Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #2256 from bunnei/gpu-vmm

Browse source 
gpu: Rewrite MemoryManager based on the VMManager implementation.
This commit is contained in:
bunnei 2019-03-22 18:41:12 -04:00 • committed by GitHub
commit e5893db3e6
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
28 changed files with 561 additions and 268 deletions
Download patch file Download diff file Expand all files Collapse all files

4
src/video_core/renderer_opengl/gl_buffer_cache.cpp
View file

@ -21,8 +21,8 @@ CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr
OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size)
: RasterizerCache{rasterizer}, stream_buffer(size, true) {}
GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size,
std::size_t alignment, bool cache) {
GLintptr OGLBufferCache::UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment,
bool cache) {
auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
// Cache management is a big overhead, so only cache entries with a given size.

2
src/video_core/renderer_opengl/gl_buffer_cache.h
View file

@ -58,7 +58,7 @@ public:
/// Uploads data from a guest GPU address. Returns host's buffer offset where it's been
/// allocated.
GLintptr UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
GLintptr UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
bool cache = true);
/// Uploads from a host memory. Returns host's buffer offset where it's been allocated.

6
src/video_core/renderer_opengl/gl_global_cache.cpp
View file

@ -46,7 +46,7 @@ GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr,
return search->second;
}
GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size,
GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(GPUVAddr addr, u32 size,
u8* host_ptr) {
GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)};
if (!region) {
@ -76,8 +76,8 @@ GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<u64>(stage)]};
const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address +
global_region.GetCbufOffset()};
const auto actual_addr{memory_manager.Read64(addr)};
const auto size{memory_manager.Read32(addr + 8)};
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)};

2
src/video_core/renderer_opengl/gl_global_cache.h
View file

@ -66,7 +66,7 @@ public:
private:
GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const;
GlobalRegion GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size, u8* host_ptr);
GlobalRegion GetUncachedGlobalRegion(GPUVAddr addr, u32 size, u8* host_ptr);
void ReserveGlobalRegion(GlobalRegion region);
std::unordered_map<CacheAddr, GlobalRegion> reserve;

3
src/video_core/renderer_opengl/gl_primitive_assembler.cpp
View file

@ -40,8 +40,7 @@ GLintptr PrimitiveAssembler::MakeQuadArray(u32 first, u32 count) {
return index_offset;
}
GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size_t index_size,
u32 count) {
GLintptr PrimitiveAssembler::MakeQuadIndexed(GPUVAddr gpu_addr, std::size_t index_size, u32 count) {
const std::size_t map_size{CalculateQuadSize(count)};
auto [dst_pointer, index_offset] = buffer_cache.ReserveMemory(map_size);

2
src/video_core/renderer_opengl/gl_primitive_assembler.h
View file

@ -24,7 +24,7 @@ public:
GLintptr MakeQuadArray(u32 first, u32 count);
GLintptr MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size_t index_size, u32 count);
GLintptr MakeQuadIndexed(GPUVAddr gpu_addr, std::size_t index_size, u32 count);
private:
OGLBufferCache& buffer_cache;

8
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

@ -225,8 +225,8 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
if (!vertex_array.IsEnabled())
continue;
const Tegra::GPUVAddr start = vertex_array.StartAddress();
const Tegra::GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
const GPUVAddr start = vertex_array.StartAddress();
const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
ASSERT(end > start);
const u64 size = end - start + 1;
@ -421,8 +421,8 @@ std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
if (!regs.vertex_array[index].IsEnabled())
continue;
const Tegra::GPUVAddr start = regs.vertex_array[index].StartAddress();
const Tegra::GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
const GPUVAddr start = regs.vertex_array[index].StartAddress();
const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
ASSERT(end > start);
size += end - start + 1;

