Merge pull request #2735 from FernandoS27/pipeline-rework

Rework Dirty Flags in GPU Pipeline, Optimize CBData and Redo Clearing mechanism
merge-requests/60/head
bunnei 2019-07-21 00:59:52 +07:00 committed by GitHub
commit 27e10e0442
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
14 changed files with 528 additions and 116 deletions

@ -22,7 +22,7 @@ void DmaPusher::DispatchCalls() {
MICROPROFILE_SCOPE(DispatchCalls);
// On entering GPU code, assume all memory may be touched by the ARM core.
gpu.Maxwell3D().dirty_flags.OnMemoryWrite();
gpu.Maxwell3D().dirty.OnMemoryWrite();
dma_pushbuffer_subindex = 0;

@ -37,7 +37,7 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
const bool is_last_call = method_call.IsLastCall();
upload_state.ProcessData(method_call.argument, is_last_call);
if (is_last_call) {
system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
system.GPU().Maxwell3D().dirty.OnMemoryWrite();
}
break;
}

@ -34,7 +34,7 @@ void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
const bool is_last_call = method_call.IsLastCall();
upload_state.ProcessData(method_call.argument, is_last_call);
if (is_last_call) {
system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
system.GPU().Maxwell3D().dirty.OnMemoryWrite();
}
break;
}

