pica/command_processor: build geometry pipeline and run geometry shader
The geometry pipeline manages data transfer between VS, GS and primitive assembler. It has known four modes:
- no GS mode: sends VS output directly to the primitive assembler (what citra currently does)
- GS mode 0: sends VS output to GS input registers, and sends GS output to primitive assembler
- GS mode 1: sends VS output to GS uniform registers, and sends GS output to primitive assembler. It also takes an index from the index buffer at the beginning of each primitive for determine the primitive size.
- GS mode 2: similar to mode 1, but doesn't take the index and uses a fixed primitive size.
hwtest shows that immediate mode also supports GS (at least for mode 0), so the geometry pipeline gets refactored into its own class for supporting both drawing mode.
In the immediate mode, some games don't set the pipeline registers to a valid value until the first attribute input, so a geometry pipeline reset flag is set in `pipeline.vs_default_attributes_setup.index` trigger, and the actual pipeline reconfigure is triggered in the first attribute input.
In the normal drawing mode with index buffer, the vertex cache is a little bit modified to support the geometry pipeline. Instead of OutputVertex, it now holds AttributeBuffer, which is the input to the geometry pipeline. The AttributeBuffer->OutputVertex conversion is done inside the pipeline vertex handler. The actual hardware vertex cache is believed to be implemented in a similar way (because this is the only way that makes sense).
Both geometry pipeline and GS unit rely on states preservation across drawing call, so they are put into the global state. In the future, the other three vertex shader units should be also placed in the global state, and a scheduler should be implemented on top of the four units. Note that the current gs_unit already allows running VS on it in the future.
2017-08-04 16:03:17 +02:00
|
|
|
// Copyright 2017 Citra Emulator Project
|
|
|
|
// Licensed under GPLv2 or any later version
|
|
|
|
// Refer to the license.txt file included.
|
|
|
|
|
|
|
|
#include "video_core/geometry_pipeline.h"
|
|
|
|
#include "video_core/pica_state.h"
|
|
|
|
#include "video_core/regs.h"
|
|
|
|
#include "video_core/renderer_base.h"
|
|
|
|
#include "video_core/video_core.h"
|
|
|
|
|
|
|
|
namespace Pica {
|
|
|
|
|
|
|
|
/// An attribute buffering interface for different pipeline modes
|
|
|
|
class GeometryPipelineBackend {
|
|
|
|
public:
|
|
|
|
virtual ~GeometryPipelineBackend() = default;
|
|
|
|
|
|
|
|
/// Checks if there is no incomplete data transfer
|
|
|
|
virtual bool IsEmpty() const = 0;
|
|
|
|
|
|
|
|
/// Checks if the pipeline needs a direct input from index buffer
|
|
|
|
virtual bool NeedIndexInput() const = 0;
|
|
|
|
|
|
|
|
/// Submits an index from index buffer
|
|
|
|
virtual void SubmitIndex(unsigned int val) = 0;
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Submits vertex attributes
|
|
|
|
* @param input attributes of a vertex output from vertex shader
|
|
|
|
* @return if the buffer is full and the geometry shader should be invoked
|
|
|
|
*/
|
|
|
|
virtual bool SubmitVertex(const Shader::AttributeBuffer& input) = 0;
|
|
|
|
};
|
|
|
|
|
|
|
|
// In the Point mode, vertex attributes are sent to the input registers in the geometry shader unit.
|
|
|
|
// The size of vertex shader outputs and geometry shader inputs are constants. Geometry shader is
|
|
|
|
// invoked upon inputs buffer filled up by vertex shader outputs. For example, if we have a geometry
|
|
|
|
// shader that takes 6 inputs, and the vertex shader outputs 2 attributes, it would take 3 vertices
|
|
|
|
// for one geometry shader invocation.
|
|
|
|
// TODO: what happens when the input size is not divisible by the output size?
