From 2971db21e33e168b9ed2665564db93957035c4ca Mon Sep 17 00:00:00 2001
From: Bartosz Taudul <wolf@nereid.pl>
Date: Fri, 10 Mar 2023 00:23:09 +0100
Subject: [PATCH] Read and report power usage.

---
 public/client/TracySysPower.cpp | 33 +++++++++++++++++++++++++++++++++
 public/client/TracySysPower.hpp |  1 +
 public/common/TracyProtocol.hpp |  2 +-
 public/common/TracyQueue.hpp    | 10 ++++++++++
 4 files changed, 45 insertions(+), 1 deletion(-)

diff --git a/public/client/TracySysPower.cpp b/public/client/TracySysPower.cpp
index d26eab19..bd5939da 100644
--- a/public/client/TracySysPower.cpp
+++ b/public/client/TracySysPower.cpp
@@ -4,11 +4,13 @@
 
 #include <sys/types.h>
 #include <dirent.h>
+#include <chrono>
 #include <inttypes.h>
 #include <stdio.h>
 #include <string.h>
 
 #include "TracyDebug.hpp"
+#include "TracyProfiler.hpp"
 #include "../common/TracyAlloc.hpp"
 
 namespace tracy
@@ -16,6 +18,7 @@ namespace tracy
 
 SysPower::SysPower()
     : m_domains( 4 )
+    , m_lastTime( 0 )
 {
     ScanDirectory( "/sys/devices/virtual/powercap/intel-rapl", -1 );
 }
@@ -31,6 +34,36 @@ SysPower::~SysPower()
 
 void SysPower::Tick()
 {
+    auto t = std::chrono::high_resolution_clock::now().time_since_epoch().count();
+    if( t - m_lastTime > 10000000 )    // 10 ms
+    {
+        m_lastTime = t;
+        for( auto& v : m_domains )
+        {
+            char tmp[32];
+            if( fread( tmp, 1, 32, v.handle ) > 0 )
+            {
+                rewind( v.handle );
+                auto p = (uint64_t)atoll( tmp );
+                uint64_t delta;
+                if( p >= v.value )
+                {
+                    delta = p - v.value;
+                }
+                else
+                {
+                    delta = v.overflow - v.value + p;
+                }
+                v.value = p;
+
+                TracyLfqPrepare( QueueType::SysPowerReport );
+                MemWrite( &item->sysPower.time, Profiler::GetTime() );
+                MemWrite( &item->sysPower.delta, delta );
+                MemWrite( &item->sysPower.name, (uint64_t)v.name );
+                TracyLfqCommit;
+            }
+        }
+    }
 }
 
 void SysPower::ScanDirectory( const char* path, int parent )
diff --git a/public/client/TracySysPower.hpp b/public/client/TracySysPower.hpp
index 01fa75c9..210123bc 100644
--- a/public/client/TracySysPower.hpp
+++ b/public/client/TracySysPower.hpp
@@ -35,6 +35,7 @@ private:
     void ScanDirectory( const char* path, int parent );
 
     FastVector<Domain> m_domains;
+    uint64_t m_lastTime;
 };
 
 }
diff --git a/public/common/TracyProtocol.hpp b/public/common/TracyProtocol.hpp
index dd30e539..04c98690 100644
--- a/public/common/TracyProtocol.hpp
+++ b/public/common/TracyProtocol.hpp
@@ -9,7 +9,7 @@ namespace tracy
 
 constexpr unsigned Lz4CompressBound( unsigned isize ) { return isize + ( isize / 255 ) + 16; }
 
-enum : uint32_t { ProtocolVersion = 63 };
+enum : uint32_t { ProtocolVersion = 64 };
 enum : uint16_t { BroadcastVersion = 3 };
 
 using lz4sz_t = uint32_t;
diff --git a/public/common/TracyQueue.hpp b/public/common/TracyQueue.hpp
index 8443193a..051d412a 100644
--- a/public/common/TracyQueue.hpp
+++ b/public/common/TracyQueue.hpp
@@ -90,6 +90,7 @@ enum class QueueType : uint8_t
     GpuNewContext,
     CallstackFrame,
     SysTimeReport,
+    SysPowerReport,
     TidToPid,
     HwSampleCpuCycle,
     HwSampleInstructionRetired,
@@ -563,6 +564,13 @@ struct QueueSysTime
     float sysTime;
 };
 
+struct QueueSysPower
+{
+    int64_t time;
+    uint64_t delta;
+    uint64_t name;  // ptr
+};
+
 struct QueueContextSwitch
 {
     int64_t time;
@@ -729,6 +737,7 @@ struct QueueItem
         QueueCrashReport crashReport;
         QueueCrashReportThread crashReportThread;
         QueueSysTime sysTime;
+        QueueSysPower sysPower;
         QueueContextSwitch contextSwitch;
         QueueThreadWakeup threadWakeup;
         QueueTidToPid tidToPid;
@@ -832,6 +841,7 @@ static constexpr size_t QueueDataSize[] = {
     sizeof( QueueHeader ) + sizeof( QueueGpuNewContext ),
     sizeof( QueueHeader ) + sizeof( QueueCallstackFrame ),
     sizeof( QueueHeader ) + sizeof( QueueSysTime ),
+    sizeof( QueueHeader ) + sizeof( QueueSysPower ),
     sizeof( QueueHeader ) + sizeof( QueueTidToPid ),
     sizeof( QueueHeader ) + sizeof( QueueHwSample ),        // cpu cycle
     sizeof( QueueHeader ) + sizeof( QueueHwSample ),        // instruction retired