Merge branch 'shadps4-emu:main' into main

src/common/config.cpp

@@ -25,7 +25,9 @@ static bool shouldDumpPM4 = false;
 static u32 vblankDivider = 1;
 static bool vkValidation = false;
 static bool vkValidationSync = false;
+static bool vkValidationGpu = false;
 static bool rdocEnable = false;
+static bool rdocMarkersEnable = false;
 // Gui
 std::string settings_install_dir = "";
 u32 main_window_geometry_x = 400;
@@ -102,6 +104,10 @@ bool isRdocEnabled() {
     return rdocEnable;
 }
 
+bool isMarkersEnabled() {
+    return rdocMarkersEnable;
+}
+
 u32 vblankDiv() {
     return vblankDivider;
 }
@@ -114,6 +120,10 @@ bool vkValidationSyncEnabled() {
     return vkValidationSync;
 }
 
+bool vkValidationGpuEnabled() {
+    return vkValidationGpu;
+}
+
 void setScreenWidth(u32 width) {
     screenWidth = width;
 }
@@ -319,7 +329,9 @@ void load(const std::filesystem::path& path) {
         gpuId = toml::find_or<int>(vk, "gpuId", -1);
         vkValidation = toml::find_or<bool>(vk, "validation", false);
         vkValidationSync = toml::find_or<bool>(vk, "validation_sync", false);
+        vkValidationGpu = toml::find_or<bool>(vk, "validation_gpu", true);
         rdocEnable = toml::find_or<bool>(vk, "rdocEnable", false);
+        rdocMarkersEnable = toml::find_or<bool>(vk, "rdocMarkersEnable", false);
     }
 
     if (data.contains("Debug")) {
@@ -394,7 +406,9 @@ void save(const std::filesystem::path& path) {
     data["Vulkan"]["gpuId"] = gpuId;
     data["Vulkan"]["validation"] = vkValidation;
     data["Vulkan"]["validation_sync"] = vkValidationSync;
+    data["Vulkan"]["validation_gpu"] = vkValidationGpu;
     data["Vulkan"]["rdocEnable"] = rdocEnable;
+    data["Vulkan"]["rdocMarkersEnable"] = rdocMarkersEnable;
     data["Debug"]["DebugDump"] = isDebugDump;
     data["LLE"]["libc"] = isLibc;
     data["GUI"]["theme"] = mw_themes;

src/common/config.h

@@ -27,6 +27,7 @@ bool nullGpu();
 bool dumpShaders();
 bool dumpPM4();
 bool isRdocEnabled();
+bool isMarkersEnabled();
 u32 vblankDiv();
 
 void setDebugDump(bool enable);
@@ -50,6 +51,7 @@ void setRdocEnabled(bool enable);
 
 bool vkValidationEnabled();
 bool vkValidationSyncEnabled();
+bool vkValidationGpuEnabled();
 
 // Gui
 void setMainWindowGeometry(u32 x, u32 y, u32 w, u32 h);

src/core/libraries/gnmdriver/gnmdriver.cpp

@@ -956,9 +956,9 @@ int PS4_SYSV_ABI sceGnmGetGpuBlockStatus() {
     return ORBIS_OK;
 }
 
-int PS4_SYSV_ABI sceGnmGetGpuCoreClockFrequency() {
+u32 PS4_SYSV_ABI sceGnmGetGpuCoreClockFrequency() {
-    LOG_DEBUG(Lib_GnmDriver, "(STUBBED) called");
+    LOG_TRACE(Lib_GnmDriver, "called");
-    return ORBIS_OK;
+    return Config::isNeoMode() ? 911'000'000 : 800'000'000;
 }
 
 int PS4_SYSV_ABI sceGnmGetGpuInfoStatus() {
@@ -1706,8 +1706,18 @@ int PS4_SYSV_ABI sceGnmSetupMipStatsReport() {
     return ORBIS_OK;
 }
 
-int PS4_SYSV_ABI sceGnmSetVgtControl() {
+s32 PS4_SYSV_ABI sceGnmSetVgtControl(u32* cmdbuf, u32 size, u32 prim_group_sz_minus_one,
-    LOG_ERROR(Lib_GnmDriver, "(STUBBED) called");
+                                     u32 partial_vs_wave_mode, u32 wd_switch_only_on_eop_mode) {
+    LOG_TRACE(Lib_GnmDriver, "called");
+
+    if (!cmdbuf || size != 3 || (prim_group_sz_minus_one >= 0x100) ||
+        ((wd_switch_only_on_eop_mode | partial_vs_wave_mode) >= 2)) {
+        return -1;
+    }
+
+    const u32 reg_value =
+        ((partial_vs_wave_mode & 1) << 0x10) | (prim_group_sz_minus_one & 0xffffu);
+    PM4CmdSetData::SetContextReg(cmdbuf, 0x2aau, reg_value); // IA_MULTI_VGT_PARAM
     return ORBIS_OK;
 }
 