|
|
|
|
class GeometryPipeline_Point : public GeometryPipelineBackend {
|
|
|
|
public:
|
|
|
|
GeometryPipeline_Point(const Regs& regs, Shader::GSUnitState& unit) : regs(regs), unit(unit) {
|
|
|
|
ASSERT(regs.pipeline.variable_primitive == 0);
|
|
|
|
ASSERT(regs.gs.input_to_uniform == 0);
|
|
|
|
vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
|
2018-09-06 22:03:28 +02:00
|
|
|
std::size_t gs_input_num = regs.gs.max_input_attribute_index + 1;
|
pica/command_processor: build geometry pipeline and run geometry shader
The geometry pipeline manages data transfer between VS, GS and primitive assembler. It has known four modes:
- no GS mode: sends VS output directly to the primitive assembler (what citra currently does)
- GS mode 0: sends VS output to GS input registers, and sends GS output to primitive assembler
- GS mode 1: sends VS output to GS uniform registers, and sends GS output to primitive assembler. It also takes an index from the index buffer at the beginning of each primitive for determine the primitive size.
- GS mode 2: similar to mode 1, but doesn't take the index and uses a fixed primitive size.
hwtest shows that immediate mode also supports GS (at least for mode 0), so the geometry pipeline gets refactored into its own class for supporting both drawing mode.
In the immediate mode, some games don't set the pipeline registers to a valid value until the first attribute input, so a geometry pipeline reset flag is set in `pipeline.vs_default_attributes_setup.index` trigger, and the actual pipeline reconfigure is triggered in the first attribute input.
In the normal drawing mode with index buffer, the vertex cache is a little bit modified to support the geometry pipeline. Instead of OutputVertex, it now holds AttributeBuffer, which is the input to the geometry pipeline. The AttributeBuffer->OutputVertex conversion is done inside the pipeline vertex handler. The actual hardware vertex cache is believed to be implemented in a similar way (because this is the only way that makes sense).
Both geometry pipeline and GS unit rely on states preservation across drawing call, so they are put into the global state. In the future, the other three vertex shader units should be also placed in the global state, and a scheduler should be implemented on top of the four units. Note that the current gs_unit already allows running VS on it in the future.
2017-08-04 16:03:17 +02:00
|
|
|
ASSERT(gs_input_num % vs_output_num == 0);
|
|
|
|
buffer_cur = attribute_buffer.attr;
|
|
|
|
buffer_end = attribute_buffer.attr + gs_input_num;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool IsEmpty() const override {
|
|
|
|
return buffer_cur == attribute_buffer.attr;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool NeedIndexInput() const override {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
void SubmitIndex(unsigned int val) override {
|
|
|
|
UNREACHABLE();
|
|
|
|
}
|
|
|
|
|
|
|
|
bool SubmitVertex(const Shader::AttributeBuffer& input) override {
|
|
|
|
buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
|
|
|
|
if (buffer_cur == buffer_end) {
|
|
|
|
buffer_cur = attribute_buffer.attr;
|
|
|
|
unit.LoadInput(regs.gs, attribute_buffer);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
private:
|
|
|
|
const Regs& regs;
|
|
|
|
Shader::GSUnitState& unit;
|
|
|
|
Shader::AttributeBuffer attribute_buffer;
|
|
|
|
Math::Vec4<float24>* buffer_cur;
|
|
|
|
Math::Vec4<float24>* buffer_end;
|
|
|
|
unsigned int vs_output_num;
|
|
|
|
};
|
|
|
|
|
|
|
|
// In VariablePrimitive mode, vertex attributes are buffered into the uniform registers in the
|
|
|
|
// geometry shader unit. The number of vertex is variable, which is specified by the first index
|
|
|
|
// value in the batch. This mode is usually used for subdivision.
