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

Merge pull request #2244 from bunnei/gpu-mem-refactor

Browse source 
video_core: Refactor to use MemoryManager interface for all memory access.
This commit is contained in:
bunnei 2019-03-16 21:59:45 -04:00 • committed by GitHub
commit 2392e146b0
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
20 changed files with 196 additions and 189 deletions
Download patch file Download diff file Expand all files Collapse all files

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

@ -24,14 +24,12 @@ OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size)
GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size,
std::size_t alignment, bool cache) {
auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
ASSERT_MSG(cpu_addr, "Invalid GPU address");
// Cache management is a big overhead, so only cache entries with a given size.
// TODO: Figure out which size is the best for given games.
cache &= size >= 2048;
const auto& host_ptr{Memory::GetPointer(*cpu_addr)};
const auto& host_ptr{memory_manager.GetPointer(gpu_addr)};
if (cache) {
auto entry = TryGet(host_ptr);
if (entry) {
@ -54,8 +52,8 @@ GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size
buffer_offset += size;
if (cache) {
auto entry = std::make_shared<CachedBufferEntry>(*cpu_addr, size, uploaded_offset,
alignment, host_ptr);
auto entry = std::make_shared<CachedBufferEntry>(
*memory_manager.GpuToCpuAddress(gpu_addr), size, uploaded_offset, alignment, host_ptr);
Register(entry);
}

33
src/video_core/renderer_opengl/gl_global_cache.cpp
View file

@ -7,7 +7,6 @@
#include "common/assert.h"
#include "common/logging/log.h"
#include "core/core.h"
#include "core/memory.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"
@ -39,7 +38,7 @@ void CachedGlobalRegion::Reload(u32 size_) {
glBufferData(GL_SHADER_STORAGE_BUFFER, size, GetHostPtr(), GL_DYNAMIC_DRAW);
}
GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(VAddr addr, u32 size) const {
GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const {
const auto search{reserve.find(addr)};
if (search == reserve.end()) {
return {};
@ -47,11 +46,14 @@ GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(VAddr addr, u32
return search->second;
}
GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 size, u8* host_ptr) {
GlobalRegion region{TryGetReservedGlobalRegion(addr, size)};
GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size,
u8* host_ptr) {
GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)};
if (!region) {
// No reserved surface available, create a new one and reserve it
region = std::make_shared<CachedGlobalRegion>(addr, size, host_ptr);
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
const auto cpu_addr = *memory_manager.GpuToCpuAddress(addr);
region = std::make_shared<CachedGlobalRegion>(cpu_addr, size, host_ptr);
ReserveGlobalRegion(region);
}
region->Reload(size);
@ -59,7 +61,7 @@ GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 si
}
void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) {
reserve.insert_or_assign(region->GetCpuAddr(), std::move(region));
reserve.insert_or_assign(region->GetCacheAddr(), std::move(region));
}
GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer)
@ -70,23 +72,20 @@ GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
Tegra::Engines::Maxwell3D::Regs::ShaderStage stage) {
auto& gpu{Core::System::GetInstance().GPU()};
const auto cbufs = gpu.Maxwell3D().state.shader_stages[static_cast<u64>(stage)];
const auto cbuf_addr = gpu.MemoryManager().GpuToCpuAddress(
cbufs.const_buffers[global_region.GetCbufIndex()].address + global_region.GetCbufOffset());
ASSERT(cbuf_addr);
const auto actual_addr_gpu = Memory::Read64(*cbuf_addr);
const auto size = Memory::Read32(*cbuf_addr + 8);
const auto actual_addr = gpu.MemoryManager().GpuToCpuAddress(actual_addr_gpu);
ASSERT(actual_addr);
auto& memory_manager{gpu.MemoryManager()};
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)};
// Look up global region in the cache based on address
const auto& host_ptr{Memory::GetPointer(*actual_addr)};
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, size, host_ptr);
region = GetUncachedGlobalRegion(actual_addr, size, host_ptr);
Register(region);
}

