Revert "Extracted the attribute setup and draw commands into their own functions"

This reverts commit b3b34a1e76. This
commit causes a performance regression for not enough benefits
This commit is contained in:
James Rowe 2017-11-16 11:46:17 -07:00
parent 947067de81
commit 9d9693c13d

View file

@ -119,224 +119,6 @@ static void WriteUniformFloatReg(ShaderRegs& config, Shader::ShaderSetup& setup,
}
}
static void LoadDefaultVertexAttributes(u32 register_value) {
auto& regs = g_state.regs;
// TODO: Does actual hardware indeed keep an intermediate buffer or does
// it directly write the values?
default_attr_write_buffer[default_attr_counter++] = register_value;
// Default attributes are written in a packed format such that four float24 values are encoded
// in three 32-bit numbers.
// We write to internal memory once a full such vector is written.
if (default_attr_counter >= 3) {
default_attr_counter = 0;
auto& setup = regs.pipeline.vs_default_attributes_setup;
if (setup.index >= 16) {
LOG_ERROR(HW_GPU, "Invalid VS default attribute index %d", (int)setup.index);
return;
}
Math::Vec4<float24> attribute;
// NOTE: The destination component order indeed is "backwards"
attribute.w = float24::FromRaw(default_attr_write_buffer[0] >> 8);
attribute.z = float24::FromRaw(((default_attr_write_buffer[0] & 0xFF) << 16) |
((default_attr_write_buffer[1] >> 16) & 0xFFFF));
attribute.y = float24::FromRaw(((default_attr_write_buffer[1] & 0xFFFF) << 8) |
((default_attr_write_buffer[2] >> 24) & 0xFF));
attribute.x = float24::FromRaw(default_attr_write_buffer[2] & 0xFFFFFF);
LOG_TRACE(HW_GPU, "Set default VS attribute %x to (%f %f %f %f)", (int)setup.index,
attribute.x.ToFloat32(), attribute.y.ToFloat32(), attribute.z.ToFloat32(),
attribute.w.ToFloat32());
// TODO: Verify that this actually modifies the register!
if (setup.index < 15) {
g_state.input_default_attributes.attr[setup.index] = attribute;
setup.index++;
} else {
// Put each attribute into an immediate input buffer. When all specified immediate
// attributes are present, the Vertex Shader is invoked and everything is sent to
// the primitive assembler.
auto& immediate_input = g_state.immediate.input_vertex;
auto& immediate_attribute_id = g_state.immediate.current_attribute;
immediate_input.attr[immediate_attribute_id] = attribute;
if (immediate_attribute_id < regs.pipeline.max_input_attrib_index) {
immediate_attribute_id += 1;
} else {
MICROPROFILE_SCOPE(GPU_Drawing);
immediate_attribute_id = 0;
auto* shader_engine = Shader::GetEngine();
shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
// Send to vertex shader
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
static_cast<void*>(&immediate_input));
Shader::UnitState shader_unit;
Shader::AttributeBuffer output{};
shader_unit.LoadInput(regs.vs, immediate_input);
shader_engine->Run(g_state.vs, shader_unit);
shader_unit.WriteOutput(regs.vs, output);
// Send to geometry pipeline
if (g_state.immediate.reset_geometry_pipeline) {
g_state.geometry_pipeline.Reconfigure();
g_state.immediate.reset_geometry_pipeline = false;
}
ASSERT(!g_state.geometry_pipeline.NeedIndexInput());
g_state.geometry_pipeline.Setup(shader_engine);
g_state.geometry_pipeline.SubmitVertex(output);
// TODO: If drawing after every immediate mode triangle kills performance,
// change it to flush triangles whenever a drawing config register changes
// See: https://github.com/citra-emu/citra/pull/2866#issuecomment-327011550
VideoCore::g_renderer->Rasterizer()->DrawTriangles();
if (g_debug_context) {
g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr);
}
}
}
}
}
static void Draw(u32 command_id) {
MICROPROFILE_SCOPE(GPU_Drawing);
auto& regs = g_state.regs;
#if PICA_LOG_TEV
DebugUtils::DumpTevStageConfig(regs.GetTevStages());
#endif
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::IncomingPrimitiveBatch, nullptr);
// Processes information about internal vertex attributes to figure out how a vertex is
// loaded.
