7 files changed, 35 insertions, 206 deletions

diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 6839abe71..7aefd4035 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -43,8 +43,6 @@ add_library(video_core STATIC
     renderer_opengl/gl_device.h
     renderer_opengl/gl_framebuffer_cache.cpp
     renderer_opengl/gl_framebuffer_cache.h
-    renderer_opengl/gl_global_cache.cpp
-    renderer_opengl/gl_global_cache.h
     renderer_opengl/gl_rasterizer.cpp
     renderer_opengl/gl_rasterizer.h
     renderer_opengl/gl_resource_manager.cpp
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
index 1219ca6ea..2f603e3d7 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
@@ -49,7 +49,8 @@ void OGLBufferCache::Unregister(const std::shared_ptr<CachedBufferEntry>& entry)
 }
 
 OGLBufferCache::BufferInfo OGLBufferCache::UploadMemory(GPUVAddr gpu_addr, std::size_t size,
-                                                        std::size_t alignment, bool internalize) {
+                                                        std::size_t alignment, bool internalize,
+                                                        bool is_written) {
     std::lock_guard lock{mutex};
 
     auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
@@ -68,18 +69,22 @@ OGLBufferCache::BufferInfo OGLBufferCache::UploadMemory(GPUVAddr gpu_addr, std::
 
     auto entry = TryGet(host_ptr);
     if (!entry) {
-        return FixedBufferUpload(gpu_addr, host_ptr, size, internalize);
+        return FixedBufferUpload(gpu_addr, host_ptr, size, internalize, is_written);
     }
 
     if (entry->GetSize() < size) {
         GrowBuffer(entry, size);
     }
+    if (is_written) {
+        entry->MarkAsModified(true, *this);
+    }
     return {entry->GetBuffer(), CachedBufferOffset};
 }
 
 OGLBufferCache::BufferInfo OGLBufferCache::UploadHostMemory(const void* raw_pointer,
                                                             std::size_t size,
                                                             std::size_t alignment) {
+    std::lock_guard lock{mutex};
     return StreamBufferUpload(raw_pointer, size, alignment);
 }
 
@@ -108,10 +113,8 @@ OGLBufferCache::BufferInfo OGLBufferCache::StreamBufferUpload(const void* raw_po
 }
 
 OGLBufferCache::BufferInfo OGLBufferCache::FixedBufferUpload(GPUVAddr gpu_addr, u8* host_ptr,
-                                                             std::size_t size, bool internalize) {
-    if (internalize) {
-        internalized_entries.emplace(ToCacheAddr(host_ptr));
-    }
+                                                             std::size_t size, bool internalize,
+                                                             bool is_written) {
     auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
     const auto cpu_addr = *memory_manager.GpuToCpuAddress(gpu_addr);
     auto entry = GetUncachedBuffer(cpu_addr, host_ptr);
@@ -119,6 +122,13 @@ OGLBufferCache::BufferInfo OGLBufferCache::FixedBufferUpload(GPUVAddr gpu_addr,
     entry->SetInternalState(internalize);
     Register(entry);
 
+    if (internalize) {
+        internalized_entries.emplace(ToCacheAddr(host_ptr));
+    }
+    if (is_written) {
+        entry->MarkAsModified(true, *this);
+    }
+
     if (entry->GetCapacity() < size) {
         entry->SetCapacity(CreateBuffer(size, GL_STATIC_DRAW), size);
     }
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h
index 00bc6008a..b4fbd201d 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.h
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.h
@@ -82,7 +82,7 @@ public:
     /// Uploads data from a guest GPU address. Returns the OpenGL buffer where it's located and its
     /// offset.
     BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
-                            bool internalize = false);
+                            bool internalize = false, bool is_written = false);
 
     /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
     BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size,
@@ -99,7 +99,7 @@ private:
     BufferInfo StreamBufferUpload(const void* raw_pointer, std::size_t size, std::size_t alignment);
 
     BufferInfo FixedBufferUpload(GPUVAddr gpu_addr, u8* host_ptr, std::size_t size,
-                                 bool internalize);
+                                 bool internalize, bool is_written);
 
     void GrowBuffer(std::shared_ptr<CachedBufferEntry>& entry, std::size_t new_size);
 