src/core/libraries/gnmdriver/gnmdriver.h

@@ -85,7 +85,7 @@ int PS4_SYSV_ABI sceGnmGetDebugTimestamp();
 int PS4_SYSV_ABI sceGnmGetEqEventType();
 int PS4_SYSV_ABI sceGnmGetEqTimeStamp();
 int PS4_SYSV_ABI sceGnmGetGpuBlockStatus();
-int PS4_SYSV_ABI sceGnmGetGpuCoreClockFrequency();
+u32 PS4_SYSV_ABI sceGnmGetGpuCoreClockFrequency();
 int PS4_SYSV_ABI sceGnmGetGpuInfoStatus();
 int PS4_SYSV_ABI sceGnmGetLastWaitedAddress();
 int PS4_SYSV_ABI sceGnmGetNumTcaUnits();
@@ -161,7 +161,8 @@ int PS4_SYSV_ABI sceGnmSetResourceUserData();
 int PS4_SYSV_ABI sceGnmSetSpiEnableSqCounters();
 int PS4_SYSV_ABI sceGnmSetSpiEnableSqCountersForUnitInstance();
 int PS4_SYSV_ABI sceGnmSetupMipStatsReport();
-int PS4_SYSV_ABI sceGnmSetVgtControl();
+s32 PS4_SYSV_ABI sceGnmSetVgtControl(u32* cmdbuf, u32 size, u32 prim_group_sz_minus_one,
+                                     u32 partial_vs_wave_mode, u32 wd_switch_only_on_eop_mode);
 s32 PS4_SYSV_ABI sceGnmSetVsShader(u32* cmdbuf, u32 size, const u32* vs_regs, u32 shader_modifier);
 int PS4_SYSV_ABI sceGnmSetWaveLimitMultiplier();
 int PS4_SYSV_ABI sceGnmSetWaveLimitMultipliers();

src/core/libraries/kernel/threads/semaphore.cpp

@@ -174,10 +174,16 @@ s32 PS4_SYSV_ABI sceKernelCreateSema(OrbisKernelSema* sem, const char* pName, u3
 }
 
 s32 PS4_SYSV_ABI sceKernelWaitSema(OrbisKernelSema sem, s32 needCount, u32* pTimeout) {
+    if (!sem) {
+        return ORBIS_KERNEL_ERROR_ESRCH;
+    }
     return sem->Wait(true, needCount, pTimeout);
 }
 
 s32 PS4_SYSV_ABI sceKernelSignalSema(OrbisKernelSema sem, s32 signalCount) {
+    if (!sem) {
+        return ORBIS_KERNEL_ERROR_ESRCH;
+    }
     if (!sem->Signal(signalCount)) {
         return ORBIS_KERNEL_ERROR_EINVAL;
     }
@@ -185,10 +191,16 @@ s32 PS4_SYSV_ABI sceKernelSignalSema(OrbisKernelSema sem, s32 signalCount) {
 }
 
 s32 PS4_SYSV_ABI sceKernelPollSema(OrbisKernelSema sem, s32 needCount) {
+    if (!sem) {
+        return ORBIS_KERNEL_ERROR_ESRCH;
+    }
     return sem->Wait(false, needCount, nullptr);
 }
 
 int PS4_SYSV_ABI sceKernelCancelSema(OrbisKernelSema sem, s32 setCount, s32* pNumWaitThreads) {
+    if (!sem) {
+        return ORBIS_KERNEL_ERROR_ESRCH;
+    }
     return sem->Cancel(setCount, pNumWaitThreads);
 }
 

src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp

@@ -386,19 +386,12 @@ static Id GetBufferFormatValue(EmitContext& ctx, u32 handle, Id address, u32 com
                 if (is_signed) {
                     value = ctx.OpBitFieldSExtract(ctx.S32[1], value, comp_offset,
                                                    ctx.ConstU32(bit_width));
-                    value = ctx.OpConvertSToF(ctx.F32[1], value);
                 } else {
                     value = ctx.OpBitFieldUExtract(ctx.U32[1], value, comp_offset,
                                                    ctx.ConstU32(bit_width));
-                    value = ctx.OpConvertUToF(ctx.F32[1], value);
-                }
-            } else {
-                if (is_signed) {
-                    value = ctx.OpConvertSToF(ctx.F32[1], value);
-                } else {
-                    value = ctx.OpConvertUToF(ctx.F32[1], value);
                 }
             }
+            value = ctx.OpBitcast(ctx.F32[1], value);
             return ConvertValue(ctx, value, num_format, bit_width);
         }
         break;

src/shader_recompiler/backend/spirv/emit_spirv_image.cpp

@@ -33,14 +33,14 @@ Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id c
                                         operands.operands);
 }
 
-Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
+Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id lod,
                               Id offset) {
     const auto& texture = ctx.images[handle & 0xFFFF];
     const Id image = ctx.OpLoad(texture.image_type, texture.id);
     const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
     const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
     return ctx.OpImageSampleExplicitLod(ctx.F32[4], sampled_image, coords,
-                                        spv::ImageOperandsMask::Lod, ctx.ConstF32(0.f));
+                                        spv::ImageOperandsMask::Lod, lod);
 }
 
 Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,

src/shader_recompiler/backend/spirv/emit_spirv_instructions.h

@@ -359,7 +359,7 @@ Id EmitConvertU32U16(EmitContext& ctx, Id value);
 
 Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
                               Id offset);
-Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
+Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id lod,
                               Id offset);
 Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,
                                   Id bias_lc, const IR::Value& offset);

src/shader_recompiler/ir/passes/resource_tracking_pass.cpp

@@ -376,9 +376,11 @@ s32 TryHandleInlineCbuf(IR::Inst& inst, Info& info, Descriptors& descriptors,
         return -1;
     }
     // We have found this pattern. Build the sharp.
-    std::array<u64, 2> buffer;
+    std::array<u32, 4> buffer;
     buffer[0] = info.pgm_base + p0->Arg(0).U32() + p0->Arg(1).U32();
-    buffer[1] = handle->Arg(2).U32() | handle->Arg(3).U64() << 32;
+    buffer[1] = 0;
+    buffer[2] = handle->Arg(2).U32();
+    buffer[3] = handle->Arg(3).U32();
     cbuf = std::bit_cast<AmdGpu::Buffer>(buffer);
     // Assign a binding to this sharp.
     return descriptors.Add(BufferResource{

src/shader_recompiler/runtime_info.h

@@ -116,7 +116,7 @@ struct PushData {
     std::array<u8, 32> buf_offsets;
 
     void AddOffset(u32 binding, u32 offset) {
-        ASSERT(offset < 64 && binding < 32);
+        ASSERT(offset < 256 && binding < buf_offsets.size());
         buf_offsets[binding] = offset;
     }
 };

src/video_core/amdgpu/liverpool.cpp

@@ -180,6 +180,17 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                 Platform::IrqC::Instance()->Signal(Platform::InterruptId::GfxFlip);
                 break;
             }
+            case PM4CmdNop::PayloadType::DebugMarkerPush: {
+                const auto marker_sz = nop->header.count.Value() * 2;
+                const std::string_view label{reinterpret_cast<const char*>(&nop->data_block[1]),
+                                             marker_sz};
+                rasterizer->ScopeMarkerBegin(label);
+                break;
+            }
+            case PM4CmdNop::PayloadType::DebugMarkerPop: {
+                rasterizer->ScopeMarkerEnd();
+                break;
+            }
             default:
                 break;
             }
@@ -226,7 +237,7 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                 ASSERT(col_buf_id < NumColorBuffers);
 