// Later, these can be compiled and cached.
const u32 base_address = regs.pipeline.vertex_attributes.GetPhysicalBaseAddress();
VertexLoader loader(regs.pipeline);
// Load vertices
bool is_indexed = (command_id == PICA_REG_INDEX(pipeline.trigger_draw_indexed));
const auto& index_info = regs.pipeline.index_array;
const u8* index_address_8 = Memory::GetPhysicalPointer(base_address + index_info.offset);
const u16* index_address_16 = reinterpret_cast<const u16*>(index_address_8);
bool index_u16 = index_info.format != 0;
PrimitiveAssembler<Shader::OutputVertex>& primitive_assembler = g_state.primitive_assembler;
if (g_debug_context && g_debug_context->recorder) {
for (int i = 0; i < 3; ++i) {
const auto texture = regs.texturing.GetTextures()[i];
if (!texture.enabled)
continue;
u8* texture_data = Memory::GetPhysicalPointer(texture.config.GetPhysicalAddress());
g_debug_context->recorder->MemoryAccessed(
texture_data, Pica::TexturingRegs::NibblesPerPixel(texture.format) *
texture.config.width / 2 * texture.config.height,
texture.config.GetPhysicalAddress());
}
}
DebugUtils::MemoryAccessTracker memory_accesses;
// Simple circular-replacement vertex cache
// The size has been tuned for optimal balance between hit-rate and the cost of lookup
const size_t VERTEX_CACHE_SIZE = 32;
std::array<u16, VERTEX_CACHE_SIZE> vertex_cache_ids;
std::array<Shader::AttributeBuffer, VERTEX_CACHE_SIZE> vertex_cache;
Shader::AttributeBuffer vs_output;
unsigned int vertex_cache_pos = 0;
vertex_cache_ids.fill(-1);
auto* shader_engine = Shader::GetEngine();
Shader::UnitState shader_unit;
shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
g_state.geometry_pipeline.Reconfigure();
g_state.geometry_pipeline.Setup(shader_engine);
if (g_state.geometry_pipeline.NeedIndexInput())
ASSERT(is_indexed);
for (unsigned int index = 0; index < regs.pipeline.num_vertices; ++index) {
// Indexed rendering doesn't use the start offset
unsigned int vertex = is_indexed
? (index_u16 ? index_address_16[index] : index_address_8[index])
: (index + regs.pipeline.vertex_offset);
// -1 is a common special value used for primitive restart. Since it's unknown if
// the PICA supports it, and it would mess up the caching, guard against it here.
ASSERT(vertex != -1);
bool vertex_cache_hit = false;
if (is_indexed) {
if (g_state.geometry_pipeline.NeedIndexInput()) {
g_state.geometry_pipeline.SubmitIndex(vertex);
continue;
}
if (g_debug_context && Pica::g_debug_context->recorder) {
int size = index_u16 ? 2 : 1;
memory_accesses.AddAccess(base_address + index_info.offset + size * index, size);
}
for (unsigned int i = 0; i < VERTEX_CACHE_SIZE; ++i) {
if (vertex == vertex_cache_ids[i]) {
vs_output = vertex_cache[i];
vertex_cache_hit = true;
break;
}
}
}
if (!vertex_cache_hit) {
// Initialize data for the current vertex
Shader::AttributeBuffer input;
loader.LoadVertex(base_address, index, vertex, input, memory_accesses);
// Send to vertex shader
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
(void*)&input);
shader_unit.LoadInput(regs.vs, input);
shader_engine->Run(g_state.vs, shader_unit);
shader_unit.WriteOutput(regs.vs, vs_output);
if (is_indexed) {
vertex_cache[vertex_cache_pos] = vs_output;
vertex_cache_ids[vertex_cache_pos] = vertex;
vertex_cache_pos = (vertex_cache_pos + 1) % VERTEX_CACHE_SIZE;
}
}
// Send to geometry pipeline
g_state.geometry_pipeline.SubmitVertex(vs_output);
}
for (auto& range : memory_accesses.ranges) {
g_debug_context->recorder->MemoryAccessed(Memory::GetPhysicalPointer(range.first),
range.second, range.first);
}
VideoCore::g_renderer->Rasterizer()->DrawTriangles();
if (g_debug_context) {
g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr);
}
}
static void WritePicaReg(u32 id, u32 value, u32 mask) {
auto& regs = g_state.regs;
@ -386,9 +168,95 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
// Load default vertex input attributes
case PICA_REG_INDEX_WORKAROUND(pipeline.vs_default_attributes_setup.set_value[0], 0x233):
case PICA_REG_INDEX_WORKAROUND(pipeline.vs_default_attributes_setup.set_value[1], 0x234):
case PICA_REG_INDEX_WORKAROUND(pipeline.vs_default_attributes_setup.set_value[2], 0x235):
LoadDefaultVertexAttributes(value);
case PICA_REG_INDEX_WORKAROUND(pipeline.vs_default_attributes_setup.set_value[2], 0x235): {
// TODO: Does actual hardware indeed keep an intermediate buffer or does
// it directly write the values?
