mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2024-12-28 18:46:06 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

hot-fix: Tracy operation restored; memory leak fix as a bonus

Browse source
This commit is contained in:
psucien 2024-12-15 20:53:29 +01:00
parent 3001b007f6
commit e7c4ffe032
5 changed files with 16 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
externals/CMakeLists.txt vendored
View file

@ -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_NO_CRASH_HANDLER "" ON) # Otherwise texture cache exceptions will be treaten as a crash
option(TRACY_ON_DEMAND "" ON) option(TRACY_ON_DEMAND "" ON)
option(TRACY_NO_FRAME_IMAGE "" 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_SYSTEM_TRACING "" ON)
option(TRACY_NO_CALLSTACK "" ON) option(TRACY_NO_CALLSTACK "" ON)
option(TRACY_NO_CODE_TRANSFER "" ON) option(TRACY_NO_CODE_TRANSFER "" ON)

7
src/common/debug.h
View file

@ -17,6 +17,8 @@ static inline bool IsProfilerConnected() {
return tracy::GetProfiler().IsConnected(); return tracy::GetProfiler().IsConnected();
} }
#define TRACY_GPU_ENABLED 0
#define CUSTOM_LOCK(type, varname) \ #define CUSTOM_LOCK(type, varname) \
tracy::LockableCtx varname { \ tracy::LockableCtx varname { \
[]() -> const tracy::SourceLocationData* { \ []() -> const tracy::SourceLocationData* { \
@ -58,5 +60,10 @@ enum MarkersPalette : int {
#define FRAME_END FrameMark #define FRAME_END FrameMark
#ifdef TRACY_FIBERS
#define FIBER_ENTER(name) TracyFiberEnter(name) #define FIBER_ENTER(name) TracyFiberEnter(name)
#define FIBER_EXIT TracyFiberLeave #define FIBER_EXIT TracyFiberLeave
#else
#define FIBER_ENTER(name)
#define FIBER_EXIT
#endif

3
src/core/libraries/gnmdriver/gnmdriver.cpp
View file

@ -488,6 +488,7 @@ int PS4_SYSV_ABI sceGnmDestroyWorkloadStream() {
} }
void PS4_SYSV_ABI sceGnmDingDong(u32 gnm_vqid, u32 next_offs_dw) { 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); LOG_DEBUG(Lib_GnmDriver, "vqid {}, offset_dw {}", gnm_vqid, next_offs_dw);
if (gnm_vqid == 0) { if (gnm_vqid == 0) {
@ -2166,6 +2167,7 @@ int PS4_SYSV_ABI sceGnmSubmitCommandBuffersForWorkload(u32 workload, u32 count,
u32* dcb_sizes_in_bytes, u32* dcb_sizes_in_bytes,
const u32* ccb_gpu_addrs[], const u32* ccb_gpu_addrs[],
u32* ccb_sizes_in_bytes) { u32* ccb_sizes_in_bytes) {
HLE_TRACE;
LOG_DEBUG(Lib_GnmDriver, "called"); LOG_DEBUG(Lib_GnmDriver, "called");
if (!dcb_gpu_addrs || !dcb_sizes_in_bytes) { 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() { int PS4_SYSV_ABI sceGnmSubmitDone() {
HLE_TRACE;
LOG_DEBUG(Lib_GnmDriver, "called"); LOG_DEBUG(Lib_GnmDriver, "called");
WaitGpuIdle(); WaitGpuIdle();
if (!liverpool->IsGpuIdle()) { if (!liverpool->IsGpuIdle()) {

3
src/video_core/amdgpu/liverpool.cpp
View file

@ -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 // there are no other submits to yield to we can sleep the thread
// instead and allow other tasks to run. // instead and allow other tasks to run.
const u64* wait_addr = wait_reg_mem->Address<u64*>(); 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(); }); vo_port->WaitVoLabel([&] { return wait_reg_mem->Test(); });
} }
while (!wait_reg_mem->Test()) { while (!wait_reg_mem->Test()) {

4
src/video_core/renderer_vulkan/vk_instance.cpp
View file

@ -9,6 +9,7 @@
#include "common/assert.h" #include "common/assert.h"
#include "common/config.h" #include "common/config.h"
#include "common/debug.h"
#include "sdl_window.h" #include "sdl_window.h"
#include "video_core/renderer_vulkan/liverpool_to_vk.h" #include "video_core/renderer_vulkan/liverpool_to_vk.h"
#include "video_core/renderer_vulkan/vk_instance.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 // 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_COLOR_WRITE_ENABLE_EXTENSION_NAME);
color_write_en &= add_extension(VK_EXT_EXTENDED_DYNAMIC_STATE_3_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); const bool robustness = add_extension(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME);
list_restart = add_extension(VK_EXT_PRIMITIVE_TOPOLOGY_LIST_RESTART_EXTENSION_NAME); list_restart = add_extension(VK_EXT_PRIMITIVE_TOPOLOGY_LIST_RESTART_EXTENSION_NAME);
maintenance5 = add_extension(VK_KHR_MAINTENANCE_5_EXTENSION_NAME); maintenance5 = add_extension(VK_KHR_MAINTENANCE_5_EXTENSION_NAME);