@ -22,6 +22,7 @@ Maxwell3D::Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& raste
MemoryManager& memory_manager)
: system{system}, rasterizer{rasterizer}, memory_manager{memory_manager},
macro_interpreter{*this}, upload_state{memory_manager, regs.upload} {
InitDirtySettings();
InitializeRegisterDefaults();
}
@ -69,6 +70,10 @@ void Maxwell3D::InitializeRegisterDefaults() {
regs.stencil_back_func_mask = 0xFFFFFFFF;
regs.stencil_back_mask = 0xFFFFFFFF;
regs.depth_test_func = Regs::ComparisonOp::Always;
regs.cull.front_face = Regs::Cull::FrontFace::CounterClockWise;
regs.cull.cull_face = Regs::Cull::CullFace::Back;
// TODO(Rodrigo): Most games do not set a point size. I think this is a case of a
// register carrying a default value. Assume it's OpenGL's default (1).
regs.point_size = 1.0f;
@ -86,6 +91,159 @@ void Maxwell3D::InitializeRegisterDefaults() {
regs.rt_separate_frag_data = 1;
}
#define DIRTY_REGS_POS(field_name) (offsetof(Maxwell3D::DirtyRegs, field_name))
void Maxwell3D::InitDirtySettings() {
const auto set_block = [this](const u32 start, const u32 range, const u8 position) {
const auto start_itr = dirty_pointers.begin() + start;
const auto end_itr = start_itr + range;
std::fill(start_itr, end_itr, position);
};
dirty.regs.fill(true);
// Init Render Targets
constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
constexpr u32 rt_start_reg = MAXWELL3D_REG_INDEX(rt);
constexpr u32 rt_end_reg = rt_start_reg + registers_per_rt * 8;
u32 rt_dirty_reg = DIRTY_REGS_POS(render_target);
for (u32 rt_reg = rt_start_reg; rt_reg < rt_end_reg; rt_reg += registers_per_rt) {
set_block(rt_reg, registers_per_rt, rt_dirty_reg);
rt_dirty_reg++;
}
constexpr u32 depth_buffer_flag = DIRTY_REGS_POS(depth_buffer);
dirty_pointers[MAXWELL3D_REG_INDEX(zeta_enable)] = depth_buffer_flag;
dirty_pointers[MAXWELL3D_REG_INDEX(zeta_width)] = depth_buffer_flag;
dirty_pointers[MAXWELL3D_REG_INDEX(zeta_height)] = depth_buffer_flag;
constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
constexpr u32 zeta_reg = MAXWELL3D_REG_INDEX(zeta);
set_block(zeta_reg, registers_in_zeta, depth_buffer_flag);
// Init Vertex Arrays
constexpr u32 vertex_array_start = MAXWELL3D_REG_INDEX(vertex_array);
constexpr u32 vertex_array_size = sizeof(regs.vertex_array[0]) / sizeof(u32);
constexpr u32 vertex_array_end = vertex_array_start + vertex_array_size * Regs::NumVertexArrays;
u32 va_reg = DIRTY_REGS_POS(vertex_array);
u32 vi_reg = DIRTY_REGS_POS(vertex_instance);
for (u32 vertex_reg = vertex_array_start; vertex_reg < vertex_array_end;
vertex_reg += vertex_array_size) {
set_block(vertex_reg, 3, va_reg);
// The divisor concerns vertex array instances
dirty_pointers[vertex_reg + 3] = vi_reg;
va_reg++;
vi_reg++;
}
constexpr u32 vertex_limit_start = MAXWELL3D_REG_INDEX(vertex_array_limit);
constexpr u32 vertex_limit_size = sizeof(regs.vertex_array_limit[0]) / sizeof(u32);
constexpr u32 vertex_limit_end = vertex_limit_start + vertex_limit_size * Regs::NumVertexArrays;
va_reg = DIRTY_REGS_POS(vertex_array);
for (u32 vertex_reg = vertex_limit_start; vertex_reg < vertex_limit_end;
vertex_reg += vertex_limit_size) {
set_block(vertex_reg, vertex_limit_size, va_reg);
va_reg++;
}
constexpr u32 vertex_instance_start = MAXWELL3D_REG_INDEX(instanced_arrays);
constexpr u32 vertex_instance_size =
sizeof(regs.instanced_arrays.is_instanced[0]) / sizeof(u32);
constexpr u32 vertex_instance_end =
vertex_instance_start + vertex_instance_size * Regs::NumVertexArrays;
vi_reg = DIRTY_REGS_POS(vertex_instance);
for (u32 vertex_reg = vertex_instance_start; vertex_reg < vertex_instance_end;
vertex_reg += vertex_instance_size) {
set_block(vertex_reg, vertex_instance_size, vi_reg);
vi_reg++;
}
set_block(MAXWELL3D_REG_INDEX(vertex_attrib_format), regs.vertex_attrib_format.size(),
DIRTY_REGS_POS(vertex_attrib_format));
// Init Shaders
constexpr u32 shader_registers_count =
sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);
set_block(MAXWELL3D_REG_INDEX(shader_config[0]), shader_registers_count,
DIRTY_REGS_POS(shaders));
// State
// Viewport
constexpr u32 viewport_dirty_reg = DIRTY_REGS_POS(viewport);
constexpr u32 viewport_start = MAXWELL3D_REG_INDEX(viewports);
constexpr u32 viewport_size = sizeof(regs.viewports) / sizeof(u32);
set_block(viewport_start, viewport_size, viewport_dirty_reg);
constexpr u32 view_volume_start = MAXWELL3D_REG_INDEX(view_volume_clip_control);
constexpr u32 view_volume_size = sizeof(regs.view_volume_clip_control) / sizeof(u32);
set_block(view_volume_start, view_volume_size, viewport_dirty_reg);
// Viewport transformation
constexpr u32 viewport_trans_start = MAXWELL3D_REG_INDEX(viewport_transform);
constexpr u32 viewport_trans_size = sizeof(regs.viewport_transform) / sizeof(u32);
set_block(viewport_trans_start, viewport_trans_size, DIRTY_REGS_POS(viewport_transform));
// Cullmode
constexpr u32 cull_mode_start = MAXWELL3D_REG_INDEX(cull);
constexpr u32 cull_mode_size = sizeof(regs.cull) / sizeof(u32);
set_block(cull_mode_start, cull_mode_size, DIRTY_REGS_POS(cull_mode));
// Screen y control
dirty_pointers[MAXWELL3D_REG_INDEX(screen_y_control)] = DIRTY_REGS_POS(screen_y_control);
// Primitive Restart
constexpr u32 primitive_restart_start = MAXWELL3D_REG_INDEX(primitive_restart);
constexpr u32 primitive_restart_size = sizeof(regs.primitive_restart) / sizeof(u32);
set_block(primitive_restart_start, primitive_restart_size, DIRTY_REGS_POS(primitive_restart));
// Depth Test
constexpr u32 depth_test_dirty_reg = DIRTY_REGS_POS(depth_test);
dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_enable)] = depth_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(depth_write_enabled)] = depth_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(depth_test_func)] = depth_test_dirty_reg;
// Stencil Test
constexpr u32 stencil_test_dirty_reg = DIRTY_REGS_POS(stencil_test);
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_enable)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_func)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_ref)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_func_mask)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_fail)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_zfail)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_op_zpass)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_front_mask)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_two_side_enable)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_func)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_ref)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_func_mask)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_fail)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_zfail)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_op_zpass)] = stencil_test_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(stencil_back_mask)] = stencil_test_dirty_reg;
// Color Mask
constexpr u32 color_mask_dirty_reg = DIRTY_REGS_POS(color_mask);