|
|
|
|
class GeometryPipeline_VariablePrimitive : public GeometryPipelineBackend {
|
|
|
|
public:
|
|
|
|
GeometryPipeline_VariablePrimitive(const Regs& regs, Shader::ShaderSetup& setup)
|
|
|
|
: regs(regs), setup(setup) {
|
|
|
|
ASSERT(regs.pipeline.variable_primitive == 1);
|
|
|
|
ASSERT(regs.gs.input_to_uniform == 1);
|
|
|
|
vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool IsEmpty() const override {
|
|
|
|
return need_index;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool NeedIndexInput() const override {
|
|
|
|
return need_index;
|
|
|
|
}
|
|
|
|
|
|
|
|
void SubmitIndex(unsigned int val) override {
|
|
|
|
DEBUG_ASSERT(need_index);
|
|
|
|
|
|
|
|
// The number of vertex input is put to the uniform register
|
2017-09-27 01:26:09 +02:00
|
|
|
float24 vertex_num = float24::FromFloat32(static_cast<float>(val));
|
pica/command_processor: build geometry pipeline and run geometry shader
The geometry pipeline manages data transfer between VS, GS and primitive assembler. It has known four modes:
- no GS mode: sends VS output directly to the primitive assembler (what citra currently does)
- GS mode 0: sends VS output to GS input registers, and sends GS output to primitive assembler
- GS mode 1: sends VS output to GS uniform registers, and sends GS output to primitive assembler. It also takes an index from the index buffer at the beginning of each primitive for determine the primitive size.
- GS mode 2: similar to mode 1, but doesn't take the index and uses a fixed primitive size.
hwtest shows that immediate mode also supports GS (at least for mode 0), so the geometry pipeline gets refactored into its own class for supporting both drawing mode.
In the immediate mode, some games don't set the pipeline registers to a valid value until the first attribute input, so a geometry pipeline reset flag is set in `pipeline.vs_default_attributes_setup.index` trigger, and the actual pipeline reconfigure is triggered in the first attribute input.
In the normal drawing mode with index buffer, the vertex cache is a little bit modified to support the geometry pipeline. Instead of OutputVertex, it now holds AttributeBuffer, which is the input to the geometry pipeline. The AttributeBuffer->OutputVertex conversion is done inside the pipeline vertex handler. The actual hardware vertex cache is believed to be implemented in a similar way (because this is the only way that makes sense).
Both geometry pipeline and GS unit rely on states preservation across drawing call, so they are put into the global state. In the future, the other three vertex shader units should be also placed in the global state, and a scheduler should be implemented on top of the four units. Note that the current gs_unit already allows running VS on it in the future.
2017-08-04 16:03:17 +02:00
|
|
|
setup.uniforms.f[0] = Math::MakeVec(vertex_num, vertex_num, vertex_num, vertex_num);
|
|
|
|
|
|
|
|
// The second uniform register and so on are used for receiving input vertices
|
|
|
|
buffer_cur = setup.uniforms.f + 1;
|
|
|
|
|
|
|
|
main_vertex_num = regs.pipeline.variable_vertex_main_num_minus_1 + 1;
|
|
|
|
total_vertex_num = val;
|
|
|
|
need_index = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool SubmitVertex(const Shader::AttributeBuffer& input) override {
|
|
|
|
DEBUG_ASSERT(!need_index);
|
|
|
|
if (main_vertex_num != 0) {
|
|
|
|
// For main vertices, receive all attributes
|
|
|
|
buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
|
|
|
|
--main_vertex_num;
|
|
|
|
} else {
|
|
|
|
// For other vertices, only receive the first attribute (usually the position)
|
|
|
|
*(buffer_cur++) = input.attr[0];
|
|
|
|
}
|
|
|
|
--total_vertex_num;
|
|
|
|
|
|
|
|
if (total_vertex_num == 0) {
|
|
|
|
need_index = true;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
private:
|
|
|
|
bool need_index = true;
|
|
|
|
const Regs& regs;
|
|
|
|
Shader::ShaderSetup& setup;
|
|
|
|
unsigned int main_vertex_num;
|
|
|
|
unsigned int total_vertex_num;
|
|
|
|
Math::Vec4<float24>* buffer_cur;
|
|
|
|
unsigned int vs_output_num;
|
|
|
|
};
|
|
|
|
|
|
|
|
// In FixedPrimitive mode, vertex attributes are buffered into the uniform registers in the geometry
|
|
|
|
// shader unit. The number of vertex per shader invocation is constant. This is usually used for
|
|
|
|
// particle system.
