early-access version 3301

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 3300.
+This is the source code for early-access 3301.
 
 ## Legal Notice
 

src/common/settings.h

@@ -415,7 +415,7 @@ struct Values {
     // Renderer
     SwitchableSetting<RendererBackend, true> renderer_backend{
         RendererBackend::Vulkan, RendererBackend::OpenGL, RendererBackend::Null, "backend"};
-    SwitchableSetting<bool> renderer_force_max_clock{true, "force_max_clock"};
+    SwitchableSetting<bool> renderer_force_max_clock{false, "force_max_clock"};
     Setting<bool> renderer_debug{false, "debug"};
     Setting<bool> renderer_shader_feedback{false, "shader_feedback"};
     Setting<bool> enable_nsight_aftermath{false, "nsight_aftermath"};

src/shader_recompiler/frontend/maxwell/translate_program.cpp

@@ -292,7 +292,7 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
 
     Optimization::PositionPass(env, program);
 
-    Optimization::GlobalMemoryToStorageBufferPass(program, host_info);
+    Optimization::GlobalMemoryToStorageBufferPass(program);
     Optimization::TexturePass(env, program, host_info);
 
     if (Settings::values.resolution_info.active) {

src/shader_recompiler/host_translate_info.h

@@ -15,7 +15,6 @@ struct HostTranslateInfo {
     bool needs_demote_reorder{}; ///< True when the device needs DemoteToHelperInvocation reordered
     bool support_snorm_render_buffer{};  ///< True when the device supports SNORM render buffers
     bool support_viewport_index_layer{}; ///< True when the device supports gl_Layer in VS
-    u32 min_ssbo_alignment{};            ///< Minimum alignment supported by the device for SSBOs
     bool support_geometry_shader_passthrough{}; ///< True when the device supports geometry
                                                 ///< passthrough shaders
 };

src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp

@@ -11,7 +11,6 @@
 #include "shader_recompiler/frontend/ir/breadth_first_search.h"
 #include "shader_recompiler/frontend/ir/ir_emitter.h"
 #include "shader_recompiler/frontend/ir/value.h"
-#include "shader_recompiler/host_translate_info.h"
 #include "shader_recompiler/ir_opt/passes.h"
 
 namespace Shader::Optimization {
@@ -403,7 +402,7 @@ void CollectStorageBuffers(IR::Block& block, IR::Inst& inst, StorageInfo& info)
 }
 
 /// Returns the offset in indices (not bytes) for an equivalent storage instruction
-IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer, u32 alignment) {
+IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer) {
     IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
     IR::U32 offset;
     if (const std::optional<LowAddrInfo> low_addr{TrackLowAddress(&inst)}) {
@@ -416,10 +415,7 @@ IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer
     }
     // Subtract the least significant 32 bits from the guest offset. The result is the storage
     // buffer offset in bytes.
-    IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))};
+    const IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))};
-
-    // Align the offset base to match the host alignment requirements
-    low_cbuf = ir.BitwiseAnd(low_cbuf, ir.Imm32(~(alignment - 1U)));
     return ir.ISub(offset, low_cbuf);
 }
 
@@ -514,7 +510,7 @@ void Replace(IR::Block& block, IR::Inst& inst, const IR::U32& storage_index,
 }
 } // Anonymous namespace
 
-void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info) {
+void GlobalMemoryToStorageBufferPass(IR::Program& program) {
     StorageInfo info;
     for (IR::Block* const block : program.post_order_blocks) {
         for (IR::Inst& inst : block->Instructions()) {
@@ -538,8 +534,7 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateIn
         const IR::U32 index{IR::Value{static_cast<u32>(info.set.index_of(it))}};
         IR::Block* const block{storage_inst.block};
         IR::Inst* const inst{storage_inst.inst};
-        const IR::U32 offset{
+        const IR::U32 offset{StorageOffset(*block, *inst, storage_buffer)};
-            StorageOffset(*block, *inst, storage_buffer, host_info.min_ssbo_alignment)};
         Replace(*block, *inst, index, offset);
     }
 }

src/shader_recompiler/ir_opt/passes.h

@@ -15,7 +15,7 @@ namespace Shader::Optimization {
 void CollectShaderInfoPass(Environment& env, IR::Program& program);
 void ConstantPropagationPass(Environment& env, IR::Program& program);
 void DeadCodeEliminationPass(IR::Program& program);
-void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info);
+void GlobalMemoryToStorageBufferPass(IR::Program& program);
 void IdentityRemovalPass(IR::Program& program);
 void LowerFp16ToFp32(IR::Program& program);
 void LowerInt64ToInt32(IR::Program& program);