dirty_pointers[MAXWELL3D_REG_INDEX(color_mask_common)] = color_mask_dirty_reg;
set_block(MAXWELL3D_REG_INDEX(color_mask), sizeof(regs.color_mask) / sizeof(u32),
color_mask_dirty_reg);
// Blend State
constexpr u32 blend_state_dirty_reg = DIRTY_REGS_POS(blend_state);
set_block(MAXWELL3D_REG_INDEX(blend_color), sizeof(regs.blend_color) / sizeof(u32),
blend_state_dirty_reg);
dirty_pointers[MAXWELL3D_REG_INDEX(independent_blend_enable)] = blend_state_dirty_reg;
set_block(MAXWELL3D_REG_INDEX(blend), sizeof(regs.blend) / sizeof(u32), blend_state_dirty_reg);
set_block(MAXWELL3D_REG_INDEX(independent_blend), sizeof(regs.independent_blend) / sizeof(u32),
blend_state_dirty_reg);
// Scissor State
constexpr u32 scissor_test_dirty_reg = DIRTY_REGS_POS(scissor_test);
set_block(MAXWELL3D_REG_INDEX(scissor_test), sizeof(regs.scissor_test) / sizeof(u32),
scissor_test_dirty_reg);
// Polygon Offset
constexpr u32 polygon_offset_dirty_reg = DIRTY_REGS_POS(polygon_offset);
dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_fill_enable)] = polygon_offset_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_line_enable)] = polygon_offset_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_point_enable)] = polygon_offset_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_units)] = polygon_offset_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_factor)] = polygon_offset_dirty_reg;
dirty_pointers[MAXWELL3D_REG_INDEX(polygon_offset_clamp)] = polygon_offset_dirty_reg;
}
void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
// Reset the current macro.
executing_macro = 0;
@ -108,6 +266,14 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
const u32 method = method_call.method;
if (method == cb_data_state.current) {
regs.reg_array[method] = method_call.argument;
ProcessCBData(method_call.argument);
return;
} else if (cb_data_state.current != null_cb_data) {
FinishCBData();
}
// It is an error to write to a register other than the current macro's ARG register before it
// has finished execution.
if (executing_macro != 0) {
@ -143,49 +309,19 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
if (regs.reg_array[method] != method_call.argument) {
regs.reg_array[method] = method_call.argument;
// Color buffers
constexpr u32 first_rt_reg = MAXWELL3D_REG_INDEX(rt);
constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
if (method >= first_rt_reg &&
method < first_rt_reg + registers_per_rt * Regs::NumRenderTargets) {
const std::size_t rt_index = (method - first_rt_reg) / registers_per_rt;
dirty_flags.color_buffer.set(rt_index);
}
// Zeta buffer
constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
if (method == MAXWELL3D_REG_INDEX(zeta_enable) ||
method == MAXWELL3D_REG_INDEX(zeta_width) ||
method == MAXWELL3D_REG_INDEX(zeta_height) ||
(method >= MAXWELL3D_REG_INDEX(zeta) &&
method < MAXWELL3D_REG_INDEX(zeta) + registers_in_zeta)) {
dirty_flags.zeta_buffer = true;
}
// Shader
constexpr u32 shader_registers_count =
sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);
if (method >= MAXWELL3D_REG_INDEX(shader_config[0]) &&
method < MAXWELL3D_REG_INDEX(shader_config[0]) + shader_registers_count) {
dirty_flags.shaders = true;
}
// Vertex format
if (method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) &&
method < MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) {
dirty_flags.vertex_attrib_format = true;
}
// Vertex buffer
if (method >= MAXWELL3D_REG_INDEX(vertex_array) &&
method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * Regs::NumVertexArrays) {
dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2);
} else if (method >= MAXWELL3D_REG_INDEX(vertex_array_limit) &&
method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * Regs::NumVertexArrays) {
dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1);
} else if (method >= MAXWELL3D_REG_INDEX(instanced_arrays) &&
method < MAXWELL3D_REG_INDEX(instanced_arrays) + Regs::NumVertexArrays) {
dirty_flags.vertex_array.set(method - MAXWELL3D_REG_INDEX(instanced_arrays));
const std::size_t dirty_reg = dirty_pointers[method];
if (dirty_reg) {
dirty.regs[dirty_reg] = true;
if (dirty_reg >= DIRTY_REGS_POS(vertex_array) &&
dirty_reg < DIRTY_REGS_POS(vertex_array_buffers)) {
dirty.vertex_array_buffers = true;
} else if (dirty_reg >= DIRTY_REGS_POS(vertex_instance) &&
dirty_reg < DIRTY_REGS_POS(vertex_instances)) {
dirty.vertex_instances = true;
} else if (dirty_reg >= DIRTY_REGS_POS(render_target) &&
dirty_reg < DIRTY_REGS_POS(render_settings)) {
dirty.render_settings = true;
}
}
}
@ -214,7 +350,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): {
ProcessCBData(method_call.argument);
StartCBData(method);
break;
}
case MAXWELL3D_REG_INDEX(cb_bind[0].raw_config): {
@ -261,7 +397,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
const bool is_last_call = method_call.IsLastCall();
upload_state.ProcessData(method_call.argument, is_last_call);
if (is_last_call) {
dirty_flags.OnMemoryWrite();
dirty.OnMemoryWrite();
}
break;
}
@ -333,7 +469,6 @@ void Maxwell3D::ProcessQueryGet() {
query_result.timestamp = system.CoreTiming().GetTicks();
memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result));
}
dirty_flags.OnMemoryWrite();
break;
}
default:
@ -405,23 +540,39 @@ void Maxwell3D::ProcessCBBind(Regs::ShaderStage stage) {
}
void Maxwell3D::ProcessCBData(u32 value) {
const u32 id = cb_data_state.id;
cb_data_state.buffer[id][cb_data_state.counter] = value;
// Increment the current buffer position.
regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
cb_data_state.counter++;
}
void Maxwell3D::StartCBData(u32 method) {
constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]);
cb_data_state.start_pos = regs.const_buffer.cb_pos;
cb_data_state.id = method - first_cb_data;
cb_data_state.current = method;
cb_data_state.counter = 0;
ProcessCBData(regs.const_buffer.cb_data[cb_data_state.id]);
}
void Maxwell3D::FinishCBData() {
// Write the input value to the current const buffer at the current position.
const GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
ASSERT(buffer_address != 0);
// Don't allow writing past the end of the buffer.
ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size);
ASSERT(regs.const_buffer.cb_pos <= regs.const_buffer.cb_size);
const GPUVAddr address{buffer_address + regs.const_buffer.cb_pos};
const GPUVAddr address{buffer_address + cb_data_state.start_pos};
const std::size_t size = regs.const_buffer.cb_pos - cb_data_state.start_pos;
u8* ptr{memory_manager.GetPointer(address)};
rasterizer.InvalidateRegion(ToCacheAddr(ptr), sizeof(u32));
memory_manager.Write<u32>(address, value);
const u32 id = cb_data_state.id;
memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size);
dirty.OnMemoryWrite();
dirty_flags.OnMemoryWrite();
// Increment the current buffer position.
regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
cb_data_state.id = null_cb_data;
cb_data_state.current = null_cb_data;
}
Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {

@ -1124,23 +1124,77 @@ public:
State state{};
struct DirtyFlags {
std::bitset<8> color_buffer{0xFF};
std::bitset<32> vertex_array{0xFFFFFFFF};
struct DirtyRegs {
static constexpr std::size_t NUM_REGS = 256;
union {
struct {
bool null_dirty;
bool vertex_attrib_format = true;
bool zeta_buffer = true;
bool shaders = true;
// Vertex Attributes
bool vertex_attrib_format;
// Vertex Arrays
std::array<bool, 32> vertex_array;
bool vertex_array_buffers;
// Vertex Instances
std::array<bool, 32> vertex_instance;
bool vertex_instances;
// Render Targets
std::array<bool, 8> render_target;
bool depth_buffer;
bool render_settings;
// Shaders
bool shaders;
// Rasterizer State
bool viewport;
bool clip_coefficient;
bool cull_mode;
bool primitive_restart;
bool depth_test;
bool stencil_test;
bool blend_state;
bool scissor_test;
bool transform_feedback;
bool color_mask;
bool polygon_offset;
// Complementary
bool viewport_transform;
bool screen_y_control;
bool memory_general;
};
std::array<bool, NUM_REGS> regs;
};
void ResetVertexArrays() {
vertex_array.fill(true);
vertex_array_buffers = true;
}
void ResetRenderTargets() {
depth_buffer = true;
render_target.fill(true);
render_settings = true;
}
void OnMemoryWrite() {
zeta_buffer = true;
shaders = true;
color_buffer.set();
vertex_array.set();
memory_general = true;
ResetRenderTargets();
ResetVertexArrays();
}
};
DirtyFlags dirty_flags;
} dirty{};
std::array<u8, Regs::NUM_REGS> dirty_pointers{};
/// Reads a register value located at the input method address
u32 GetRegisterValue(u32 method) const;
@ -1192,6 +1246,15 @@ private:
/// Interpreter for the macro codes uploaded to the GPU.
MacroInterpreter macro_interpreter;
static constexpr u32 null_cb_data = 0xFFFFFFFF;
struct {
std::array<std::array<u32, 0x4000>, 16> buffer;
u32 current{null_cb_data};
u32 id{null_cb_data};
u32 start_pos{};
u32 counter{};
} cb_data_state;
Upload::State upload_state;
/// Retrieves information about a specific TIC entry from the TIC buffer.
@ -1200,6 +1263,8 @@ private:
/// Retrieves information about a specific TSC entry from the TSC buffer.
Texture::TSCEntry GetTSCEntry(u32 tsc_index) const;
void InitDirtySettings();
/**
* Call a macro on this engine.
* @param method Method to call
@ -1223,7 +1288,9 @@ private:
void ProcessSyncPoint();
/// Handles a write to the CB_DATA[i] register.
void StartCBData(u32 method);
void ProcessCBData(u32 value);
void FinishCBData();
/// Handles a write to the CB_BIND register.
void ProcessCBBind(Regs::ShaderStage stage);