|
|
|
|
class GeometryPipeline_FixedPrimitive : public GeometryPipelineBackend {
|
|
|
|
public:
|
|
|
|
GeometryPipeline_FixedPrimitive(const Regs& regs, Shader::ShaderSetup& setup)
|
|
|
|
: regs(regs), setup(setup) {
|
|
|
|
ASSERT(regs.pipeline.variable_primitive == 0);
|
|
|
|
ASSERT(regs.gs.input_to_uniform == 1);
|
|
|
|
vs_output_num = regs.pipeline.vs_outmap_total_minus_1_a + 1;
|
|
|
|
ASSERT(vs_output_num == regs.pipeline.gs_config.stride_minus_1 + 1);
|
2018-09-06 22:03:28 +02:00
|
|
|
std::size_t vertex_num = regs.pipeline.gs_config.fixed_vertex_num_minus_1 + 1;
|
pica/command_processor: build geometry pipeline and run geometry shader
The geometry pipeline manages data transfer between VS, GS and primitive assembler. It has known four modes:
- no GS mode: sends VS output directly to the primitive assembler (what citra currently does)
- GS mode 0: sends VS output to GS input registers, and sends GS output to primitive assembler
- GS mode 1: sends VS output to GS uniform registers, and sends GS output to primitive assembler. It also takes an index from the index buffer at the beginning of each primitive for determine the primitive size.
- GS mode 2: similar to mode 1, but doesn't take the index and uses a fixed primitive size.
hwtest shows that immediate mode also supports GS (at least for mode 0), so the geometry pipeline gets refactored into its own class for supporting both drawing mode.
In the immediate mode, some games don't set the pipeline registers to a valid value until the first attribute input, so a geometry pipeline reset flag is set in `pipeline.vs_default_attributes_setup.index` trigger, and the actual pipeline reconfigure is triggered in the first attribute input.
In the normal drawing mode with index buffer, the vertex cache is a little bit modified to support the geometry pipeline. Instead of OutputVertex, it now holds AttributeBuffer, which is the input to the geometry pipeline. The AttributeBuffer->OutputVertex conversion is done inside the pipeline vertex handler. The actual hardware vertex cache is believed to be implemented in a similar way (because this is the only way that makes sense).
Both geometry pipeline and GS unit rely on states preservation across drawing call, so they are put into the global state. In the future, the other three vertex shader units should be also placed in the global state, and a scheduler should be implemented on top of the four units. Note that the current gs_unit already allows running VS on it in the future.
2017-08-04 16:03:17 +02:00
|
|
|
buffer_cur = buffer_begin = setup.uniforms.f + regs.pipeline.gs_config.start_index;
|
|
|
|
buffer_end = buffer_begin + vs_output_num * vertex_num;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool IsEmpty() const override {
|
|
|
|
return buffer_cur == buffer_begin;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool NeedIndexInput() const override {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
void SubmitIndex(unsigned int val) override {
|
|
|
|
UNREACHABLE();
|
|
|
|
}
|
|
|
|
|
|
|
|
bool SubmitVertex(const Shader::AttributeBuffer& input) override {
|
|
|
|
buffer_cur = std::copy(input.attr, input.attr + vs_output_num, buffer_cur);
|
|
|
|
if (buffer_cur == buffer_end) {
|
|
|
|
buffer_cur = buffer_begin;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
