CodeXL 2.5 Release

Author: chesik-amd

CodeXL/AMDTApplicationFramework/Include/afAppStringConstants.h

@@ -779,8 +779,8 @@
 #define AF_STR_SystemInformationCommandVendorID L"Vendor ID"
 #define AF_STR_SystemInformationCommandCpuFamily L"CPU Family"
 #define AF_STR_SystemInformationCommandPipePrefix L"AMDT_SI_PIPE_NAME"
-#define AF_STR_SystemInformationCommandHelper32Name L"AMDTSystemInformationHelper"
-#define AF_STR_SystemInformationCommandHelper64Name L"AMDTSystemInformationHelper-x64"
+#define AF_STR_SystemInformationCommandHelper32Name L"CXLSystemInformationHelper"
+#define AF_STR_SystemInformationCommandHelper64Name L"CXLSystemInformationHelper-x64"
 #define AF_STR_SystemInformationCommandTimeOut L"System Information gathering timed out after 10s"
 
 // Context Information dialog

CodeXL/App/LinuxResources/CodeXL_Release_Notes.pdf

@@ -55,12 +55,12 @@ const QString CounterManager::ms_GFX7_FAMILY_ENV_VAR_VALUE = "GFX7";
 const QString CounterManager::ms_GFX8_FAMILY_ENV_VAR_VALUE = "GFX8";
 const QString CounterManager::ms_GFX9_FAMILY_ENV_VAR_VALUE = "GFX9";
 
-const int CounterManager::ms_GFX6_PLACEHOLDER_DEVICE_ID = 0x6798; // a Tahiti device
-const int CounterManager::ms_GFX7_PLACEHOLDER_DEVICE_ID = 0x6650; // a Bonaire device
-const int CounterManager::ms_GFX8_PLACEHOLDER_DEVICE_ID = 0x6900; // an Iceland device
-const int CounterManager::ms_GFX9_PLACEHOLDER_DEVICE_ID = 0x6860; // a Vega device
+const size_t CounterManager::ms_GFX6_PLACEHOLDER_DEVICE_ID = 0x6798; // a Tahiti device
+const size_t CounterManager::ms_GFX7_PLACEHOLDER_DEVICE_ID = 0x6649; // a Bonaire device
+const size_t CounterManager::ms_GFX8_PLACEHOLDER_DEVICE_ID = 0x6900; // an Iceland device
+const size_t CounterManager::ms_GFX9_PLACEHOLDER_DEVICE_ID = 0x6860; // a Vega device
 
-const int CounterManager::ms_UNSPECIFIED_REV_ID = 0;
+const size_t CounterManager::ms_UNSPECIFIED_REV_ID = static_cast<size_t>(REVISION_ID_ANY);
 
 const int CounterManager::ms_AMD_VENDOR_ID = 0x1002;
 
@@ -359,7 +359,7 @@ void CounterManager::LoadCountersModule()
     }
 }
 
-void CounterManager::AddDeviceId(int deviceId, int revId)
+void CounterManager::AddDeviceId(size_t deviceId, size_t revId)
 {
     GDT_GfxCardInfo gfxCardInfo;
 
@@ -394,7 +394,7 @@ void CounterManager::AddDeviceId(int deviceId, int revId)
     }
 }
 
-void CounterManager::AddDeviceIdToFamily(HardwareFamily hardwareFamily, int deviceId, int revId)
+void CounterManager::AddDeviceIdToFamily(HardwareFamily hardwareFamily, size_t deviceId, size_t revId)
 {
     QList<DeviceAndRevInfo> deviceInfoList;
 

