yuzu-emu
/
yuzu
Archived
1
0
Fork 0
Code

Merge pull request #12094 from ameerj/gl-buffer-cache-batch-vtx 

gl_buffer_cache: Batch vertex/tfb buffer binding
This commit is contained in:
liamwhite 2023-12-03 16:27:22 -05:00 committed by GitHub
parent 6da8301773 a595ed499d
commit 5fb1a83e4c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 36 additions and 64 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
src/video_core/renderer_opengl/gl_buffer_cache.cpp
View File

@ -53,13 +53,11 @@ Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rast
VAddr cpu_addr_, u64 size_bytes_)
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) {
buffer.Create();
if (runtime.device.HasDebuggingToolAttached()) {
const std::string name = fmt::format("Buffer 0x{:x}", CpuAddr());
glObjectLabel(GL_BUFFER, buffer.handle, static_cast<GLsizei>(name.size()), name.data());
glNamedBufferData(buffer.handle, SizeBytes(), nullptr, GL_DYNAMIC_DRAW);
if (runtime.has_unified_vertex_buffers) {
glGetNamedBufferParameterui64vNV(buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &address);
}
glNamedBufferData(buffer.handle, SizeBytes(), nullptr, GL_DYNAMIC_DRAW);
}
void Buffer::ImmediateUpload(size_t offset, std::span<const u8> data) noexcept {
@ -111,7 +109,6 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_,
: device{device_}, staging_buffer_pool{staging_buffer_pool_},
has_fast_buffer_sub_data{device.HasFastBufferSubData()},
use_assembly_shaders{device.UseAssemblyShaders()},
has_unified_vertex_buffers{device.HasVertexBufferUnifiedMemory()},
stream_buffer{has_fast_buffer_sub_data ? std::nullopt : std::make_optional<StreamBuffer>()} {
GLint gl_max_attributes;
glGetIntegerv(GL_MAX_VERTEX_ATTRIBS, &gl_max_attributes);
@ -123,6 +120,7 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_,
GL_STREAM_DRAW);
}
}
if (use_assembly_shaders) {
for (auto& stage_uniforms : copy_uniforms) {
for (OGLBuffer& buffer : stage_uniforms) {
buffer.Create();
@ -133,6 +131,7 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_,
buffer.Create();
glNamedBufferData(buffer.handle, 0x10'000, nullptr, GL_STREAM_COPY);
}
}
device_access_memory = [this]() -> u64 {
if (device.CanReportMemoryUsage()) {
@ -211,39 +210,32 @@ void BufferCacheRuntime::ClearBuffer(Buffer& dest_buffer, u32 offset, size_t siz
}
void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) {
if (has_unified_vertex_buffers) {
buffer.MakeResident(GL_READ_ONLY);
glBufferAddressRangeNV(GL_ELEMENT_ARRAY_ADDRESS_NV, 0, buffer.HostGpuAddr() + offset,
static_cast<GLsizeiptr>(Common::AlignUp(size, 4)));
} else {
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer.Handle());
index_buffer_offset = offset;
}
}
void BufferCacheRuntime::BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size,
u32 stride) {
if (index >= max_attributes) {
return;
}
if (has_unified_vertex_buffers) {
buffer.MakeResident(GL_READ_ONLY);
glBindVertexBuffer(index, 0, 0, static_cast<GLsizei>(stride));
glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, index,
buffer.HostGpuAddr() + offset, static_cast<GLsizeiptr>(size));
} else {
glBindVertexBuffer(index, buffer.Handle(), static_cast<GLintptr>(offset),
static_cast<GLsizei>(stride));
}
}
void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings) {
for (u32 index = 0; index < bindings.buffers.size(); ++index) {
BindVertexBuffer(bindings.min_index + index, *bindings.buffers[index],
static_cast<u32>(bindings.offsets[index]),
static_cast<u32>(bindings.sizes[index]),
static_cast<u32>(bindings.strides[index]));
}
// TODO: Should HostBindings provide the correct runtime types to avoid these transforms?
std::array<GLuint, 32> buffer_handles;
std::array<GLsizei, 32> buffer_strides;
std::ranges::transform(bindings.buffers, buffer_handles.begin(),
[](const Buffer* const buffer) { return buffer->Handle(); });
std::ranges::transform(bindings.strides, buffer_strides.begin(),
[](u64 stride) { return static_cast<GLsizei>(stride); });
const u32 count =
std::min(static_cast<u32>(bindings.buffers.size()), max_attributes - bindings.min_index);
glBindVertexBuffers(bindings.min_index, static_cast<GLsizei>(count), buffer_handles.data(),
reinterpret_cast<const GLintptr*>(bindings.offsets.data()),
buffer_strides.data());
}
void BufferCacheRuntime::BindUniformBuffer(size_t stage, u32 binding_index, Buffer& buffer,
@ -335,11 +327,13 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, Buffer& buffer,
}
void BufferCacheRuntime::BindTransformFeedbackBuffers(VideoCommon::HostBindings<Buffer>& bindings) {
for (u32 index = 0; index < bindings.buffers.size(); ++index) {
glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, index, bindings.buffers[index]->Handle(),
static_cast<GLintptr>(bindings.offsets[index]),
static_cast<GLsizeiptr>(bindings.sizes[index]));
}
std::array<GLuint, 4> buffer_handles;
std::ranges::transform(bindings.buffers, buffer_handles.begin(),
[](const Buffer* const buffer) { return buffer->Handle(); });
glBindBuffersRange(GL_TRANSFORM_FEEDBACK_BUFFER, 0,
static_cast<GLsizei>(bindings.buffers.size()), buffer_handles.data(),
reinterpret_cast<const GLintptr*>(bindings.offsets.data()),
reinterpret_cast<const GLsizeiptr*>(bindings.strides.data()));
}
void BufferCacheRuntime::BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,