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

@ -65,11 +65,11 @@ public:
Tegra::Engines::Maxwell3D::Regs::ShaderStage stage);
private:
GlobalRegion TryGetReservedGlobalRegion(VAddr addr, u32 size) const;
GlobalRegion GetUncachedGlobalRegion(VAddr addr, u32 size, u8* host_ptr);
GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const;
GlobalRegion GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size, u8* host_ptr);
void ReserveGlobalRegion(GlobalRegion region);
std::unordered_map<VAddr, GlobalRegion> reserve;
std::unordered_map<CacheAddr, GlobalRegion> reserve;
};
} // namespace OpenGL

7
src/video_core/renderer_opengl/gl_primitive_assembler.cpp
View file

@ -46,10 +46,7 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size
auto [dst_pointer, index_offset] = buffer_cache.ReserveMemory(map_size);
auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
ASSERT_MSG(cpu_addr, "Invalid GPU address");
const u8* source{Memory::GetPointer(*cpu_addr)};
const u8* source{memory_manager.GetPointer(gpu_addr)};
for (u32 primitive = 0; primitive < count / 4; ++primitive) {
for (std::size_t i = 0; i < TRIANGLES_PER_QUAD; ++i) {
@ -64,4 +61,4 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size
return index_offset;
}
} // namespace OpenGL
} // namespace OpenGL

39
src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
View file

@ -57,11 +57,9 @@ static void ApplyTextureDefaults(GLuint texture, u32 max_mip_level) {
void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) {
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr_)};
addr = cpu_addr ? *cpu_addr : 0;
gpu_addr = gpu_addr_;
host_ptr = Memory::GetPointer(addr);
host_ptr = memory_manager.GetPointer(gpu_addr_);
size_in_bytes = SizeInBytesRaw();
if (IsPixelFormatASTC(pixel_format)) {
@ -447,7 +445,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level),
params.MipBlockHeight(mip_level), params.MipHeight(mip_level),
params.MipBlockDepth(mip_level), 1, params.tile_width_spacing,
gl_buffer.data() + offset_gl, params.addr + offset);
gl_buffer.data() + offset_gl, params.host_ptr + offset);
offset += layer_size;
offset_gl += gl_size;
}
@ -456,7 +454,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level),
params.MipBlockHeight(mip_level), params.MipHeight(mip_level),
params.MipBlockDepth(mip_level), depth, params.tile_width_spacing,
gl_buffer.data(), params.addr + offset);
gl_buffer.data(), params.host_ptr + offset);
}
}
@ -514,9 +512,9 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac
"reinterpretation but the texture is tiled.");
}
const std::size_t remaining_size = dst_params.size_in_bytes - src_params.size_in_bytes;
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes, remaining_size,
Memory::GetPointer(dst_params.addr + src_params.size_in_bytes));
memory_manager.GetPointer(dst_params.gpu_addr + src_params.size_in_bytes));
}
glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
@ -604,7 +602,7 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
ApplyTextureDefaults(texture.handle, params.max_mip_level);
OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.addr, params.IdentityString());
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
@ -617,6 +615,8 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
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 +925,7 @@ void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {
}
Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool preserve_contents) {
if (params.addr == 0 || params.height * params.width == 0) {
if (params.gpu_addr == 0 || params.height * params.width == 0) {
return {};
}
@ -979,14 +979,16 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
const Surface& dst_surface) {
const auto& init_params{src_surface->GetSurfaceParams()};
const auto& dst_params{dst_surface->GetSurfaceParams()};
VAddr address = init_params.addr;
const std::size_t layer_size = dst_params.LayerMemorySize();
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
Tegra::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 VAddr sub_address = address + dst_params.GetMipmapLevelOffset(mipmap);
const Surface& copy = TryGet(Memory::GetPointer(sub_address));
if (!copy)
const Tegra::GPUVAddr sub_address{address + dst_params.GetMipmapLevelOffset(mipmap)};
const Surface& copy{TryGet(memory_manager.GetPointer(sub_address))};
if (!copy) {
continue;
}
const auto& src_params{copy->GetSurfaceParams()};
const u32 width{std::min(src_params.width, dst_params.MipWidth(mipmap))};
const u32 height{std::min(src_params.height, dst_params.MipHeight(mipmap))};
@ -1242,9 +1244,10 @@ static std::optional<u32> TryFindBestMipMap(std::size_t memory, const SurfacePar
return {};
}
static std::optional<u32> TryFindBestLayer(VAddr addr, const SurfaceParams params, u32 mipmap) {
const std::size_t size = params.LayerMemorySize();
VAddr start = params.addr + params.GetMipmapLevelOffset(mipmap);
static std::optional<u32> TryFindBestLayer(Tegra::GPUVAddr addr, const SurfaceParams params,
u32 mipmap) {
const std::size_t size{params.LayerMemorySize()};
Tegra::GPUVAddr start{params.gpu_addr + params.GetMipmapLevelOffset(mipmap)};
for (u32 i = 0; i < params.depth; i++) {
if (start == addr) {
return {i};
@ -1266,7 +1269,7 @@ static bool LayerFitReinterpretSurface(RasterizerCacheOpenGL& cache, const Surfa
src_params.height == dst_params.MipHeight(*level) &&
src_params.block_height >= dst_params.MipBlockHeight(*level)) {
const std::optional<u32> slot =
TryFindBestLayer(render_surface->GetCpuAddr(), dst_params, *level);
TryFindBestLayer(render_surface->GetSurfaceParams().gpu_addr, dst_params, *level);
if (slot.has_value()) {
glCopyImageSubData(render_surface->Texture().handle,
SurfaceTargetToGL(src_params.target), 0, 0, 0, 0,