                 const auto nop_offset = header->type3.count;
-                if (nop_offset == 0x0e || nop_offset == 0x0d) {
+                if (nop_offset == 0x0e || nop_offset == 0x0d || nop_offset == 0x0b) {
                     ASSERT_MSG(payload[nop_offset] == 0xc0001000,
                                "NOP hint is missing in CB setup sequence");
                     last_cb_extent[col_buf_id].raw = payload[nop_offset + 1];
@@ -295,8 +306,9 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             regs.num_indices = draw_index->index_count;
             regs.draw_initiator = draw_index->draw_initiator;
             if (rasterizer) {
-                rasterizer->ScopeMarkerBegin(
+                const auto cmd_address = reinterpret_cast<const void*>(header);
-                    fmt::format("dcb:{}:DrawIndex2", reinterpret_cast<const void*>(dcb.data())));
+                rasterizer->ScopeMarkerBegin(fmt::format("dcb:{}:DrawIndex2", cmd_address));
+                rasterizer->Breadcrumb(u64(cmd_address));
                 rasterizer->Draw(true);
                 rasterizer->ScopeMarkerEnd();
             }
@@ -308,8 +320,9 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             regs.num_indices = draw_index_off->index_count;
             regs.draw_initiator = draw_index_off->draw_initiator;
             if (rasterizer) {
-                rasterizer->ScopeMarkerBegin(fmt::format(
+                const auto cmd_address = reinterpret_cast<const void*>(header);
-                    "dcb:{}:DrawIndexOffset2", reinterpret_cast<const void*>(dcb.data())));
+                rasterizer->ScopeMarkerBegin(fmt::format("dcb:{}:DrawIndexOffset2", cmd_address));
+                rasterizer->Breadcrumb(u64(cmd_address));
                 rasterizer->Draw(true, draw_index_off->index_offset);
                 rasterizer->ScopeMarkerEnd();
             }
@@ -320,8 +333,9 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             regs.num_indices = draw_index->index_count;
             regs.draw_initiator = draw_index->draw_initiator;
             if (rasterizer) {
-                rasterizer->ScopeMarkerBegin(
+                const auto cmd_address = reinterpret_cast<const void*>(header);
-                    fmt::format("dcb:{}:DrawIndexAuto", reinterpret_cast<const void*>(dcb.data())));
+                rasterizer->ScopeMarkerBegin(fmt::format("dcb:{}:DrawIndexAuto", cmd_address));
+                rasterizer->Breadcrumb(u64(cmd_address));
                 rasterizer->Draw(false);
                 rasterizer->ScopeMarkerEnd();
             }
@@ -334,8 +348,9 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
             regs.cs_program.dim_z = dispatch_direct->dim_z;
             regs.cs_program.dispatch_initiator = dispatch_direct->dispatch_initiator;
             if (rasterizer && (regs.cs_program.dispatch_initiator & 1)) {
-                rasterizer->ScopeMarkerBegin(
+                const auto cmd_address = reinterpret_cast<const void*>(header);
-                    fmt::format("dcb:{}:Dispatch", reinterpret_cast<const void*>(dcb.data())));
+                rasterizer->ScopeMarkerBegin(fmt::format("dcb:{}:Dispatch", cmd_address));
+                rasterizer->Breadcrumb(u64(cmd_address));
                 rasterizer->DispatchDirect();
                 rasterizer->ScopeMarkerEnd();
             }
@@ -486,8 +501,9 @@ Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, int vqid) {
             regs.cs_program.dim_z = dispatch_direct->dim_z;
             regs.cs_program.dispatch_initiator = dispatch_direct->dispatch_initiator;
             if (rasterizer && (regs.cs_program.dispatch_initiator & 1)) {
-                rasterizer->ScopeMarkerBegin(fmt::format(
+                const auto cmd_address = reinterpret_cast<const void*>(header);
-                    "acb[{}]:{}:Dispatch", vqid, reinterpret_cast<const void*>(acb.data())));
+                rasterizer->ScopeMarkerBegin(fmt::format("acb[{}]:{}:Dispatch", vqid, cmd_address));
+                rasterizer->Breadcrumb(u64(cmd_address));
                 rasterizer->DispatchDirect();
                 rasterizer->ScopeMarkerEnd();
             }

src/video_core/amdgpu/liverpool.h

@@ -766,7 +766,8 @@ struct Liverpool {
         }
 
         TilingMode GetTilingMode() const {
-            return attrib.tile_mode_index;
+            return info.linear_general ? TilingMode::Display_Linear
+                                       : attrib.tile_mode_index.Value();
         }
 
         bool IsTiled() const {

src/video_core/amdgpu/pm4_cmds.h

@@ -282,6 +282,13 @@ enum class InterruptSelect : u32 {
     IrqUndocumented = 3,
 };
 
+static u64 GetGpuClock64() {
+    auto now = std::chrono::high_resolution_clock::now();
+    auto duration = now.time_since_epoch();
+    auto ticks = std::chrono::duration_cast<std::chrono::nanoseconds>(duration).count();
+    return static_cast<u64>(ticks);
+}
+
 struct PM4CmdEventWriteEop {
     PM4Type3Header header;
     union {
@@ -325,6 +332,10 @@ struct PM4CmdEventWriteEop {
             *Address<u64>() = DataQWord();
             break;
         }
+        case DataSelect::GpuClock64: {
+            *Address<u64>() = GetGpuClock64();
+            break;
+        }
         case DataSelect::PerfCounter: {
             *Address<u64>() = Common::FencedRDTSC();
             break;
@@ -652,13 +663,6 @@ struct PM4CmdReleaseMem {
         return data_lo | u64(data_hi) << 32;
     }
 
