Merge pull request #11995 from FernandoS27/you-dont-need-the-new-iphone

Revert PR #11806 and do a proper fix to the memory handling.

src/core/memory.cpp

@@ -1,8 +1,10 @@
 // SPDX-FileCopyrightText: 2015 Citra Emulator Project
+// SPDX-FileCopyrightText: 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
 #include <cstring>
+#include <mutex>
 #include <span>
 
 #include "common/assert.h"
@@ -10,6 +12,7 @@
 #include "common/common_types.h"
 #include "common/logging/log.h"
 #include "common/page_table.h"
+#include "common/scope_exit.h"
 #include "common/settings.h"
 #include "common/swap.h"
 #include "core/core.h"
@@ -318,7 +321,7 @@ struct Memory::Impl {
             [&](const Common::ProcessAddress current_vaddr, const std::size_t copy_amount,
                 u8* const host_ptr) {
                 if constexpr (!UNSAFE) {
-                    system.GPU().InvalidateRegion(GetInteger(current_vaddr), copy_amount);
+                    HandleRasterizerWrite(GetInteger(current_vaddr), copy_amount);
                 }
                 std::memcpy(host_ptr, src_buffer, copy_amount);
             },
@@ -351,7 +354,7 @@ struct Memory::Impl {
             },
             [&](const Common::ProcessAddress current_vaddr, const std::size_t copy_amount,
                 u8* const host_ptr) {
-                system.GPU().InvalidateRegion(GetInteger(current_vaddr), copy_amount);
+                HandleRasterizerWrite(GetInteger(current_vaddr), copy_amount);
                 std::memset(host_ptr, 0, copy_amount);
             },
             [](const std::size_t copy_amount) {});
@@ -420,7 +423,7 @@ struct Memory::Impl {
                                  const std::size_t block_size) {
             // dc cvac: Store to point of coherency
             // CPU flush -> GPU invalidate
-            system.GPU().InvalidateRegion(GetInteger(current_vaddr), block_size);
+            HandleRasterizerWrite(GetInteger(current_vaddr), block_size);
         };
         return PerformCacheOperation(dest_addr, size, on_rasterizer);
     }
@@ -430,7 +433,7 @@ struct Memory::Impl {
                                  const std::size_t block_size) {
             // dc civac: Store to point of coherency, and invalidate from cache
             // CPU flush -> GPU invalidate
-            system.GPU().InvalidateRegion(GetInteger(current_vaddr), block_size);
+            HandleRasterizerWrite(GetInteger(current_vaddr), block_size);
         };
         return PerformCacheOperation(dest_addr, size, on_rasterizer);
     }
@@ -767,7 +770,18 @@ struct Memory::Impl {
     }
 
     void HandleRasterizerWrite(VAddr address, size_t size) {
-        const size_t core = system.GetCurrentHostThreadID();
+        constexpr size_t sys_core = Core::Hardware::NUM_CPU_CORES - 1;
+        const size_t core = std::min(system.GetCurrentHostThreadID(),
+                                     sys_core); // any other calls threads go to syscore.
+        // Guard on sys_core;
+        if (core == sys_core) [[unlikely]] {
+            sys_core_guard.lock();
+        }
+        SCOPE_EXIT({
+            if (core == sys_core) [[unlikely]] {
+                sys_core_guard.unlock();
+            }
+        });
         auto& current_area = rasterizer_write_areas[core];
         VAddr subaddress = address >> YUZU_PAGEBITS;
         bool do_collection = current_area.last_address == subaddress;
@@ -799,6 +813,7 @@ struct Memory::Impl {
         rasterizer_read_areas{};
     std::array<GPUDirtyState, Core::Hardware::NUM_CPU_CORES> rasterizer_write_areas{};
     std::span<Core::GPUDirtyMemoryManager> gpu_dirty_managers;
+    std::mutex sys_core_guard;
 };
 
 Memory::Memory(Core::System& system_) : system{system_} {

src/video_core/fence_manager.h

@@ -86,10 +86,7 @@ public:
             uncommitted_operations.emplace_back(std::move(func));
         }
         pending_operations.emplace_back(std::move(uncommitted_operations));
-        {
-            std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
-            QueueFence(new_fence);
-        }
+        QueueFence(new_fence);
         if (!delay_fence) {
             func();
         }