4
src/video_core/renderer_opengl/gl_rasterizer_cache.h
View file

@ -296,7 +296,6 @@ struct SurfaceParams {
bool is_array;
bool srgb_conversion;
// Parameters used for caching
VAddr addr;
u8* host_ptr;
Tegra::GPUVAddr gpu_addr;
std::size_t size_in_bytes;
@ -349,7 +348,7 @@ public:
explicit CachedSurface(const SurfaceParams& params);
VAddr GetCpuAddr() const override {
return params.addr;
return cpu_addr;
}
std::size_t GetSizeInBytes() const override {
@ -433,6 +432,7 @@ private:
std::size_t memory_size;
bool reinterpreted = false;
bool must_reload = false;
VAddr cpu_addr{};
};
class RasterizerCacheOpenGL final : public RasterizerCache<Surface> {

37
src/video_core/renderer_opengl/gl_shader_cache.cpp
View file

@ -32,13 +32,10 @@ struct UnspecializedShader {
namespace {
/// Gets the address for the specified shader stage program
VAddr 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)];
const auto address = gpu.memory_manager.GpuToCpuAddress(gpu.regs.code_address.CodeAddress() +
shader_config.offset);
ASSERT_MSG(address, "Invalid GPU address");
return *address;
Tegra::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;
}
/// Gets the shader program code from memory for the specified address
@ -214,11 +211,11 @@ std::set<GLenum> GetSupportedFormats() {
} // namespace
CachedShader::CachedShader(VAddr guest_addr, u64 unique_identifier,
CachedShader::CachedShader(VAddr cpu_addr, u64 unique_identifier,
Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
const PrecompiledPrograms& precompiled_programs,
ProgramCode&& program_code, ProgramCode&& program_code_b, u8* host_ptr)
: host_ptr{host_ptr}, guest_addr{guest_addr}, unique_identifier{unique_identifier},
: host_ptr{host_ptr}, cpu_addr{cpu_addr}, unique_identifier{unique_identifier},
program_type{program_type}, disk_cache{disk_cache},
precompiled_programs{precompiled_programs}, RasterizerCacheObject{host_ptr} {
@ -244,11 +241,11 @@ CachedShader::CachedShader(VAddr guest_addr, u64 unique_identifier,
disk_cache.SaveRaw(raw);
}
CachedShader::CachedShader(VAddr guest_addr, u64 unique_identifier,
CachedShader::CachedShader(VAddr cpu_addr, u64 unique_identifier,
Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
const PrecompiledPrograms& precompiled_programs,
GLShader::ProgramResult result, u8* host_ptr)
: guest_addr{guest_addr}, unique_identifier{unique_identifier}, program_type{program_type},
: cpu_addr{cpu_addr}, unique_identifier{unique_identifier}, program_type{program_type},
disk_cache{disk_cache}, precompiled_programs{precompiled_programs}, RasterizerCacheObject{
host_ptr} {
@ -273,7 +270,7 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive
disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
}
LabelGLObject(GL_PROGRAM, program->handle, guest_addr);
LabelGLObject(GL_PROGRAM, program->handle, cpu_addr);
}
handle = program->handle;
@ -325,7 +322,7 @@ GLuint CachedShader::LazyGeometryProgram(CachedProgram& target_program, BaseBind
disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
}