-    uint64_t GetGpuClock64() const {
-        auto now = std::chrono::high_resolution_clock::now();
-        auto duration = now.time_since_epoch();
-        auto ticks = std::chrono::duration_cast<std::chrono::nanoseconds>(duration).count();
-        return static_cast<uint64_t>(ticks);
-    }
-
     void SignalFence(Platform::InterruptId irq_id) const {
         switch (data_sel.Value()) {
         case DataSelect::Data32Low: {

src/video_core/amdgpu/pm4_opcodes.h

@@ -41,6 +41,7 @@ enum class PM4ItOpcode : u32 {
     CondIndirectBuffer = 0x3F,
     CopyData = 0x40,
     CommandProcessorDma = 0x41,
+    PfpSyncMe = 0x42,
     SurfaceSync = 0x43,
     CondWrite = 0x45,
     EventWrite = 0x46,

src/video_core/buffer_cache/buffer.cpp

@@ -106,10 +106,8 @@ Buffer::Buffer(const Vulkan::Instance& instance_, MemoryUsage usage_, VAddr cpu_
     VmaAllocationInfo alloc_info{};
     buffer.Create(buffer_ci, usage, &alloc_info);
 
-    if (instance->HasDebuggingToolAttached()) {
+    const auto device = instance->GetDevice();
-        const auto device = instance->GetDevice();
+    Vulkan::SetObjectName(device, Handle(), "Buffer {:#x}:{:#x}", cpu_addr, size_bytes);
-        Vulkan::SetObjectName(device, Handle(), "Buffer {:#x} {} KiB", cpu_addr, size_bytes / 1024);
-    }
 
     // Map it if it is host visible.
     VkMemoryPropertyFlags property_flags{};
@@ -152,10 +150,8 @@ StreamBuffer::StreamBuffer(const Vulkan::Instance& instance, Vulkan::Scheduler&
     ReserveWatches(current_watches, WATCHES_INITIAL_RESERVE);
     ReserveWatches(previous_watches, WATCHES_INITIAL_RESERVE);
     const auto device = instance.GetDevice();
-    if (instance.HasDebuggingToolAttached()) {
+    Vulkan::SetObjectName(device, Handle(), "StreamBuffer({}):{:#x}", BufferTypeName(usage),
-        Vulkan::SetObjectName(device, Handle(), "StreamBuffer({}): {} KiB", BufferTypeName(usage),
+                          size_bytes);
-                              size_bytes / 1024);
-    }
 }
 
 std::pair<u8*, u64> StreamBuffer::Map(u64 size, u64 alignment) {

src/video_core/buffer_cache/buffer.h

@@ -146,6 +146,10 @@ public:
         return offset;
     }
 
+    u64 GetFreeSize() const {
+        return size_bytes - offset - mapped_size;
+    }
+
 private:
     struct Watch {
         u64 tick{};

src/video_core/renderer_vulkan/vk_instance.cpp

@@ -8,6 +8,7 @@
 #include <fmt/ranges.h>
 
 #include "common/assert.h"
+#include "common/config.h"
 #include "sdl_window.h"
 #include "video_core/renderer_vulkan/liverpool_to_vk.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
@@ -213,6 +214,13 @@ bool Instance::CreateDevice() {
     add_extension(VK_KHR_MAINTENANCE_4_EXTENSION_NAME);
     add_extension(VK_KHR_DYNAMIC_RENDERING_EXTENSION_NAME);
     add_extension(VK_EXT_SHADER_DEMOTE_TO_HELPER_INVOCATION_EXTENSION_NAME);
+    const bool has_sync2 = add_extension(VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME);
+
+    if (has_sync2) {
+        has_nv_checkpoints = Config::isMarkersEnabled()
+                                 ? add_extension(VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME)
+                                 : false;
+    }
 
     const auto family_properties = physical_device.getQueueFamilyProperties();
     if (family_properties.empty()) {
@@ -308,6 +316,9 @@ bool Instance::CreateDevice() {
         vk::PhysicalDeviceRobustness2FeaturesEXT{
             .nullDescriptor = true,
         },
+        vk::PhysicalDeviceSynchronization2Features{
+            .synchronization2 = has_sync2,
+        },
     };
 
     if (!color_write_en) {

src/video_core/renderer_vulkan/vk_instance.h

@@ -88,6 +88,10 @@ public:
         return profiler_context;
     }
 
+    bool HasNvCheckpoints() const {
+        return has_nv_checkpoints;
+    }
+
     /// Returns true when a known debugging tool is attached.
     bool HasDebuggingToolAttached() const {
         return has_renderdoc || has_nsight_graphics;
@@ -259,6 +263,7 @@ private:
     bool debug_utils_supported{};
     bool has_nsight_graphics{};
     bool has_renderdoc{};
+    bool has_nv_checkpoints{};
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_platform.cpp