default_attr_write_buffer[default_attr_counter++] = value;
// Default attributes are written in a packed format such that four float24 values are
// encoded in
// three 32-bit numbers. We write to internal memory once a full such vector is
// written.
if (default_attr_counter >= 3) {
default_attr_counter = 0;
auto& setup = regs.pipeline.vs_default_attributes_setup;
if (setup.index >= 16) {
LOG_ERROR(HW_GPU, "Invalid VS default attribute index %d", (int)setup.index);
break;
}
Math::Vec4<float24> attribute;
// NOTE: The destination component order indeed is "backwards"
attribute.w = float24::FromRaw(default_attr_write_buffer[0] >> 8);
attribute.z = float24::FromRaw(((default_attr_write_buffer[0] & 0xFF) << 16) |
((default_attr_write_buffer[1] >> 16) & 0xFFFF));
attribute.y = float24::FromRaw(((default_attr_write_buffer[1] & 0xFFFF) << 8) |
((default_attr_write_buffer[2] >> 24) & 0xFF));
attribute.x = float24::FromRaw(default_attr_write_buffer[2] & 0xFFFFFF);
LOG_TRACE(HW_GPU, "Set default VS attribute %x to (%f %f %f %f)", (int)setup.index,
attribute.x.ToFloat32(), attribute.y.ToFloat32(), attribute.z.ToFloat32(),
attribute.w.ToFloat32());
// TODO: Verify that this actually modifies the register!
if (setup.index < 15) {
g_state.input_default_attributes.attr[setup.index] = attribute;
setup.index++;
} else {
// Put each attribute into an immediate input buffer. When all specified immediate
// attributes are present, the Vertex Shader is invoked and everything is sent to
// the primitive assembler.
auto& immediate_input = g_state.immediate.input_vertex;
auto& immediate_attribute_id = g_state.immediate.current_attribute;
immediate_input.attr[immediate_attribute_id] = attribute;
if (immediate_attribute_id < regs.pipeline.max_input_attrib_index) {
immediate_attribute_id += 1;
} else {
MICROPROFILE_SCOPE(GPU_Drawing);
immediate_attribute_id = 0;
auto* shader_engine = Shader::GetEngine();
shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
// Send to vertex shader
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
static_cast<void*>(&immediate_input));
Shader::UnitState shader_unit;
Shader::AttributeBuffer output{};
shader_unit.LoadInput(regs.vs, immediate_input);
shader_engine->Run(g_state.vs, shader_unit);
shader_unit.WriteOutput(regs.vs, output);
// Send to geometry pipeline
if (g_state.immediate.reset_geometry_pipeline) {
g_state.geometry_pipeline.Reconfigure();
g_state.immediate.reset_geometry_pipeline = false;
}
ASSERT(!g_state.geometry_pipeline.NeedIndexInput());
g_state.geometry_pipeline.Setup(shader_engine);
g_state.geometry_pipeline.SubmitVertex(output);
// TODO: If drawing after every immediate mode triangle kills performance,
// change it to flush triangles whenever a drawing config register changes
// See: https://github.com/citra-emu/citra/pull/2866#issuecomment-327011550
VideoCore::g_renderer->Rasterizer()->DrawTriangles();
if (g_debug_context) {
g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch,
nullptr);
}
}
}
}
break;
}
case PICA_REG_INDEX(pipeline.gpu_mode):
// This register likely just enables vertex processing and doesn't need any special handling
@ -407,9 +275,136 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
// It seems like these trigger vertex rendering
case PICA_REG_INDEX(pipeline.trigger_draw):
case PICA_REG_INDEX(pipeline.trigger_draw_indexed):
Draw(id);
case PICA_REG_INDEX(pipeline.trigger_draw_indexed): {
MICROPROFILE_SCOPE(GPU_Drawing);
#if PICA_LOG_TEV
DebugUtils::DumpTevStageConfig(regs.GetTevStages());
#endif
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::IncomingPrimitiveBatch, nullptr);
// Processes information about internal vertex attributes to figure out how a vertex is
// loaded.
