hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 4028

Browse Source
This commit is contained in:
pineappleEA
2023-12-20 02:37:09 +01:00
parent 4539458956
commit 7bd2421ff9
51 changed files with 1669 additions and 645 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
src/video_core/engines/maxwell_3d.cpp
View File

@@ -586,14 +586,22 @@ void Maxwell3D::ProcessQueryCondition() {
}
void Maxwell3D::ProcessCounterReset() {
switch (regs.clear_report_value) {
case Regs::ClearReport::ZPassPixelCount:
rasterizer->ResetCounter(VideoCommon::QueryType::ZPassPixelCount64);
break;
default:
LOG_DEBUG(Render_OpenGL, "Unimplemented counter reset={}", regs.clear_report_value);
break;
}
const auto query_type = [clear_report = regs.clear_report_value]() {
switch (clear_report) {
case Tegra::Engines::Maxwell3D::Regs::ClearReport::ZPassPixelCount:
return VideoCommon::QueryType::ZPassPixelCount64;
case Tegra::Engines::Maxwell3D::Regs::ClearReport::StreamingPrimitivesSucceeded:
return VideoCommon::QueryType::StreamingPrimitivesSucceeded;
case Tegra::Engines::Maxwell3D::Regs::ClearReport::PrimitivesGenerated:
return VideoCommon::QueryType::PrimitivesGenerated;
case Tegra::Engines::Maxwell3D::Regs::ClearReport::VtgPrimitivesOut:
return VideoCommon::QueryType::VtgPrimitivesOut;
default:
LOG_DEBUG(HW_GPU, "Unimplemented counter reset={}", clear_report);
return VideoCommon::QueryType::Payload;
}
}();
rasterizer->ResetCounter(query_type);
}
void Maxwell3D::ProcessSyncPoint() {

39
src/video_core/query_cache.h
View File

@@ -28,8 +28,11 @@
namespace VideoCore {
enum class QueryType {
SamplesPassed,
PrimitivesGenerated,
TfbPrimitivesWritten,
Count,
};
constexpr std::size_t NumQueryTypes = 1;
constexpr std::size_t NumQueryTypes = static_cast<size_t>(QueryType::Count);
} // namespace VideoCore
namespace VideoCommon {
@@ -44,15 +47,6 @@ public:
explicit CounterStreamBase(QueryCache& cache_, VideoCore::QueryType type_)
: cache{cache_}, type{type_} {}
/// Updates the state of the stream, enabling or disabling as needed.
void Update(bool enabled) {
if (enabled) {
Enable();
} else {
Disable();
}
}
/// Resets the stream to zero. It doesn't disable the query after resetting.
void Reset() {
if (current) {
@@ -80,7 +74,6 @@ public:
return current != nullptr;
}
private:
/// Enables the stream.
void Enable() {
if (current) {
@@ -97,6 +90,7 @@ private:
last = std::exchange(current, nullptr);
}
private:
QueryCache& cache;
const VideoCore::QueryType type;
@@ -112,8 +106,14 @@ public:
: rasterizer{rasterizer_},
// Use reinterpret_cast instead of static_cast as workaround for
// UBSan bug (https://github.com/llvm/llvm-project/issues/59060)
cpu_memory{cpu_memory_}, streams{{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::SamplesPassed}}} {
cpu_memory{cpu_memory_}, streams{{
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::SamplesPassed}},
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::PrimitivesGenerated}},
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::TfbPrimitivesWritten}},
}} {
(void)slot_async_jobs.insert(); // Null value
}
@@ -157,12 +157,11 @@ public:
AsyncFlushQuery(query, timestamp, lock);
}
/// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch.
void UpdateCounters() {
/// Enables all available GPU counters
void EnableCounters() {
std::unique_lock lock{mutex};
if (maxwell3d) {
const auto& regs = maxwell3d->regs;
Stream(VideoCore::QueryType::SamplesPassed).Update(regs.zpass_pixel_count_enable);
for (auto& stream : streams) {
stream.Enable();
}
}
@@ -176,7 +175,7 @@ public:
void DisableStreams() {
std::unique_lock lock{mutex};
for (auto& stream : streams) {
stream.Update(false);
stream.Disable();
}
}
@@ -353,7 +352,7 @@ private:
std::shared_ptr<std::vector<AsyncJobId>> uncommitted_flushes{};
std::list<std::shared_ptr<std::vector<AsyncJobId>>> committed_flushes;
};
}; // namespace VideoCommon
template <class QueryCache, class HostCounter>
class HostCounterBase {

