shader_bytecode: Decode instructions based on bit strings.

This commit is contained in:
bunnei 2018-04-20 20:49:05 -04:00
parent 8ac3a3f45e
commit 9f6d305eab
2 changed files with 201 additions and 205 deletions

View File

@ -4,10 +4,16 @@
#pragma once
#include <bitset>
#include <cstring>
#include <map>
#include <string>
#include <vector>
#include <boost/optional.hpp>
#include "common/bit_field.h"
#include "common/common_types.h"
namespace Tegra {
namespace Shader {
@ -89,188 +95,12 @@ union Uniform {
BitField<34, 5, u64> index;
};
union OpCode {
enum class Id : u64 {
TEXS = 0x6C,
IPA = 0xE0,
FMUL32_IMM = 0x1E,
FFMA_IMM = 0x65,
FFMA_CR = 0x93,
FFMA_RC = 0xA3,
FFMA_RR = 0xB3,
FADD_C = 0x98B,
FMUL_C = 0x98D,
MUFU = 0xA10,
FADD_R = 0xB8B,
FMUL_R = 0xB8D,
LD_A = 0x1DFB,
ST_A = 0x1DFE,
FSETP_R = 0x5BB,
FSETP_C = 0x4BB,
FSETP_IMM = 0x36B,
FSETP_NEG_IMM = 0x37B,
EXIT = 0xE30,
KIL = 0xE33,
FMUL_IMM = 0x70D,
FMUL_IMM_x = 0x72D,
FADD_IMM = 0x70B,
FADD_IMM_x = 0x72B,
};
enum class Type {
Trivial,
Arithmetic,
Ffma,
Flow,
Memory,
FloatPredicate,
Unknown,
};
struct Info {
Type type;
std::string name;
};
OpCode() = default;
constexpr OpCode(Id value) : value(static_cast<u64>(value)) {}
constexpr OpCode(u64 value) : value{value} {}
constexpr Id EffectiveOpCode() const {
switch (op1) {
case Id::TEXS:
return op1;
}
switch (op2) {
case Id::IPA:
case Id::FMUL32_IMM:
return op2;
}
switch (op3) {
case Id::FFMA_IMM:
case Id::FFMA_CR:
case Id::FFMA_RC:
case Id::FFMA_RR:
return op3;
}
switch (op4) {
case Id::EXIT:
case Id::FSETP_R:
case Id::FSETP_C:
case Id::KIL:
return op4;
case Id::FSETP_IMM:
case Id::FSETP_NEG_IMM:
return Id::FSETP_IMM;
}
switch (op5) {
case Id::MUFU:
case Id::LD_A:
case Id::ST_A:
case Id::FADD_R:
case Id::FADD_C:
case Id::FMUL_R:
case Id::FMUL_C:
return op5;
case Id::FMUL_IMM:
case Id::FMUL_IMM_x:
return Id::FMUL_IMM;
case Id::FADD_IMM:
case Id::FADD_IMM_x:
return Id::FADD_IMM;
}
return static_cast<Id>(value);
}
static const Info& GetInfo(const OpCode& opcode) {
static const std::map<Id, Info> info_table{BuildInfoTable()};
const auto& search{info_table.find(opcode.EffectiveOpCode())};
if (search != info_table.end()) {
return search->second;
}
static const Info unknown{Type::Unknown, "UNK"};
return unknown;
}
constexpr operator Id() const {
return static_cast<Id>(value);
}
constexpr OpCode operator<<(size_t bits) const {
return value << bits;
}
constexpr OpCode operator>>(size_t bits) const {
return value >> bits;
}
template <typename T>
constexpr u64 operator-(const T& oth) const {
return value - oth;
}
constexpr u64 operator&(const OpCode& oth) const {
return value & oth.value;
}
constexpr u64 operator~() const {
return ~value;
}
static std::map<Id, Info> BuildInfoTable() {
std::map<Id, Info> info_table;
info_table[Id::TEXS] = {Type::Memory, "texs"};
info_table[Id::LD_A] = {Type::Memory, "ld_a"};
info_table[Id::ST_A] = {Type::Memory, "st_a"};
info_table[Id::MUFU] = {Type::Arithmetic, "mufu"};
info_table[Id::FFMA_IMM] = {Type::Ffma, "ffma_imm"};
info_table[Id::FFMA_CR] = {Type::Ffma, "ffma_cr"};
info_table[Id::FFMA_RC] = {Type::Ffma, "ffma_rc"};
info_table[Id::FFMA_RR] = {Type::Ffma, "ffma_rr"};