CodeXL/Components/GpuProfiling/AMDTGpuProfiling/CounterManager.cpp

@@ -192,12 +192,12 @@ class AMDT_GPU_PROF_API CounterManager : public TSingleton<CounterManager>
     static const QString ms_GFX7_FAMILY_NAME;           ///< string constant for the GFX7 hardware family name
     static const QString ms_GFX8_FAMILY_NAME;           ///< string constant for the GFX8 hardware family name
     static const QString ms_GFX9_FAMILY_NAME;           ///< string constant for the GFX9 hardware family name
-    static const int     ms_GFX6_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX6 device
-    static const int     ms_GFX7_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX7 device
-    static const int     ms_GFX8_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX8 device
-    static const int     ms_GFX9_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX9 device
-    static const int     ms_UNSPECIFIED_REV_ID;       ///< unspecified device revision id
-    static const int     ms_AMD_VENDOR_ID;            ///< AMD vendor id
+    static const size_t  ms_GFX6_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX6 device
+    static const size_t  ms_GFX7_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX7 device
+    static const size_t  ms_GFX8_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX8 device
+    static const size_t  ms_GFX9_PLACEHOLDER_DEVICE_ID; ///< placeholder device id for a GFX9 device
+    static const size_t  ms_UNSPECIFIED_REV_ID;         ///< unspecified device revision id
+    static const int     ms_AMD_VENDOR_ID;              ///< AMD vendor id
 
     /// typedef used for counter indices
     struct CounterIndices
@@ -260,13 +260,13 @@ class AMDT_GPU_PROF_API CounterManager : public TSingleton<CounterManager>
     /// NA_HARDWARE_FAMILY family
     /// \param deviceId the device id to add to the hardware family
     /// \param revId the revision id to add to the hardware family
-    void AddDeviceId(int deviceId, int revId);
+    void AddDeviceId(size_t deviceId, size_t revId);
 
     /// Adds the specified device id to the list of devices for the specified hardware family
     /// \param hardwareFamily the hardware family to which the device should be added
     /// \param deviceId the device id to add to the hardware family
     /// \param revId the revision id to add to the hardware family
-    void AddDeviceIdToFamily(HardwareFamily hardwareFamily, int deviceId, int revId);
+    void AddDeviceIdToFamily(HardwareFamily hardwareFamily, size_t deviceId, size_t revId);
 
     /// Updates the m_counterIndices after the counters have been sorted
     /// \param counterNames the sorted list of counters
@@ -276,8 +276,8 @@ class AMDT_GPU_PROF_API CounterManager : public TSingleton<CounterManager>
     // structure for holding device id and rev id for installed devices
     struct DeviceAndRevInfo
     {
-        int m_deviceID; ///< device id
-        int m_revID;    ///< rev id
+        size_t m_deviceID; ///< device id
+        size_t m_revID;    ///< rev id
     };
 
     osModuleHandle                                 m_gpaCountersModuleHandle;         ///< module handle for GPACounters DLL

CodeXL/Components/GpuProfiling/AMDTGpuProfiling/CounterManager.h

@@ -115,16 +115,30 @@ void HSAMemoryTimelineItem::tooltipItems(acTimelineItemToolTip& tooltip) const
     tooltip.add(tr("Size"), QString(tr("%1")).arg(m_size));
 }
 
-HSAMemoryTransferTimelineItem::HSAMemoryTransferTimelineItem(quint64 transferStartTime, quint64 transferEndTime, int apiIndex, size_t size, QString srcAgent, QString dstAgent)
-    : HSAMemoryTimelineItem(transferStartTime, transferEndTime, apiIndex, size), m_srcAgent(srcAgent), m_dstAgent(dstAgent)
+HSAMemoryTransferTimelineItem::HSAMemoryTransferTimelineItem(
+    quint64 transferStartTime,
+    quint64 transferEndTime,
+    int apiIndex,
+    size_t size,
+    QString srcAgentHandle,
+    QString srcAgentName,
+    QString dstAgentHandle,
+    QString dstAgentName):
+    HSAMemoryTimelineItem(transferStartTime, transferEndTime, apiIndex, size),
+    m_srcAgentHandle(srcAgentHandle),
+    m_srcAgentName(srcAgentName),
+    m_dstAgentHandle(dstAgentHandle),
+    m_dstAgentName(dstAgentName)
 {
 }
 
 void HSAMemoryTransferTimelineItem::tooltipItems(acTimelineItemToolTip& tooltip) const
 {
     tooltip.add(tr("Name"), m_strText);
-    tooltip.add(tr("Source Agent Handle"), m_srcAgent);
-    tooltip.add(tr("Destination Agent Handle"), m_dstAgent);
+    tooltip.add(tr("Source Agent Handle"), m_srcAgentHandle);
+    tooltip.add(tr("Source Agent Name"), m_srcAgentName);
+    tooltip.add(tr("Destination Agent Handle"), m_dstAgentHandle);
+    tooltip.add(tr("Destination Agent Name"), m_dstAgentName);
 
     acTimeline* timeline = m_pParentBranch->parentTimeline();
 