45
src/video_core/renderer_opengl/gl_buffer_cache.cpp
View File

@@ -58,6 +58,9 @@ Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rast
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);
}
}
void Buffer::ImmediateUpload(size_t offset, std::span<const u8> data) noexcept {
@@ -109,6 +112,7 @@ 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);
@@ -210,8 +214,14 @@ void BufferCacheRuntime::ClearBuffer(Buffer& dest_buffer, u32 offset, size_t siz
}
void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) {
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer.Handle());
index_buffer_offset = offset;
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,
@@ -219,8 +229,15 @@ void BufferCacheRuntime::BindVertexBuffer(u32 index, Buffer& buffer, u32 offset,
if (index >= max_attributes) {
return;
}
glBindVertexBuffer(index, buffer.Handle(), static_cast<GLintptr>(offset),
static_cast<GLsizei>(stride));
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) {
@@ -233,9 +250,23 @@ void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bi
[](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());
if (has_unified_vertex_buffers) {
for (u32 index = 0; index < count; ++index) {
Buffer& buffer = *bindings.buffers[index];
buffer.MakeResident(GL_READ_ONLY);
glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, bindings.min_index + index,
buffer.HostGpuAddr() + bindings.offsets[index],
static_cast<GLsizeiptr>(bindings.sizes[index]));
}
static constexpr std::array<size_t, 32> ZEROS{};
glBindVertexBuffers(bindings.min_index, static_cast<GLsizei>(count),
reinterpret_cast<const GLuint*>(ZEROS.data()),
reinterpret_cast<const GLintptr*>(ZEROS.data()), buffer_strides.data());
} else {
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,

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

@@ -209,6 +209,7 @@ 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

@@ -200,6 +200,7 @@ 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

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

23
src/video_core/renderer_opengl/gl_query_cache.cpp
View File

@@ -18,16 +18,27 @@ namespace OpenGL {
namespace {
constexpr std::array<GLenum, VideoCore::NumQueryTypes> QueryTargets = {GL_SAMPLES_PASSED};
constexpr GLenum GetTarget(VideoCore::QueryType type) {
return QueryTargets[static_cast<std::size_t>(type)];
switch (type) {
case VideoCore::QueryType::SamplesPassed:
return GL_SAMPLES_PASSED;
case VideoCore::QueryType::PrimitivesGenerated:
return GL_PRIMITIVES_GENERATED;
case VideoCore::QueryType::TfbPrimitivesWritten:
return GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN;
default:
break;
}
UNIMPLEMENTED_MSG("Query type {}", type);
return 0;
}
} // Anonymous namespace
QueryCache::QueryCache(RasterizerOpenGL& rasterizer_, Core::Memory::Memory& cpu_memory_)
: QueryCacheLegacy(rasterizer_, cpu_memory_), gl_rasterizer{rasterizer_} {}
: QueryCacheLegacy(rasterizer_, cpu_memory_), gl_rasterizer{rasterizer_} {
EnableCounters();
}
QueryCache::~QueryCache() = default;
@@ -103,13 +114,13 @@ u64 CachedQuery::Flush([[maybe_unused]] bool async) {
auto& stream = cache->Stream(type);
const bool slice_counter = WaitPending() && stream.IsEnabled();
if (slice_counter) {
stream.Update(false);
stream.Disable();
}
auto result = VideoCommon::CachedQueryBase<HostCounter>::Flush();
if (slice_counter) {
stream.Update(true);
stream.Enable();
}
return result;

