Pica/VertexShader: Remove (now) duplicated shader bytecode definitions in favor of nihstro's ones.

merge-requests/60/head
Tony Wasserka 2014-12-16 00:32:49 +07:00
parent 056a8f9dfa
commit 8ce1d32460
2 changed files with 30 additions and 222 deletions

@ -8,11 +8,18 @@
#include <core/mem_map.h> #include <core/mem_map.h>
#include <nihstro/shader_bytecode.h>
#include "debug_utils/debug_utils.h" #include "debug_utils/debug_utils.h"
#include "pica.h" #include "pica.h"
#include "vertex_shader.h" #include "vertex_shader.h"
using nihstro::Instruction;
using nihstro::RegisterType;
using nihstro::SourceRegister;
using nihstro::SwizzlePattern;
namespace Pica { namespace Pica {
namespace VertexShader { namespace VertexShader {
@ -70,19 +77,28 @@ static void ProcessShaderCode(VertexShaderState& state) {
const Instruction& instr = *(const Instruction*)state.program_counter; const Instruction& instr = *(const Instruction*)state.program_counter;
state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + (state.program_counter - shader_memory)); state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + (state.program_counter - shader_memory));
const float24* src1_ = (instr.common.src1 < 0x10) ? state.input_register_table[instr.common.src1.GetIndex()] auto LookupSourceRegister = [&](const SourceRegister& source_reg) -> const float24* {
: (instr.common.src1 < 0x20) ? &state.temporary_registers[instr.common.src1.GetIndex()].x switch (source_reg.GetRegisterType()) {
: (instr.common.src1 < 0x80) ? &shader_uniforms.f[instr.common.src1.GetIndex()].x case RegisterType::Input:
: nullptr; return state.input_register_table[source_reg.GetIndex()];
const float24* src2_ = (instr.common.src2 < 0x10) ? state.input_register_table[instr.common.src2.GetIndex()]
: &state.temporary_registers[instr.common.src2.GetIndex()].x; case RegisterType::Temporary:
return &state.temporary_registers[source_reg.GetIndex()].x;
case RegisterType::FloatUniform:
return &shader_uniforms.f[source_reg.GetIndex()].x;
}
};
bool is_inverted = 0 != (instr.opcode.GetInfo().subtype & Instruction::OpCodeInfo::SrcInversed);
const float24* src1_ = LookupSourceRegister(instr.common.GetSrc1(is_inverted));
const float24* src2_ = LookupSourceRegister(instr.common.GetSrc2(is_inverted));
float24* dest = (instr.common.dest < 0x08) ? state.output_register_table[4*instr.common.dest.GetIndex()] float24* dest = (instr.common.dest < 0x08) ? state.output_register_table[4*instr.common.dest.GetIndex()]
: (instr.common.dest < 0x10) ? nullptr : (instr.common.dest < 0x10) ? nullptr
: (instr.common.dest < 0x20) ? &state.temporary_registers[instr.common.dest.GetIndex()][0] : (instr.common.dest < 0x20) ? &state.temporary_registers[instr.common.dest.GetIndex()][0]
: nullptr; : nullptr;
const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.common.operand_desc_id]; const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.common.operand_desc_id];
const bool negate_src1 = (swizzle.negate != 0); const bool negate_src1 = (swizzle.negate_src1 != 0);
float24 src1[4] = { float24 src1[4] = {
src1_[(int)swizzle.GetSelectorSrc1(0)], src1_[(int)swizzle.GetSelectorSrc1(0)],
@ -192,7 +208,9 @@ static void ProcessShaderCode(VertexShaderState& state) {
break; break;
} }
case Instruction::OpCode::RET: // NOP is currently used as a heuristic for leaving from a function.
// TODO: This is completely incorrect.
case Instruction::OpCode::NOP:
if (*state.call_stack_pointer == VertexShaderState::INVALID_ADDRESS) { if (*state.call_stack_pointer == VertexShaderState::INVALID_ADDRESS) {
exit_loop = true; exit_loop = true;
} else { } else {
@ -209,17 +227,16 @@ static void ProcessShaderCode(VertexShaderState& state) {
_dbg_assert_(HW_GPU, state.call_stack_pointer - state.call_stack < sizeof(state.call_stack)); _dbg_assert_(HW_GPU, state.call_stack_pointer - state.call_stack < sizeof(state.call_stack));
*++state.call_stack_pointer = state.program_counter - shader_memory; *++state.call_stack_pointer = state.program_counter - shader_memory;
// TODO: Does this offset refer to the beginning of shader memory? state.program_counter = &shader_memory[instr.flow_control.dest_offset];
state.program_counter = &shader_memory[instr.flow_control.offset_words];
break; break;
case Instruction::OpCode::FLS: case Instruction::OpCode::END:
// TODO: Do whatever needs to be done here? // TODO
break; break;
default: default:
LOG_ERROR(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x", LOG_ERROR(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value(), instr.GetOpCodeName().c_str(), instr.hex); (int)instr.opcode.Value(), instr.opcode.GetInfo().name, instr.hex);
break; break;
} }