@ -58,7 +58,7 @@ void MaxwellDMA::HandleCopy() {
}
// All copies here update the main memory, so mark all rasterizer states as invalid.
system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
system.GPU().Maxwell3D().dirty.OnMemoryWrite();
if (regs.exec.is_dst_linear && regs.exec.is_src_linear) {
// When the enable_2d bit is disabled, the copy is performed as if we were copying a 1D

@ -105,6 +105,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind
shader_program_manager = std::make_unique<GLShader::ProgramManager>();
state.draw.shader_program = 0;
state.Apply();
clear_framebuffer.Create();
LOG_DEBUG(Render_OpenGL, "Sync fixed function OpenGL state here");
CheckExtensions();
@ -124,10 +125,10 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
auto& gpu = system.GPU().Maxwell3D();
const auto& regs = gpu.regs;
if (!gpu.dirty_flags.vertex_attrib_format) {
if (!gpu.dirty.vertex_attrib_format) {
return state.draw.vertex_array;
}
gpu.dirty_flags.vertex_attrib_format = false;
gpu.dirty.vertex_attrib_format = false;
MICROPROFILE_SCOPE(OpenGL_VAO);
@ -181,7 +182,7 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
}
// Rebinding the VAO invalidates the vertex buffer bindings.
gpu.dirty_flags.vertex_array.set();
gpu.dirty.ResetVertexArrays();
state.draw.vertex_array = vao_entry.handle;
return vao_entry.handle;
@ -189,17 +190,20 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
auto& gpu = system.GPU().Maxwell3D();
const auto& regs = gpu.regs;
if (gpu.dirty_flags.vertex_array.none())
if (!gpu.dirty.vertex_array_buffers)
return;
gpu.dirty.vertex_array_buffers = false;
const auto& regs = gpu.regs;
MICROPROFILE_SCOPE(OpenGL_VB);
// Upload all guest vertex arrays sequentially to our buffer
for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
if (!gpu.dirty_flags.vertex_array[index])
if (!gpu.dirty.vertex_array[index])
continue;
gpu.dirty.vertex_array[index] = false;
gpu.dirty.vertex_instance[index] = false;
const auto& vertex_array = regs.vertex_array[index];
if (!vertex_array.IsEnabled())
@ -224,8 +228,32 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
glVertexArrayBindingDivisor(vao, index, 0);
}
}
}
gpu.dirty_flags.vertex_array.reset();
void RasterizerOpenGL::SetupVertexInstances(GLuint vao) {
auto& gpu = system.GPU().Maxwell3D();
if (!gpu.dirty.vertex_instances)
return;
gpu.dirty.vertex_instances = false;
const auto& regs = gpu.regs;
// Upload all guest vertex arrays sequentially to our buffer
for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
if (!gpu.dirty.vertex_instance[index])
continue;
gpu.dirty.vertex_instance[index] = false;
if (regs.instanced_arrays.IsInstancingEnabled(index) &&
regs.vertex_array[index].divisor != 0) {
// Enable vertex buffer instancing with the specified divisor.
glVertexArrayBindingDivisor(vao, index, regs.vertex_array[index].divisor);
} else {
// Disable the vertex buffer instancing.
glVertexArrayBindingDivisor(vao, index, 0);
}
}
}
GLintptr RasterizerOpenGL::SetupIndexBuffer() {
@ -341,7 +369,7 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
SyncClipEnabled(clip_distances);
gpu.dirty_flags.shaders = false;
gpu.dirty.shaders = false;
}
std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
@ -424,13 +452,13 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
const FramebufferConfigState fb_config_state{using_color_fb, using_depth_fb, preserve_contents,
single_color_target};
if (fb_config_state == current_framebuffer_config_state &&
gpu.dirty_flags.color_buffer.none() && !gpu.dirty_flags.zeta_buffer) {
if (fb_config_state == current_framebuffer_config_state && !gpu.dirty.render_settings) {
// Only skip if the previous ConfigureFramebuffers call was from the same kind (multiple or
// single color targets). This is done because the guest registers may not change but the
// host framebuffer may contain different attachments
return current_depth_stencil_usage;
}
gpu.dirty.render_settings = false;
current_framebuffer_config_state = fb_config_state;
texture_cache.GuardRenderTargets(true);
@ -519,13 +547,65 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
return current_depth_stencil_usage = {static_cast<bool>(depth_surface), fbkey.stencil_enable};
}
void RasterizerOpenGL::ConfigureClearFramebuffer(OpenGLState& current_state, bool using_color_fb,
bool using_depth_fb, bool using_stencil_fb) {
auto& gpu = system.GPU().Maxwell3D();
const auto& regs = gpu.regs;
texture_cache.GuardRenderTargets(true);
View color_surface{};
if (using_color_fb) {
color_surface = texture_cache.GetColorBufferSurface(regs.clear_buffers.RT, false);
}
View depth_surface{};
if (using_depth_fb || using_stencil_fb) {
depth_surface = texture_cache.GetDepthBufferSurface(false);
}
texture_cache.GuardRenderTargets(false);
current_state.draw.draw_framebuffer = clear_framebuffer.handle;
current_state.ApplyFramebufferState();
if (color_surface) {
color_surface->Attach(GL_COLOR_ATTACHMENT0, GL_DRAW_FRAMEBUFFER);
} else {
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
}
if (depth_surface) {
const auto& params = depth_surface->GetSurfaceParams();
switch (params.type) {
case VideoCore::Surface::SurfaceType::Depth: {
depth_surface->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
break;
}
case VideoCore::Surface::SurfaceType::DepthStencil: {
depth_surface->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
break;
}
default: { UNIMPLEMENTED(); }
}
} else {
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
0);
}
}
void RasterizerOpenGL::Clear() {
const auto& regs = system.GPU().Maxwell3D().regs;
bool use_color{};
bool use_depth{};
bool use_stencil{};
OpenGLState clear_state;
OpenGLState prev_state{OpenGLState::GetCurState()};
SCOPE_EXIT({
prev_state.AllDirty();
prev_state.Apply();
});
OpenGLState clear_state{OpenGLState::GetCurState()};
clear_state.SetDefaultViewports();
if (regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B ||
regs.clear_buffers.A) {
use_color = true;
@ -545,6 +625,7 @@ void RasterizerOpenGL::Clear() {
