Initial instancing and asynchronous compute queues (#207)

* gnm_driver: added `sceGnmRegisterOwner` and `sceGnmRegisterResource`

* video_out: `sceVideoOutGetDeviceCapabilityInfo` for sdk runtime

* gnm_driver: correct vqid index range

* amdgpu: indirect buffer, release mem and some additional irq modes

* amdgpu: added ASC commands processor

* shader_recompiler: added support for fetch instance id

* amdgpu: classic bitfields for T# representation (debugging experience)

* renderer_vulkan: skip zero sized VBs from binding

* texture_cache: image upload logic moved into `Image` object

* gnm_driver: `sceGnmDingDong` implementation

* texture_cache: `Image` usage flags moved; correct VO buffer pitch

src/core/libraries/error_codes.h

@@ -252,6 +252,7 @@ constexpr int ORBIS_GNM_ERROR_COMPUTEQUEUE_INVALID_QUEUE_ID = 0x80D17001;
 constexpr int ORBIS_GNM_ERROR_COMPUTEQUEUE_INVALID_RING_BASE_ADDR = 0x80D17003;
 constexpr int ORBIS_GNM_ERROR_COMPUTEQUEUE_INVALID_RING_SIZE = 0x80D17002;
 constexpr int ORBIS_GNM_ERROR_COMPUTEQUEUE_INVALID_READ_PTR_ADDR = 0x80D17004;
+constexpr int ORBIS_GNM_ERROR_FAILURE = 0x8EEE00FF;
 
 // Generic
 constexpr int ORBIS_OK = 0x00000000;

src/core/libraries/gnmdriver/gnmdriver.cpp

@@ -31,6 +31,7 @@ static constexpr bool g_fair_hw_init = false;
 
 // In case if `submitDone` is issued we need to block submissions until GPU idle
 static u32 submission_lock{};
+static std::mutex m_submission{};
 static u64 frames_submitted{}; // frame counter
 
 struct AscQueueInfo {
@@ -211,9 +212,32 @@ int PS4_SYSV_ABI sceGnmDestroyWorkloadStream() {
     return ORBIS_OK;
 }
 
-int PS4_SYSV_ABI sceGnmDingDong() {
-    LOG_ERROR(Lib_GnmDriver, "(STUBBED) called");
-    return ORBIS_OK;
+void PS4_SYSV_ABI sceGnmDingDong(u32 gnm_vqid, u32 next_offs_dw) {
+    LOG_INFO(Lib_GnmDriver, "vqid {}, offset_dw {}", gnm_vqid, next_offs_dw);
+
+    if (gnm_vqid == 0) {
+        return;
+    }
+
+    std::unique_lock lock{m_submission};
+    if (submission_lock != 0) {
+        liverpool->WaitGpuIdle();
+
+        // Suspend logic goes here
+
+        submission_lock = 0;
+    }
+
+    auto vqid = gnm_vqid - 1;
+    auto& asc_queue = asc_queues[{vqid}];
+    const auto* acb_ptr = reinterpret_cast<const u32*>(asc_queue.map_addr + *asc_queue.read_addr);
+    const auto acb_size = next_offs_dw ? (next_offs_dw << 2u) - *asc_queue.read_addr
+                                       : (asc_queue.ring_size_dw << 2u) - *asc_queue.read_addr;
+
+    liverpool->SubmitAsc(vqid, {acb_ptr, acb_size >> 2u});
+
+    *asc_queue.read_addr += acb_size;
+    *asc_queue.read_addr %= asc_queue.ring_size_dw * 4;
 }
 
 int PS4_SYSV_ABI sceGnmDingDongForWorkload() {
@@ -764,10 +788,12 @@ int PS4_SYSV_ABI sceGnmMapComputeQueue(u32 pipe_id, u32 queue_id, VAddr ring_bas
     }
 
     auto vqid = asc_queues.insert(VAddr(ring_base_addr), read_ptr_addr, ring_size_dw);
+    // We need to offset index as `dingDong` assumes it to be from the range [1..64]
+    const auto gnm_vqid = vqid.index + 1;
     LOG_INFO(Lib_GnmDriver, "ASC pipe {} queue {} mapped to vqueue {}", pipe_id, queue_id,
-             vqid.index);
+             gnm_vqid);
 
