mirror of
https://git.suyu.dev/suyu/suyu
synced 2024-11-01 21:07:52 +00:00
Merge pull request #2002 from ReinUsesLisp/dsa-vao-buffer
gl_rasterizer: Use DSA for VAOs and buffers
This commit is contained in:
commit
cbf8bea9d5
9 changed files with 73 additions and 103 deletions
|
@ -14,7 +14,7 @@
|
|||
namespace OpenGL {
|
||||
|
||||
OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size)
|
||||
: RasterizerCache{rasterizer}, stream_buffer(GL_ARRAY_BUFFER, size) {}
|
||||
: RasterizerCache{rasterizer}, stream_buffer(size, true) {}
|
||||
|
||||
GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size,
|
||||
std::size_t alignment, bool cache) {
|
||||
|
|
|
@ -135,27 +135,31 @@ void RasterizerOpenGL::CheckExtensions() {
|
|||
}
|
||||
}
|
||||
|
||||
void RasterizerOpenGL::SetupVertexFormat() {
|
||||
GLuint RasterizerOpenGL::SetupVertexFormat() {
|
||||
auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
|
||||
const auto& regs = gpu.regs;
|
||||
|
||||
if (!gpu.dirty_flags.vertex_attrib_format)
|
||||
return;
|
||||
if (!gpu.dirty_flags.vertex_attrib_format) {
|
||||
return state.draw.vertex_array;
|
||||
}
|
||||
gpu.dirty_flags.vertex_attrib_format = false;
|
||||
|
||||
MICROPROFILE_SCOPE(OpenGL_VAO);
|
||||
|
||||
auto [iter, is_cache_miss] = vertex_array_cache.try_emplace(regs.vertex_attrib_format);
|
||||
auto& VAO = iter->second;
|
||||
auto& vao_entry = iter->second;
|
||||
|
||||
if (is_cache_miss) {
|
||||
VAO.Create();
|
||||
state.draw.vertex_array = VAO.handle;
|
||||
state.ApplyVertexBufferState();
|
||||
vao_entry.Create();
|
||||
const GLuint vao = vao_entry.handle;
|
||||
|
||||
// The index buffer binding is stored within the VAO. Stupid OpenGL, but easy to work
|
||||
// around.
|
||||
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer_cache.GetHandle());
|
||||
// Eventhough we are using DSA to create this vertex array, there is a bug on Intel's blob
|
||||
// that fails to properly create the vertex array if it's not bound even after creating it
|
||||
// with glCreateVertexArrays
|
||||
state.draw.vertex_array = vao;
|
||||
state.ApplyVertexArrayState();
|
||||
|
||||
glVertexArrayElementBuffer(vao, buffer_cache.GetHandle());
|
||||
|
||||
// Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL.
|
||||
// Enables the first 16 vertex attributes always, as we don't know which ones are actually
|
||||
|
@ -163,7 +167,7 @@ void RasterizerOpenGL::SetupVertexFormat() {
|
|||
// for now to avoid OpenGL errors.
|
||||
// TODO(Subv): Analyze the shader to identify which attributes are actually used and don't
|
||||
// assume every shader uses them all.
|
||||
for (unsigned index = 0; index < 16; ++index) {
|
||||
for (u32 index = 0; index < 16; ++index) {
|
||||
const auto& attrib = regs.vertex_attrib_format[index];
|
||||
|
||||
// Ignore invalid attributes.