@@ -221,12 +221,61 @@ vk::UniqueInstance CreateInstance(vk::DynamicLoader& dl, Frontend::WindowSystemT
 
     vk::Bool32 enable_sync =
         enable_validation && Config::vkValidationSyncEnabled() ? vk::True : vk::False;
-    vk::LayerSettingEXT layer_set = {
+    vk::Bool32 enable_gpuav =
-        .pLayerName = VALIDATION_LAYER_NAME,
+        enable_validation && Config::vkValidationSyncEnabled() ? vk::True : vk::False;
-        .pSettingName = "validate_sync",
+    const char* gpuav_mode = enable_validation && Config::vkValidationGpuEnabled()
-        .type = vk::LayerSettingTypeEXT::eBool32,
+                                 ? "GPU_BASED_GPU_ASSISTED"
-        .valueCount = 1,
+                                 : "GPU_BASED_NONE";
-        .pValues = &enable_sync,
+    const std::array layer_setings = {
+        vk::LayerSettingEXT{
+            .pLayerName = VALIDATION_LAYER_NAME,
+            .pSettingName = "validate_sync",
+            .type = vk::LayerSettingTypeEXT::eBool32,
+            .valueCount = 1,
+            .pValues = &enable_sync,
+        },
+        vk::LayerSettingEXT{
+            .pLayerName = VALIDATION_LAYER_NAME,
+            .pSettingName = "sync_queue_submit",
+            .type = vk::LayerSettingTypeEXT::eBool32,
+            .valueCount = 1,
+            .pValues = &enable_sync,
+        },
+        vk::LayerSettingEXT{
+            .pLayerName = VALIDATION_LAYER_NAME,
+            .pSettingName = "validate_gpu_based",
+            .type = vk::LayerSettingTypeEXT::eString,
+            .valueCount = 1,
+            .pValues = &gpuav_mode,
+        },
+        vk::LayerSettingEXT{
+            .pLayerName = VALIDATION_LAYER_NAME,
+            .pSettingName = "gpuav_reserve_binding_slot",
+            .type = vk::LayerSettingTypeEXT::eBool32,
+            .valueCount = 1,
+            .pValues = &enable_gpuav,
+        },
+        vk::LayerSettingEXT{
+            .pLayerName = VALIDATION_LAYER_NAME,
+            .pSettingName = "gpuav_descriptor_checks",
+            .type = vk::LayerSettingTypeEXT::eBool32,
+            .valueCount = 1,
+            .pValues = &enable_gpuav,
+        },
+        vk::LayerSettingEXT{
+            .pLayerName = VALIDATION_LAYER_NAME,
+            .pSettingName = "gpuav_validate_indirect_buffer",
+            .type = vk::LayerSettingTypeEXT::eBool32,
+            .valueCount = 1,
+            .pValues = &enable_gpuav,
+        },
+        vk::LayerSettingEXT{
+            .pLayerName = VALIDATION_LAYER_NAME,
+            .pSettingName = "gpuav_buffer_copies",
+            .type = vk::LayerSettingTypeEXT::eBool32,
+            .valueCount = 1,
+            .pValues = &enable_gpuav,
+        },
     };
 
     vk::StructureChain<vk::InstanceCreateInfo, vk::LayerSettingsCreateInfoEXT> instance_ci_chain = {
@@ -238,8 +287,8 @@ vk::UniqueInstance CreateInstance(vk::DynamicLoader& dl, Frontend::WindowSystemT
             .ppEnabledExtensionNames = extensions.data(),
         },
         vk::LayerSettingsCreateInfoEXT{
-            .settingCount = 1,
+            .settingCount = layer_setings.size(),
-            .pSettings = &layer_set,
+            .pSettings = layer_setings.data(),
         },
     };
 

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -152,7 +152,8 @@ void Rasterizer::BeginRendering() {
                                                           .stencil = regs.stencil_clear}},
         };
         texture_cache.TouchMeta(htile_address, false);
-        state.num_depth_attachments++;
+        state.has_depth = true;
+        state.has_stencil = image.info.usage.stencil;
     }
     scheduler.BeginRendering(state);
 }
@@ -230,16 +231,42 @@ void Rasterizer::UpdateDepthStencilState() {
     cmdbuf.setDepthBoundsTestEnable(depth.depth_bounds_enable);
 }
 