// true.
clear_state.depth.test_enabled = true;
clear_state.depth.test_func = GL_ALWAYS;
clear_state.depth.write_mask = GL_TRUE;
}
if (regs.clear_buffers.S) {
ASSERT_MSG(regs.zeta_enable != 0, "Tried to clear stencil but buffer is not enabled!");
@ -581,8 +662,9 @@ void RasterizerOpenGL::Clear() {
return;
}
const auto [clear_depth, clear_stencil] = ConfigureFramebuffers(
clear_state, use_color, use_depth || use_stencil, false, regs.clear_buffers.RT.Value());
ConfigureClearFramebuffer(clear_state, use_color, use_depth, use_stencil);
SyncViewport(clear_state);
if (regs.clear_flags.scissor) {
SyncScissorTest(clear_state);
}
@ -591,21 +673,18 @@ void RasterizerOpenGL::Clear() {
clear_state.EmulateViewportWithScissor();
}
clear_state.ApplyColorMask();
clear_state.ApplyDepth();
clear_state.ApplyStencilTest();
clear_state.ApplyViewport();
clear_state.ApplyFramebufferState();
clear_state.AllDirty();
clear_state.Apply();
if (use_color) {
glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color);
glClearBufferfv(GL_COLOR, 0, regs.clear_color);
}
if (clear_depth && clear_stencil) {
if (use_depth && use_stencil) {
glClearBufferfi(GL_DEPTH_STENCIL, 0, regs.clear_depth, regs.clear_stencil);
} else if (clear_depth) {
} else if (use_depth) {
glClearBufferfv(GL_DEPTH, 0, &regs.clear_depth);
} else if (clear_stencil) {
} else if (use_stencil) {
glClearBufferiv(GL_STENCIL, 0, &regs.clear_stencil);
}
}
@ -661,6 +740,7 @@ void RasterizerOpenGL::DrawArrays() {
// Upload vertex and index data.
SetupVertexBuffer(vao);
SetupVertexInstances(vao);
const GLintptr index_buffer_offset = SetupIndexBuffer();
// Setup draw parameters. It will automatically choose what glDraw* method to use.
@ -687,7 +767,7 @@ void RasterizerOpenGL::DrawArrays() {
if (invalidate) {
// As all cached buffers are invalidated, we need to recheck their state.
gpu.dirty_flags.vertex_array.set();
gpu.dirty.ResetVertexArrays();
}
shader_program_manager->ApplyTo(state);
@ -700,6 +780,7 @@ void RasterizerOpenGL::DrawArrays() {
params.DispatchDraw();
accelerate_draw = AccelDraw::Disabled;
gpu.dirty.memory_general = false;
}
void RasterizerOpenGL::FlushAll() {}
@ -907,10 +988,11 @@ void RasterizerOpenGL::SyncClipCoef() {
}
void RasterizerOpenGL::SyncCullMode() {
const auto& regs = system.GPU().Maxwell3D().regs;
auto& maxwell3d = system.GPU().Maxwell3D();
const auto& regs = maxwell3d.regs;
state.cull.enabled = regs.cull.enabled != 0;
if (state.cull.enabled) {
state.cull.front_face = MaxwellToGL::FrontFace(regs.cull.front_face);
state.cull.mode = MaxwellToGL::CullFace(regs.cull.cull_face);
@ -943,16 +1025,21 @@ void RasterizerOpenGL::SyncDepthTestState() {
state.depth.test_enabled = regs.depth_test_enable != 0;
state.depth.write_mask = regs.depth_write_enabled ? GL_TRUE : GL_FALSE;
if (!state.depth.test_enabled)
if (!state.depth.test_enabled) {
return;
}
state.depth.test_func = MaxwellToGL::ComparisonOp(regs.depth_test_func);
}
void RasterizerOpenGL::SyncStencilTestState() {
const auto& regs = system.GPU().Maxwell3D().regs;
state.stencil.test_enabled = regs.stencil_enable != 0;
auto& maxwell3d = system.GPU().Maxwell3D();
if (!maxwell3d.dirty.stencil_test) {
return;
}
const auto& regs = maxwell3d.regs;
state.stencil.test_enabled = regs.stencil_enable != 0;
if (!regs.stencil_enable) {
return;
}
@ -981,10 +1068,17 @@ void RasterizerOpenGL::SyncStencilTestState() {
state.stencil.back.action_depth_fail = GL_KEEP;
state.stencil.back.action_depth_pass = GL_KEEP;
}
state.MarkDirtyStencilState();
maxwell3d.dirty.stencil_test = false;
}
void RasterizerOpenGL::SyncColorMask() {
const auto& regs = system.GPU().Maxwell3D().regs;
auto& maxwell3d = system.GPU().Maxwell3D();
if (!maxwell3d.dirty.color_mask) {
return;
}
const auto& regs = maxwell3d.regs;
const std::size_t count =
regs.independent_blend_enable ? Tegra::Engines::Maxwell3D::Regs::NumRenderTargets : 1;
for (std::size_t i = 0; i < count; i++) {
@ -995,6 +1089,9 @@ void RasterizerOpenGL::SyncColorMask() {
dest.blue_enabled = (source.B == 0) ? GL_FALSE : GL_TRUE;
dest.alpha_enabled = (source.A == 0) ? GL_FALSE : GL_TRUE;
}
state.MarkDirtyColorMask();
maxwell3d.dirty.color_mask = false;
}
void RasterizerOpenGL::SyncMultiSampleState() {
@ -1009,7 +1106,11 @@ void RasterizerOpenGL::SyncFragmentColorClampState() {
}
void RasterizerOpenGL::SyncBlendState() {
const auto& regs = system.GPU().Maxwell3D().regs;
auto& maxwell3d = system.GPU().Maxwell3D();
if (!maxwell3d.dirty.blend_state) {
return;
}
const auto& regs = maxwell3d.regs;
state.blend_color.red = regs.blend_color.r;
state.blend_color.green = regs.blend_color.g;
@ -1032,6 +1133,8 @@ void RasterizerOpenGL::SyncBlendState() {
for (std::size_t i = 1; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; i++) {
state.blend[i].enabled = false;
}
maxwell3d.dirty.blend_state = false;
state.MarkDirtyBlendState();
return;
}
@ -1048,6 +1151,9 @@ void RasterizerOpenGL::SyncBlendState() {
blend.src_a_func = MaxwellToGL::BlendFunc(src.factor_source_a);
blend.dst_a_func = MaxwellToGL::BlendFunc(src.factor_dest_a);
}
state.MarkDirtyBlendState();
maxwell3d.dirty.blend_state = false;
}
void RasterizerOpenGL::SyncLogicOpState() {
@ -1099,13 +1205,21 @@ void RasterizerOpenGL::SyncPointState() {
}
void RasterizerOpenGL::SyncPolygonOffset() {
const auto& regs = system.GPU().Maxwell3D().regs;
auto& maxwell3d = system.GPU().Maxwell3D();
if (!maxwell3d.dirty.polygon_offset) {
return;
}
const auto& regs = maxwell3d.regs;
state.polygon_offset.fill_enable = regs.polygon_offset_fill_enable != 0;
state.polygon_offset.line_enable = regs.polygon_offset_line_enable != 0;
state.polygon_offset.point_enable = regs.polygon_offset_point_enable != 0;
state.polygon_offset.units = regs.polygon_offset_units;
state.polygon_offset.factor = regs.polygon_offset_factor;
state.polygon_offset.clamp = regs.polygon_offset_clamp;
state.MarkDirtyPolygonOffset();
maxwell3d.dirty.polygon_offset = false;
}
void RasterizerOpenGL::SyncAlphaTest() {

