From e7c4ffe032a0dd5605322363b9766425ea5531f9 Mon Sep 17 00:00:00 2001
From: psucien <bad_cast@protonmail.com>
Date: Sun, 15 Dec 2024 20:53:29 +0100
Subject: [PATCH] hot-fix: Tracy operation restored; memory leak fix as a bonus

---
 externals/CMakeLists.txt                       | 2 +-
 src/common/debug.h                             | 7 +++++++
 src/core/libraries/gnmdriver/gnmdriver.cpp     | 3 +++
 src/video_core/amdgpu/liverpool.cpp            | 3 ++-
 src/video_core/renderer_vulkan/vk_instance.cpp | 4 +++-
 5 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/externals/CMakeLists.txt b/externals/CMakeLists.txt
index 1ab23a40..dbe6794d 100644
--- a/externals/CMakeLists.txt
+++ b/externals/CMakeLists.txt
@@ -193,7 +193,7 @@ option(TRACY_ENABLE "" ON)
 option(TRACY_NO_CRASH_HANDLER "" ON) # Otherwise texture cache exceptions will be treaten as a crash
 option(TRACY_ON_DEMAND "" ON)
 option(TRACY_NO_FRAME_IMAGE "" ON)
-option(TRACY_FIBERS "" ON) # For AmdGpu frontend profiling
+option(TRACY_FIBERS "" OFF) # For AmdGpu frontend profiling, disabled due to instability
 option(TRACY_NO_SYSTEM_TRACING "" ON)
 option(TRACY_NO_CALLSTACK "" ON)
 option(TRACY_NO_CODE_TRANSFER "" ON)
diff --git a/src/common/debug.h b/src/common/debug.h
index 4d42aa4a..882e9e5c 100644
--- a/src/common/debug.h
+++ b/src/common/debug.h
@@ -17,6 +17,8 @@ static inline bool IsProfilerConnected() {
     return tracy::GetProfiler().IsConnected();
 }
 
+#define TRACY_GPU_ENABLED 0
+
 #define CUSTOM_LOCK(type, varname)                                                                 \
     tracy::LockableCtx varname {                                                                   \
         []() -> const tracy::SourceLocationData* {                                                 \
@@ -58,5 +60,10 @@ enum MarkersPalette : int {
 
 #define FRAME_END FrameMark
 
+#ifdef TRACY_FIBERS
 #define FIBER_ENTER(name) TracyFiberEnter(name)
 #define FIBER_EXIT TracyFiberLeave
+#else
+#define FIBER_ENTER(name)
+#define FIBER_EXIT
+#endif
diff --git a/src/core/libraries/gnmdriver/gnmdriver.cpp b/src/core/libraries/gnmdriver/gnmdriver.cpp
index 583339dd..1a6007bf 100644
--- a/src/core/libraries/gnmdriver/gnmdriver.cpp
+++ b/src/core/libraries/gnmdriver/gnmdriver.cpp
@@ -488,6 +488,7 @@ int PS4_SYSV_ABI sceGnmDestroyWorkloadStream() {
 }
 
 void PS4_SYSV_ABI sceGnmDingDong(u32 gnm_vqid, u32 next_offs_dw) {
+    HLE_TRACE;
     LOG_DEBUG(Lib_GnmDriver, "vqid {}, offset_dw {}", gnm_vqid, next_offs_dw);
 
     if (gnm_vqid == 0) {
@@ -2166,6 +2167,7 @@ int PS4_SYSV_ABI sceGnmSubmitCommandBuffersForWorkload(u32 workload, u32 count,
                                                        u32* dcb_sizes_in_bytes,
                                                        const u32* ccb_gpu_addrs[],
                                                        u32* ccb_sizes_in_bytes) {
+    HLE_TRACE;
     LOG_DEBUG(Lib_GnmDriver, "called");
 
     if (!dcb_gpu_addrs || !dcb_sizes_in_bytes) {
@@ -2258,6 +2260,7 @@ s32 PS4_SYSV_ABI sceGnmSubmitCommandBuffers(u32 count, const u32* dcb_gpu_addrs[
 }
 
 int PS4_SYSV_ABI sceGnmSubmitDone() {
+    HLE_TRACE;
     LOG_DEBUG(Lib_GnmDriver, "called");
     WaitGpuIdle();
     if (!liverpool->IsGpuIdle()) {
diff --git a/src/video_core/amdgpu/liverpool.cpp b/src/video_core/amdgpu/liverpool.cpp
index 8cca636c..5dd3edd6 100644
--- a/src/video_core/amdgpu/liverpool.cpp
+++ b/src/video_core/amdgpu/liverpool.cpp
@@ -659,7 +659,8 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                 // there are no other submits to yield to we can sleep the thread
                 // instead and allow other tasks to run.
                 const u64* wait_addr = wait_reg_mem->Address<u64*>();
-                if (vo_port->IsVoLabel(wait_addr) && num_submits == 1) {
+                if (vo_port->IsVoLabel(wait_addr) &&
+                    num_submits == mapped_queues[GfxQueueId].submits.size()) {
                     vo_port->WaitVoLabel([&] { return wait_reg_mem->Test(); });
                 }
                 while (!wait_reg_mem->Test()) {
diff --git a/src/video_core/renderer_vulkan/vk_instance.cpp b/src/video_core/renderer_vulkan/vk_instance.cpp
index d7bfaee4..b479c146 100644
--- a/src/video_core/renderer_vulkan/vk_instance.cpp
+++ b/src/video_core/renderer_vulkan/vk_instance.cpp
@@ -9,6 +9,7 @@
 
 #include "common/assert.h"
 #include "common/config.h"
+#include "common/debug.h"
 #include "sdl_window.h"
 #include "video_core/renderer_vulkan/liverpool_to_vk.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
@@ -261,7 +262,8 @@ bool Instance::CreateDevice() {
     // The next two extensions are required to be available together in order to support write masks
     color_write_en = add_extension(VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
     color_write_en &= add_extension(VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME);
-    const bool calibrated_timestamps = add_extension(VK_EXT_CALIBRATED_TIMESTAMPS_EXTENSION_NAME);
+    const bool calibrated_timestamps =
+        TRACY_GPU_ENABLED ? add_extension(VK_EXT_CALIBRATED_TIMESTAMPS_EXTENSION_NAME) : false;
     const bool robustness = add_extension(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME);
     list_restart = add_extension(VK_EXT_PRIMITIVE_TOPOLOGY_LIST_RESTART_EXTENSION_NAME);
     maintenance5 = add_extension(VK_KHR_MAINTENANCE_5_EXTENSION_NAME);