35
src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
View file

@ -55,7 +55,7 @@ static void ApplyTextureDefaults(GLuint texture, u32 max_mip_level) {
}
}
void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) {
void SurfaceParams::InitCacheParameters(GPUVAddr gpu_addr_) {
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
gpu_addr = gpu_addr_;
@ -222,7 +222,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
}
/*static*/ SurfaceParams SurfaceParams::CreateForDepthBuffer(
u32 zeta_width, u32 zeta_height, Tegra::GPUVAddr zeta_address, Tegra::DepthFormat format,
u32 zeta_width, u32 zeta_height, GPUVAddr zeta_address, Tegra::DepthFormat format,
u32 block_width, u32 block_height, u32 block_depth,
Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type) {
SurfaceParams params{};
@ -564,6 +564,12 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac
CachedSurface::CachedSurface(const SurfaceParams& params)
: params{params}, gl_target{SurfaceTargetToGL(params.target)},
cached_size_in_bytes{params.size_in_bytes}, RasterizerCacheObject{params.host_ptr} {
const auto optional_cpu_addr{
Core::System::GetInstance().GPU().MemoryManager().GpuToCpuAddress(params.gpu_addr)};
ASSERT_MSG(optional_cpu_addr, "optional_cpu_addr is invalid");
cpu_addr = *optional_cpu_addr;
texture.Create(gl_target);
// TODO(Rodrigo): Using params.GetRect() returns a different size than using its Mip*(0)
@ -603,20 +609,6 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
ApplyTextureDefaults(texture.handle, params.max_mip_level);
OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.gpu_addr, params.IdentityString());
// Clamp size to mapped GPU memory region
// TODO(bunnei): Super Mario Odyssey maps a 0x40000 byte region and then uses it for a 0x80000
// R32F render buffer. We do not yet know if this is a game bug or something else, but this
// check is necessary to prevent flushing from overwriting unmapped memory.
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
const u64 max_size{memory_manager.GetRegionEnd(params.gpu_addr) - params.gpu_addr};
if (cached_size_in_bytes > max_size) {
LOG_ERROR(HW_GPU, "Surface size {} exceeds region size {}", params.size_in_bytes, max_size);
cached_size_in_bytes = max_size;
}
cpu_addr = *memory_manager.GpuToCpuAddress(params.gpu_addr);
}
MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 192, 64));
@ -925,7 +917,7 @@ void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {
}
Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool preserve_contents) {
if (params.gpu_addr == 0 || params.height * params.width == 0) {
if (!params.IsValid()) {
return {};
}
@ -980,11 +972,11 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
const auto& init_params{src_surface->GetSurfaceParams()};
const auto& dst_params{dst_surface->GetSurfaceParams()};
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
Tegra::GPUVAddr address{init_params.gpu_addr};
GPUVAddr address{init_params.gpu_addr};
const std::size_t layer_size{dst_params.LayerMemorySize()};
for (u32 layer = 0; layer < dst_params.depth; layer++) {
for (u32 mipmap = 0; mipmap < dst_params.max_mip_level; mipmap++) {
const Tegra::GPUVAddr sub_address{address + dst_params.GetMipmapLevelOffset(mipmap)};
const GPUVAddr sub_address{address + dst_params.GetMipmapLevelOffset(mipmap)};
const Surface& copy{TryGet(memory_manager.GetPointer(sub_address))};
if (!copy) {
continue;
@ -1244,10 +1236,9 @@ static std::optional<u32> TryFindBestMipMap(std::size_t memory, const SurfacePar
return {};
}
static std::optional<u32> TryFindBestLayer(Tegra::GPUVAddr addr, const SurfaceParams params,
u32 mipmap) {
static std::optional<u32> TryFindBestLayer(GPUVAddr addr, const SurfaceParams params, u32 mipmap) {
const std::size_t size{params.LayerMemorySize()};
Tegra::GPUVAddr start{params.gpu_addr + params.GetMipmapLevelOffset(mipmap)};
GPUVAddr start{params.gpu_addr + params.GetMipmapLevelOffset(mipmap)};
for (u32 i = 0; i < params.depth; i++) {
if (start == addr) {
return {i};

11
src/video_core/renderer_opengl/gl_rasterizer_cache.h
View file

@ -109,6 +109,11 @@ struct SurfaceParams {
return size;
}
/// Returns true if the parameters constitute a valid rasterizer surface.
bool IsValid() const {
return gpu_addr && host_ptr && height && width;
}
/// Returns the exact size of the memory occupied by a layer in a texture in VRAM, including
/// mipmaps.
std::size_t LayerMemorySize() const {
@ -210,7 +215,7 @@ struct SurfaceParams {
/// Creates SurfaceParams for a depth buffer configuration
static SurfaceParams CreateForDepthBuffer(
u32 zeta_width, u32 zeta_height, Tegra::GPUVAddr zeta_address, Tegra::DepthFormat format,
u32 zeta_width, u32 zeta_height, GPUVAddr zeta_address, Tegra::DepthFormat format,
u32 block_width, u32 block_height, u32 block_depth,
Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type);
@ -232,7 +237,7 @@ struct SurfaceParams {
}
/// Initializes parameters for caching, should be called after everything has been initialized
void InitCacheParameters(Tegra::GPUVAddr gpu_addr);
void InitCacheParameters(GPUVAddr gpu_addr);
std::string TargetName() const {
switch (target) {
@ -297,7 +302,7 @@ struct SurfaceParams {
bool srgb_conversion;
// Parameters used for caching
u8* host_ptr;
Tegra::GPUVAddr gpu_addr;
GPUVAddr gpu_addr;
std::size_t size_in_bytes;
std::size_t size_in_bytes_gl;

4
src/video_core/renderer_opengl/gl_shader_cache.cpp
View file

@ -32,7 +32,7 @@ struct UnspecializedShader {
namespace {
/// Gets the address for the specified shader stage program
Tegra::GPUVAddr GetShaderAddress(Maxwell::ShaderProgram program) {
GPUVAddr GetShaderAddress(Maxwell::ShaderProgram program) {
const auto& gpu{Core::System::GetInstance().GPU().Maxwell3D()};
const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]};
return gpu.regs.code_address.CodeAddress() + shader_config.offset;
@ -486,7 +486,7 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
}
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
const Tegra::GPUVAddr program_addr{GetShaderAddress(program)};
const GPUVAddr program_addr{GetShaderAddress(program)};
// Look up shader in the cache based on address
const auto& host_ptr{memory_manager.GetPointer(program_addr)};