@ -108,6 +108,9 @@ private:
OpenGLState& current_state, bool using_color_fb = true, bool using_depth_fb = true,
bool preserve_contents = true, std::optional<std::size_t> single_color_target = {});
void ConfigureClearFramebuffer(OpenGLState& current_state, bool using_color_fb,
bool using_depth_fb, bool using_stencil_fb);
/// Configures the current constbuffers to use for the draw command.
void SetupDrawConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
const Shader& shader);
@ -216,6 +219,7 @@ private:
GLuint SetupVertexFormat();
void SetupVertexBuffer(GLuint vao);
void SetupVertexInstances(GLuint vao);
GLintptr SetupIndexBuffer();
@ -226,6 +230,8 @@ private:
enum class AccelDraw { Disabled, Arrays, Indexed };
AccelDraw accelerate_draw = AccelDraw::Disabled;
OGLFramebuffer clear_framebuffer;
using CachedPageMap = boost::icl::interval_map<u64, int>;
CachedPageMap cached_pages;
};

@ -572,7 +572,7 @@ std::unordered_map<u64, UnspecializedShader> ShaderCacheOpenGL::GenerateUnspecia
}
Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
if (!system.GPU().Maxwell3D().dirty_flags.shaders) {
if (!system.GPU().Maxwell3D().dirty.shaders) {
return last_shaders[static_cast<std::size_t>(program)];
}

@ -165,6 +165,25 @@ OpenGLState::OpenGLState() {
alpha_test.ref = 0.0f;
}
void OpenGLState::SetDefaultViewports() {
for (auto& item : viewports) {
item.x = 0;
item.y = 0;
item.width = 0;
item.height = 0;
item.depth_range_near = 0.0f;
item.depth_range_far = 1.0f;
item.scissor.enabled = false;
item.scissor.x = 0;
item.scissor.y = 0;
item.scissor.width = 0;
item.scissor.height = 0;
}
depth_clamp.far_plane = false;
depth_clamp.near_plane = false;
}
void OpenGLState::ApplyDefaultState() {
glEnable(GL_BLEND);
glDisable(GL_FRAMEBUFFER_SRGB);
@ -526,7 +545,7 @@ void OpenGLState::ApplySamplers() const {
}
}
void OpenGLState::Apply() const {
void OpenGLState::Apply() {
MICROPROFILE_SCOPE(OpenGL_State);
ApplyFramebufferState();
ApplyVertexArrayState();
@ -536,19 +555,31 @@ void OpenGLState::Apply() const {
ApplyPointSize();
ApplyFragmentColorClamp();
ApplyMultisample();
if (dirty.color_mask) {
ApplyColorMask();
dirty.color_mask = false;
}
ApplyDepthClamp();
ApplyColorMask();
ApplyViewport();
ApplyStencilTest();
if (dirty.stencil_state) {
ApplyStencilTest();
dirty.stencil_state = false;
}
ApplySRgb();
ApplyCulling();
ApplyDepth();
ApplyPrimitiveRestart();
ApplyBlending();
if (dirty.blend_state) {
ApplyBlending();
dirty.blend_state = false;
}
ApplyLogicOp();
ApplyTextures();
ApplySamplers();
ApplyPolygonOffset();
if (dirty.polygon_offset) {
ApplyPolygonOffset();
dirty.polygon_offset = false;
}
ApplyAlphaTest();
}