// Later, these can be compiled and cached.
const u32 base_address = regs.pipeline.vertex_attributes.GetPhysicalBaseAddress();
VertexLoader loader(regs.pipeline);
// Load vertices
bool is_indexed = (id == PICA_REG_INDEX(pipeline.trigger_draw_indexed));
const auto& index_info = regs.pipeline.index_array;
const u8* index_address_8 = Memory::GetPhysicalPointer(base_address + index_info.offset);
const u16* index_address_16 = reinterpret_cast<const u16*>(index_address_8);
bool index_u16 = index_info.format != 0;
PrimitiveAssembler<Shader::OutputVertex>& primitive_assembler = g_state.primitive_assembler;
if (g_debug_context && g_debug_context->recorder) {
for (int i = 0; i < 3; ++i) {
const auto texture = regs.texturing.GetTextures()[i];
if (!texture.enabled)
continue;
u8* texture_data = Memory::GetPhysicalPointer(texture.config.GetPhysicalAddress());
g_debug_context->recorder->MemoryAccessed(
texture_data, Pica::TexturingRegs::NibblesPerPixel(texture.format) *
texture.config.width / 2 * texture.config.height,
texture.config.GetPhysicalAddress());
}
}
DebugUtils::MemoryAccessTracker memory_accesses;
// Simple circular-replacement vertex cache
// The size has been tuned for optimal balance between hit-rate and the cost of lookup
const size_t VERTEX_CACHE_SIZE = 32;
std::array<u16, VERTEX_CACHE_SIZE> vertex_cache_ids;
std::array<Shader::AttributeBuffer, VERTEX_CACHE_SIZE> vertex_cache;
Shader::AttributeBuffer vs_output;
unsigned int vertex_cache_pos = 0;
vertex_cache_ids.fill(-1);
auto* shader_engine = Shader::GetEngine();
Shader::UnitState shader_unit;
shader_engine->SetupBatch(g_state.vs, regs.vs.main_offset);
g_state.geometry_pipeline.Reconfigure();
g_state.geometry_pipeline.Setup(shader_engine);
if (g_state.geometry_pipeline.NeedIndexInput())
ASSERT(is_indexed);
for (unsigned int index = 0; index < regs.pipeline.num_vertices; ++index) {
// Indexed rendering doesn't use the start offset
unsigned int vertex =
is_indexed ? (index_u16 ? index_address_16[index] : index_address_8[index])
: (index + regs.pipeline.vertex_offset);
// -1 is a common special value used for primitive restart. Since it's unknown if
// the PICA supports it, and it would mess up the caching, guard against it here.
ASSERT(vertex != -1);
bool vertex_cache_hit = false;
if (is_indexed) {
if (g_state.geometry_pipeline.NeedIndexInput()) {
g_state.geometry_pipeline.SubmitIndex(vertex);
continue;
}
if (g_debug_context && Pica::g_debug_context->recorder) {
int size = index_u16 ? 2 : 1;
memory_accesses.AddAccess(base_address + index_info.offset + size * index,
size);
}
for (unsigned int i = 0; i < VERTEX_CACHE_SIZE; ++i) {
if (vertex == vertex_cache_ids[i]) {
vs_output = vertex_cache[i];
vertex_cache_hit = true;
break;
}
}
}
if (!vertex_cache_hit) {
// Initialize data for the current vertex
Shader::AttributeBuffer input;
loader.LoadVertex(base_address, index, vertex, input, memory_accesses);
// Send to vertex shader
if (g_debug_context)
g_debug_context->OnEvent(DebugContext::Event::VertexShaderInvocation,
(void*)&input);
shader_unit.LoadInput(regs.vs, input);
shader_engine->Run(g_state.vs, shader_unit);
shader_unit.WriteOutput(regs.vs, vs_output);
if (is_indexed) {
vertex_cache[vertex_cache_pos] = vs_output;
vertex_cache_ids[vertex_cache_pos] = vertex;
vertex_cache_pos = (vertex_cache_pos + 1) % VERTEX_CACHE_SIZE;
}
}
// Send to geometry pipeline
g_state.geometry_pipeline.SubmitVertex(vs_output);
}
for (auto& range : memory_accesses.ranges) {
g_debug_context->recorder->MemoryAccessed(Memory::GetPhysicalPointer(range.first),
range.second, range.first);
}
VideoCore::g_renderer->Rasterizer()->DrawTriangles();
if (g_debug_context) {
g_debug_context->OnEvent(DebugContext::Event::FinishedPrimitiveBatch, nullptr);
}
break;
}
case PICA_REG_INDEX(gs.bool_uniforms):
WriteUniformBoolReg(g_state.gs, g_state.regs.gs.bool_uniforms.Value());