@ -66,215 +66,6 @@ struct OutputVertex {
static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD"); static_assert(std::is_pod<OutputVertex>::value, "Structure is not POD");
static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size"); static_assert(sizeof(OutputVertex) == 32 * sizeof(float), "OutputVertex has invalid size");
union Instruction {
enum class OpCode : u32 {
ADD = 0x0,
DP3 = 0x1,
DP4 = 0x2,
MUL = 0x8,
MAX = 0xC,
MIN = 0xD,
RCP = 0xE,
RSQ = 0xF,
MOV = 0x13,
RET = 0x21,
FLS = 0x22, // Flush
CALL = 0x24,
};
std::string GetOpCodeName() const {
std::map<OpCode, std::string> map = {
{ OpCode::ADD, "ADD" },
{ OpCode::DP3, "DP3" },
{ OpCode::DP4, "DP4" },
{ OpCode::MUL, "MUL" },
{ OpCode::MAX, "MAX" },
{ OpCode::MIN, "MIN" },
{ OpCode::RCP, "RCP" },
{ OpCode::RSQ, "RSQ" },
{ OpCode::MOV, "MOV" },
{ OpCode::RET, "RET" },
{ OpCode::FLS, "FLS" },
};
auto it = map.find(opcode);
if (it == map.end())
return "UNK";
else
return it->second;
}
u32 hex;
BitField<0x1a, 0x6, OpCode> opcode;
// General notes:
//
// When two input registers are used, one of them uses a 5-bit index while the other
// one uses a 7-bit index. This is because at most one floating point uniform may be used
// as an input.
// Format used e.g. by arithmetic instructions and comparisons
// "src1" and "src2" specify register indices (i.e. indices referring to groups of 4 floats),
// while "dest" addresses individual floats.
union {
BitField<0x00, 0x5, u32> operand_desc_id;
template<class BitFieldType>
struct SourceRegister : BitFieldType {
enum RegisterType {
Input,
Temporary,
FloatUniform
};
RegisterType GetRegisterType() const {
if (BitFieldType::Value() < 0x10)
return Input;
else if (BitFieldType::Value() < 0x20)
return Temporary;
else
return FloatUniform;
}
int GetIndex() const {
if (GetRegisterType() == Input)
return BitFieldType::Value();
else if (GetRegisterType() == Temporary)
return BitFieldType::Value() - 0x10;
else // if (GetRegisterType() == FloatUniform)
return BitFieldType::Value() - 0x20;
}
std::string GetRegisterName() const {
std::map<RegisterType, std::string> type = {
{ Input, "i" },
{ Temporary, "t" },
{ FloatUniform, "f" },
};
return type[GetRegisterType()] + std::to_string(GetIndex());
}
};
SourceRegister<BitField<0x07, 0x5, u32>> src2;
SourceRegister<BitField<0x0c, 0x7, u32>> src1;
struct : BitField<0x15, 0x5, u32>
{
enum RegisterType {
Output,
Temporary,
Unknown
};
RegisterType GetRegisterType() const {
if (Value() < 0x8)
return Output;
else if (Value() < 0x10)
return Unknown;
else
return Temporary;
}
int GetIndex() const {
if (GetRegisterType() == Output)
return Value();
else if (GetRegisterType() == Temporary)
return Value() - 0x10;
else
return Value();
}
std::string GetRegisterName() const {
std::map<RegisterType, std::string> type = {
{ Output, "o" },
{ Temporary, "t" },
{ Unknown, "u" }
};
return type[GetRegisterType()] + std::to_string(GetIndex());
}
} dest;
} common;
// Format used for flow control instructions ("if")
union {
BitField<0x00, 0x8, u32> num_instructions;
BitField<0x0a, 0xc, u32> offset_words;
} flow_control;
};
static_assert(std::is_standard_layout<Instruction>::value, "Structure is not using standard layout!");
union SwizzlePattern {
u32 hex;
enum class Selector : u32 {
x = 0,
y = 1,
z = 2,
w = 3
};
Selector GetSelectorSrc1(int comp) const {
Selector selectors[] = {
src1_selector_0, src1_selector_1, src1_selector_2, src1_selector_3
};
return selectors[comp];
}
Selector GetSelectorSrc2(int comp) const {
Selector selectors[] = {
src2_selector_0, src2_selector_1, src2_selector_2, src2_selector_3
};
return selectors[comp];
}
bool DestComponentEnabled(int i) const {
return (dest_mask & (0x8 >> i)) != 0;
}
std::string SelectorToString(bool src2) const {
std::map<Selector, std::string> map = {
{ Selector::x, "x" },
{ Selector::y, "y" },
{ Selector::z, "z" },
{ Selector::w, "w" }
};
std::string ret;
for (int i = 0; i < 4; ++i) {
ret += map.at(src2 ? GetSelectorSrc2(i) : GetSelectorSrc1(i));
}
return ret;
}
std::string DestMaskToString() const {
std::string ret;
for (int i = 0; i < 4; ++i) {
if (!DestComponentEnabled(i))
ret += "_";
else
ret += "xyzw"[i];
}
return ret;
}
// Components of "dest" that should be written to: LSB=dest.w, MSB=dest.x
BitField< 0, 4, u32> dest_mask;
BitField< 4, 1, u32> negate; // negates src1
BitField< 5, 2, Selector> src1_selector_3;
BitField< 7, 2, Selector> src1_selector_2;
BitField< 9, 2, Selector> src1_selector_1;
BitField<11, 2, Selector> src1_selector_0;
BitField<14, 2, Selector> src2_selector_3;
BitField<16, 2, Selector> src2_selector_2;
BitField<18, 2, Selector> src2_selector_1;
BitField<20, 2, Selector> src2_selector_0;
BitField<31, 1, u32> flag; // not sure what this means, maybe it's the sign?
};
void SubmitShaderMemoryChange(u32 addr, u32 value); void SubmitShaderMemoryChange(u32 addr, u32 value);
void SubmitSwizzleDataChange(u32 addr, u32 value); void SubmitSwizzleDataChange(u32 addr, u32 value);