-    return vqid.index;
+    return gnm_vqid;
 }
 
 int PS4_SYSV_ABI sceGnmMapComputeQueueWithPriority(u32 pipe_id, u32 queue_id, VAddr ring_base_addr,
@@ -814,14 +840,16 @@ int PS4_SYSV_ABI sceGnmRegisterGnmLiveCallbackConfig() {
     return ORBIS_OK;
 }
 
-int PS4_SYSV_ABI sceGnmRegisterOwner() {
-    LOG_ERROR(Lib_GnmDriver, "(STUBBED) called");
-    return ORBIS_OK;
+s32 PS4_SYSV_ABI sceGnmRegisterOwner(void* handle, const char* name) {
+    LOG_TRACE(Lib_GnmDriver, "called");
+    return ORBIS_GNM_ERROR_FAILURE; // PA Debug is always disabled in retail FW
 }
 
-int PS4_SYSV_ABI sceGnmRegisterResource() {
-    LOG_ERROR(Lib_GnmDriver, "(STUBBED) called");
-    return ORBIS_OK;
+s32 PS4_SYSV_ABI sceGnmRegisterResource(void* res_handle, void* owner_handle, const void* addr,
+                                        size_t size, const char* name, int res_type,
+                                        u64 user_data) {
+    LOG_TRACE(Lib_GnmDriver, "called");
+    return ORBIS_GNM_ERROR_FAILURE; // PA Debug is always disabled in retail FW
 }
 
 int PS4_SYSV_ABI sceGnmRequestFlipAndSubmitDone() {

src/core/libraries/gnmdriver/gnmdriver.h

@@ -33,7 +33,7 @@ int PS4_SYSV_ABI sceGnmDebuggerWriteSqIndirectRegister();
 int PS4_SYSV_ABI sceGnmDebugHardwareStatus();
 s32 PS4_SYSV_ABI sceGnmDeleteEqEvent(SceKernelEqueue eq, u64 id);
 int PS4_SYSV_ABI sceGnmDestroyWorkloadStream();
-int PS4_SYSV_ABI sceGnmDingDong();
+void PS4_SYSV_ABI sceGnmDingDong(u32 gnm_vqid, u32 next_offs_dw);
 int PS4_SYSV_ABI sceGnmDingDongForWorkload();
 int PS4_SYSV_ABI sceGnmDisableMipStatsReport();
 s32 PS4_SYSV_ABI sceGnmDispatchDirect(u32* cmdbuf, u32 size, u32 threads_x, u32 threads_y,
@@ -125,8 +125,9 @@ int PS4_SYSV_ABI sceGnmQueryResourceRegistrationUserMemoryRequirements();
 int PS4_SYSV_ABI sceGnmRaiseUserExceptionEvent();
 int PS4_SYSV_ABI sceGnmRegisterGdsResource();
 int PS4_SYSV_ABI sceGnmRegisterGnmLiveCallbackConfig();
-int PS4_SYSV_ABI sceGnmRegisterOwner();
-int PS4_SYSV_ABI sceGnmRegisterResource();
+s32 PS4_SYSV_ABI sceGnmRegisterOwner(void* handle, const char* name);
+s32 PS4_SYSV_ABI sceGnmRegisterResource(void* res_handle, void* owner_handle, const void* addr,
+                                        size_t size, const char* name, int res_type, u64 user_data);
 int PS4_SYSV_ABI sceGnmRequestFlipAndSubmitDone();
 int PS4_SYSV_ABI sceGnmRequestFlipAndSubmitDoneForWorkload();
 int PS4_SYSV_ABI sceGnmRequestMipStatsReportAndReset();

src/core/libraries/videoout/video_out.cpp

@@ -288,6 +288,8 @@ void RegisterLib(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("uquVH4-Du78", "libSceVideoOut", 1, "libSceVideoOut", 0, 0, sceVideoOutClose);
     LIB_FUNCTION("1FZBKy8HeNU", "libSceVideoOut", 1, "libSceVideoOut", 0, 0,
                  sceVideoOutGetVblankStatus);
+    LIB_FUNCTION("kGVLc3htQE8", "libSceVideoOut", 1, "libSceVideoOut", 0, 0,
+                 sceVideoOutGetDeviceCapabilityInfo);
 
     // openOrbis appears to have libSceVideoOut_v1 module libSceVideoOut_v1.1
     LIB_FUNCTION("Up36PTk687E", "libSceVideoOut", 1, "libSceVideoOut", 1, 1, sceVideoOutOpen);