-void Rasterizer::ScopeMarkerBegin(const std::string& str) {
+void Rasterizer::ScopeMarkerBegin(const std::string_view& str) {
+    if (!Config::isMarkersEnabled()) {
+        return;
+    }
+
     const auto cmdbuf = scheduler.CommandBuffer();
     cmdbuf.beginDebugUtilsLabelEXT(vk::DebugUtilsLabelEXT{
-        .pLabelName = str.c_str(),
+        .pLabelName = str.data(),
     });
 }
 
 void Rasterizer::ScopeMarkerEnd() {
+    if (!Config::isMarkersEnabled()) {
+        return;
+    }
+
     const auto cmdbuf = scheduler.CommandBuffer();
     cmdbuf.endDebugUtilsLabelEXT();
 }
 
+void Rasterizer::ScopedMarkerInsert(const std::string_view& str) {
+    if (!Config::isMarkersEnabled()) {
+        return;
+    }
+
+    const auto cmdbuf = scheduler.CommandBuffer();
+    cmdbuf.insertDebugUtilsLabelEXT(vk::DebugUtilsLabelEXT{
+        .pLabelName = str.data(),
+    });
+}
+
+void Rasterizer::Breadcrumb(u64 id) {
+    if (!instance.HasNvCheckpoints()) {
+        return;
+    }
+    scheduler.CommandBuffer().setCheckpointNV(id);
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -35,8 +35,10 @@ public:
 
     void DispatchDirect();
 
-    void ScopeMarkerBegin(const std::string& str);
+    void ScopeMarkerBegin(const std::string_view& str);
     void ScopeMarkerEnd();
+    void ScopedMarkerInsert(const std::string_view& str);
+    void Breadcrumb(u64 id);
 
     void InvalidateMemory(VAddr addr, u64 size);
     void MapMemory(VAddr addr, u64 size);

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -38,8 +38,7 @@ void Scheduler::BeginRendering(const RenderState& new_state) {
         .layerCount = 1,
         .colorAttachmentCount = render_state.num_color_attachments,
         .pColorAttachments = render_state.color_attachments.data(),
-        .pDepthAttachment =
+        .pDepthAttachment = render_state.has_depth ? &render_state.depth_attachment : nullptr,
-            render_state.num_depth_attachments ? &render_state.depth_attachment : nullptr,
     };
 
     current_cmdbuf.beginRendering(rendering_info);
@@ -50,6 +49,8 @@ void Scheduler::EndRendering() {
         return;
     }
     is_rendering = false;
+    current_cmdbuf.endRendering();
+
     boost::container::static_vector<vk::ImageMemoryBarrier, 9> barriers;
     for (size_t i = 0; i < render_state.num_color_attachments; ++i) {
         barriers.push_back(vk::ImageMemoryBarrier{
@@ -70,10 +71,35 @@ void Scheduler::EndRendering() {
                 },
         });
     }
-    current_cmdbuf.endRendering();
+    if (render_state.has_depth) {
+        barriers.push_back(vk::ImageMemoryBarrier{
+            .srcAccessMask = vk::AccessFlagBits::eDepthStencilAttachmentWrite,
+            .dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite,
+            .oldLayout = render_state.depth_attachment.imageLayout,
+            .newLayout = render_state.depth_attachment.imageLayout,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = render_state.depth_image,
+            .subresourceRange =
+                {
+                    .aspectMask = vk::ImageAspectFlagBits::eDepth |
+                                  (render_state.has_stencil ? vk::ImageAspectFlagBits::eStencil
+                                                            : vk::ImageAspectFlagBits::eNone),
+                    .baseMipLevel = 0,
+                    .levelCount = VK_REMAINING_MIP_LEVELS,
+                    .baseArrayLayer = 0,
+                    .layerCount = VK_REMAINING_ARRAY_LAYERS,
+                },
+        });
+    }
+
     if (!barriers.empty()) {
-        current_cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eColorAttachmentOutput,
+        const auto src_stages =
-                                       vk::PipelineStageFlagBits::eFragmentShader,
+            vk::PipelineStageFlagBits::eColorAttachmentOutput |
+            (render_state.has_depth ? vk::PipelineStageFlagBits::eLateFragmentTests |
+                                          vk::PipelineStageFlagBits::eEarlyFragmentTests
+                                    : vk::PipelineStageFlagBits::eNone);
+        current_cmdbuf.pipelineBarrier(src_stages, vk::PipelineStageFlagBits::eFragmentShader,
                                        vk::DependencyFlagBits::eByRegion, {}, {}, barriers);
     }
 }
@@ -158,6 +184,13 @@ void Scheduler::SubmitExecution(SubmitInfo& info) {
     try {
         instance.GetGraphicsQueue().submit(submit_info, info.fence);
     } catch (vk::DeviceLostError& err) {
+        if (instance.HasNvCheckpoints()) {
+            const auto checkpoint_data = instance.GetGraphicsQueue().getCheckpointData2NV();
+            for (const auto& cp : checkpoint_data) {
+                LOG_CRITICAL(Render_Vulkan, "{}: {:#x}", vk::to_string(cp.stage),
+                             reinterpret_cast<u64>(cp.pCheckpointMarker));
+            }
+        }
         UNREACHABLE_MSG("Device lost during submit: {}", err.what());
     }
 

