Merge pull request #1783 from JayFoxRox/cleanup-shadersetup

Cleanup ShaderSetup (Part 1)
This commit is contained in:
bunnei 2016-05-12 19:32:44 -04:00
commit d474d117f6
7 changed files with 47 additions and 50 deletions

View file

@ -74,7 +74,7 @@ void GraphicsTracingWidget::StartRecording() {
std::array<u32, 4 * 16> default_attributes; std::array<u32, 4 * 16> default_attributes;
for (unsigned i = 0; i < 16; ++i) { for (unsigned i = 0; i < 16; ++i) {
for (unsigned comp = 0; comp < 3; ++comp) { for (unsigned comp = 0; comp < 3; ++comp) {
default_attributes[4 * i + comp] = nihstro::to_float24(Pica::g_state.vs.default_attributes[i][comp].ToFloat32()); default_attributes[4 * i + comp] = nihstro::to_float24(Pica::g_state.vs_default_attributes[i][comp].ToFloat32());
} }
} }

View file

@ -128,7 +128,7 @@ static void WritePicaReg(u32 id, u32 value, u32 mask) {
// TODO: Verify that this actually modifies the register! // TODO: Verify that this actually modifies the register!
if (setup.index < 15) { if (setup.index < 15) {
g_state.vs.default_attributes[setup.index] = attribute; g_state.vs_default_attributes[setup.index] = attribute;
setup.index++; setup.index++;
} else { } else {
// Put each attribute into an immediate input buffer. // Put each attribute into an immediate input buffer.

View file

@ -25,6 +25,8 @@ struct State {
Shader::ShaderSetup vs; Shader::ShaderSetup vs;
Shader::ShaderSetup gs; Shader::ShaderSetup gs;
std::array<Math::Vec4<float24>, 16> vs_default_attributes;
struct { struct {
union LutEntry { union LutEntry {
// Used for raw access // Used for raw access

View file

@ -67,7 +67,6 @@ OutputVertex ShaderSetup::Run(UnitState<false>& state, const InputVertex& input,
MICROPROFILE_SCOPE(GPU_Shader); MICROPROFILE_SCOPE(GPU_Shader);
state.program_counter = config.main_offset;
state.debug.max_offset = 0; state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0; state.debug.max_opdesc_id = 0;
@ -143,7 +142,6 @@ OutputVertex ShaderSetup::Run(UnitState<false>& state, const InputVertex& input,
DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup) { DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup) {
UnitState<true> state; UnitState<true> state;
state.program_counter = config.main_offset;
state.debug.max_offset = 0; state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0; state.debug.max_opdesc_id = 0;

View file

@ -272,29 +272,12 @@ struct UnitState {
} registers; } registers;
static_assert(std::is_pod<Registers>::value, "Structure is not POD"); static_assert(std::is_pod<Registers>::value, "Structure is not POD");
u32 program_counter;
bool conditional_code[2]; bool conditional_code[2];
// Two Address registers and one loop counter // Two Address registers and one loop counter
// TODO: How many bits do these actually have? // TODO: How many bits do these actually have?
s32 address_registers[3]; s32 address_registers[3];
enum {
INVALID_ADDRESS = 0xFFFFFFFF
};
struct CallStackElement {
u32 final_address; // Address upon which we jump to return_address
u32 return_address; // Where to jump when leaving scope
u8 repeat_counter; // How often to repeat until this call stack element is removed
u8 loop_increment; // Which value to add to the loop counter after an iteration
// TODO: Should this be a signed value? Does it even matter?
u32 loop_address; // The address where we'll return to after each loop iteration
};
// TODO: Is there a maximal size for this?
boost::container::static_vector<CallStackElement, 16> call_stack;
DebugData<Debug> debug; DebugData<Debug> debug;
static size_t InputOffset(const SourceRegister& reg) { static size_t InputOffset(const SourceRegister& reg) {
@ -340,8 +323,6 @@ struct ShaderSetup {
std::array<Math::Vec4<u8>, 4> i; std::array<Math::Vec4<u8>, 4> i;
} uniforms; } uniforms;
Math::Vec4<float24> default_attributes[16];
std::array<u32, 1024> program_code; std::array<u32, 1024> program_code;
std::array<u32, 1024> swizzle_data; std::array<u32, 1024> swizzle_data;