info_table[Id::FADD_R] = {Type::Arithmetic, "fadd_r"};
info_table[Id::FADD_C] = {Type::Arithmetic, "fadd_c"};
info_table[Id::FADD_IMM] = {Type::Arithmetic, "fadd_imm"};
info_table[Id::FMUL_R] = {Type::Arithmetic, "fmul_r"};
info_table[Id::FMUL_C] = {Type::Arithmetic, "fmul_c"};
info_table[Id::FMUL_IMM] = {Type::Arithmetic, "fmul_imm"};
info_table[Id::FMUL32_IMM] = {Type::Arithmetic, "fmul32_imm"};
info_table[Id::FSETP_C] = {Type::FloatPredicate, "fsetp_c"};
info_table[Id::FSETP_R] = {Type::FloatPredicate, "fsetp_r"};
info_table[Id::FSETP_IMM] = {Type::FloatPredicate, "fsetp_imm"};
info_table[Id::EXIT] = {Type::Trivial, "exit"};
info_table[Id::IPA] = {Type::Trivial, "ipa"};
info_table[Id::KIL] = {Type::Flow, "kil"};
return info_table;
}
BitField<57, 7, Id> op1;
BitField<56, 8, Id> op2;
BitField<55, 9, Id> op3;
BitField<52, 12, Id> op4;
BitField<51, 13, Id> op5;
u64 value{};
};
static_assert(sizeof(OpCode) == 0x8, "Incorrect structure size");
} // namespace Shader
} // namespace Tegra
namespace std {
// TODO(bunne): The below is forbidden by the C++ standard, but works fine. See #330.
// TODO(bunnei): The below is forbidden by the C++ standard, but works fine. See #330.
template <>
struct make_unsigned<Tegra::Shader::Attribute> {
using type = Tegra::Shader::Attribute;
@ -281,11 +111,6 @@ struct make_unsigned<Tegra::Shader::Register> {
using type = Tegra::Shader::Register;
};
template <>
struct make_unsigned<Tegra::Shader::OpCode> {
using type = Tegra::Shader::OpCode;
};
} // namespace std
namespace Tegra {
@ -324,11 +149,12 @@ enum class SubOp : u64 {
union Instruction {
Instruction& operator=(const Instruction& instr) {
hex = instr.hex;
value = instr.value;
return *this;
}
OpCode opcode;
constexpr Instruction(u64 value) : value{value} {}
BitField<0, 8, Register> gpr0;
BitField<8, 8, Register> gpr8;
union {
@ -340,6 +166,7 @@ union Instruction {
BitField<20, 7, SubOp> sub_op;
BitField<28, 8, Register> gpr28;
BitField<39, 8, Register> gpr39;
BitField<48, 16, u64> opcode;
union {
BitField<20, 19, u64> imm20_19;
@ -395,11 +222,171 @@ union Instruction {
Uniform uniform;
Sampler sampler;
u64 hex;
u64 value;
};
static_assert(sizeof(Instruction) == 0x8, "Incorrect structure size");
static_assert(std::is_standard_layout<Instruction>::value,
"Structure does not have standard layout");
class OpCode {
public:
enum class Id {
KIL,
LD_A,
ST_A,
TEXS,
EXIT,
IPA,
FFMA_IMM,
FFMA_CR,
FFMA_RC,
FFMA_RR,
FADD_C,
FADD_R,
FADD_IMM,
FMUL_C,
FMUL_R,
FMUL_IMM,
FMUL32_IMM,
MUFU,
FSETP_R,
FSETP_C,
FSETP_IMM,
};
enum class Type {
Trivial,
Arithmetic,
Ffma,
Flow,
Memory,
FloatPredicate,
Unknown,
};
class Matcher {
public:
Matcher(const char* const name, u16 mask, u16 expected, OpCode::Id id, OpCode::Type type)
: name{name}, mask{mask}, expected{expected}, id{id}, type{type} {}
const char* GetName() const {
return name;
}
u16 GetMask() const {
return mask;
}
Id GetId() const {
return id;
}
Type GetType() const {
return type;
}
/**
* Tests to see if the given instruction is the instruction this matcher represents.
* @param instruction The instruction to test
* @returns true if the given instruction matches.