1
src/video_core/renderer_opengl/gl_buffer_cache.h
View File

@ -209,7 +209,6 @@ private:
bool has_fast_buffer_sub_data = false;
bool use_assembly_shaders = false;
bool has_unified_vertex_buffers = false;
bool use_storage_buffers = false;

1
src/video_core/renderer_opengl/gl_device.cpp
View File

@ -200,7 +200,6 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
has_broken_texture_view_formats = is_amd || (!is_linux && is_intel);
has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2;
has_derivative_control = GLAD_GL_ARB_derivative_control;
has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory;
has_debugging_tool_attached = IsDebugToolAttached(extensions);
has_depth_buffer_float = HasExtension(extensions, "GL_NV_depth_buffer_float");
has_geometry_shader_passthrough = GLAD_GL_NV_geometry_shader_passthrough;

5
src/video_core/renderer_opengl/gl_device.h
View File

@ -72,10 +72,6 @@ public:
return has_texture_shadow_lod;
}
bool HasVertexBufferUnifiedMemory() const {
return has_vertex_buffer_unified_memory;
}
bool HasASTC() const {
return has_astc;
}
@ -215,7 +211,6 @@ private:
bool has_vertex_viewport_layer{};
bool has_image_load_formatted{};
bool has_texture_shadow_lod{};
bool has_vertex_buffer_unified_memory{};
bool has_astc{};
bool has_variable_aoffi{};
bool has_component_indexing_bug{};

15
src/video_core/renderer_opengl/renderer_opengl.cpp
View File

@ -168,15 +168,6 @@ RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_,
if (!GLAD_GL_ARB_seamless_cubemap_per_texture && !GLAD_GL_AMD_seamless_cubemap_per_texture) {
glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS);
}
// Enable unified vertex attributes and query vertex buffer address when the driver supports it
if (device.HasVertexBufferUnifiedMemory()) {
glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV);
glEnableClientState(GL_ELEMENT_ARRAY_UNIFIED_NV);
glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY);
glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV,
&vertex_buffer_address);
}
}
RendererOpenGL::~RendererOpenGL() = default;
@ -680,13 +671,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
offsetof(ScreenRectVertex, tex_coord));
glVertexAttribBinding(PositionLocation, 0);
glVertexAttribBinding(TexCoordLocation, 0);
if (device.HasVertexBufferUnifiedMemory()) {
glBindVertexBuffer(0, 0, 0, sizeof(ScreenRectVertex));
glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, 0, vertex_buffer_address,
sizeof(vertices));
} else {
glBindVertexBuffer(0, vertex_buffer.handle, 0, sizeof(ScreenRectVertex));
}
if (Settings::values.scaling_filter.GetValue() != Settings::ScalingFilter::NearestNeighbor) {
glBindSampler(0, present_sampler.handle);