CodeXL/Components/GpuProfiling/AMDTGpuProfiling/HSATimelineItems.cpp

@@ -80,7 +80,7 @@ class HSADispatchTimelineItem : public HostAPITimelineItem
 
     /// Fill in a TimelineItemToolTip instance with a set of name/value pairs that will be displayed in the tooltip for this timeline item
     /// \param tooltip acTimelineItemToolTip instance that should get populated with name/value pairs
-    virtual void tooltipItems(acTimelineItemToolTip& tooltip) const;
+    virtual void tooltipItems(acTimelineItemToolTip& tooltip) const override;
 
 private:
     QString                         m_strQueueHandle;    ///< the queue handle for this API
@@ -104,7 +104,7 @@ class HSAMemoryTimelineItem : public HostAPITimelineItem
 
     /// Fill in a TimelineItemToolTip instance with a set of name/value pairs that will be displayed in the tooltip for this timeline item
     /// \param tooltip acTimelineItemToolTip instance that should get populated with name/value pairs
-    virtual void tooltipItems(acTimelineItemToolTip& tooltip) const;
+    virtual void tooltipItems(acTimelineItemToolTip& tooltip) const override;
 
 protected:
     size_t m_size;                ///< the memory size for this API
@@ -119,17 +119,29 @@ class HSAMemoryTransferTimelineItem : public HSAMemoryTimelineItem
     /// \param transferEndTime the end time for this data transfer.
     /// \param apiIndex the index of this api in the application's call sequence
     /// \param size the size of the memory operation
-    /// \param srcAgent the source agent of the memory transfer
-    /// \param dstAgent the destination agent of the memory transfer
-    HSAMemoryTransferTimelineItem(quint64 transferStartTime, quint64 transferEndTime, int apiIndex, size_t size, QString srcAgent, QString dstAgent);
+    /// \param srcAgentHandle the source agent handle of the memory transfer
+    /// \param srcAgentName the source agent name of the memory transfer
+    /// \param dstAgentHandle the destination agent handle of the memory transfer
+    /// \param dstAgentName the destination agent name of the memory transfer
+    HSAMemoryTransferTimelineItem(
+        quint64 transferStartTime,
+        quint64 transferEndTime,
+        int apiIndex,
+        size_t size,
+        QString srcAgentHandle,
+        QString srcAgentName,
+        QString dstAgentHandle,
+        QString dstAgentName);
 
     /// Fill in a TimelineItemToolTip instance with a set of name/value pairs that will be displayed in the tooltip for this timeline item
     /// \param tooltip acTimelineItemToolTip instance that should get populated with name/value pairs
-    virtual void tooltipItems(acTimelineItemToolTip& tooltip) const;
+    virtual void tooltipItems(acTimelineItemToolTip& tooltip) const override;
 
 protected:
-    QString m_srcAgent;  ///< the source agent of the transfer
-    QString m_dstAgent;  ///< the destination agent of the transfer
+    QString m_srcAgentHandle;   ///< the source agent handle of the transfer
+    QString m_srcAgentName;     ///< the source agent name of the transfer
+    QString m_dstAgentHandle;   ///< the destination agent handle of the transfer
+    QString m_dstAgentName;     ///< the destination agent name of the transfer
 };
 
 #endif // _HSA_TIMELINE_ITEMS_H_

CodeXL/Components/GpuProfiling/AMDTGpuProfiling/HSATimelineItems.h

@@ -1681,7 +1681,7 @@ bool ProfileManager::GenerateOccupancyPage(GPUSessionTreeItemData* pSessionData,
         // Generate the path of the output file
         m_strOutputOccHTMLPage = QString("%1%2_%3_%4_Occupancy.html")
             .arg(sessionDirStr)
-            .arg(QString().fromStdString(pOccInfo->GetKernelName()))
+            .arg(QString::fromStdString(pOccInfo->GetKernelName()))
             .arg(pOccInfo->GetThreadId())
             .arg(callIndex);
 