LabelGLObject(GL_PROGRAM, target_program->handle, guest_addr, debug_name);
LabelGLObject(GL_PROGRAM, target_program->handle, cpu_addr, debug_name);
return target_program->handle;
};
@ -488,31 +485,31 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
return last_shaders[static_cast<u32>(program)];
}
const VAddr program_addr{GetShaderAddress(program)};
auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
const Tegra::GPUVAddr program_addr{GetShaderAddress(program)};
// Look up shader in the cache based on address
const auto& host_ptr{Memory::GetPointer(program_addr)};
const auto& host_ptr{memory_manager.GetPointer(program_addr)};
Shader shader{TryGet(host_ptr)};
if (!shader) {
// No shader found - create a new one
const auto& host_ptr{Memory::GetPointer(program_addr)};
ProgramCode program_code{GetShaderCode(host_ptr)};
ProgramCode program_code_b;
if (program == Maxwell::ShaderProgram::VertexA) {
program_code_b = GetShaderCode(
Memory::GetPointer(GetShaderAddress(Maxwell::ShaderProgram::VertexB)));
memory_manager.GetPointer(GetShaderAddress(Maxwell::ShaderProgram::VertexB)));
}
const u64 unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b);
const VAddr cpu_addr{*memory_manager.GpuToCpuAddress(program_addr)};
const auto found = precompiled_shaders.find(unique_identifier);
if (found != precompiled_shaders.end()) {
shader =
std::make_shared<CachedShader>(program_addr, unique_identifier, program, disk_cache,
std::make_shared<CachedShader>(cpu_addr, unique_identifier, program, disk_cache,
precompiled_programs, found->second, host_ptr);
} else {
shader = std::make_shared<CachedShader>(
program_addr, unique_identifier, program, disk_cache, precompiled_programs,
cpu_addr, unique_identifier, program, disk_cache, precompiled_programs,
std::move(program_code), std::move(program_code_b), host_ptr);
}
Register(shader);

8
src/video_core/renderer_opengl/gl_shader_cache.h
View file

@ -39,18 +39,18 @@ using PrecompiledShaders = std::unordered_map<u64, GLShader::ProgramResult>;
class CachedShader final : public RasterizerCacheObject {
public:
explicit CachedShader(VAddr guest_addr, u64 unique_identifier,
explicit CachedShader(VAddr cpu_addr, u64 unique_identifier,
Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
const PrecompiledPrograms& precompiled_programs,
ProgramCode&& program_code, ProgramCode&& program_code_b, u8* host_ptr);
explicit CachedShader(VAddr guest_addr, u64 unique_identifier,
explicit CachedShader(VAddr cpu_addr, u64 unique_identifier,
Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
const PrecompiledPrograms& precompiled_programs,
GLShader::ProgramResult result, u8* host_ptr);
VAddr GetCpuAddr() const override {
return guest_addr;
return cpu_addr;
}
std::size_t GetSizeInBytes() const override {
@ -92,7 +92,7 @@ private:
ShaderDiskCacheUsage GetUsage(GLenum primitive_mode, BaseBindings base_bindings) const;
u8* host_ptr{};
VAddr guest_addr{};
VAddr cpu_addr{};
u64 unique_identifier{};
Maxwell::ShaderProgram program_type{};
ShaderDiskCacheOpenGL& disk_cache;