OpenPunk/citra
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Shader: Use a POD struct for registers.

Browse Source
This commit is contained in:
bunnei
2015-08-15 16:51:32 -04:00
parent b39c053785
commit db97090cad
5 changed files with 43 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
src/video_core/shader/shader.cpp
View File

@@ -67,29 +67,29 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes)
// Setup input register table // Setup input register table
const auto& attribute_register_map = config.input_register_map; const auto& attribute_register_map = config.input_register_map;
if (num_attributes > 0) state.input_registers[attribute_register_map.attribute0_register] = input.attr[0]; if (num_attributes > 0) state.registers.input[attribute_register_map.attribute0_register] = input.attr[0];
if (num_attributes > 1) state.input_registers[attribute_register_map.attribute1_register] = input.attr[1]; if (num_attributes > 1) state.registers.input[attribute_register_map.attribute1_register] = input.attr[1];
if (num_attributes > 2) state.input_registers[attribute_register_map.attribute2_register] = input.attr[2]; if (num_attributes > 2) state.registers.input[attribute_register_map.attribute2_register] = input.attr[2];
if (num_attributes > 3) state.input_registers[attribute_register_map.attribute3_register] = input.attr[3]; if (num_attributes > 3) state.registers.input[attribute_register_map.attribute3_register] = input.attr[3];
if (num_attributes > 4) state.input_registers[attribute_register_map.attribute4_register] = input.attr[4]; if (num_attributes > 4) state.registers.input[attribute_register_map.attribute4_register] = input.attr[4];
if (num_attributes > 5) state.input_registers[attribute_register_map.attribute5_register] = input.attr[5]; if (num_attributes > 5) state.registers.input[attribute_register_map.attribute5_register] = input.attr[5];
if (num_attributes > 6) state.input_registers[attribute_register_map.attribute6_register] = input.attr[6]; if (num_attributes > 6) state.registers.input[attribute_register_map.attribute6_register] = input.attr[6];
if (num_attributes > 7) state.input_registers[attribute_register_map.attribute7_register] = input.attr[7]; if (num_attributes > 7) state.registers.input[attribute_register_map.attribute7_register] = input.attr[7];
if (num_attributes > 8) state.input_registers[attribute_register_map.attribute8_register] = input.attr[8]; if (num_attributes > 8) state.registers.input[attribute_register_map.attribute8_register] = input.attr[8];
if (num_attributes > 9) state.input_registers[attribute_register_map.attribute9_register] = input.attr[9]; if (num_attributes > 9) state.registers.input[attribute_register_map.attribute9_register] = input.attr[9];
if (num_attributes > 10) state.input_registers[attribute_register_map.attribute10_register] = input.attr[10]; if (num_attributes > 10) state.registers.input[attribute_register_map.attribute10_register] = input.attr[10];
if (num_attributes > 11) state.input_registers[attribute_register_map.attribute11_register] = input.attr[11]; if (num_attributes > 11) state.registers.input[attribute_register_map.attribute11_register] = input.attr[11];
if (num_attributes > 12) state.input_registers[attribute_register_map.attribute12_register] = input.attr[12]; if (num_attributes > 12) state.registers.input[attribute_register_map.attribute12_register] = input.attr[12];
if (num_attributes > 13) state.input_registers[attribute_register_map.attribute13_register] = input.attr[13]; if (num_attributes > 13) state.registers.input[attribute_register_map.attribute13_register] = input.attr[13];
if (num_attributes > 14) state.input_registers[attribute_register_map.attribute14_register] = input.attr[14]; if (num_attributes > 14) state.registers.input[attribute_register_map.attribute14_register] = input.attr[14];
if (num_attributes > 15) state.input_registers[attribute_register_map.attribute15_register] = input.attr[15]; if (num_attributes > 15) state.registers.input[attribute_register_map.attribute15_register] = input.attr[15];
state.conditional_code[0] = false; state.conditional_code[0] = false;
state.conditional_code[1] = false; state.conditional_code[1] = false;
#ifdef ARCHITECTURE_x86_64 #ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled) if (VideoCore::g_shader_jit_enabled)
jit_shader(&state); jit_shader(&state.registers);
else else
RunInterpreter(state); RunInterpreter(state);
#else #else
@@ -117,7 +117,7 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes)
for (int comp = 0; comp < 4; ++comp) { for (int comp = 0; comp < 4; ++comp) {
float24* out = ((float24*)&ret) + semantics[comp]; float24* out = ((float24*)&ret) + semantics[comp];
if (semantics[comp] != Regs::VSOutputAttributes::INVALID) { if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
*out = state.output_registers[i][comp]; *out = state.registers.output[i][comp];
} else { } else {
// Zero output so that attributes which aren't output won't have denormals in them, // Zero output so that attributes which aren't output won't have denormals in them,
// which would slow us down later. // which would slow us down later.