src/video_core/buffer_cache/buffer_cache.h

@@ -1938,21 +1938,14 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s
                                                                       bool is_written) const {
     const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr);
     const u32 size = gpu_memory->Read<u32>(ssbo_addr + 8);
-    const u32 alignment = runtime.GetStorageBufferAlignment();
+    const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
-
-    const GPUVAddr aligned_gpu_addr = Common::AlignDown(gpu_addr, alignment);
-    const u32 aligned_size =
-        Common::AlignUp(static_cast<u32>(gpu_addr - aligned_gpu_addr) + size, alignment);
-
-    const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(aligned_gpu_addr);
     if (!cpu_addr || size == 0) {
         return NULL_BINDING;
     }
-
+    const VAddr cpu_end = Common::AlignUp(*cpu_addr + size, Core::Memory::YUZU_PAGESIZE);
-    const VAddr cpu_end = Common::AlignUp(*cpu_addr + aligned_size, Core::Memory::YUZU_PAGESIZE);
     const Binding binding{
         .cpu_addr = *cpu_addr,
-        .size = is_written ? aligned_size : static_cast<u32>(cpu_end - *cpu_addr),
+        .size = is_written ? size : static_cast<u32>(cpu_end - *cpu_addr),
         .buffer_id = BufferId{},
     };
     return binding;

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -160,10 +160,6 @@ public:
         return device.CanReportMemoryUsage();
     }
 
-    u32 GetStorageBufferAlignment() const {
-        return static_cast<u32>(device.GetShaderStorageBufferAlignment());
-    }
-
 private:
     static constexpr std::array PABO_LUT{
         GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV,          GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV,

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -140,6 +140,7 @@ void RasterizerOpenGL::LoadDiskResources(u64 title_id, std::stop_token stop_load
 void RasterizerOpenGL::Clear(u32 layer_count) {
     MICROPROFILE_SCOPE(OpenGL_Clears);
 
+    gpu_memory->FlushCaching();
     const auto& regs = maxwell3d->regs;
     bool use_color{};
     bool use_depth{};
@@ -208,6 +209,7 @@ void RasterizerOpenGL::PrepareDraw(bool is_indexed, Func&& draw_func) {
     MICROPROFILE_SCOPE(OpenGL_Drawing);
 
     SCOPE_EXIT({ gpu.TickWork(); });
+    gpu_memory->FlushCaching();
     query_cache.UpdateCounters();
 
     GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()};
@@ -361,6 +363,7 @@ void RasterizerOpenGL::DrawTexture() {
 }
 
 void RasterizerOpenGL::DispatchCompute() {
+    gpu_memory->FlushCaching();
     ComputePipeline* const pipeline{shader_cache.CurrentComputePipeline()};
     if (!pipeline) {
         return;
@@ -568,6 +571,7 @@ void RasterizerOpenGL::TickFrame() {
 }
 
 bool RasterizerOpenGL::AccelerateConditionalRendering() {
+    gpu_memory->FlushCaching();
     if (Settings::IsGPULevelHigh()) {
         // Reimplement Host conditional rendering.
         return false;

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -236,7 +236,6 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
           .needs_demote_reorder = device.IsAmd(),
           .support_snorm_render_buffer = false,
           .support_viewport_index_layer = device.HasVertexViewportLayer(),
-          .min_ssbo_alignment = static_cast<u32>(device.GetShaderStorageBufferAlignment()),
           .support_geometry_shader_passthrough = device.HasGeometryShaderPassthrough(),
       } {
     if (use_asynchronous_shaders) {

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -99,6 +99,7 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
                  state_tracker, scheduler) {
     if (Settings::values.renderer_force_max_clock.GetValue() && device.ShouldBoostClocks()) {
         turbo_mode.emplace(instance, dld);
+        scheduler.RegisterOnSubmit([this] { turbo_mode->QueueSubmitted(); });
     }
     Report();
 } catch (const vk::Exception& exception) {
@@ -107,6 +108,7 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
 }
 
 RendererVulkan::~RendererVulkan() {
+    scheduler.RegisterOnSubmit([] {});
     void(device.GetLogical().WaitIdle());
 }
 

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -330,10 +330,6 @@ bool BufferCacheRuntime::CanReportMemoryUsage() const {
     return device.CanReportMemoryUsage();
 }
 
-u32 BufferCacheRuntime::GetStorageBufferAlignment() const {
-    return static_cast<u32>(device.GetStorageBufferAlignment());
-}
-
 void BufferCacheRuntime::Finish() {
     scheduler.Finish();
 }