diff --git a/src/video_core/renderer_opengl/gl_global_cache.cpp b/src/video_core/renderer_opengl/gl_global_cache.cpp
deleted file mode 100644
index d5e385151..000000000
--- a/src/video_core/renderer_opengl/gl_global_cache.cpp
+++ /dev/null
@@ -1,102 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <glad/glad.h>
-
-#include "common/logging/log.h"
-#include "core/core.h"
-#include "video_core/memory_manager.h"
-#include "video_core/renderer_opengl/gl_global_cache.h"
-#include "video_core/renderer_opengl/gl_rasterizer.h"
-#include "video_core/renderer_opengl/gl_shader_decompiler.h"
-#include "video_core/renderer_opengl/utils.h"
-
-namespace OpenGL {
-
-CachedGlobalRegion::CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size)
-    : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, host_ptr{host_ptr}, size{size},
-      max_size{max_size} {
-    buffer.Create();
-    LabelGLObject(GL_BUFFER, buffer.handle, cpu_addr, "GlobalMemory");
-}
-
-CachedGlobalRegion::~CachedGlobalRegion() = default;
-
-void CachedGlobalRegion::Reload(u32 size_) {
-    size = size_;
-    if (size > max_size) {
-        size = max_size;
-        LOG_CRITICAL(HW_GPU, "Global region size {} exceeded the supported size {}!", size_,
-                     max_size);
-    }
-    glNamedBufferData(buffer.handle, size, host_ptr, GL_STREAM_DRAW);
-}
-
-void CachedGlobalRegion::Flush() {
-    LOG_DEBUG(Render_OpenGL, "Flushing {} bytes to CPU memory address 0x{:16}", size, cpu_addr);
-    glGetNamedBufferSubData(buffer.handle, 0, static_cast<GLsizeiptr>(size), host_ptr);
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const {
-    const auto search{reserve.find(addr)};
-    if (search == reserve.end()) {
-        return {};
-    }
-    return search->second;
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr,
-                                                              u32 size) {
-    GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)};
-    if (!region) {
-        // No reserved surface available, create a new one and reserve it
-        auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
-        const auto cpu_addr{memory_manager.GpuToCpuAddress(addr)};
-        ASSERT(cpu_addr);
-
-        region = std::make_shared<CachedGlobalRegion>(*cpu_addr, host_ptr, size, max_ssbo_size);
-        ReserveGlobalRegion(region);
-    }
-    region->Reload(size);
-    return region;
-}
-
-void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) {
-    reserve.insert_or_assign(region->GetCacheAddr(), std::move(region));
-}
-
-GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer)
-    : RasterizerCache{rasterizer} {
-    GLint max_ssbo_size_;
-    glGetIntegerv(GL_MAX_SHADER_STORAGE_BLOCK_SIZE, &max_ssbo_size_);
-    max_ssbo_size = static_cast<u32>(max_ssbo_size_);
-}
-
-GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
-    const GLShader::GlobalMemoryEntry& global_region,
-    Tegra::Engines::Maxwell3D::Regs::ShaderStage stage) {
-    std::lock_guard lock{mutex};
-
-    auto& gpu{Core::System::GetInstance().GPU()};
-    auto& memory_manager{gpu.MemoryManager()};
-    const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
-    const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address +
-                    global_region.GetCbufOffset()};
-    const auto actual_addr{memory_manager.Read<u64>(addr)};
-    const auto size{memory_manager.Read<u32>(addr + 8)};
-
-    // Look up global region in the cache based on address
-    const auto& host_ptr{memory_manager.GetPointer(actual_addr)};
-    GlobalRegion region{TryGet(host_ptr)};
-
-    if (!region) {
-        // No global region found - create a new one
-        region = GetUncachedGlobalRegion(actual_addr, host_ptr, size);
-        Register(region);
-    }
-
-    return region;
-}
-
-} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_global_cache.h b/src/video_core/renderer_opengl/gl_global_cache.h
deleted file mode 100644
index 2d467a240..000000000
--- a/src/video_core/renderer_opengl/gl_global_cache.h
+++ /dev/null
@@ -1,82 +0,0 @@
-// Copyright 2018 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <memory>
-#include <unordered_map>
-
-#include <glad/glad.h>
-
-#include "common/assert.h"
-#include "common/common_types.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/rasterizer_cache.h"
-#include "video_core/renderer_opengl/gl_resource_manager.h"
-
-namespace OpenGL {
-
-namespace GLShader {
-class GlobalMemoryEntry;
-}
-
-class RasterizerOpenGL;
-class CachedGlobalRegion;
-using GlobalRegion = std::shared_ptr<CachedGlobalRegion>;
-
-class CachedGlobalRegion final : public RasterizerCacheObject {
-public:
-    explicit CachedGlobalRegion(VAddr cpu_addr, u8* host_ptr, u32 size, u32 max_size);
-    ~CachedGlobalRegion();
-
-    VAddr GetCpuAddr() const override {
-        return cpu_addr;
-    }
-
-    std::size_t GetSizeInBytes() const override {
-        return size;
-    }
-
-    /// Gets the GL program handle for the buffer
-    GLuint GetBufferHandle() const {
-        return buffer.handle;
-    }
-
-    /// Reloads the global region from guest memory
-    void Reload(u32 size_);
-
-    void Flush();
-
-private:
-    VAddr cpu_addr{};
-    u8* host_ptr{};
-    u32 size{};
-    u32 max_size{};
-
-    OGLBuffer buffer;
-};
-
-class GlobalRegionCacheOpenGL final : public RasterizerCache<GlobalRegion> {
-public:
-    explicit GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer);
-
-    /// Gets the current specified shader stage program
-    GlobalRegion GetGlobalRegion(const GLShader::GlobalMemoryEntry& descriptor,
-                                 Tegra::Engines::Maxwell3D::Regs::ShaderStage stage);
-
-protected:
-    void FlushObjectInner(const GlobalRegion& object) override {
-        object->Flush();
-    }
-
-private:
-    GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const;
-    GlobalRegion GetUncachedGlobalRegion(GPUVAddr addr, u8* host_ptr, u32 size);
-    void ReserveGlobalRegion(GlobalRegion region);
-
-    std::unordered_map<CacheAddr, GlobalRegion> reserve;
-    u32 max_ssbo_size{};
-};
-
-} // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index e216163e1..d1790f24f 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -20,6 +20,7 @@
 #include "core/hle/kernel/process.h"
 #include "core/settings.h"
 #include "video_core/engines/maxwell_3d.h"
+#include "video_core/memory_manager.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
 #include "video_core/renderer_opengl/gl_shader_gen.h"
@@ -82,8 +83,8 @@ struct DrawParameters {
 
 RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWindow& emu_window,
                                    ScreenInfo& info)
-    : texture_cache{system, *this, device}, shader_cache{*this, system, emu_window, device},
-      global_cache{*this}, system{system}, screen_info{info},
+    : texture_cache{system, *this, device},
+      shader_cache{*this, system, emu_window, device}, system{system}, screen_info{info},
       buffer_cache(*this, STREAM_BUFFER_SIZE) {
     OpenGLState::ApplyDefaultState();
 
@@ -689,7 +690,7 @@ void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
         return;
     }
     texture_cache.FlushRegion(addr, size);
-    global_cache.FlushRegion(addr, size);
+    buffer_cache.FlushRegion(addr, size);
 }
 
 void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
@@ -699,7 +700,6 @@ void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
     }
     texture_cache.InvalidateRegion(addr, size);
     shader_cache.InvalidateRegion(addr, size);
-    global_cache.InvalidateRegion(addr, size);
     buffer_cache.InvalidateRegion(addr, size);
 }
 
@@ -797,15 +797,22 @@ void RasterizerOpenGL::SetupConstBuffer(const Tegra::Engines::ConstBufferInfo& b
 
 void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
                                           const Shader& shader) {
+    auto& gpu{system.GPU()};
+    auto& memory_manager{gpu.MemoryManager()};
+    const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<std::size_t>(stage)]};
+    const auto alignment{device.GetShaderStorageBufferAlignment()};
+
     const auto& entries = shader->GetShaderEntries().global_memory_entries;
     for (std::size_t bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
         const auto& entry{entries[bindpoint]};
-        const auto& region{global_cache.GetGlobalRegion(entry, stage)};
-        if (entry.IsWritten()) {
-            region->MarkAsModified(true, global_cache);
-        }
-        bind_ssbo_pushbuffer.Push(region->GetBufferHandle(), 0,
-                                  static_cast<GLsizeiptr>(region->GetSizeInBytes()));
+
+        const auto addr{cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset()};
+        const auto actual_addr{memory_manager.Read<u64>(addr)};
+        const auto size{memory_manager.Read<u32>(addr + 8)};
+
+        const auto [ssbo, buffer_offset] =
+            buffer_cache.UploadMemory(actual_addr, size, alignment, true, entry.IsWritten());
+        bind_ssbo_pushbuffer.Push(ssbo, buffer_offset, static_cast<GLsizeiptr>(size));
     }
 }
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index a03bc759f..bc988727b 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -24,7 +24,6 @@
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_framebuffer_cache.h"
-#include "video_core/renderer_opengl/gl_global_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_sampler_cache.h"
 #include "video_core/renderer_opengl/gl_shader_cache.h"
@@ -189,7 +188,6 @@ private:
 
     TextureCacheOpenGL texture_cache;
     ShaderCacheOpenGL shader_cache;
-    GlobalRegionCacheOpenGL global_cache;
     SamplerCacheOpenGL sampler_cache;
     FramebufferCacheOpenGL framebuffer_cache;