21
src/video_core/shader/shader.h
View File

@@ -79,11 +79,14 @@ static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has inva
* here will make it easier for us to parallelize the shader processing later. * here will make it easier for us to parallelize the shader processing later.
*/ */
struct UnitState { struct UnitState {
// The registers are accessed by the shader JIT using SSE instructions, and are therefore struct Registers {
// required to be 16-byte aligned. // The registers are accessed by the shader JIT using SSE instructions, and are therefore
Math::Vec4<float24> MEMORY_ALIGNED16(input_registers[16]); // required to be 16-byte aligned.
Math::Vec4<float24> MEMORY_ALIGNED16(output_registers[16]); Math::Vec4<float24> MEMORY_ALIGNED16(input[16]);
Math::Vec4<float24> MEMORY_ALIGNED16(temporary_registers[16]); Math::Vec4<float24> MEMORY_ALIGNED16(output[16]);
Math::Vec4<float24> MEMORY_ALIGNED16(temporary[16]);
} registers;
static_assert(std::is_pod<Registers>::value, "Structure is not POD");
u32 program_counter; u32 program_counter;
bool conditional_code[2]; bool conditional_code[2];
@@ -116,10 +119,10 @@ struct UnitState {
static int InputOffset(const SourceRegister& reg) { static int InputOffset(const SourceRegister& reg) {
switch (reg.GetRegisterType()) { switch (reg.GetRegisterType()) {
case RegisterType::Input: case RegisterType::Input:
return (int)offsetof(UnitState, input_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>); return (int)offsetof(UnitState::Registers, input) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
case RegisterType::Temporary: case RegisterType::Temporary:
return (int)offsetof(UnitState, temporary_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>); return (int)offsetof(UnitState::Registers, temporary) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
default: default:
UNREACHABLE(); UNREACHABLE();
@@ -130,10 +133,10 @@ struct UnitState {
static int OutputOffset(const DestRegister& reg) { static int OutputOffset(const DestRegister& reg) {
switch (reg.GetRegisterType()) { switch (reg.GetRegisterType()) {
case RegisterType::Output: case RegisterType::Output:
return (int)offsetof(UnitState, output_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>); return (int)offsetof(UnitState::Registers, output) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
case RegisterType::Temporary: case RegisterType::Temporary:
return (int)offsetof(UnitState, temporary_registers) + reg.GetIndex()*sizeof(Math::Vec4<float24>); return (int)offsetof(UnitState::Registers, temporary) + reg.GetIndex()*sizeof(Math::Vec4<float24>);
default: default:
UNREACHABLE(); UNREACHABLE();

12
src/video_core/shader/shader_interpreter.cpp
View File

@@ -62,10 +62,10 @@ void RunInterpreter(UnitState& state) {
auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* { auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
switch (source_reg.GetRegisterType()) { switch (source_reg.GetRegisterType()) {
case RegisterType::Input: case RegisterType::Input:
return &state.input_registers[source_reg.GetIndex()].x; return &state.registers.input[source_reg.GetIndex()].x;
case RegisterType::Temporary: case RegisterType::Temporary:
return &state.temporary_registers[source_reg.GetIndex()].x; return &state.registers.temporary[source_reg.GetIndex()].x;
case RegisterType::FloatUniform: case RegisterType::FloatUniform:
return &uniforms.f[source_reg.GetIndex()].x; return &uniforms.f[source_reg.GetIndex()].x;
@@ -114,8 +114,8 @@ void RunInterpreter(UnitState& state) {
src2[3] = src2[3] * float24::FromFloat32(-1); src2[3] = src2[3] * float24::FromFloat32(-1);
} }
float24* dest = (instr.common.dest.Value() < 0x10) ? &state.output_registers[instr.common.dest.Value().GetIndex()][0] float24* dest = (instr.common.dest.Value() < 0x10) ? &state.registers.output[instr.common.dest.Value().GetIndex()][0]
: (instr.common.dest.Value() < 0x20) ? &state.temporary_registers[instr.common.dest.Value().GetIndex()][0] : (instr.common.dest.Value() < 0x20) ? &state.registers.temporary[instr.common.dest.Value().GetIndex()][0]
: dummy_vec4_float24; : dummy_vec4_float24;
state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id); state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id);
@@ -355,8 +355,8 @@ void RunInterpreter(UnitState& state) {
src3[3] = src3[3] * float24::FromFloat32(-1); src3[3] = src3[3] * float24::FromFloat32(-1);
} }
float24* dest = (instr.mad.dest.Value() < 0x10) ? &state.output_registers[instr.mad.dest.Value().GetIndex()][0] float24* dest = (instr.mad.dest.Value() < 0x10) ? &state.registers.output[instr.mad.dest.Value().GetIndex()][0]
: (instr.mad.dest.Value() < 0x20) ? &state.temporary_registers[instr.mad.dest.Value().GetIndex()][0] : (instr.mad.dest.Value() < 0x20) ? &state.registers.temporary[instr.mad.dest.Value().GetIndex()][0]
: dummy_vec4_float24; : dummy_vec4_float24;
for (int i = 0; i < 4; ++i) { for (int i = 0; i < 4; ++i) {