private:
|
|
|
|
const Regs& regs;
|
|
|
|
Shader::ShaderSetup& setup;
|
|
|
|
Math::Vec4<float24>* buffer_begin;
|
|
|
|
Math::Vec4<float24>* buffer_cur;
|
|
|
|
Math::Vec4<float24>* buffer_end;
|
|
|
|
unsigned int vs_output_num;
|
|
|
|
};
|
|
|
|
|
|
|
|
GeometryPipeline::GeometryPipeline(State& state) : state(state) {}
|
|
|
|
|
|
|
|
GeometryPipeline::~GeometryPipeline() = default;
|
|
|
|
|
|
|
|
void GeometryPipeline::SetVertexHandler(Shader::VertexHandler vertex_handler) {
|
|
|
|
this->vertex_handler = vertex_handler;
|
|
|
|
}
|
|
|
|
|
|
|
|
void GeometryPipeline::Setup(Shader::ShaderEngine* shader_engine) {
|
|
|
|
if (!backend)
|
|
|
|
return;
|
|
|
|
|
|
|
|
this->shader_engine = shader_engine;
|
|
|
|
shader_engine->SetupBatch(state.gs, state.regs.gs.main_offset);
|
|
|
|
}
|
|
|
|
|
|
|
|
void GeometryPipeline::Reconfigure() {
|
|
|
|
ASSERT(!backend || backend->IsEmpty());
|
|
|
|
|
|
|
|
if (state.regs.pipeline.use_gs == PipelineRegs::UseGS::No) {
|
|
|
|
backend = nullptr;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
ASSERT(state.regs.pipeline.use_gs == PipelineRegs::UseGS::Yes);
|
|
|
|
|
|
|
|
// The following assumes that when geometry shader is in use, the shader unit 3 is configured as
|
|
|
|
// a geometry shader unit.
|
|
|
|
// TODO: what happens if this is not true?
|
|
|
|
ASSERT(state.regs.pipeline.gs_unit_exclusive_configuration == 1);
|
|
|
|
ASSERT(state.regs.gs.shader_mode == ShaderRegs::ShaderMode::GS);
|
|
|
|
|
|
|
|
state.gs_unit.ConfigOutput(state.regs.gs);
|
|
|
|
|
|
|
|
ASSERT(state.regs.pipeline.vs_outmap_total_minus_1_a ==
|
|
|
|
state.regs.pipeline.vs_outmap_total_minus_1_b);
|
|
|
|
|
|
|
|
switch (state.regs.pipeline.gs_config.mode) {
|
|
|
|
case PipelineRegs::GSMode::Point:
|
|
|
|
backend = std::make_unique<GeometryPipeline_Point>(state.regs, state.gs_unit);
|
|
|
|
break;
|
|
|
|
case PipelineRegs::GSMode::VariablePrimitive:
|
|
|
|
backend = std::make_unique<GeometryPipeline_VariablePrimitive>(state.regs, state.gs);
|
|
|
|
break;
|
|
|
|
case PipelineRegs::GSMode::FixedPrimitive:
|
|
|
|
backend = std::make_unique<GeometryPipeline_FixedPrimitive>(state.regs, state.gs);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
UNREACHABLE();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
bool GeometryPipeline::NeedIndexInput() const {
|
|
|
|
if (!backend)
|
|
|
|
return false;
|
|
|
|
return backend->NeedIndexInput();
|
|
|
|
}
|
|
|
|
|
|
|
|
void GeometryPipeline::SubmitIndex(unsigned int val) {
|
|
|
|
backend->SubmitIndex(val);
|
|
|
|
}
|
|
|
|
|
|
|
|
void GeometryPipeline::SubmitVertex(const Shader::AttributeBuffer& input) {
|
|
|
|
if (!backend) {
|
|
|
|
// No backend means the geometry shader is disabled, so we send the vertex shader output
|
|
|
|
// directly to the primitive assembler.
|
|
|
|
vertex_handler(input);
|
|
|
|
} else {
|
|
|
|
if (backend->SubmitVertex(input)) {
|
|
|
|
shader_engine->Run(state.gs, state.gs_unit);
|
|
|
|
|
|
|
|
// The uniform b15 is set to true after every geometry shader invocation. This is useful
|
|
|
|
// for the shader to know if this is the first invocation in a batch, if the program set
|
|
|
|
// b15 to false first.
|
|
|
|
state.gs.uniforms.b[15] = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace Pica
|