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -73,8 +73,6 @@ public:
 
     bool CanReportMemoryUsage() const;
 
-    u32 GetStorageBufferAlignment() const;
-
     [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size);
 
     [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size);

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -344,7 +344,6 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device
             driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE,
         .support_snorm_render_buffer = true,
         .support_viewport_index_layer = device.IsExtShaderViewportIndexLayerSupported(),
-        .min_ssbo_alignment = static_cast<u32>(device.GetStorageBufferAlignment()),
         .support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
     };
 

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -298,6 +298,7 @@ void RasterizerVulkan::Clear(u32 layer_count) {
     MICROPROFILE_SCOPE(Vulkan_Clearing);
 
     FlushWork();
+    gpu_memory->FlushCaching();
 
     query_cache.UpdateCounters();
 
@@ -656,6 +657,7 @@ void RasterizerVulkan::TickFrame() {
 }
 
 bool RasterizerVulkan::AccelerateConditionalRendering() {
+    gpu_memory->FlushCaching();
     if (Settings::IsGPULevelHigh()) {
         // TODO(Blinkhawk): Reimplement Host conditional rendering.
         return false;

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -213,6 +213,11 @@ void Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_s
             .signalSemaphoreCount = num_signal_semaphores,
             .pSignalSemaphores = signal_semaphores.data(),
         };
+
+        if (on_submit) {
+            on_submit();
+        }
+
         switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info)) {
         case VK_SUCCESS:
             break;

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -5,6 +5,7 @@
 
 #include <condition_variable>
 #include <cstddef>
+#include <functional>
 #include <memory>
 #include <thread>
 #include <utility>
@@ -66,6 +67,11 @@ public:
         query_cache = &query_cache_;
     }
 
+    // Registers a callback to perform on queue submission.
+    void RegisterOnSubmit(std::function<void()>&& func) {
+        on_submit = std::move(func);
+    }
+
     /// Send work to a separate thread.
     template <typename T>
     void Record(T&& command) {
@@ -216,6 +222,7 @@ private:
     vk::CommandBuffer current_cmdbuf;
 
     std::unique_ptr<CommandChunk> chunk;
+    std::function<void()> on_submit;
 
     State state;
 

src/video_core/renderer_vulkan/vk_turbo_mode.cpp

@@ -14,11 +14,21 @@ using namespace Common::Literals;
 
 TurboMode::TurboMode(const vk::Instance& instance, const vk::InstanceDispatch& dld)
     : m_device{CreateDevice(instance, dld, VK_NULL_HANDLE)}, m_allocator{m_device, false} {
+    {
+        std::scoped_lock lk{m_submission_lock};
+        m_submission_time = std::chrono::steady_clock::now();
+    }
     m_thread = std::jthread([&](auto stop_token) { Run(stop_token); });
 }
 
 TurboMode::~TurboMode() = default;
 
+void TurboMode::QueueSubmitted() {
+    std::scoped_lock lk{m_submission_lock};
+    m_submission_time = std::chrono::steady_clock::now();
+    m_submission_cv.notify_one();
+}
+
 void TurboMode::Run(std::stop_token stop_token) {
     auto& dld = m_device.GetLogical();
 
@@ -199,6 +209,13 @@ void TurboMode::Run(std::stop_token stop_token) {
 
         // Wait for completion.
         fence.Wait();
+
+        // Wait for the next graphics queue submission if necessary.
+        std::unique_lock lk{m_submission_lock};
+        Common::CondvarWait(m_submission_cv, lk, stop_token, [this] {
+            return (std::chrono::steady_clock::now() - m_submission_time) <=
+                   std::chrono::milliseconds{100};
+        });
     }
 }
 

src/video_core/renderer_vulkan/vk_turbo_mode.h

@@ -3,6 +3,9 @@
 
 #pragma once
 
+#include <chrono>
+#include <mutex>
+
 #include "common/polyfill_thread.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_memory_allocator.h"
@@ -15,11 +18,17 @@ public:
     explicit TurboMode(const vk::Instance& instance, const vk::InstanceDispatch& dld);
     ~TurboMode();
 
+    void QueueSubmitted();
+
 private:
     void Run(std::stop_token stop_token);
 
     Device m_device;
     MemoryAllocator m_allocator;
+    std::mutex m_submission_lock;
+    std::condition_variable_any m_submission_cv;
+    std::chrono::time_point<std::chrono::steady_clock> m_submission_time{};
+
     std::jthread m_thread;
 };