|
||||
|
@ -178,28 +182,29 @@ void RasterizerOpenGL::SetupVertexFormat() {
|
|||
|
||||
ASSERT(buffer.IsEnabled());
|
||||
|
||||
glEnableVertexAttribArray(index);
|
||||
glEnableVertexArrayAttrib(vao, index);
|
||||
if (attrib.type == Tegra::Engines::Maxwell3D::Regs::VertexAttribute::Type::SignedInt ||
|
||||
attrib.type ==
|
||||
Tegra::Engines::Maxwell3D::Regs::VertexAttribute::Type::UnsignedInt) {
|
||||
glVertexAttribIFormat(index, attrib.ComponentCount(),
|
||||
glVertexArrayAttribIFormat(vao, index, attrib.ComponentCount(),
|
||||
MaxwellToGL::VertexType(attrib), attrib.offset);
|
||||
} else {
|
||||
glVertexAttribFormat(index, attrib.ComponentCount(),
|
||||
MaxwellToGL::VertexType(attrib),
|
||||
glVertexArrayAttribFormat(
|
||||
vao, index, attrib.ComponentCount(), MaxwellToGL::VertexType(attrib),
|
||||
attrib.IsNormalized() ? GL_TRUE : GL_FALSE, attrib.offset);
|
||||
}
|
||||
glVertexAttribBinding(index, attrib.buffer);
|
||||
glVertexArrayAttribBinding(vao, index, attrib.buffer);
|
||||
}
|
||||
}
|
||||
state.draw.vertex_array = VAO.handle;
|
||||
state.ApplyVertexBufferState();
|
||||
|
||||
// Rebinding the VAO invalidates the vertex buffer bindings.
|
||||
gpu.dirty_flags.vertex_array = 0xFFFFFFFF;
|
||||
|
||||
state.draw.vertex_array = vao_entry.handle;
|
||||
return vao_entry.handle;
|
||||
}
|
||||
|
||||
void RasterizerOpenGL::SetupVertexBuffer() {
|
||||
void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
|
||||
auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
|
||||
const auto& regs = gpu.regs;
|
||||
|
||||
|
@ -217,7 +222,7 @@ void RasterizerOpenGL::SetupVertexBuffer() {
|
|||
if (!vertex_array.IsEnabled())
|
||||
continue;
|
||||
|
||||
Tegra::GPUVAddr start = vertex_array.StartAddress();
|
||||
const Tegra::GPUVAddr start = vertex_array.StartAddress();
|
||||
const Tegra::GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
|
||||
|
||||
ASSERT(end > start);
|
||||
|
@ -225,21 +230,18 @@ void RasterizerOpenGL::SetupVertexBuffer() {
|
|||
const GLintptr vertex_buffer_offset = buffer_cache.UploadMemory(start, size);
|
||||
|
||||
// Bind the vertex array to the buffer at the current offset.
|
||||
glBindVertexBuffer(index, buffer_cache.GetHandle(), vertex_buffer_offset,
|
||||
glVertexArrayVertexBuffer(vao, index, buffer_cache.GetHandle(), vertex_buffer_offset,
|
||||
vertex_array.stride);
|
||||
|
||||
if (regs.instanced_arrays.IsInstancingEnabled(index) && vertex_array.divisor != 0) {
|
||||
// Enable vertex buffer instancing with the specified divisor.
|
||||
glVertexBindingDivisor(index, vertex_array.divisor);
|
||||
glVertexArrayBindingDivisor(vao, index, vertex_array.divisor);
|
||||
} else {
|
||||
// Disable the vertex buffer instancing.
|
||||
glVertexBindingDivisor(index, 0);
|
||||
glVertexArrayBindingDivisor(vao, index, 0);
|
||||
}
|
||||
}
|
||||
|
||||
// Implicit set by glBindVertexBuffer. Stupid glstate handling...
