From 5fbb811f5d93253e4344d2bdbdd2723e93dee35c Mon Sep 17 00:00:00 2001
From: Bartosz Taudul <wolf.pld@gmail.com>
Date: Wed, 14 Aug 2019 16:19:02 +0200
Subject: [PATCH] Degrade ARM timer to monotonic raw clock.

The monotonic raw clock has the same accuracy as reading cntvct registers, but
using clock_gettime() has a measurable impact on queueing time (135 us vs
83 us).

This change is needed to enable ftrace time readings on ARM linux, which
doesn't provide any way to get raw cntvct readings, like x86-tsc on x86.
---
 client/TracyProfiler.cpp | 103 ++-------------------------------------
 client/TracyProfiler.hpp |  14 +++---
 2 files changed, 12 insertions(+), 105 deletions(-)

diff --git a/client/TracyProfiler.cpp b/client/TracyProfiler.cpp
index e0e21eb6..17e665f7 100644
--- a/client/TracyProfiler.cpp
+++ b/client/TracyProfiler.cpp
@@ -151,89 +151,7 @@ struct ThreadHandleWrapper
 #endif
 
 
-#if defined TRACY_HW_TIMER && __ARM_ARCH >= 6 && !defined TARGET_OS_IOS
-int64_t (*GetTimeImpl)();
-
-int64_t GetTimeImplFallback()
-{
-#  ifdef CLOCK_MONOTONIC_RAW
-    struct timespec ts;
-    clock_gettime( CLOCK_MONOTONIC_RAW, &ts );
-    return int64_t( ts.tv_sec ) * 1000000000 + int64_t( ts.tv_nsec );
-#  else
-    return std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::high_resolution_clock::now().time_since_epoch() ).count();
-#  endif
-}
-
-int64_t GetTimeImplCntvct()
-{
-    int64_t t;
-#  ifdef __aarch64__
-    asm volatile ( "mrs %0, cntvct_el0" : "=r" (t) );
-#  else
-    asm volatile ( "mrrc p15, 1, %Q0, %R0, c14" : "=r" (t) );
-#  endif
-    return t;
-}
-
-static sigjmp_buf SigIllEnv;
-
-static int SetupHwTimerFailed()
-{
-    return sigsetjmp( SigIllEnv, 1 );
-}
-
-static void SetupHwTimerSigIllHandler( int /*signum*/ )
-{
-    siglongjmp( SigIllEnv, 1 );
-}
-
-static int64_t SetupHwTimer()
-{
-    struct sigaction act, oldact;
-    memset( &act, 0, sizeof( act ) );
-    act.sa_handler = SetupHwTimerSigIllHandler;
-
-    if( sigaction( SIGILL, &act, &oldact ) )
-    {
-        GetTimeImpl = GetTimeImplFallback;
-        return Profiler::GetTime();
-    }
-
-    if( SetupHwTimerFailed() )
-    {
-        sigaction( SIGILL, &oldact, nullptr );
-        GetTimeImpl = GetTimeImplFallback;
-        return Profiler::GetTime();
-    }
-
-    GetTimeImplCntvct();
-
-    sigaction( SIGILL, &oldact, nullptr );
-    GetTimeImpl = GetTimeImplCntvct;
-
-    // Check if cntvct is monotonic (there is faulty hw out there)
-    enum { NumProbes = 32 * 1024 };
-    int64_t probe[NumProbes];
-    for( int j=0; j<10; j++ )
-    {
-        for( int i=0; i<NumProbes; i++ )
-        {
-            probe[i] = Profiler::GetTime();
-        }
-        for( int i=1; i<NumProbes; i++ )
-        {
-            if( probe[i] < probe[i-1] )
-            {
-                GetTimeImpl = GetTimeImplFallback;
-                return Profiler::GetTime();
-            }
-        }
-    }
-
-    return Profiler::GetTime();
-}
-#elif defined TRACY_HW_TIMER && ( defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64 )
+#if defined TRACY_HW_TIMER && ( defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64 )
 static inline void CpuId( uint32_t* regs, uint32_t leaf )
 {
 #if defined _WIN32 || defined __CYGWIN__
@@ -2141,24 +2059,12 @@ void Profiler::HandleDisconnect()
     }
 }
 
-bool Profiler::IsTimerMonotonicRaw()
-{
-#if defined TRACY_HW_TIMER && !defined TARGET_OS_IOS && __ARM_ARCH >= 6
-    return GetTimeImpl == GetTimeImplFallback;
-#else
-    return false;
-#endif
-}
-
 void Profiler::CalibrateTimer()
 {
 #ifdef TRACY_HW_TIMER
-    if( IsTimerMonotonicRaw() )
-    {
-        m_timerMul = 1.;
-        return;
-    }
-
+#  if !defined TARGET_OS_IOS && __ARM_ARCH >= 6
+    m_timerMul = 1.;
+#  else
     std::atomic_signal_fence( std::memory_order_acq_rel );
     const auto t0 = std::chrono::high_resolution_clock::now();
     const auto r0 = GetTime();
@@ -2173,6 +2079,7 @@ void Profiler::CalibrateTimer()
     const auto dr = r1 - r0;
 
     m_timerMul = double( dt ) / double( dr );
+#  endif
 #else
     m_timerMul = 1.;
 #endif
diff --git a/client/TracyProfiler.hpp b/client/TracyProfiler.hpp
index 415e2576..97bf3890 100644
--- a/client/TracyProfiler.hpp
+++ b/client/TracyProfiler.hpp
@@ -78,10 +78,6 @@ struct LuaZoneState
 
 using Magic = moodycamel::ConcurrentQueueDefaultTraits::index_t;
 
-#if __ARM_ARCH >= 6 && !defined TARGET_OS_IOS
-extern int64_t (*GetTimeImpl)();
-#endif
-
 
 class Profiler
 {
@@ -105,7 +101,13 @@ public:
 #  if TARGET_OS_IOS == 1
         return mach_absolute_time();
 #  elif __ARM_ARCH >= 6
-        return GetTimeImpl();
+#    ifdef CLOCK_MONOTONIC_RAW
+        struct timespec ts;
+        clock_gettime( CLOCK_MONOTONIC_RAW, &ts );
+        return int64_t( ts.tv_sec ) * 1000000000ll + int64_t( ts.tv_nsec );
+#    else
+        return std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::high_resolution_clock::now().time_since_epoch() ).count();
+#    endif
 #  elif defined _WIN32 || defined __CYGWIN__
         unsigned int dontcare;
         const auto t = int64_t( __rdtscp( &dontcare ) );
@@ -440,8 +442,6 @@ public:
     void RequestShutdown() { m_shutdown.store( true, std::memory_order_relaxed ); m_shutdownManual.store( true, std::memory_order_relaxed ); }
     bool HasShutdownFinished() const { return m_shutdownFinished.load( std::memory_order_relaxed ); }
 
-    static bool IsTimerMonotonicRaw();
-
 private:
     enum class DequeueStatus { Success, ConnectionLost, QueueEmpty };