1
0
mirror of https://github.com/TomHarte/CLK.git synced 2024-11-29 12:50:28 +00:00
CLK/InstructionSets/x86/Implementation/PerformImplementation.hpp
2023-11-14 16:23:24 -05:00

533 lines
21 KiB
C++
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

//
//
// PerformImplementation.hpp
// Clock Signal
//
// Created by Thomas Harte on 05/10/2023.
// Copyright © 2023 Thomas Harte. All rights reserved.
//
#ifndef PerformImplementation_h
#define PerformImplementation_h
#include "Arithmetic.hpp"
#include "BCD.hpp"
#include "FlowControl.hpp"
#include "InOut.hpp"
#include "LoadStore.hpp"
#include "Logical.hpp"
#include "Repetition.hpp"
#include "Resolver.hpp"
#include "ShiftRoll.hpp"
#include "Stack.hpp"
#include "../Interrupts.hpp"
#include "../AccessType.hpp"
//
// Comments throughout headers above come from the 1997 edition of the
// Intel Architecture Software Developers Manual; that year all such
// definitions still fitted within a single volume, Volume 2.
//
// Order Number 243191; e.g. https://www.ardent-tool.com/CPU/docs/Intel/IA/243191-002.pdf
//
namespace InstructionSet::x86 {
template <
DataSize data_size,
AddressSize address_size,
typename InstructionT,
typename ContextT
> void perform(
const InstructionT &instruction,
ContextT &context
) {
using IntT = typename DataSizeType<data_size>::type;
using AddressT = typename AddressSizeType<address_size>::type;
// Establish source() and destination() shorthands to fetch data if necessary.
//
// C++17, which this project targets at the time of writing, does not provide templatised lambdas.
// So the following division is in part a necessity.
//
// (though GCC offers C++20 syntax as an extension, and Clang seems to follow along, so maybe I'm overthinking)
IntT immediate;
const auto source_r = [&]() -> read_t<IntT> {
return resolve<IntT, AccessType::Read>(
instruction,
instruction.source().source(),
instruction.source(),
context,
nullptr,
&immediate);
};
const auto source_rmw = [&]() -> modify_t<IntT> {
return resolve<IntT, AccessType::ReadModifyWrite>(
instruction,
instruction.source().source(),
instruction.source(),
context,
nullptr,
&immediate);
};
const auto destination_r = [&]() -> read_t<IntT> {
return resolve<IntT, AccessType::Read>(
instruction,
instruction.destination().source(),
instruction.destination(),
context,
nullptr,
&immediate);
};
const auto destination_w = [&]() -> write_t<IntT> {
return resolve<IntT, AccessType::Write>(
instruction,
instruction.destination().source(),
instruction.destination(),
context,
nullptr,
&immediate);
};
const auto destination_rmw = [&]() -> modify_t<IntT> {
return resolve<IntT, AccessType::ReadModifyWrite>(
instruction,
instruction.destination().source(),
instruction.destination(),
context,
nullptr,
&immediate);
};
// Performs a displacement jump only if @c condition is true.
const auto jcc = [&](bool condition) {
Primitive::jump(
condition,
instruction.displacement(),
context);
};
const auto shift_count = [&]() -> uint8_t {
static constexpr uint8_t mask = (ContextT::model != Model::i8086) ? 0x1f : 0xff;
switch(instruction.source().source()) {
case Source::None: return 1;
case Source::Immediate: return uint8_t(instruction.operand()) & mask;
default: return context.registers.cl() & mask;
}
};
// Some instructions use a pair of registers as an extended accumulator — DX:AX or EDX:EAX.
// The two following return the high and low parts of that pair; they also work in Byte mode to return AH:AL,
// i.e. AX split into high and low parts.
const auto pair_high = [&]() -> IntT& {
if constexpr (data_size == DataSize::Byte) return context.registers.ah();
else if constexpr (data_size == DataSize::Word) return context.registers.dx();
else if constexpr (data_size == DataSize::DWord) return context.registers.edx();
};
const auto pair_low = [&]() -> IntT& {
if constexpr (data_size == DataSize::Byte) return context.registers.al();
else if constexpr (data_size == DataSize::Word) return context.registers.ax();
else if constexpr (data_size == DataSize::DWord) return context.registers.eax();
};
// For the string operations, evaluate to either SI and DI or ESI and EDI, depending on the address size.
const auto eSI = [&]() -> AddressT& {
if constexpr (std::is_same_v<AddressT, uint16_t>) {
return context.registers.si();
} else {
return context.registers.esi();
}
};
const auto eDI = [&]() -> AddressT& {
if constexpr (std::is_same_v<AddressT, uint16_t>) {
return context.registers.di();
} else {
return context.registers.edi();
}
};
// For counts, provide either eCX or CX depending on address size.
const auto eCX = [&]() -> AddressT& {
if constexpr (std::is_same_v<AddressT, uint16_t>) {
return context.registers.cx();
} else {
return context.registers.ecx();
}
};
// Gets the port for an IN or OUT; these are always 16-bit.