|
||||
state.draw.vertex_buffer = buffer_cache.GetHandle();
|
||||
|
||||
gpu.dirty_flags.vertex_array = 0;
|
||||
}
|
||||
|
||||
|
@ -691,9 +693,6 @@ void RasterizerOpenGL::DrawArrays() {
|
|||
// Draw the vertex batch
|
||||
const bool is_indexed = accelerate_draw == AccelDraw::Indexed;
|
||||
|
||||
state.draw.vertex_buffer = buffer_cache.GetHandle();
|
||||
state.ApplyVertexBufferState();
|
||||
|
||||
std::size_t buffer_size = CalculateVertexArraysSize();
|
||||
|
||||
// Add space for index buffer (keeping in mind non-core primitives)
|
||||
|
@ -723,8 +722,9 @@ void RasterizerOpenGL::DrawArrays() {
|
|||
gpu.dirty_flags.vertex_array = 0xFFFFFFFF;
|
||||
}
|
||||
|
||||
SetupVertexFormat();
|
||||
SetupVertexBuffer();
|
||||
const GLuint vao = SetupVertexFormat();
|
||||
SetupVertexBuffer(vao);
|
||||
|
||||
DrawParameters params = SetupDraw();
|
||||
SetupShaders(params.primitive_mode);
|
||||
|
||||
|
|
|
@ -215,8 +215,10 @@ private:
|
|||
|
||||
std::size_t CalculateIndexBufferSize() const;
|
||||
|
||||
void SetupVertexFormat();
|
||||
void SetupVertexBuffer();
|
||||
/// Updates and returns a vertex array object representing current vertex format
|
||||
GLuint SetupVertexFormat();
|
||||
|
||||
void SetupVertexBuffer(GLuint vao);
|
||||
|
||||
DrawParameters SetupDraw();
|
||||
|
||||
|
|
|
@ -117,7 +117,7 @@ void OGLBuffer::Create() {
|
|||
return;
|
||||
|
||||
MICROPROFILE_SCOPE(OpenGL_ResourceCreation);
|
||||
glGenBuffers(1, &handle);
|
||||
glCreateBuffers(1, &handle);
|
||||
}
|
||||
|
||||
void OGLBuffer::Release() {
|
||||
|
@ -126,7 +126,6 @@ void OGLBuffer::Release() {
|
|||
|
||||
MICROPROFILE_SCOPE(OpenGL_ResourceDeletion);
|
||||
glDeleteBuffers(1, &handle);
|
||||
OpenGLState::GetCurState().ResetBuffer(handle).Apply();
|
||||
handle = 0;
|
||||
}
|
||||
|
||||
|
@ -152,7 +151,7 @@ void OGLVertexArray::Create() {
|
|||
return;
|
||||
|
||||
MICROPROFILE_SCOPE(OpenGL_ResourceCreation);
|
||||
glGenVertexArrays(1, &handle);
|
||||
glCreateVertexArrays(1, &handle);
|
||||
}
|
||||
|
||||
void OGLVertexArray::Release() {
|
||||
|
|
|
@ -83,8 +83,6 @@ OpenGLState::OpenGLState() {
|
|||
draw.read_framebuffer = 0;
|
||||
draw.draw_framebuffer = 0;
|
||||
draw.vertex_array = 0;
|
||||
draw.vertex_buffer = 0;
|
||||
draw.uniform_buffer = 0;
|
||||
draw.shader_program = 0;
|
||||
draw.program_pipeline = 0;
|
||||
|
||||
|
@ -505,7 +503,6 @@ void OpenGLState::ApplySamplers() const {
|
|||
}
|
||||
|
||||
void OpenGLState::ApplyFramebufferState() const {
|
||||
// Framebuffer
|
||||
if (draw.read_framebuffer != cur_state.draw.read_framebuffer) {
|
||||
glBindFramebuffer(GL_READ_FRAMEBUFFER, draw.read_framebuffer);
|
||||
}
|
||||
|
@ -514,16 +511,10 @@ void OpenGLState::ApplyFramebufferState() const {
|
|||
}
|
||||
}
|
||||
|
||||
void OpenGLState::ApplyVertexBufferState() const {
|
||||
// Vertex array
|
||||
void OpenGLState::ApplyVertexArrayState() const {
|
||||
if (draw.vertex_array != cur_state.draw.vertex_array) {
|
||||
glBindVertexArray(draw.vertex_array);
|
||||
}
|
||||
|
||||
// Vertex buffer
|
||||
