H3cJP/citra
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #344 from bunnei/shader-decompiler-p2

Browse Source 
Shader decompiler changes part 2
This commit is contained in:
bunnei 2018-04-17 22:10:53 -04:00 committed by GitHub
commit 71b4a3b9f6
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 182 additions and 75 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/common/bit_field.h
View File

@ -192,11 +192,6 @@ private:
static_assert(position < 8 * sizeof(T), "Invalid position");
static_assert(bits <= 8 * sizeof(T), "Invalid number of bits");
static_assert(bits > 0, "Invalid number of bits");
static_assert(std::is_pod<T>::value, "Invalid base type");
static_assert(std::is_trivially_copyable_v<T>, "T must be trivially copyable in a BitField");
};
#pragma pack()
#if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)
static_assert(std::is_trivially_copyable<BitField<0, 1, unsigned>>::value,
"BitField must be trivially copyable");
#endif

43
src/video_core/engines/shader_bytecode.h
View File

@ -4,6 +4,7 @@
#pragma once
#include <cstring>
#include <map>
#include <string>
#include "common/bit_field.h"
@ -12,14 +13,10 @@ namespace Tegra {
namespace Shader {
struct Register {
Register() = default;
constexpr Register() = default;
constexpr Register(u64 value) : value(value) {}
constexpr u64 GetIndex() const {
return value;
}
constexpr operator u64() const {
return value;
}
@ -43,13 +40,13 @@ struct Register {
}
private:
u64 value;
u64 value{};
};
union Attribute {
Attribute() = default;
constexpr Attribute(u64 value) : value(value) {}
constexpr explicit Attribute(u64 value) : value(value) {}
enum class Index : u64 {
Position = 7,
@ -68,7 +65,20 @@ union Attribute {
} fmt28;
BitField<39, 8, u64> reg;
u64 value;
u64 value{};
};
union Sampler {
Sampler() = default;
constexpr explicit Sampler(u64 value) : value(value) {}
enum class Index : u64 {
Sampler_0 = 8,
};
BitField<36, 13, Index> index;
u64 value{};
};
union Uniform {
@ -238,7 +248,7 @@ union OpCode {
BitField<55, 9, Id> op3;
BitField<52, 12, Id> op4;
BitField<51, 13, Id> op5;
u64 value;
u64 value{};
};
static_assert(sizeof(OpCode) == 0x8, "Incorrect structure size");
@ -280,6 +290,7 @@ enum class SubOp : u64 {
Lg2 = 0x3,
Rcp = 0x4,
Rsq = 0x5,
Min = 0x8,
};
union Instruction {
@ -295,15 +306,25 @@ union Instruction {
BitField<20, 8, Register> gpr20;
BitField<20, 7, SubOp> sub_op;
BitField<28, 8, Register> gpr28;
BitField<36, 13, u64> imm36;
BitField<39, 8, Register> gpr39;
union {
BitField<20, 19, u64> imm20;
BitField<45, 1, u64> negate_b;
BitField<46, 1, u64> abs_a;
BitField<48, 1, u64> negate_a;
BitField<49, 1, u64> abs_b;
BitField<50, 1, u64> abs_d;
BitField<56, 1, u64> negate_imm;
float GetImm20() const {
float result{};
u32 imm{static_cast<u32>(imm20)};
imm <<= 12;
imm |= negate_imm ? 0x80000000 : 0;
std::memcpy(&result, &imm, sizeof(imm));
return result;
}
} alu;
union {
@ -311,11 +332,13 @@ union Instruction {
BitField<49, 1, u64> negate_c;
} ffma;
BitField<61, 1, u64> is_b_imm;
BitField<60, 1, u64> is_b_gpr;
BitField<59, 1, u64> is_c_gpr;
Attribute attribute;
Uniform uniform;
Sampler sampler;
u64 hex;
};