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -20,7 +20,8 @@ struct RenderState {
     vk::RenderingAttachmentInfo depth_attachment{};
     vk::Image depth_image{};
     u32 num_color_attachments{};
-    u32 num_depth_attachments{};
+    bool has_depth{};
+    bool has_stencil{};
     u32 width = std::numeric_limits<u32>::max();
     u32 height = std::numeric_limits<u32>::max();
 

src/video_core/texture_cache/image.cpp

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include "common/assert.h"
+#include "common/config.h"
 #include "video_core/renderer_vulkan/liverpool_to_vk.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -154,6 +155,9 @@ Image::Image(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_,
     };
 
     image.Create(image_ci);
+
+    Vulkan::SetObjectName(instance->GetDevice(), (vk::Image)image, "Image {:#x}:{:#x}",
+                          info.guest_address, info.guest_size_bytes);
 }
 
 void Image::Transit(vk::ImageLayout dst_layout, vk::Flags<vk::AccessFlagBits> dst_mask,

src/video_core/texture_cache/image_info.cpp

@@ -189,6 +189,8 @@ ImageInfo::ImageInfo(const AmdGpu::Liverpool::DepthBuffer& buffer, u32 num_slice
     resources.layers = num_slices;
     meta_info.htile_addr = buffer.z_info.tile_surface_en ? htile_address : 0;
     usage.depth_target = true;
+    usage.stencil =
+        buffer.stencil_info.format != AmdGpu::Liverpool::DepthBuffer::StencilFormat::Invalid;
 
     guest_address = buffer.Address();
     const auto depth_slice_sz = buffer.GetDepthSliceSize();
@@ -260,7 +262,7 @@ ImageInfo::ImageInfo(const AmdGpu::Image& image) noexcept {
         case AmdGpu::TilingMode::Display_MacroTiled:
         case AmdGpu::TilingMode::Texture_MacroTiled:
         case AmdGpu::TilingMode::Depth_MacroTiled: {
-            // ASSERT(!props.is_cube && !props.is_block);
+            ASSERT(!props.is_block);
             ASSERT(num_samples == 1);
             std::tie(mip_info.pitch, mip_info.size) =
                 ImageSizeMacroTiled(mip_w, mip_h, bpp, num_samples, image.tiling_index);

src/video_core/texture_cache/image_view.cpp

@@ -92,6 +92,8 @@ ImageViewInfo::ImageViewInfo(const AmdGpu::Liverpool::ColorBuffer& col_buffer,
                              bool is_vo_surface) noexcept {
     const auto base_format =
         Vulkan::LiverpoolToVK::SurfaceFormat(col_buffer.info.format, col_buffer.NumFormat());
+    range.base.layer = col_buffer.view.slice_start;
+    range.extent.layers = col_buffer.NumSlices();
     format = Vulkan::LiverpoolToVK::AdjustColorBufferFormat(
         base_format, col_buffer.info.comp_swap.Value(), is_vo_surface);
 }

src/video_core/texture_cache/tile_manager.cpp

@@ -194,6 +194,7 @@ vk::Format DemoteImageFormatForDetiling(vk::Format format) {
     case vk::Format::eR32G32Sfloat:
     case vk::Format::eR32G32Uint:
     case vk::Format::eR16G16B16A16Unorm:
+    case vk::Format::eR16G16B16A16Sfloat:
         return vk::Format::eR32G32Uint;
     case vk::Format::eBc2SrgbBlock:
     case vk::Format::eBc2UnormBlock:
@@ -397,7 +398,7 @@ std::optional<vk::Buffer> TileManager::TryDetile(Image& image) {
     const u32 image_size = image.info.guest_size_bytes;
     const auto [in_buffer, in_offset] = [&] -> std::pair<vk::Buffer, u32> {
         // Use stream buffer for smaller textures.
-        if (image_size <= StreamBufferSize) {
+        if (image_size <= stream_buffer.GetFreeSize()) {
             u32 offset = stream_buffer.Copy(image.info.guest_address, image_size);
             return {stream_buffer.Handle(), offset};
         }