citra-emu
/
citra-canary
Archived
1
0
Fork 0

Videocore: Simplify variables in vertex shader interpreter

Simplifies the code and gives a tiny speed-up.
This commit is contained in:
Yuri Kunde Schlesner 2015-07-26 07:40:34 -03:00
parent 877d2a0e48
commit 4909a1486e
1 changed files with 21 additions and 24 deletions

View File

@ -26,7 +26,7 @@ namespace Pica {
namespace VertexShader { namespace VertexShader {
struct VertexShaderState { struct VertexShaderState {
const u32* program_counter; u32 program_counter;
const float24* input_register_table[16]; const float24* input_register_table[16];
Math::Vec4<float24> output_registers[16]; Math::Vec4<float24> output_registers[16];
@ -71,14 +71,14 @@ static void ProcessShaderCode(VertexShaderState& state) {
while (true) { while (true) {
if (!state.call_stack.empty()) { if (!state.call_stack.empty()) {
auto& top = state.call_stack.back(); auto& top = state.call_stack.back();
if (state.program_counter - program_code.data() == top.final_address) { if (state.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 = &program_code[top.return_address]; state.program_counter = top.return_address;
state.call_stack.pop_back(); state.call_stack.pop_back();
} else { } else {
state.program_counter = &program_code[top.loop_address]; state.program_counter = top.loop_address;
} }
// TODO: Is "trying again" accurate to hardware? // TODO: Is "trying again" accurate to hardware?
@ -87,18 +87,16 @@ static void ProcessShaderCode(VertexShaderState& state) {
} }
bool exit_loop = false; bool exit_loop = false;
const Instruction& instr = *(const Instruction*)state.program_counter; const Instruction instr = { program_code[state.program_counter] };
const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.common.operand_desc_id]; const SwizzlePattern swizzle = { swizzle_data[instr.common.operand_desc_id] };
static auto call = [&program_code](VertexShaderState& state, u32 offset, u32 num_instructions, static auto call = [](VertexShaderState& 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 = &program_code[offset] - 1; // -1 to make sure when incrementing the PC we end up at the correct offset state.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(state.call_stack.size() < state.call_stack.capacity());
state.call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset }); state.call_stack.push_back({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset });
}; };
u32 binary_offset = state.program_counter - program_code.data(); 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 + binary_offset);
auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* { auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
switch (source_reg.GetRegisterType()) { switch (source_reg.GetRegisterType()) {
@ -442,13 +440,13 @@ static void ProcessShaderCode(VertexShaderState& state) {
case OpCode::Id::JMPC: case OpCode::Id::JMPC:
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 = &program_code[instr.flow_control.dest_offset] - 1; state.program_counter = instr.flow_control.dest_offset - 1;
} }
break; break;
case OpCode::Id::JMPU: case OpCode::Id::JMPU:
if (uniforms.b[instr.flow_control.bool_uniform_id]) { if (uniforms.b[instr.flow_control.bool_uniform_id]) {
state.program_counter = &program_code[instr.flow_control.dest_offset] - 1; state.program_counter = instr.flow_control.dest_offset - 1;
} }
break; break;
@ -456,7 +454,7 @@ static void ProcessShaderCode(VertexShaderState& 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,
binary_offset + 1, 0, 0); state.program_counter + 1, 0, 0);
break; break;
case OpCode::Id::CALLU: case OpCode::Id::CALLU:
@ -464,7 +462,7 @@ static void ProcessShaderCode(VertexShaderState& 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,
binary_offset + 1, 0, 0); state.program_counter + 1, 0, 0);
} }
break; break;
@ -473,7 +471,7 @@ static void ProcessShaderCode(VertexShaderState& 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,
binary_offset + 1, 0, 0); state.program_counter + 1, 0, 0);
} }
break; break;
@ -483,8 +481,8 @@ static void ProcessShaderCode(VertexShaderState& state) {
case OpCode::Id::IFU: case OpCode::Id::IFU:
if (uniforms.b[instr.flow_control.bool_uniform_id]) { if (uniforms.b[instr.flow_control.bool_uniform_id]) {
call(state, call(state,
binary_offset + 1, state.program_counter + 1,
instr.flow_control.dest_offset - binary_offset - 1, instr.flow_control.dest_offset - state.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,
@ -501,8 +499,8 @@ static void ProcessShaderCode(VertexShaderState& state) {
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,
binary_offset + 1, state.program_counter + 1,
instr.flow_control.dest_offset - binary_offset - 1, instr.flow_control.dest_offset - state.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,
@ -519,8 +517,8 @@ static void ProcessShaderCode(VertexShaderState& state) {
state.address_registers[2] = uniforms.i[instr.flow_control.int_uniform_id].y; state.address_registers[2] = uniforms.i[instr.flow_control.int_uniform_id].y;
call(state, call(state,
binary_offset + 1, state.program_counter + 1,
instr.flow_control.dest_offset - binary_offset + 1, instr.flow_control.dest_offset - state.program_counter + 1,
instr.flow_control.dest_offset + 1, instr.flow_control.dest_offset + 1,
uniforms.i[instr.flow_control.int_uniform_id].x, uniforms.i[instr.flow_control.int_uniform_id].x,
uniforms.i[instr.flow_control.int_uniform_id].z); uniforms.i[instr.flow_control.int_uniform_id].z);
@ -551,8 +549,7 @@ OutputVertex RunShader(const InputVertex& input, int num_attributes, const Regs:
VertexShaderState state; VertexShaderState state;
const u32* main = &setup.program_code[config.main_offset]; state.program_counter = config.main_offset;
state.program_counter = (u32*)main;
state.debug.max_offset = 0; state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0; state.debug.max_opdesc_id = 0;