@ -195,8 +195,9 @@ public:
s_rgb_used = false;
}
void SetDefaultViewports();
/// Apply this state as the current OpenGL state
void Apply() const;
void Apply();
void ApplyFramebufferState() const;
void ApplyVertexArrayState() const;
@ -237,11 +238,41 @@ public:
/// Viewport does not affects glClearBuffer so emulate viewport using scissor test
void EmulateViewportWithScissor();
void MarkDirtyBlendState() {
dirty.blend_state = true;
}
void MarkDirtyStencilState() {
dirty.stencil_state = true;
}
void MarkDirtyPolygonOffset() {
dirty.polygon_offset = true;
}
void MarkDirtyColorMask() {
dirty.color_mask = true;
}
void AllDirty() {
dirty.blend_state = true;
dirty.stencil_state = true;
dirty.polygon_offset = true;
dirty.color_mask = true;
}
private:
static OpenGLState cur_state;
// Workaround for sRGB problems caused by QT not supporting srgb output
static bool s_rgb_used;
struct {
bool blend_state;
bool stencil_state;
bool viewport_state;
bool polygon_offset;
bool color_mask;
} dirty{};
};
} // namespace OpenGL

@ -485,11 +485,15 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
const auto& dst_params{dst_view->GetSurfaceParams()};
OpenGLState prev_state{OpenGLState::GetCurState()};
SCOPE_EXIT({ prev_state.Apply(); });
SCOPE_EXIT({
prev_state.AllDirty();
prev_state.Apply();
});
OpenGLState state;
state.draw.read_framebuffer = src_framebuffer.handle;
state.draw.draw_framebuffer = dst_framebuffer.handle;
state.AllDirty();
state.Apply();
u32 buffers{};

@ -108,6 +108,7 @@ void RendererOpenGL::SwapBuffers(
// Maintain the rasterizer's state as a priority
OpenGLState prev_state = OpenGLState::GetCurState();
state.AllDirty();
state.Apply();
if (framebuffer) {
@ -140,6 +141,7 @@ void RendererOpenGL::SwapBuffers(
system.GetPerfStats().BeginSystemFrame();
// Restore the rasterizer state
prev_state.AllDirty();
prev_state.Apply();
}
@ -206,6 +208,7 @@ void RendererOpenGL::InitOpenGLObjects() {
// Link shaders and get variable locations
shader.CreateFromSource(vertex_shader, nullptr, fragment_shader);
state.draw.shader_program = shader.handle;
state.AllDirty();
state.Apply();
uniform_modelview_matrix = glGetUniformLocation(shader.handle, "modelview_matrix");
uniform_color_texture = glGetUniformLocation(shader.handle, "color_texture");
@ -338,12 +341,14 @@ void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x,
// 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
state.framebuffer_srgb.enabled = OpenGLState::GetsRGBUsed();
state.AllDirty();
state.Apply();
glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), vertices.data());
glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
// Restore default state
state.framebuffer_srgb.enabled = false;
state.texture_units[0].texture = 0;
state.AllDirty();
state.Apply();
// Clear sRGB state for the next frame
OpenGLState::ClearsRGBUsed();
@ -388,6 +393,7 @@ void RendererOpenGL::CaptureScreenshot() {
GLuint old_read_fb = state.draw.read_framebuffer;
GLuint old_draw_fb = state.draw.draw_framebuffer;
state.draw.read_framebuffer = state.draw.draw_framebuffer = screenshot_framebuffer.handle;
state.AllDirty();
state.Apply();
Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout};
@ -407,6 +413,7 @@ void RendererOpenGL::CaptureScreenshot() {
screenshot_framebuffer.Release();
state.draw.read_framebuffer = old_read_fb;
state.draw.draw_framebuffer = old_draw_fb;
state.AllDirty();
state.Apply();
glDeleteRenderbuffers(1, &renderbuffer);

@ -116,10 +116,10 @@ public:
std::lock_guard lock{mutex};
auto& maxwell3d = system.GPU().Maxwell3D();
if (!maxwell3d.dirty_flags.zeta_buffer) {
if (!maxwell3d.dirty.depth_buffer) {
return depth_buffer.view;
}
maxwell3d.dirty_flags.zeta_buffer = false;
maxwell3d.dirty.depth_buffer = false;
const auto& regs{maxwell3d.regs};
const auto gpu_addr{regs.zeta.Address()};
@ -145,10 +145,10 @@ public:
std::lock_guard lock{mutex};
ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
auto& maxwell3d = system.GPU().Maxwell3D();
if (!maxwell3d.dirty_flags.color_buffer[index]) {
if (!maxwell3d.dirty.render_target[index]) {
return render_targets[index].view;
}
maxwell3d.dirty_flags.color_buffer.reset(index);
maxwell3d.dirty.render_target[index] = false;
const auto& regs{maxwell3d.regs};
if (index >= regs.rt_control.count || regs.rt[index].Address() == 0 ||
@ -274,10 +274,11 @@ protected:
auto& maxwell3d = system.GPU().Maxwell3D();
const u32 index = surface->GetRenderTarget();
if (index == DEPTH_RT) {
maxwell3d.dirty_flags.zeta_buffer = true;
maxwell3d.dirty.depth_buffer = true;
} else {
maxwell3d.dirty_flags.color_buffer.set(index, true);
maxwell3d.dirty.render_target[index] = true;
}
maxwell3d.dirty.render_settings = true;
}
void Register(TSurface surface) {