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -555,7 +555,7 @@ void RasterizerOpenGL::OnCacheInvalidation(VAddr addr, u64 size) {
     }
     {
         std::scoped_lock lock{buffer_cache.mutex};
-        buffer_cache.CachedWriteMemory(addr, size);
+        buffer_cache.WriteMemory(addr, size);
     }
     shader_cache.InvalidateRegion(addr, size);
 }

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -132,16 +132,12 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     const bool use_accelerated =
         rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride);
     const bool is_srgb = use_accelerated && screen_info.is_srgb;
+    RenderScreenshot(*framebuffer, use_accelerated);
 
-    {
-        std::scoped_lock lock{rasterizer.LockCaches()};
-        RenderScreenshot(*framebuffer, use_accelerated);
-
-        Frame* frame = present_manager.GetRenderFrame();
-        blit_screen.DrawToSwapchain(frame, *framebuffer, use_accelerated, is_srgb);
-        scheduler.Flush(*frame->render_ready);
-        present_manager.Present(frame);
-    }
+    Frame* frame = present_manager.GetRenderFrame();
+    blit_screen.DrawToSwapchain(frame, *framebuffer, use_accelerated, is_srgb);
+    scheduler.Flush(*frame->render_ready);
+    present_manager.Present(frame);
 
     gpu.RendererFrameEndNotify();
     rasterizer.TickFrame();

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -199,7 +199,7 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
     if (!pipeline) {
         return;
     }
-    std::scoped_lock lock{LockCaches()};
+    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
     // update engine as channel may be different.
     pipeline->SetEngine(maxwell3d, gpu_memory);
     pipeline->Configure(is_indexed);
@@ -621,7 +621,7 @@ void RasterizerVulkan::OnCacheInvalidation(VAddr addr, u64 size) {
     }
     {
         std::scoped_lock lock{buffer_cache.mutex};
-        buffer_cache.CachedWriteMemory(addr, size);
+        buffer_cache.WriteMemory(addr, size);
     }
     pipeline_cache.InvalidateRegion(addr, size);
 }
@@ -710,7 +710,6 @@ void RasterizerVulkan::TiledCacheBarrier() {
 }
 
 void RasterizerVulkan::FlushCommands() {
-    std::scoped_lock lock{LockCaches()};
     if (draw_counter == 0) {
         return;
     }
@@ -808,7 +807,6 @@ void RasterizerVulkan::FlushWork() {
     if ((++draw_counter & 7) != 7) {
         return;
     }
-    std::scoped_lock lock{LockCaches()};
     if (draw_counter < DRAWS_TO_DISPATCH) {
         // Send recorded tasks to the worker thread
         scheduler.DispatchWork();
@@ -1507,7 +1505,7 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs)
 void RasterizerVulkan::InitializeChannel(Tegra::Control::ChannelState& channel) {
     CreateChannel(channel);
     {
-        std::scoped_lock lock{LockCaches()};
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.CreateChannel(channel);
         buffer_cache.CreateChannel(channel);
     }
@@ -1520,7 +1518,7 @@ void RasterizerVulkan::BindChannel(Tegra::Control::ChannelState& channel) {
     const s32 channel_id = channel.bind_id;
     BindToChannel(channel_id);
     {
-        std::scoped_lock lock{LockCaches()};
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.BindToChannel(channel_id);
         buffer_cache.BindToChannel(channel_id);
     }
@@ -1533,7 +1531,7 @@ void RasterizerVulkan::BindChannel(Tegra::Control::ChannelState& channel) {
 void RasterizerVulkan::ReleaseChannel(s32 channel_id) {
     EraseChannel(channel_id);
     {
-        std::scoped_lock lock{LockCaches()};
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.EraseChannel(channel_id);
         buffer_cache.EraseChannel(channel_id);
     }

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -133,10 +133,6 @@ public:
 
     void ReleaseChannel(s32 channel_id) override;
 
-    std::scoped_lock<std::recursive_mutex, std::recursive_mutex> LockCaches() {
-        return std::scoped_lock{buffer_cache.mutex, texture_cache.mutex};
-    }
-
 private:
     static constexpr size_t MAX_TEXTURES = 192;
     static constexpr size_t MAX_IMAGES = 48;