From 6944fdbe781cd30da109eb47dea86158ec1f8ebf Mon Sep 17 00:00:00 2001
From: RSDuck <RSDuck@users.noreply.github.com>
Date: Sat, 24 Jul 2021 16:20:03 +0200
Subject: [PATCH] add support for JIT profiling with VTune

---
 CMakeLists.txt                     |  7 +++++
 cmake/FindVTune.cmake              |  5 ++++
 src/ARMJIT_x64/ARMJIT_Compiler.cpp | 43 ++++++++++++++++++++++++++++++
 src/ARMJIT_x64/ARMJIT_Compiler.h   |  8 ++++++
 src/CMakeLists.txt                 |  4 +++
 5 files changed, 67 insertions(+)
 create mode 100644 cmake/FindVTune.cmake

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 8f89c314..18f79671 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -67,6 +67,13 @@ endif()
 
 if (ENABLE_JIT)
 	add_definitions(-DJIT_ENABLED)
+
+	option(ENABLE_JIT_PROFILING "Enable JIT profiling with VTune" OFF)
+
+	if (ENABLE_JIT_PROFILING)
+		include(cmake/FindVTune.cmake)
+		add_definitions(-DJIT_PROFILING_ENABLED)
+	endif()
 endif()
 
 if (CMAKE_BUILD_TYPE STREQUAL Release)
diff --git a/cmake/FindVTune.cmake b/cmake/FindVTune.cmake
new file mode 100644
index 00000000..3c16dba6
--- /dev/null
+++ b/cmake/FindVTune.cmake
@@ -0,0 +1,5 @@
+
+find_path(VTUNE_PATH "")
+
+include_directories("${VTUNE_PATH}/include")
+link_directories("${VTUNE_PATH}/lib64")
diff --git a/src/ARMJIT_x64/ARMJIT_Compiler.cpp b/src/ARMJIT_x64/ARMJIT_Compiler.cpp
index aef8b91b..5e01077d 100644
--- a/src/ARMJIT_x64/ARMJIT_Compiler.cpp
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.cpp
@@ -22,6 +22,7 @@
 #include "../Config.h"
 
 #include <assert.h>
+#include <stdarg.h>
 
 #include "../dolphin/CommonFuncs.h"
 
@@ -298,6 +299,10 @@ Compiler::Compiler()
         SetJumpTarget(und);
         MOV(32, R(RSCRATCH3), MComplex(RCPU, RSCRATCH2, SCALE_4, offsetof(ARM, R_UND)));
         RET();
+
+#ifdef JIT_PROFILING_ENABLED
+        CreateMethod("ReadBanked", ReadBanked);
+#endif
     }
     {
         // RSCRATCH  mode
@@ -341,6 +346,10 @@ Compiler::Compiler()
         MOV(32, MComplex(RCPU, RSCRATCH2, SCALE_4, offsetof(ARM, R_UND)), R(RSCRATCH3));
         CLC();
         RET();
+
+#ifdef JIT_PROFILING_ENABLED
+        CreateMethod("WriteBanked", WriteBanked);
+#endif
     }
 
     for (int consoleType = 0; consoleType < 2; consoleType++)
@@ -401,6 +410,10 @@ Compiler::Compiler()
                     ABI_PopRegistersAndAdjustStack(CallerSavedPushRegs, 8);
                     RET();
 
+#ifdef JIT_PROFILING_ENABLED
+                    CreateMethod("FastMemStorePatch%d_%d_%d", PatchedStoreFuncs[consoleType][num][size][reg], num, size, reg);
+#endif
+
                     for (int signextend = 0; signextend < 2; signextend++)
                     {
                         PatchedLoadFuncs[consoleType][num][size][signextend][reg] = GetWritableCodePtr();
@@ -439,6 +452,10 @@ Compiler::Compiler()
                         else
                             MOVZX(32, 8 << size, rdMapped, R(RSCRATCH));
                         RET();
+
+#ifdef JIT_PROFILING_ENABLED
+                        CreateMethod("FastMemLoadPatch%d_%d_%d_%d", PatchedLoadFuncs[consoleType][num][size][signextend][reg], num, size, reg, signextend);
+#endif
                     }
                 }
             }
@@ -668,6 +685,28 @@ void Compiler::Comp_SpecialBranchBehaviour(bool taken)
     }
 }
 
+#ifdef JIT_PROFILING_ENABLED
+void Compiler::CreateMethod(const char* namefmt, void* start, ...)
+{
+    if (iJIT_IsProfilingActive())
+    {
+        va_list args;
+        va_start(args, start);
+        char name[64];
+        vsprintf(name, namefmt, args);
+        va_end(args);
+
+        iJIT_Method_Load method = {0};
+        method.method_id = iJIT_GetNewMethodID();
+        method.method_name = name;
+        method.method_load_address = start;
+        method.method_size = GetWritableCodePtr() - (u8*)start;
+
+        iJIT_NotifyEvent(iJVM_EVENT_TYPE_METHOD_LOAD_FINISHED, (void*)&method);
+    }
+}
+#endif
+
 JitBlockEntry Compiler::CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[], int instrsCount, bool hasMemoryInstr)
 {
     if (NearSize - (GetCodePtr() - NearStart) < 1024 * 32) // guess...
@@ -805,6 +844,10 @@ JitBlockEntry Compiler::CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[]
     ADD(32, MDisp(RCPU, offsetof(ARM, Cycles)), Imm32(ConstantCycles));
     JMP((u8*)ARM_Ret, true);
 
+#ifdef JIT_PROFILING_ENABLED
+    CreateMethod("JIT_Block_%d_%d_%08X", (void*)res, Num, Thumb, instrs[0].Addr);
+#endif
+
     /*FILE* codeout = fopen("codeout", "a");
     fprintf(codeout, "beginning block argargarg__ %x!!!", instrs[0].Addr);
     fwrite((u8*)res, GetWritableCodePtr() - (u8*)res, 1, codeout);
diff --git a/src/ARMJIT_x64/ARMJIT_Compiler.h b/src/ARMJIT_x64/ARMJIT_Compiler.h
index 98362821..5046c884 100644
--- a/src/ARMJIT_x64/ARMJIT_Compiler.h
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.h
@@ -25,6 +25,10 @@
 #include "../ARMJIT_Internal.h"
 #include "../ARMJIT_RegisterCache.h"
 
+#ifdef JIT_PROFILING_ENABLED
+#include <jitprofiling.h>
+#endif
+
 #include <unordered_map>
 
 namespace ARMJIT
@@ -230,6 +234,10 @@ public:
 
     u8* RewriteMemAccess(u8* pc);
 
+#ifdef JIT_PROFILING_ENABLED
+    void CreateMethod(const char* namefmt, void* start, ...);
+#endif
+
     u8* FarCode;
     u8* NearCode;
     u32 FarSize;
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 56039435..f604a396 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -124,3 +124,7 @@ else()
 		target_link_libraries(core rt)
 	endif()
 endif()
+
+if (ENABLE_JIT_PROFILING)
+	target_link_libraries(core jitprofiling)
+endif()
\ No newline at end of file