201
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View File

@ -17,6 +17,7 @@ using Tegra::Shader::Attribute;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Register;
using Tegra::Shader::Sampler;
using Tegra::Shader::SubOp;
using Tegra::Shader::Uniform;
@ -155,23 +156,27 @@ private:
/// Generates code representing an input attribute register.
std::string GetInputAttribute(Attribute::Index attribute) {
declr_input_attribute.insert(attribute);
switch (attribute) {
case Attribute::Index::Position:
return "position";
default:
const u32 index{static_cast<u32>(attribute) -
static_cast<u32>(Attribute::Index::Attribute_0)};
if (attribute >= Attribute::Index::Attribute_0) {
declr_input_attribute.insert(attribute);
return "input_attribute_" + std::to_string(index);
}
const u32 index{static_cast<u32>(attribute) -
static_cast<u32>(Attribute::Index::Attribute_0)};
if (attribute >= Attribute::Index::Attribute_0) {
return "input_attribute_" + std::to_string(index);
NGLOG_CRITICAL(HW_GPU, "Unhandled input attribute: {}", index);
UNREACHABLE();
}
LOG_CRITICAL(HW_GPU, "Unhandled input attribute: 0x%02x", index);
UNREACHABLE();
}
/// Generates code representing an output attribute register.
std::string GetOutputAttribute(Attribute::Index attribute) {
switch (attribute) {
case Attribute::Index::Position:
return "gl_Position";
return "position";
default:
const u32 index{static_cast<u32>(attribute) -
static_cast<u32>(Attribute::Index::Attribute_0)};
@ -180,22 +185,42 @@ private:
return "output_attribute_" + std::to_string(index);
}
LOG_CRITICAL(HW_GPU, "Unhandled output attribute: 0x%02x", index);
NGLOG_CRITICAL(HW_GPU, "Unhandled output attribute: {}", index);
UNREACHABLE();
}
}
/// Generates code representing an immediate value
static std::string GetImmediate(const Instruction& instr) {
return std::to_string(instr.alu.GetImm20());
}
/// Generates code representing a temporary (GPR) register.
std::string GetRegister(const Register& reg) {
return *declr_register.insert("register_" + std::to_string(reg)).first;
std::string GetRegister(const Register& reg, unsigned elem = 0) {
if (stage == Maxwell3D::Regs::ShaderStage::Fragment && reg < 4) {
// GPRs 0-3 are output color for the fragment shader
return std::string{"color."} + "rgba"[(reg + elem) & 3];
}
return *declr_register.insert("register_" + std::to_string(reg + elem)).first;
}
/// Generates code representing a uniform (C buffer) register.
std::string GetUniform(const Uniform& reg) {
declr_const_buffers[reg.index].MarkAsUsed(reg.index, reg.offset, stage);
declr_const_buffers[reg.index].MarkAsUsed(static_cast<unsigned>(reg.index),
static_cast<unsigned>(reg.offset), stage);
return 'c' + std::to_string(reg.index) + '[' + std::to_string(reg.offset) + ']';
}
/// Generates code representing a texture sampler.
std::string GetSampler(const Sampler& sampler) const {
// TODO(Subv): Support more than just texture sampler 0
ASSERT_MSG(sampler.index == Sampler::Index::Sampler_0, "unsupported");
const unsigned index{static_cast<unsigned>(sampler.index.Value()) -
static_cast<unsigned>(Sampler::Index::Sampler_0)};
return "tex[" + std::to_string(index) + "]";
}
/**
* Adds code that calls a subroutine.
* @param subroutine the subroutine to call.
@ -217,12 +242,13 @@ private:
* @param value the code representing the value to assign.
*/
void SetDest(u64 elem, const std::string& reg, const std::string& value,