View file

@ -29,8 +29,24 @@ namespace Pica {
namespace Shader { namespace Shader {
constexpr u32 INVALID_ADDRESS = 0xFFFFFFFF;
struct CallStackElement {
u32 final_address; // Address upon which we jump to return_address
u32 return_address; // Where to jump when leaving scope
u8 repeat_counter; // How often to repeat until this call stack element is removed
u8 loop_increment; // Which value to add to the loop counter after an iteration
// TODO: Should this be a signed value? Does it even matter?
u32 loop_address; // The address where we'll return to after each loop iteration
};
template<bool Debug> template<bool Debug>
void RunInterpreter(UnitState<Debug>& state) { void RunInterpreter(UnitState<Debug>& state) {
// TODO: Is there a maximal size for this?
boost::container::static_vector<CallStackElement, 16> call_stack;
u32 program_counter = g_state.regs.vs.main_offset;
const auto& uniforms = g_state.vs.uniforms; const auto& uniforms = g_state.vs.uniforms;
const auto& swizzle_data = g_state.vs.swizzle_data; const auto& swizzle_data = g_state.vs.swizzle_data;
const auto& program_code = g_state.vs.program_code; const auto& program_code = g_state.vs.program_code;
@ -41,16 +57,16 @@ void RunInterpreter(UnitState<Debug>& state) {
unsigned iteration = 0; unsigned iteration = 0;
bool exit_loop = false; bool exit_loop = false;
while (!exit_loop) { while (!exit_loop) {
if (!state.call_stack.empty()) { if (!call_stack.empty()) {
auto& top = state.call_stack.back(); auto& top = call_stack.back();
if (state.program_counter == top.final_address) { if (program_counter == top.final_address) {
state.address_registers[2] += top.loop_increment; state.address_registers[2] += top.loop_increment;
if (top.repeat_counter-- == 0) { if (top.repeat_counter-- == 0) {
state.program_counter = top.return_address; program_counter = top.return_address;
state.call_stack.pop_back(); call_stack.pop_back();
} else { } else {
state.program_counter = top.loop_address; program_counter = top.loop_address;
} }
// TODO: Is "trying again" accurate to hardware? // TODO: Is "trying again" accurate to hardware?
@ -58,20 +74,20 @@ void RunInterpreter(UnitState<Debug>& state) {
} }
} }
const Instruction instr = { program_code[state.program_counter] }; const Instruction instr = { program_code[program_counter] };
const SwizzlePattern swizzle = { swizzle_data[instr.common.operand_desc_id] }; const SwizzlePattern swizzle = { swizzle_data[instr.common.operand_desc_id] };
static auto call = [](UnitState<Debug>& state, u32 offset, u32 num_instructions, static auto call = [&program_counter, &call_stack](UnitState<Debug>& state, u32 offset, u32 num_instructions,
u32 return_offset, u8 repeat_count, u8 loop_increment) { u32 return_offset, u8 repeat_count, u8 loop_increment) {
state.program_counter = offset - 1; // -1 to make sure when incrementing the PC we end up at the correct offset program_counter = offset - 1; // -1 to make sure when incrementing the PC we end up at the correct offset
ASSERT(state.call_stack.size() < state.call_stack.capacity()); ASSERT(call_stack.size() < call_stack.capacity());
state.call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset }); call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset });
}; };
Record<DebugDataRecord::CUR_INSTR>(state.debug, iteration, state.program_counter); Record<DebugDataRecord::CUR_INSTR>(state.debug, iteration, program_counter);
if (iteration > 0) if (iteration > 0)
Record<DebugDataRecord::NEXT_INSTR>(state.debug, iteration - 1, state.program_counter); Record<DebugDataRecord::NEXT_INSTR>(state.debug, iteration - 1, program_counter);
state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + state.program_counter); state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + program_counter);
auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* { auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
switch (source_reg.GetRegisterType()) { switch (source_reg.GetRegisterType()) {
@ -519,7 +535,7 @@ void RunInterpreter(UnitState<Debug>& state) {
case OpCode::Id::JMPC: case OpCode::Id::JMPC:
Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration, state.conditional_code); Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration, state.conditional_code);
if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) {
state.program_counter = instr.flow_control.dest_offset - 1; program_counter = instr.flow_control.dest_offset - 1;
} }