if (draw.vertex_buffer != cur_state.draw.vertex_buffer) {
|
||||
glBindBuffer(GL_ARRAY_BUFFER, draw.vertex_buffer);
|
||||
}
|
||||
}
|
||||
|
||||
void OpenGLState::ApplyDepthClamp() const {
|
||||
|
@ -543,11 +534,7 @@ void OpenGLState::ApplyDepthClamp() const {
|
|||
|
||||
void OpenGLState::Apply() const {
|
||||
ApplyFramebufferState();
|
||||
ApplyVertexBufferState();
|
||||
// Uniform buffer
|
||||
if (draw.uniform_buffer != cur_state.draw.uniform_buffer) {
|
||||
glBindBuffer(GL_UNIFORM_BUFFER, draw.uniform_buffer);
|
||||
}
|
||||
ApplyVertexArrayState();
|
||||
|
||||
// Shader program
|
||||
if (draw.shader_program != cur_state.draw.shader_program) {
|
||||
|
@ -638,16 +625,6 @@ OpenGLState& OpenGLState::ResetPipeline(GLuint handle) {
|
|||
return *this;
|
||||
}
|
||||
|
||||
OpenGLState& OpenGLState::ResetBuffer(GLuint handle) {
|
||||
if (draw.vertex_buffer == handle) {
|
||||
draw.vertex_buffer = 0;
|
||||
}
|
||||
if (draw.uniform_buffer == handle) {
|
||||
draw.uniform_buffer = 0;
|
||||
}
|
||||
return *this;
|
||||
}
|
||||
|
||||
OpenGLState& OpenGLState::ResetVertexArray(GLuint handle) {
|
||||
if (draw.vertex_array == handle) {
|
||||
draw.vertex_array = 0;
|
||||
|
|
|
@ -154,8 +154,6 @@ public:
|
|||
GLuint read_framebuffer; // GL_READ_FRAMEBUFFER_BINDING
|
||||
GLuint draw_framebuffer; // GL_DRAW_FRAMEBUFFER_BINDING
|
||||
GLuint vertex_array; // GL_VERTEX_ARRAY_BINDING
|
||||
GLuint vertex_buffer; // GL_ARRAY_BUFFER_BINDING
|
||||
GLuint uniform_buffer; // GL_UNIFORM_BUFFER_BINDING
|
||||
GLuint shader_program; // GL_CURRENT_PROGRAM
|
||||
GLuint program_pipeline; // GL_PROGRAM_PIPELINE_BINDING
|
||||
} draw;
|
||||
|
@ -206,10 +204,10 @@ public:
|
|||
}
|
||||
/// Apply this state as the current OpenGL state
|
||||
void Apply() const;
|
||||
/// Apply only the state afecting the framebuffer
|
||||
/// Apply only the state affecting the framebuffer
|
||||
void ApplyFramebufferState() const;
|
||||
/// Apply only the state afecting the vertex buffer
|
||||
void ApplyVertexBufferState() const;
|
||||
/// Apply only the state affecting the vertex array
|
||||
void ApplyVertexArrayState() const;
|
||||
/// Set the initial OpenGL state
|
||||
static void ApplyDefaultState();
|
||||
/// Resets any references to the given resource
|
||||
|
@ -217,7 +215,6 @@ public:
|
|||
OpenGLState& ResetSampler(GLuint handle);
|
||||
OpenGLState& ResetProgram(GLuint handle);
|
||||
OpenGLState& ResetPipeline(GLuint handle);
|
||||
OpenGLState& ResetBuffer(GLuint handle);
|
||||
OpenGLState& ResetVertexArray(GLuint handle);
|
||||
OpenGLState& ResetFramebuffer(GLuint handle);
|
||||
void EmulateViewportWithScissor();
|
||||
|
|
|
@ -15,13 +15,12 @@ MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning",
|
|||
|
||||
namespace OpenGL {
|
||||
|
||||
OGLStreamBuffer::OGLStreamBuffer(GLenum target, GLsizeiptr size, bool prefer_coherent)
|
||||
: gl_target(target), buffer_size(size) {
|
||||
OGLStreamBuffer::OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent)
|
||||
: buffer_size(size) {
|
||||
gl_buffer.Create();
|
||||
glBindBuffer(gl_target, gl_buffer.handle);
