push_back( cpui );
+    }
+
+    // load bitset with flags for function 0x00000001
+    if ( nIds >= 1 ) {
+        std::bitset<32> f_1_ECX = data[ 1 ][ 2 ];
+        std::bitset<32> f_1_EDX = data[ 1 ][ 3 ];
+
+        if ( f_1_EDX[ 26 ] ) {
+            cpuInstructions |= CpuInstructions::SSE2;
+        }
+        if ( f_1_ECX[ 0 ] ) {
+            cpuInstructions |= CpuInstructions::SSE3;
+        }
+        if ( f_1_ECX[ 9 ] ) {
+            cpuInstructions |= CpuInstructions::SSSE3;
+        }
+        if ( f_1_ECX[ 19 ] ) {
+            cpuInstructions |= CpuInstructions::SSE41;
+        }
+
+        if ( f_1_ECX[ 23 ] ) {
+            cpuInstructions |= CpuInstructions::POPCNT;
+        }
+
+        if ( f_1_ECX[ 28 ] ) {
+            cpuInstructions |= CpuInstructions::AVX;
+        }
+    }
+
+    // load bitset with flags for function 0x00000007
+    if ( nIds >= 7 ) {
+        std::bitset<32> f_7_EBX = data[ 7 ][ 1 ];
+
+        if ( f_7_EBX[ 5 ] ) {
+            cpuInstructions |= CpuInstructions::AVX2;
+        }
+    }
+
+    return cpuInstructions;
+}
+#else
+CpuInstructions supportedCpuInstructions()
+{
+    CpuInstructions cpuInstructions = CpuInstructions::NONE;
+
+    if ( __builtin_cpu_supports( "avx512f" ) ) {
+        cpuInstructions |= CpuInstructions::SSE2;
+        cpuInstructions |= CpuInstructions::SSE3;
+        cpuInstructions |= CpuInstructions::SSSE3;
+        cpuInstructions |= CpuInstructions::SSE41;
+        cpuInstructions |= CpuInstructions::AVX;
+        cpuInstructions |= CpuInstructions::AVX2;
+    }
+    else if ( __builtin_cpu_supports( "avx2" ) ) {
+        cpuInstructions |= CpuInstructions::SSE2;
+        cpuInstructions |= CpuInstructions::SSE3;
+        cpuInstructions |= CpuInstructions::SSSE3;
+        cpuInstructions |= CpuInstructions::SSE41;
+        cpuInstructions |= CpuInstructions::AVX;
+        cpuInstructions |= CpuInstructions::AVX2;
+    }
+    else if ( __builtin_cpu_supports( "avx" ) ) {
+        cpuInstructions |= CpuInstructions::SSE2;
+        cpuInstructions |= CpuInstructions::SSE3;
+        cpuInstructions |= CpuInstructions::SSSE3;
+        cpuInstructions |= CpuInstructions::SSE41;
+        cpuInstructions |= CpuInstructions::AVX;
+    }
+    else if ( __builtin_cpu_supports( "sse4.1" ) ) {
+        cpuInstructions |= CpuInstructions::SSE2;
+        cpuInstructions |= CpuInstructions::SSE3;
+        cpuInstructions |= CpuInstructions::SSSE3;
+        cpuInstructions |= CpuInstructions::SSE41;
+    }
+    else if ( __builtin_cpu_supports( "ssse3" ) ) {
+        cpuInstructions |= CpuInstructions::SSE2;
+        cpuInstructions |= CpuInstructions::SSE3;
+        cpuInstructions |= CpuInstructions::SSSE3;
+    }
+    else if ( __builtin_cpu_supports( "sse3" ) ) {
+        cpuInstructions |= CpuInstructions::SSE3;
+        cpuInstructions |= CpuInstructions::SSE2;
+    }
+    else if ( __builtin_cpu_supports( "sse2" ) ) {
+        cpuInstructions |= CpuInstructions::SSE2;
+    }
+    if ( __builtin_cpu_supports( "popcnt" ) ) {
+        cpuInstructions |= CpuInstructions::POPCNT;
+    }
+    return cpuInstructions;
+}
+
+#endif