break; break;
@ -527,7 +543,7 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::COND_BOOL_IN>(state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]); Record<DebugDataRecord::COND_BOOL_IN>(state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id] == !(instr.flow_control.num_instructions & 1)) { if (uniforms.b[instr.flow_control.bool_uniform_id] == !(instr.flow_control.num_instructions & 1)) {
state.program_counter = instr.flow_control.dest_offset - 1; program_counter = instr.flow_control.dest_offset - 1;
} }
break; break;
@ -535,7 +551,7 @@ void RunInterpreter(UnitState<Debug>& state) {
call(state, call(state,
instr.flow_control.dest_offset, instr.flow_control.dest_offset,
instr.flow_control.num_instructions, instr.flow_control.num_instructions,
state.program_counter + 1, 0, 0); program_counter + 1, 0, 0);
break; break;
case OpCode::Id::CALLU: case OpCode::Id::CALLU:
@ -544,7 +560,7 @@ void RunInterpreter(UnitState<Debug>& state) {
call(state, call(state,
instr.flow_control.dest_offset, instr.flow_control.dest_offset,
instr.flow_control.num_instructions, instr.flow_control.num_instructions,
state.program_counter + 1, 0, 0); program_counter + 1, 0, 0);
} }
break; break;
@ -554,7 +570,7 @@ void RunInterpreter(UnitState<Debug>& state) {
call(state, call(state,
instr.flow_control.dest_offset, instr.flow_control.dest_offset,
instr.flow_control.num_instructions, instr.flow_control.num_instructions,
state.program_counter + 1, 0, 0); program_counter + 1, 0, 0);
} }
break; break;
@ -565,8 +581,8 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::COND_BOOL_IN>(state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]); Record<DebugDataRecord::COND_BOOL_IN>(state.debug, iteration, uniforms.b[instr.flow_control.bool_uniform_id]);
if (uniforms.b[instr.flow_control.bool_uniform_id]) { if (uniforms.b[instr.flow_control.bool_uniform_id]) {
call(state, call(state,
state.program_counter + 1, program_counter + 1,
instr.flow_control.dest_offset - state.program_counter - 1, instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else { } else {
call(state, call(state,
@ -584,8 +600,8 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration, state.conditional_code); Record<DebugDataRecord::COND_CMP_IN>(state.debug, iteration, state.conditional_code);
if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) { if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) {
call(state, call(state,
state.program_counter + 1, program_counter + 1,
instr.flow_control.dest_offset - state.program_counter - 1, instr.flow_control.dest_offset - program_counter - 1,
instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0); instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else { } else {
call(state, call(state,
@ -607,8 +623,8 @@ void RunInterpreter(UnitState<Debug>& state) {
Record<DebugDataRecord::LOOP_INT_IN>(state.debug, iteration, loop_param); Record<DebugDataRecord::LOOP_INT_IN>(state.debug, iteration, loop_param);
call(state, call(state,
state.program_counter + 1, program_counter + 1,
instr.flow_control.dest_offset - state.program_counter + 1, instr.flow_control.dest_offset - program_counter + 1,
instr.flow_control.dest_offset + 1, instr.flow_control.dest_offset + 1,
loop_param.x, loop_param.x,
loop_param.z); loop_param.z);
@ -625,7 +641,7 @@ void RunInterpreter(UnitState<Debug>& state) {
} }
} }
++state.program_counter; ++program_counter;
++iteration; ++iteration;
} }
} }

View file

@ -124,7 +124,7 @@ void VertexLoader::LoadVertex(u32 base_address, int index, int vertex, Shader::I
input.attr[i][0].ToFloat32(), input.attr[i][1].ToFloat32(), input.attr[i][2].ToFloat32(), input.attr[i][3].ToFloat32()); input.attr[i][0].ToFloat32(), input.attr[i][1].ToFloat32(), input.attr[i][2].ToFloat32(), input.attr[i][3].ToFloat32());
} else if (vertex_attribute_is_default[i]) { } else if (vertex_attribute_is_default[i]) {
// Load the default attribute if we're configured to do so // Load the default attribute if we're configured to do so
input.attr[i] = g_state.vs.default_attributes[i]; input.attr[i] = g_state.vs_default_attributes[i];
LOG_TRACE(HW_GPU, "Loaded default attribute %x for vertex %x (index %x): (%f, %f, %f, %f)", LOG_TRACE(HW_GPU, "Loaded default attribute %x for vertex %x (index %x): (%f, %f, %f, %f)",
i, vertex, index, i, vertex, index,
input.attr[i][0].ToFloat32(), input.attr[i][1].ToFloat32(), input.attr[i][0].ToFloat32(), input.attr[i][1].ToFloat32(),