12
src/video_core/shader/shader_jit_x64.cpp
View File

@@ -106,7 +106,7 @@ static const X64Reg COND0 = R13;
/// Result of the previous CMP instruction for the Y-component comparison /// Result of the previous CMP instruction for the Y-component comparison
static const X64Reg COND1 = R14; static const X64Reg COND1 = R14;
/// Pointer to the UnitState instance for the current VS unit /// Pointer to the UnitState instance for the current VS unit
static const X64Reg STATE = R15; static const X64Reg REGISTERS = R15;
/// SIMD scratch register /// SIMD scratch register
static const X64Reg SCRATCH = XMM0; static const X64Reg SCRATCH = XMM0;
/// Loaded with the first swizzled source register, otherwise can be used as a scratch register /// Loaded with the first swizzled source register, otherwise can be used as a scratch register
@@ -140,7 +140,7 @@ void JitCompiler::Compile_SwizzleSrc(Instruction instr, unsigned src_num, Source
src_ptr = UNIFORMS; src_ptr = UNIFORMS;
src_offset = src_reg.GetIndex() * sizeof(float24) * 4; src_offset = src_reg.GetIndex() * sizeof(float24) * 4;
} else { } else {
src_ptr = STATE; src_ptr = REGISTERS;
src_offset = UnitState::InputOffset(src_reg); src_offset = UnitState::InputOffset(src_reg);
} }
@@ -217,11 +217,11 @@ void JitCompiler::Compile_DestEnable(Instruction instr,X64Reg src) {
// If all components are enabled, write the result to the destination register // If all components are enabled, write the result to the destination register
if (swiz.dest_mask == NO_DEST_REG_MASK) { if (swiz.dest_mask == NO_DEST_REG_MASK) {
// Store dest back to memory // Store dest back to memory
MOVAPS(MDisp(STATE, UnitState::OutputOffset(dest)), src); MOVAPS(MDisp(REGISTERS, UnitState::OutputOffset(dest)), src);
} else { } else {
// Not all components are enabled, so mask the result when storing to the destination register... // Not all components are enabled, so mask the result when storing to the destination register...
MOVAPS(SCRATCH, MDisp(STATE, UnitState::OutputOffset(dest))); MOVAPS(SCRATCH, MDisp(REGISTERS, UnitState::OutputOffset(dest)));
if (Common::GetCPUCaps().sse4_1) { if (Common::GetCPUCaps().sse4_1) {
u8 mask = ((swiz.dest_mask & 1) << 3) | ((swiz.dest_mask & 8) >> 3) | ((swiz.dest_mask & 2) << 1) | ((swiz.dest_mask & 4) >> 1); u8 mask = ((swiz.dest_mask & 1) << 3) | ((swiz.dest_mask & 8) >> 3) | ((swiz.dest_mask & 2) << 1) | ((swiz.dest_mask & 4) >> 1);
@@ -240,7 +240,7 @@ void JitCompiler::Compile_DestEnable(Instruction instr,X64Reg src) {
} }
// Store dest back to memory // Store dest back to memory
MOVAPS(MDisp(STATE, UnitState::OutputOffset(dest)), SCRATCH); MOVAPS(MDisp(REGISTERS, UnitState::OutputOffset(dest)), SCRATCH);
} }
} }
@@ -635,7 +635,7 @@ CompiledShader* JitCompiler::Compile() {
ABI_PushAllCalleeSavedRegsAndAdjustStack(); ABI_PushAllCalleeSavedRegsAndAdjustStack();
MOV(PTRBITS, R(STATE), R(ABI_PARAM1)); MOV(PTRBITS, R(REGISTERS), R(ABI_PARAM1));
MOV(PTRBITS, R(UNIFORMS), ImmPtr(&g_state.vs.uniforms)); MOV(PTRBITS, R(UNIFORMS), ImmPtr(&g_state.vs.uniforms));
// Zero address/loop registers // Zero address/loop registers

2
src/video_core/shader/shader_jit_x64.h
View File

@@ -20,7 +20,7 @@ namespace Pica {
namespace Shader { namespace Shader {
using CompiledShader = void(void* state); using CompiledShader = void(void* registers);
/** /**
* This class implements the shader JIT compiler. It recompiles a Pica shader program into x86_64 * This class implements the shader JIT compiler. It recompiles a Pica shader program into x86_64