*/
bool Matches(u16 instruction) const {
return (instruction & mask) == expected;
}
private:
const char* name;
u16 mask;
u16 expected;
Id id;
Type type;
};
static boost::optional<const Matcher&> Decode(Instruction instr) {
static const auto table{GetDecodeTable()};
const auto matches_instruction = [instr](const auto& matcher) {
return matcher.Matches(static_cast<u16>(instr.opcode));
};
auto iter = std::find_if(table.begin(), table.end(), matches_instruction);
return iter != table.end() ? boost::optional<const Matcher&>(*iter) : boost::none;
}
private:
struct Detail {
private:
static constexpr size_t opcode_bitsize = 16;
/**
* Generates the mask and the expected value after masking from a given bitstring.
* A '0' in a bitstring indicates that a zero must be present at that bit position.
* A '1' in a bitstring indicates that a one must be present at that bit position.
*/
static auto GetMaskAndExpect(const char* const bitstring) {
u16 mask = 0, expect = 0;
for (size_t i = 0; i < opcode_bitsize; i++) {
const size_t bit_position = opcode_bitsize - i - 1;
switch (bitstring[i]) {
case '0':
mask |= 1 << bit_position;
break;
case '1':
expect |= 1 << bit_position;
mask |= 1 << bit_position;
break;
default:
// Ignore
break;
}
}
return std::make_tuple(mask, expect);
}
public:
/// Creates a matcher that can match and parse instructions based on bitstring.
static auto GetMatcher(const char* const bitstring, OpCode::Id op, OpCode::Type type,
const char* const name) {
const auto mask_expect = GetMaskAndExpect(bitstring);
return Matcher(name, std::get<0>(mask_expect), std::get<1>(mask_expect), op, type);
}
};
static std::vector<Matcher> GetDecodeTable() {
std::vector<Matcher> table = {
#define INST(bitstring, op, type, name) Detail::GetMatcher(bitstring, op, type, name)
INST("111000110011----", Id::KIL, Type::Flow, "KIL"),
INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
INST("1110111111110---", Id::ST_A, Type::Memory, "ST_A"),
INST("1101100---------", Id::TEXS, Type::Memory, "TEXS"),
INST("111000110000----", Id::EXIT, Type::Trivial, "EXIT"),
INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
INST("001100101-------", Id::FFMA_IMM, Type::Ffma, "FFMA_IMM"),
INST("010010011-------", Id::FFMA_CR, Type::Ffma, "FFMA_CR"),
INST("010100011-------", Id::FFMA_RC, Type::Ffma, "FFMA_RC"),
INST("010110011-------", Id::FFMA_RR, Type::Ffma, "FFMA_RR"),
INST("0100110001011---", Id::FADD_C, Type::Arithmetic, "FADD_C"),
INST("0101110001011---", Id::FADD_R, Type::Arithmetic, "FADD_R"),
INST("0011100-01011---", Id::FADD_IMM, Type::Arithmetic, "FADD_IMM"),
INST("0100110001101---", Id::FMUL_C, Type::Arithmetic, "FMUL_C"),
INST("0101110001101---", Id::FMUL_R, Type::Arithmetic, "FMUL_R"),
INST("0011100-01101---", Id::FMUL_IMM, Type::Arithmetic, "FMUL_IMM"),
INST("00011110--------", Id::FMUL32_IMM, Type::Arithmetic, "FMUL32_IMM"),
INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
INST("010110111011----", Id::FSETP_R, Type::FloatPredicate, "FSETP_R"),
INST("010010111011----", Id::FSETP_C, Type::FloatPredicate, "FSETP_C"),
INST("0011011-1011----", Id::FSETP_IMM, Type::FloatPredicate, "FSETP_IMM"),
};
#undef INST
std::stable_sort(table.begin(), table.end(), [](const auto& a, const auto& b) {
// If a matcher has more bits in its mask it is more specific, so it
// should come first.