u64 dest_num_components, u64 value_num_components) {
u64 dest_num_components, u64 value_num_components, bool is_abs = false) {
std::string swizzle = ".";
swizzle += "xyzw"[elem];
std::string dest = reg + (dest_num_components != 1 ? swizzle : "");
std::string src = "(" + value + ")" + (value_num_components != 1 ? swizzle : "");
src = is_abs ? "abs(" + src + ")" : src;
shader.AddLine(dest + " = " + src + ";");
}
@ -240,8 +266,6 @@ private:
switch (OpCode::GetInfo(instr.opcode).type) {
case OpCode::Type::Arithmetic: {
ASSERT(!instr.alu.abs_d);
std::string dest = GetRegister(instr.gpr0);
std::string op_a = instr.alu.negate_a ? "-" : "";
op_a += GetRegister(instr.gpr8);
@ -250,63 +274,109 @@ private:
}
std::string op_b = instr.alu.negate_b ? "-" : "";
if (instr.is_b_gpr) {
op_b += GetRegister(instr.gpr20);
if (instr.is_b_imm) {
op_b += GetImmediate(instr);
} else {
op_b += GetUniform(instr.uniform);
if (instr.is_b_gpr) {
op_b += GetRegister(instr.gpr20);
} else {
op_b += GetUniform(instr.uniform);
}
}
if (instr.alu.abs_b) {
op_b = "abs(" + op_b + ")";
}
switch (instr.opcode.EffectiveOpCode()) {
case OpCode::Id::FMUL_C:
case OpCode::Id::FMUL_R: {
SetDest(0, dest, op_a + " * " + op_b, 1, 1);
case OpCode::Id::FMUL_R:
case OpCode::Id::FMUL_IMM: {
SetDest(0, dest, op_a + " * " + op_b, 1, 1, instr.alu.abs_d);
break;
}
case OpCode::Id::FADD_C:
case OpCode::Id::FADD_R: {
SetDest(0, dest, op_a + " + " + op_b, 1, 1);
case OpCode::Id::FADD_R:
case OpCode::Id::FADD_IMM: {
SetDest(0, dest, op_a + " + " + op_b, 1, 1, instr.alu.abs_d);
break;
}
case OpCode::Id::MUFU: {
switch (instr.sub_op) {
case SubOp::Cos:
SetDest(0, dest, "cos(" + op_a + ")", 1, 1, instr.alu.abs_d);
break;
case SubOp::Sin:
SetDest(0, dest, "sin(" + op_a + ")", 1, 1, instr.alu.abs_d);
break;
case SubOp::Ex2:
SetDest(0, dest, "exp2(" + op_a + ")", 1, 1, instr.alu.abs_d);
break;
case SubOp::Lg2:
SetDest(0, dest, "log2(" + op_a + ")", 1, 1, instr.alu.abs_d);
break;
case SubOp::Rcp:
SetDest(0, dest, "1.0 / " + op_a, 1, 1, instr.alu.abs_d);
break;
case SubOp::Rsq:
SetDest(0, dest, "inversesqrt(" + op_a + ")", 1, 1, instr.alu.abs_d);
break;
case SubOp::Min:
SetDest(0, dest, "min(" + op_a + "," + op_b + ")", 1, 1, instr.alu.abs_d);
break;
default:
NGLOG_CRITICAL(HW_GPU, "Unhandled MUFU sub op: {}",
static_cast<unsigned>(instr.sub_op.Value()));
UNREACHABLE();
}
break;
}
default: {
LOG_CRITICAL(HW_GPU, "Unhandled arithmetic instruction: 0x%02x (%s): 0x%08x",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name.c_str(), instr.hex);
throw DecompileFail("Unhandled instruction");
break;
NGLOG_CRITICAL(HW_GPU, "Unhandled arithmetic instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
UNREACHABLE();
}
}
break;
}
case OpCode::Type::Ffma: {
ASSERT_MSG(!instr.ffma.negate_b, "untested");
ASSERT_MSG(!instr.ffma.negate_c, "untested");
std::string dest = GetRegister(instr.gpr0);
std::string op_a = GetRegister(instr.gpr8);
std::string op_b = instr.ffma.negate_b ? "-" : "";
op_b += GetUniform(instr.uniform);
std::string op_c = instr.ffma.negate_c ? "-" : "";
op_c += GetRegister(instr.gpr39);
switch (instr.opcode.EffectiveOpCode()) {
case OpCode::Id::FFMA_CR: {
SetDest(0, dest, op_a + " * " + op_b + " + " + op_c, 1, 1);
op_b += GetUniform(instr.uniform);
op_c += GetRegister(instr.gpr39);
break;
}
case OpCode::Id::FFMA_RR: {
op_b += GetRegister(instr.gpr20);
op_c += GetRegister(instr.gpr39);
break;
}
case OpCode::Id::FFMA_RC: {
op_b += GetRegister(instr.gpr39);