57
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@@ -51,6 +51,22 @@ constexpr size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16;
void oglEnable(GLenum cap, bool state) {
(state ? glEnable : glDisable)(cap);
}
std::optional<VideoCore::QueryType> MaxwellToVideoCoreQuery(VideoCommon::QueryType type) {
switch (type) {
case VideoCommon::QueryType::PrimitivesGenerated:
case VideoCommon::QueryType::VtgPrimitivesOut:
return VideoCore::QueryType::PrimitivesGenerated;
case VideoCommon::QueryType::ZPassPixelCount64:
return VideoCore::QueryType::SamplesPassed;
case VideoCommon::QueryType::StreamingPrimitivesSucceeded:
// case VideoCommon::QueryType::StreamingByteCount:
// TODO: StreamingByteCount = StreamingPrimitivesSucceeded * num_verts * vert_stride
return VideoCore::QueryType::TfbPrimitivesWritten;
default:
return std::nullopt;
}
}
} // Anonymous namespace
RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
@@ -162,14 +178,18 @@ void RasterizerOpenGL::Clear(u32 layer_count) {
SyncFramebufferSRGB();
}
if (regs.clear_surface.Z) {
ASSERT_MSG(regs.zeta_enable != 0, "Tried to clear Z but buffer is not enabled!");
if (regs.zeta_enable != 0) {
LOG_DEBUG(Render_OpenGL, "Tried to clear Z but buffer is not enabled!");
}
use_depth = true;
state_tracker.NotifyDepthMask();
glDepthMask(GL_TRUE);
}
if (regs.clear_surface.S) {
ASSERT_MSG(regs.zeta_enable, "Tried to clear stencil but buffer is not enabled!");
if (regs.zeta_enable) {
LOG_DEBUG(Render_OpenGL, "Tried to clear stencil but buffer is not enabled!");
}
use_stencil = true;
}
@@ -212,7 +232,6 @@ void RasterizerOpenGL::PrepareDraw(bool is_indexed, Func&& draw_func) {
SCOPE_EXIT({ gpu.TickWork(); });
gpu_memory->FlushCaching();
query_cache.UpdateCounters();
GraphicsPipeline* const pipeline{shader_cache.CurrentGraphicsPipeline()};
if (!pipeline) {
@@ -330,7 +349,6 @@ void RasterizerOpenGL::DrawTexture() {
MICROPROFILE_SCOPE(OpenGL_Drawing);
SCOPE_EXIT({ gpu.TickWork(); });
query_cache.UpdateCounters();
texture_cache.SynchronizeGraphicsDescriptors();
texture_cache.UpdateRenderTargets(false);
@@ -397,21 +415,28 @@ void RasterizerOpenGL::DispatchCompute() {
}
void RasterizerOpenGL::ResetCounter(VideoCommon::QueryType type) {
if (type == VideoCommon::QueryType::ZPassPixelCount64) {
query_cache.ResetCounter(VideoCore::QueryType::SamplesPassed);
const auto query_cache_type = MaxwellToVideoCoreQuery(type);
if (!query_cache_type.has_value()) {
UNIMPLEMENTED_MSG("Reset query type: {}", type);
return;
}
query_cache.ResetCounter(*query_cache_type);
}
void RasterizerOpenGL::Query(GPUVAddr gpu_addr, VideoCommon::QueryType type,
VideoCommon::QueryPropertiesFlags flags, u32 payload, u32 subreport) {
if (type == VideoCommon::QueryType::ZPassPixelCount64) {
if (True(flags & VideoCommon::QueryPropertiesFlags::HasTimeout)) {
query_cache.Query(gpu_addr, VideoCore::QueryType::SamplesPassed, {gpu.GetTicks()});
} else {
query_cache.Query(gpu_addr, VideoCore::QueryType::SamplesPassed, std::nullopt);
}
return;
const auto query_cache_type = MaxwellToVideoCoreQuery(type);
if (!query_cache_type.has_value()) {
return QueryFallback(gpu_addr, type, flags, payload, subreport);
}
const bool has_timeout = True(flags & VideoCommon::QueryPropertiesFlags::HasTimeout);
const auto timestamp = has_timeout ? std::optional<u64>{gpu.GetTicks()} : std::nullopt;
query_cache.Query(gpu_addr, *query_cache_type, timestamp);
}
void RasterizerOpenGL::QueryFallback(GPUVAddr gpu_addr, VideoCommon::QueryType type,
VideoCommon::QueryPropertiesFlags flags, u32 payload,
u32 subreport) {
if (type != VideoCommon::QueryType::Payload) {
payload = 1u;
}
@@ -1294,15 +1319,13 @@ void RasterizerOpenGL::BeginTransformFeedback(GraphicsPipeline* program, GLenum
program->ConfigureTransformFeedback();
UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderType::TessellationInit) ||
regs.IsShaderConfigEnabled(Maxwell::ShaderType::Tessellation) ||
regs.IsShaderConfigEnabled(Maxwell::ShaderType::Geometry));
UNIMPLEMENTED_IF(primitive_mode != GL_POINTS);
regs.IsShaderConfigEnabled(Maxwell::ShaderType::Tessellation));
// We may have to call BeginTransformFeedbackNV here since they seem to call different
// implementations on Nvidia's driver (the pointer is different) but we are using
// ARB_transform_feedback3 features with NV_transform_feedback interactions and the ARB
// extension doesn't define BeginTransformFeedback (without NV) interactions. It just works.
glBeginTransformFeedback(GL_POINTS);
glBeginTransformFeedback(primitive_mode);
}
void RasterizerOpenGL::EndTransformFeedback() {

3
src/video_core/renderer_opengl/gl_rasterizer.h
View File

@@ -225,6 +225,9 @@ private:
/// End a transform feedback
void EndTransformFeedback();
void QueryFallback(GPUVAddr gpu_addr, VideoCommon::QueryType type,
VideoCommon::QueryPropertiesFlags flags, u32 payload, u32 subreport);
Tegra::GPU& gpu;
const Device& device;

16
src/video_core/renderer_opengl/renderer_opengl.cpp
View File

@@ -168,6 +168,14 @@ 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;
@@ -667,7 +675,13 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
offsetof(ScreenRectVertex, tex_coord));
glVertexAttribBinding(PositionLocation, 0);
glVertexAttribBinding(TexCoordLocation, 0);
glBindVertexBuffer(0, vertex_buffer.handle, 0, sizeof(ScreenRectVertex));
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);

4
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@@ -485,6 +485,10 @@ void RasterizerVulkan::DispatchCompute() {
}
void RasterizerVulkan::ResetCounter(VideoCommon::QueryType type) {
if (type != VideoCommon::QueryType::ZPassPixelCount64) {
LOG_DEBUG(Render_Vulkan, "Unimplemented counter reset={}", type);
return;
}
query_cache.CounterReset(type);
}