@@ -1707,7 +1707,7 @@ bool ProfileManager::GenerateOccupancyPage(GPUSessionTreeItemData* pSessionData,
             }
             else
             {
-                strErrorMessageOut.fromStdString(errorMessage);
+                strErrorMessageOut = QString::fromStdString(errorMessage);
             }
         }
         else
@@ -1979,14 +1979,22 @@ void ProfileManager::HandleProfileFinished(int exitCode)
 
 void ProfileManager::HandleMissingProfileOutput(QString& strError)
 {
-    QString strProjectType = (m_pCurrentProjectSettings->m_traceOptions.m_apiToTrace == APIToTrace_OPENCL) ? "an OpenCL program" : "an HSA program";
+    QString strProjectType;
+
+    if (m_profileParameters.ProfileTypeValue() == API_TRACE)
+    {
+       strProjectType = (m_pCurrentProjectSettings->m_traceOptions.m_apiToTrace == APIToTrace_OPENCL) ? "an OpenCL program" : "an HSA program";
+    }
+    else if (m_profileParameters.ProfileTypeValue() == PERFORMANCE)
+    {
+        strProjectType = (m_pCurrentProjectSettings->m_counterOptions.m_api == APIToTrace_OPENCL) ? "an OpenCL program" : "an HSA program";
+    }
+
     QString strExtraReason;
     strExtraReason.clear();
 
     if (m_profileParameters.ProfileTypeValue() == API_TRACE)
     {
-        strProjectType = (m_pCurrentProjectSettings->m_traceOptions.m_apiToTrace == APIToTrace_OPENCL) ? "an OpenCL program" : "an HSA program";
-
         // in timeout mode, the backend does not produce an .atp file if all APIs called were filtered away
         if (m_pCurrentProjectSettings->m_traceOptions.m_mode == TIMEOUT)
         {
@@ -1997,8 +2005,14 @@ void ProfileManager::HandleMissingProfileOutput(QString& strError)
     }
     else
     {
-        strProjectType += ".</li><li>The active project is an OpenCL program, but it did not enqueue any kernels.";
-        strProjectType += ".</li><li>The active project is an OpenCL program, but it did not enqueue any kernels listed in the Profile Specific Kernels section";
+        QString thisReason = QString(".</li><li>The active project is %1,  but it did not enqueue any kernels.").arg(strProjectType);
+        thisReason += QString("</li><li>The active project is %1, but it did not enqueue any kernels listed in the Profile Specific Kernels section").arg(strProjectType);
+        strProjectType.append(thisReason);
+
+        if (Util::IsHSAEnabled() && m_pCurrentProjectSettings->m_counterOptions.m_api == APIToTrace_OPENCL)
+        {
+            strExtraReason = QString("<li>Collecting OpenCL performance counters when running on the ROCm stack might not be supported.%1 </li>").arg(Util::ms_ENABLE_HSA_OPTION);
+        }
     }
 
     strError = QString("Unable to gather profile data.  This error can occur for one of several reasons:<ul>");

CodeXL/Components/GpuProfiling/AMDTGpuProfiling/ProfileManager.cpp

@@ -592,45 +592,45 @@ const OccupancyTable& GPUSessionTreeItemData::LoadAndGetOccupancyTable()
 
             void* pPtr;
             AtpDataHandlerFunc pAtpDataHandler_func = AtpUtils::Instance()->GetAtpDataHandlerFunc();
-            IOccupancyFileInfoDataHandler* occupancyFileDataInfo = nullptr;
+            IOccupancyFileInfoDataHandler* pOccupancyFileDataInfo = nullptr;
 
             if (nullptr != pAtpDataHandler_func)
             {
                 pAtpDataHandler_func(&pPtr);
                 IAtpDataHandler* pApDataHandler = reinterpret_cast<IAtpDataHandler*>(pPtr);
-                m_occupancyFile.fromStdString(std::string(m_pParentData->m_filePath.asString().asASCIICharArray()));
                 std::string occupancyFile = m_occupancyFile.toStdString();
 
-                occupancyFileDataInfo = pApDataHandler->GetOccupancyFileInfoDataHandler(occupancyFile.c_str());
+                pOccupancyFileDataInfo = pApDataHandler->GetOccupancyFileInfoDataHandler(occupancyFile.c_str());
 