|
||||
|
||||
GLsizeiptr allocate_size = size;
|
||||
if (target == GL_ARRAY_BUFFER) {
|
||||
if (vertex_data_usage) {
|
||||
// On AMD GPU there is a strange crash in indexed drawing. The crash happens when the buffer
|
||||
// read position is near the end and is an out-of-bound access to the vertex buffer. This is
|
||||
// probably a bug in the driver and is related to the usage of vec3<byte> attributes in the
|
||||
|
@ -35,18 +34,17 @@ OGLStreamBuffer::OGLStreamBuffer(GLenum target, GLsizeiptr size, bool prefer_coh
|
|||
coherent = prefer_coherent;
|
||||
const GLbitfield flags =
|
||||
GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | (coherent ? GL_MAP_COHERENT_BIT : 0);
|
||||
glBufferStorage(gl_target, allocate_size, nullptr, flags);
|
||||
mapped_ptr = static_cast<u8*>(glMapBufferRange(
|
||||
gl_target, 0, buffer_size, flags | (coherent ? 0 : GL_MAP_FLUSH_EXPLICIT_BIT)));
|
||||
glNamedBufferStorage(gl_buffer.handle, allocate_size, nullptr, flags);
|
||||
mapped_ptr = static_cast<u8*>(glMapNamedBufferRange(
|
||||
gl_buffer.handle, 0, buffer_size, flags | (coherent ? 0 : GL_MAP_FLUSH_EXPLICIT_BIT)));
|
||||
} else {
|
||||
glBufferData(gl_target, allocate_size, nullptr, GL_STREAM_DRAW);
|
||||
glNamedBufferData(gl_buffer.handle, allocate_size, nullptr, GL_STREAM_DRAW);
|
||||
}
|
||||
}
|
||||
|
||||
OGLStreamBuffer::~OGLStreamBuffer() {
|
||||
if (persistent) {
|
||||
glBindBuffer(gl_target, gl_buffer.handle);
|
||||
glUnmapBuffer(gl_target);
|
||||
glUnmapNamedBuffer(gl_buffer.handle);
|
||||
}
|
||||
gl_buffer.Release();
|
||||
}
|
||||
|
@ -74,7 +72,7 @@ std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr a
|
|||
invalidate = true;
|
||||
|
||||
if (persistent) {
|
||||
glUnmapBuffer(gl_target);
|
||||
glUnmapNamedBuffer(gl_buffer.handle);
|
||||
}
|
||||
}
|
||||
|
||||
|
@ -84,7 +82,7 @@ std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr a
|
|||
(coherent ? GL_MAP_COHERENT_BIT : GL_MAP_FLUSH_EXPLICIT_BIT) |
|
||||
(invalidate ? GL_MAP_INVALIDATE_BUFFER_BIT : GL_MAP_UNSYNCHRONIZED_BIT);
|
||||
mapped_ptr = static_cast<u8*>(
|
||||
glMapBufferRange(gl_target, buffer_pos, buffer_size - buffer_pos, flags));
|
||||
glMapNamedBufferRange(gl_buffer.handle, buffer_pos, buffer_size - buffer_pos, flags));
|
||||
mapped_offset = buffer_pos;
|
||||
}
|
||||
|
||||
|
@ -95,11 +93,11 @@ void OGLStreamBuffer::Unmap(GLsizeiptr size) {
|
|||
ASSERT(size <= mapped_size);
|
||||
|
||||
if (!coherent && size > 0) {
|
||||
glFlushMappedBufferRange(gl_target, buffer_pos - mapped_offset, size);
|
||||
glFlushMappedNamedBufferRange(gl_buffer.handle, buffer_pos - mapped_offset, size);
|
||||
}
|
||||
|
||||
if (!persistent) {
|
||||
glUnmapBuffer(gl_target);
|
||||
glUnmapNamedBuffer(gl_buffer.handle);
|
||||
}
|
||||
|
||||
buffer_pos += size;
|
||||
|
|
|
@ -13,7 +13,7 @@ namespace OpenGL {
|
|||
|
||||
class OGLStreamBuffer : private NonCopyable {
|
||||
public:
|
||||
explicit OGLStreamBuffer(GLenum target, GLsizeiptr size, bool prefer_coherent = false);
|
||||