return std::bitset<16>(a.GetMask()).count() > std::bitset<16>(b.GetMask()).count();
});
return table;
}
};
} // namespace Shader
} // namespace Tegra

View File

@ -97,11 +97,12 @@ private:
return exit_method;
for (u32 offset = begin; offset != end && offset != PROGRAM_END; ++offset) {
const Instruction instr = {program_code[offset]};
switch (instr.opcode.EffectiveOpCode()) {
case OpCode::Id::EXIT: {
return exit_method = ExitMethod::AlwaysEnd;
}
if (const auto opcode = OpCode::Decode({program_code[offset]})) {
switch (opcode->GetId()) {
case OpCode::Id::EXIT: {
return exit_method = ExitMethod::AlwaysEnd;
}
}
}
}
return exit_method = ExitMethod::AlwaysReturn;
@ -332,12 +333,20 @@ private:
*/
u32 CompileInstr(u32 offset) {
// Ignore sched instructions when generating code.
if (IsSchedInstruction(offset))
if (IsSchedInstruction(offset)) {
return offset + 1;
}
const Instruction instr = {program_code[offset]};
const auto opcode = OpCode::Decode(instr);
shader.AddLine("// " + std::to_string(offset) + ": " + OpCode::GetInfo(instr.opcode).name);
// Decoding failure
if (!opcode) {
NGLOG_CRITICAL(HW_GPU, "Unhandled instruction: {}", instr.value);
UNREACHABLE();
}
shader.AddLine("// " + std::to_string(offset) + ": " + opcode->GetName());
using Tegra::Shader::Pred;
ASSERT_MSG(instr.pred.full_pred != Pred::NeverExecute,
@ -349,7 +358,7 @@ private:
++shader.scope;
}
switch (OpCode::GetInfo(instr.opcode).type) {
switch (opcode->GetType()) {
case OpCode::Type::Arithmetic: {
std::string dest = GetRegister(instr.gpr0);
std::string op_a = instr.alu.negate_a ? "-" : "";
@ -374,7 +383,7 @@ private:
op_b = "abs(" + op_b + ")";
}
switch (instr.opcode.EffectiveOpCode()) {
switch (opcode->GetId()) {
case OpCode::Id::FMUL_C:
case OpCode::Id::FMUL_R:
case OpCode::Id::FMUL_IMM: {
@ -424,8 +433,8 @@ private:
}
default: {
NGLOG_CRITICAL(HW_GPU, "Unhandled arithmetic instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
static_cast<unsigned>(opcode->GetId()), opcode->GetName(),
instr.value);
UNREACHABLE();
}
}
@ -437,7 +446,7 @@ private:
std::string op_b = instr.ffma.negate_b ? "-" : "";
std::string op_c = instr.ffma.negate_c ? "-" : "";
switch (instr.opcode.EffectiveOpCode()) {
switch (opcode->GetId()) {
case OpCode::Id::FFMA_CR: {
op_b += GetUniform(instr.uniform);
op_c += GetRegister(instr.gpr39);
@ -460,8 +469,8 @@ private:
}
default: {
NGLOG_CRITICAL(HW_GPU, "Unhandled FFMA instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
static_cast<unsigned>(opcode->GetId()), opcode->GetName(),
instr.value);
UNREACHABLE();
}
}
@ -473,7 +482,7 @@ private:
std::string gpr0 = GetRegister(instr.gpr0);
const Attribute::Index attribute = instr.attribute.fmt20.index;
switch (instr.opcode.EffectiveOpCode()) {
switch (opcode->GetId()) {
case OpCode::Id::LD_A: {
ASSERT_MSG(instr.attribute.fmt20.size == 0, "untested");
SetDest(instr.attribute.fmt20.element, gpr0, GetInputAttribute(attribute), 1, 4);
@ -505,8 +514,8 @@ private:
}
default: {
NGLOG_CRITICAL(HW_GPU, "Unhandled memory instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
static_cast<unsigned>(opcode->GetId()), opcode->GetName(),
instr.value);
UNREACHABLE();
}
}
@ -564,7 +573,7 @@ private:
break;
}
default: {
switch (instr.opcode.EffectiveOpCode()) {
switch (opcode->GetId()) {
case OpCode::Id::EXIT: {
ASSERT_MSG(instr.pred.pred_index == static_cast<u64>(Pred::UnusedIndex),
"Predicated exits not implemented");
@ -584,8 +593,8 @@ private:
}
default: {
NGLOG_CRITICAL(HW_GPU, "Unhandled instruction: {} ({}): {}",
static_cast<unsigned>(instr.opcode.EffectiveOpCode()),
OpCode::GetInfo(instr.opcode).name, instr.hex);
static_cast<unsigned>(opcode->GetId()), opcode->GetName(),
instr.value);
UNREACHABLE();
}
}