-                if (nullptr != occupancyFileDataInfo)
+                if (nullptr != pOccupancyFileDataInfo)
                 {
-                    m_occupancyFileIsLoaded = occupancyFileDataInfo->ParseOccupancyFile(occupancyFile.c_str());
+                    if (!pOccupancyFileDataInfo->IsDataReady())
+                    {
+                        m_occupancyFileIsLoaded = pOccupancyFileDataInfo->ParseOccupancyFile(occupancyFile.c_str());
+                    }
                 }
-
-                m_occupancyFileIsLoaded = occupancyFileDataInfo->ParseOccupancyFile(occupancyFile.c_str());
             }
 
-            if (m_occupancyFileIsLoaded)
+            if (m_occupancyFileIsLoaded && nullptr != pOccupancyFileDataInfo)
             {
-                osThreadId* osThreadIds;
+                osThreadId* pOsThreadIds;
                 unsigned int threadCount;
-                occupancyFileDataInfo->GetOccupancyThreads(&osThreadIds, threadCount);
+                pOccupancyFileDataInfo->GetOccupancyThreads(&pOsThreadIds, threadCount);
 
                 for (unsigned int i = 0; i < threadCount; i++)
                 {
                     const IOccupancyInfoDataHandler* occupancyInfo;
                     unsigned int kernelCount;
-                    occupancyFileDataInfo->GetKernelCountByThreadId(osThreadIds[i], kernelCount);
+                    pOccupancyFileDataInfo->GetKernelCountByThreadId(pOsThreadIds[i], kernelCount);
                     QList<const IOccupancyInfoDataHandler*> occupancyInfoList;
 
                     for (unsigned int j = 0; j < kernelCount; j++)
                     {
-                        occupancyInfo = occupancyFileDataInfo->GetOccupancyInfoDataHandler(osThreadIds[i], j);
+                        occupancyInfo = pOccupancyFileDataInfo->GetOccupancyInfoDataHandler(pOsThreadIds[i], j);
                         occupancyInfoList.push_back(occupancyInfo);
                     }
 
-                    m_occupancyTable.insert(osThreadIds[i], occupancyInfoList);
+                    m_occupancyTable.insert(pOsThreadIds[i], occupancyInfoList);
                 }
             }
         }

CodeXL/Components/GpuProfiling/AMDTGpuProfiling/Session.cpp

@@ -528,8 +528,7 @@ const IOccupancyInfoDataHandler* SessionControl::GetOccupancyForRow(int rowIndex
         {
             // If the occupancy info doesn't match the one in the map, this means that the user sorted the table,
             // and the information in the map is no longer reflecting the correct line numbers:
-            QString tempString;
-            tempString.fromStdString(retVal->GetKernelName());
+            QString tempString = QString::fromStdString(retVal->GetKernelName());
             if (!kernelName.startsWith(tempString))
             {
                 QMap<int, const IOccupancyInfoDataHandler*>::const_iterator iter = m_rowOccupancyInfoMap.begin();
@@ -538,8 +537,7 @@ const IOccupancyInfoDataHandler* SessionControl::GetOccupancyForRow(int rowIndex
                 {
                     if ((*iter) != nullptr)
                     {
-                        QString currentRowKernelName;
-                        currentRowKernelName.fromStdString((*iter)->GetKernelName());
+                        QString currentRowKernelName = QString::fromStdString((*iter)->GetKernelName());
                         if (kernelName.startsWith(currentRowKernelName))
                         {
                             retVal = (*iter);

CodeXL/Components/GpuProfiling/AMDTGpuProfiling/SessionControl.cpp

@@ -373,7 +373,7 @@ void GPUSessionWindow::CellContentClicked(const QModelIndex& index)
                 return;
             }
 
-            m_currentDisplayedOccupancyKernel = QString().fromStdString(occupancyInfo->GetKernelName());
+            m_currentDisplayedOccupancyKernel = QString::fromStdString(occupancyInfo->GetKernelName());
             QTableView* pPcTable = m_pControl->GetTableView();
             int columnCount = pPcTable->model()->columnCount();
             int callIndexColIndex = -1;