explicit OGLStreamBuffer(GLsizeiptr size, bool vertex_data_usage, bool prefer_coherent = false);
|
||||
~OGLStreamBuffer();
|
||||
|
||||
GLuint GetHandle() const;
|
||||
|
@ -33,7 +33,6 @@ public:
|
|||
|
||||
private:
|
||||
OGLBuffer gl_buffer;
|
||||
GLenum gl_target;
|
||||
|
||||
bool coherent = false;
|
||||
bool persistent = false;
|
||||
|
|
|
@ -245,20 +245,20 @@ void RendererOpenGL::InitOpenGLObjects() {
|
|||
|
||||
// Generate VAO
|
||||
vertex_array.Create();
|
||||
|
||||
state.draw.vertex_array = vertex_array.handle;
|
||||
state.draw.vertex_buffer = vertex_buffer.handle;
|
||||
state.draw.uniform_buffer = 0;
|
||||
state.Apply();
|
||||
|
||||
// Attach vertex data to VAO
|
||||
glBufferData(GL_ARRAY_BUFFER, sizeof(ScreenRectVertex) * 4, nullptr, GL_STREAM_DRAW);
|
||||
glVertexAttribPointer(attrib_position, 2, GL_FLOAT, GL_FALSE, sizeof(ScreenRectVertex),
|
||||
(GLvoid*)offsetof(ScreenRectVertex, position));
|
||||
glVertexAttribPointer(attrib_tex_coord, 2, GL_FLOAT, GL_FALSE, sizeof(ScreenRectVertex),
|
||||
(GLvoid*)offsetof(ScreenRectVertex, tex_coord));
|
||||
glEnableVertexAttribArray(attrib_position);
|
||||
glEnableVertexAttribArray(attrib_tex_coord);
|
||||
glNamedBufferData(vertex_buffer.handle, sizeof(ScreenRectVertex) * 4, nullptr, GL_STREAM_DRAW);
|
||||
glVertexArrayAttribFormat(vertex_array.handle, attrib_position, 2, GL_FLOAT, GL_FALSE,
|
||||
offsetof(ScreenRectVertex, position));
|
||||
glVertexArrayAttribFormat(vertex_array.handle, attrib_tex_coord, 2, GL_FLOAT, GL_FALSE,
|
||||
offsetof(ScreenRectVertex, tex_coord));
|
||||
glVertexArrayAttribBinding(vertex_array.handle, attrib_position, 0);
|
||||
glVertexArrayAttribBinding(vertex_array.handle, attrib_tex_coord, 0);
|
||||
glEnableVertexArrayAttrib(vertex_array.handle, attrib_position);
|
||||
glEnableVertexArrayAttrib(vertex_array.handle, attrib_tex_coord);
|
||||
glVertexArrayVertexBuffer(vertex_array.handle, 0, vertex_buffer.handle, 0,
|
||||
sizeof(ScreenRectVertex));
|
||||
|
||||
// Allocate textures for the screen
|
||||
screen_info.texture.resource.Create();
|
||||
|
@ -370,14 +370,12 @@ void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x,
|
|||
state.texture_units[0].texture = screen_info.display_texture;
|
||||
state.texture_units[0].swizzle = {GL_RED, GL_GREEN, GL_BLUE, GL_ALPHA};
|
||||
// Workaround brigthness problems in SMO by enabling sRGB in the final output
|
||||
// if it has been used in the frame
|
||||
// Needed because of this bug in QT
|
||||
// QTBUG-50987
|
||||
// if it has been used in the frame. Needed because of this bug in QT: QTBUG-50987
|
||||
state.framebuffer_srgb.enabled = OpenGLState::GetsRGBUsed();
|
||||
state.Apply();
|
||||
glBufferSubData(GL_ARRAY_BUFFER, 0, sizeof(vertices), vertices.data());
|
||||
glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), vertices.data());
|
||||
glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
|
||||
// restore default state
|
||||
// Restore default state
|
||||
state.framebuffer_srgb.enabled = false;
|
||||
state.texture_units[0].texture = 0;
|
||||
state.Apply();
|
||||
|
|
Loading…
Reference in a new issue