Merge pull request #4260 from Sonicadvance1/profile_win32

Profiler: Setup for usage on Windows
FEX-Emu · Jan 15, 2025 · 8e2b4a3 · 8e2b4a3
2 parents 48c03d7 + 4c9f1b1
commit 8e2b4a3
Show file tree

Hide file tree

Showing 5 changed files with 54 additions and 25 deletions.
diff --git a/FEXCore/Source/Utils/Profiler.cpp b/FEXCore/Source/Utils/Profiler.cpp
@@ -7,6 +7,7 @@
 #include <linux/magic.h>
 #include <sys/stat.h>
 #include <sys/vfs.h>
+#include <time.h>
 #endif
 
 #include <FEXCore/Utils/LogManager.h>
@@ -18,6 +19,36 @@
 #define BACKEND_GPUVIS 1
 
 #ifdef ENABLE_FEXCORE_PROFILER
+#ifndef _WIN32
+static inline uint64_t GetTime() {
+  // We want the time in the least amount of overhead possible
+  // clock_gettime will do a VDSO call with the least amount of overhead
+  struct timespec ts;
+  clock_gettime(CLOCK_MONOTONIC, &ts);
+  return ts.tv_sec * 1'000'000'000ULL + ts.tv_nsec;
+}
+#else
+
+static inline uint64_t GetTime() {
+  // GetTime needs to return nanoseconds, query the interface.
+  static uint64_t FrequencyScale = {};
+  if (!FrequencyScale) [[unlikely]] {
+    LARGE_INTEGER Frequency {};
+    while (!QueryPerformanceFrequency(&Frequency))
+      ;
+    constexpr uint64_t NanosecondsInSecond = 1'000'000'000ULL;
+
+    // On WINE this will always result in a scale of 100.
+    FrequencyScale = NanosecondsInSecond / Frequency.QuadPart;
+  }
+  LARGE_INTEGER ticks;
+  while (!QueryPerformanceCounter(&ticks))
+    ;
+  return ticks.QuadPart * FrequencyScale;
+}
+
+#endif
+
 #if FEXCORE_PROFILER_BACKEND == BACKEND_GPUVIS
 namespace FEXCore::Profiler {
 ProfilerBlock::ProfilerBlock(std::string_view const Format)
@@ -41,23 +72,18 @@ static std::array<const char*, 2> TraceFSDirectories {
   "/sys/kernel/debug/tracing",
 };
 
-static bool IsTraceFS(const char* Path) {
-  struct statfs stat;
-  if (statfs(Path, &stat)) {
-    return false;
-  }
-  return stat.f_type == TRACEFS_MAGIC;
-}
-
 void Init() {
   for (auto Path : TraceFSDirectories) {
-    if (IsTraceFS(Path)) {
-      fextl::string FilePath = fextl::fmt::format("{}/trace_marker", Path);
-      TraceFD = open(FilePath.c_str(), O_WRONLY | O_CLOEXEC);
-      if (TraceFD != -1) {
-        // Opened TraceFD, early exit
-        break;
-      }
+#ifdef _WIN32
+    constexpr auto flags = O_WRONLY;
+#else
+    constexpr auto flags = O_WRONLY | O_CLOEXEC;
+#endif
+    fextl::string FilePath = fextl::fmt::format("{}/trace_marker", Path);
+    TraceFD = open(FilePath.c_str(), flags);
+    if (TraceFD != -1) {
+      // Opened TraceFD, early exit
+      break;
     }
   }
 }

diff --git a/FEXCore/include/FEXCore/Utils/Profiler.h b/FEXCore/include/FEXCore/Utils/Profiler.h
@@ -2,7 +2,6 @@
 #pragma once
 #include <cstdint>
 #include <string_view>
-#include <time.h>
 
 #include <FEXCore/Utils/CompilerDefs.h>
 
@@ -14,14 +13,6 @@ FEX_DEFAULT_VISIBILITY void Shutdown();
 FEX_DEFAULT_VISIBILITY void TraceObject(std::string_view const Format);
 FEX_DEFAULT_VISIBILITY void TraceObject(std::string_view const Format, uint64_t Duration);
 
-static inline uint64_t GetTime() {
-  // We want the time in the least amount of overhead possible
-  // clock_gettime will do a VDSO call with the least amount of overhead
-  struct timespec ts;
-  clock_gettime(CLOCK_MONOTONIC, &ts);
-  return ts.tv_sec * 1'000'000'000ULL + ts.tv_nsec;
-}
-
 // A class that follows scoping rules to generate a profile duration block
 class ProfilerBlock final {
 public:

diff --git a/Source/Windows/ARM64EC/Module.cpp b/Source/Windows/ARM64EC/Module.cpp
@@ -17,6 +17,7 @@ desc: Implements the ARM64EC BT module API using FEXCore
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Threads.h>
+#include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/Utils/EnumOperators.h>
 #include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/FPState.h>
@@ -533,6 +534,8 @@ NTSTATUS ProcessInit() {
   // Not applicable to Windows
   FEXCore::Config::EraseSet(FEXCore::Config::ConfigOption::CONFIG_TSOAUTOMIGRATION, "0");
 
+  FEXCore::Profiler::Init();
+
   FEXCore::Context::InitializeStaticTables(FEXCore::Context::MODE_64BIT);
 
   SignalDelegator = fextl::make_unique<FEX::DummyHandlers::DummySignalDelegator>();

diff --git a/Source/Windows/Common/CRT/IO.cpp b/Source/Windows/Common/CRT/IO.cpp
@@ -130,7 +130,13 @@ DLLEXPORT_FUNC(int, _wopen, (const wchar_t* Filename, int OpenFlag, ...)) {
       Attrs = FILE_ATTRIBUTE_READONLY;
     }
   }
-  HANDLE Handle = CreateFileW(Filename, OpenFlagToAccess(OpenFlag), FILE_SHARE_READ, nullptr, OpenFlagToCreation(OpenFlag), Attrs, nullptr);
+  auto access = OpenFlagToAccess(OpenFlag);
+  ULONG sharing = FILE_SHARE_READ;
+  if (access == GENERIC_WRITE) {
+    sharing |= FILE_SHARE_WRITE;
+  }
+
+  HANDLE Handle = CreateFileW(Filename, access, sharing, nullptr, OpenFlagToCreation(OpenFlag), Attrs, nullptr);
   if (Handle != INVALID_HANDLE_VALUE) {
     return AllocateFile(std::make_unique<FILE>(Handle, -1, OpenFlag & _O_APPEND));
   }

diff --git a/Source/Windows/WOW64/Module.cpp b/Source/Windows/WOW64/Module.cpp
@@ -19,6 +19,7 @@ desc: Implements the WOW64 BT module API using FEXCore
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Threads.h>
+#include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/Utils/EnumOperators.h>
 #include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/FPState.h>
@@ -457,6 +458,8 @@ void BTCpuProcessInit() {
   // Not applicable to Windows
   FEXCore::Config::EraseSet(FEXCore::Config::ConfigOption::CONFIG_TSOAUTOMIGRATION, "0");
 
+  FEXCore::Profiler::Init();
+
   FEXCore::Context::InitializeStaticTables(FEXCore::Context::MODE_32BIT);
 
   SignalDelegator = fextl::make_unique<FEX::DummyHandlers::DummySignalDelegator>();