op_c += GetUniform(instr.uniform);
break;
}
case OpCode::Id::FFMA_IMM: {
op_b += GetImmediate(instr);
op_c += GetRegister(instr.gpr39);
break;
}
default: {
NGLOG_CRITICAL(HW_GPU, "Unhandled FFMA instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
UNREACHABLE();
}
}
default: {
LOG_CRITICAL(HW_GPU, "Unhandled arithmetic FFMA instruction: 0x%02x (%s): 0x%08x",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name.c_str(), instr.hex);
throw DecompileFail("Unhandled instruction");
break;
}
}
SetDest(0, dest, op_a + " * " + op_b + " + " + op_c, 1, 1);
break;
}
case OpCode::Type::Memory: {
@ -315,22 +385,33 @@ private:
switch (instr.opcode.EffectiveOpCode()) {
case OpCode::Id::LD_A: {
ASSERT(instr.attribute.fmt20.size == 0);
ASSERT_MSG(instr.attribute.fmt20.size == 0, "untested");
SetDest(instr.attribute.fmt20.element, gpr0, GetInputAttribute(attribute), 1, 4);
break;
}
case OpCode::Id::ST_A: {
ASSERT(instr.attribute.fmt20.size == 0);
ASSERT_MSG(instr.attribute.fmt20.size == 0, "untested");
SetDest(instr.attribute.fmt20.element, GetOutputAttribute(attribute), gpr0, 4, 1);
break;
}
default: {
LOG_CRITICAL(HW_GPU, "Unhandled memory instruction: 0x%02x (%s): 0x%08x",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name.c_str(), instr.hex);
throw DecompileFail("Unhandled instruction");
case OpCode::Id::TEXS: {
ASSERT_MSG(instr.attribute.fmt20.size == 4, "untested");
const std::string op_a = GetRegister(instr.gpr8);
const std::string op_b = GetRegister(instr.gpr20);
const std::string sampler = GetSampler(instr.sampler);
const std::string coord = "vec2(" + op_a + ", " + op_b + ")";
const std::string texture = "texture(" + sampler + ", " + coord + ")";
for (unsigned elem = 0; elem < instr.attribute.fmt20.size; ++elem) {
SetDest(elem, GetRegister(instr.gpr0, elem), texture, 1, 4);
}
break;
}
default: {
NGLOG_CRITICAL(HW_GPU, "Unhandled memory instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
UNREACHABLE();
}
}
break;
}
@ -342,14 +423,18 @@ private:
offset = PROGRAM_END - 1;
break;
}
default: {
LOG_CRITICAL(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name.c_str(), instr.hex);
throw DecompileFail("Unhandled instruction");
case OpCode::Id::IPA: {
const auto& attribute = instr.attribute.fmt28;
std::string dest = GetRegister(instr.gpr0);
SetDest(attribute.element, dest, GetInputAttribute(attribute.index), 1, 4);
break;
}
default: {
NGLOG_CRITICAL(HW_GPU, "Unhandled instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
UNREACHABLE();
}
}
break;
@ -514,7 +599,7 @@ boost::optional<ProgramResult> DecompileProgram(const ProgramCode& program_code,
GLSLGenerator generator(subroutines, program_code, main_offset, stage);
return ProgramResult{generator.GetShaderCode(), generator.GetEntries()};
} catch (const DecompileFail& exception) {
LOG_ERROR(HW_GPU, "Shader decompilation failed: %s", exception.what());
NGLOG_ERROR(HW_GPU, "Shader decompilation failed: {}", exception.what());
}
return boost::none;
}

6
src/video_core/renderer_opengl/gl_shader_gen.cpp
View File

@ -27,10 +27,13 @@ out gl_PerVertex {
vec4 gl_Position;
};
out vec4 position;
void main() {
exec_shader();
}
gl_Position = position;
}
)";
out += program.first;
return {out, program.second};
@ -46,6 +49,7 @@ ProgramResult GenerateFragmentShader(const ShaderSetup& setup, const MaxwellFSCo
.get_value_or({});
out += R"(
in vec4 position;
out vec4 color;
uniform sampler2D tex[32];