const auto port = [&](Source source) -> uint16_t {
switch(source) {
case Source::DirectAddress: return instruction.offset();
default: return context.registers.dx();
}
};
// Guide to the below:
//
// * use hard-coded register names where appropriate, otherwise use the source_X() and destination_X() lambdas;
// * return directly if there is definitely no possible write back to RAM;
// * break if there's a chance of writeback.
switch(instruction.operation()) {
default:
assert(false);
case Operation::ESC:
case Operation::NOP: return;
case Operation::AAM: Primitive::aam(context.registers.axp(), instruction.operand(), context); return;
case Operation::AAD: Primitive::aad(context.registers.axp(), instruction.operand(), context); return;
case Operation::AAA: Primitive::aaas<true>(context.registers.axp(), context); return;
case Operation::AAS: Primitive::aaas<false>(context.registers.axp(), context); return;
case Operation::DAA: Primitive::daas<true>(context.registers.al(), context); return;
case Operation::DAS: Primitive::daas<false>(context.registers.al(), context); return;
case Operation::CBW: Primitive::cbw(pair_low()); return;
case Operation::CWD: Primitive::cwd(pair_high(), pair_low()); return;
case Operation::HLT: context.flow_controller.halt(); return;
case Operation::WAIT: context.flow_controller.wait(); return;
case Operation::ADC:
Primitive::add<true, IntT>(destination_rmw(), source_r(), context);
break;
case Operation::ADD:
Primitive::add<false, IntT>(destination_rmw(), source_r(), context);
break;
case Operation::SBB:
Primitive::sub<true, AccessType::ReadModifyWrite, IntT>(destination_rmw(), source_r(), context);
break;
case Operation::SUB:
Primitive::sub<false, AccessType::ReadModifyWrite, IntT>(destination_rmw(), source_r(), context);
break;
case Operation::CMP:
Primitive::sub<false, AccessType::Read, IntT>(destination_r(), source_r(), context);
return;
case Operation::TEST:
Primitive::test<IntT>(destination_r(), source_r(), context);
return;
case Operation::MUL: Primitive::mul<IntT>(pair_high(), pair_low(), source_r(), context); return;
case Operation::IMUL_1: Primitive::imul<IntT>(pair_high(), pair_low(), source_r(), context); return;
case Operation::DIV: Primitive::div<IntT>(pair_high(), pair_low(), source_r(), context); return;
case Operation::IDIV: Primitive::idiv<false, IntT>(pair_high(), pair_low(), source_r(), context); return;
case Operation::IDIV_REP:
if constexpr (ContextT::model == Model::i8086) {
Primitive::idiv<true, IntT>(pair_high(), pair_low(), source_r(), context);
break;
} else {
static_assert(int(Operation::IDIV_REP) == int(Operation::LEAVE));
if constexpr (std::is_same_v<IntT, uint16_t> || std::is_same_v<IntT, uint32_t>) {
Primitive::leave<IntT>();
}
}
return;
case Operation::INC: Primitive::inc<IntT>(destination_rmw(), context); break;
case Operation::DEC: Primitive::dec<IntT>(destination_rmw(), context); break;
case Operation::AND: Primitive::and_<IntT>(destination_rmw(), source_r(), context); break;
case Operation::OR: Primitive::or_<IntT>(destination_rmw(), source_r(), context); break;
case Operation::XOR: Primitive::xor_<IntT>(destination_rmw(), source_r(), context); break;
case Operation::NEG: Primitive::neg<IntT>(source_rmw(), context); break; // TODO: should be a destination.
case Operation::NOT: Primitive::not_<IntT>(source_rmw()); break; // TODO: should be a destination.
case Operation::CALLrel: Primitive::call_relative<AddressT>(instruction.displacement(), context); return;
case Operation::CALLabs: Primitive::call_absolute<IntT>(destination_r(), context); return;
case Operation::CALLfar: Primitive::call_far(instruction, context); return;
case Operation::JMPrel: jcc(true); return;
case Operation::JMPabs: Primitive::jump_absolute<IntT>(destination_r(), context); return;
case Operation::JMPfar: Primitive::jump_far(instruction, context); return;
case Operation::JCXZ: jcc(!eCX()); return;
case Operation::LOOP: Primitive::loop<AddressT>(eCX(), instruction.offset(), context); return;
case Operation::LOOPE: Primitive::loope<AddressT>(eCX(), instruction.offset(), context); return;
case Operation::LOOPNE: Primitive::loopne<AddressT>(eCX(), instruction.offset(), context); return;
case Operation::IRET: Primitive::iret(context); return;
case Operation::RETnear: Primitive::ret_near(instruction, context); return;
case Operation::RETfar: Primitive::ret_far(instruction, context); return;
case Operation::INT: interrupt(instruction.operand(), context); return;
case Operation::INTO: Primitive::into(context); return;
case Operation::SAHF: Primitive::sahf(context.registers.ah(), context); return;
case Operation::LAHF: Primitive::lahf(context.registers.ah(), context); return;
case Operation::LDS:
if constexpr (data_size == DataSize::Word) {
Primitive::ld<Source::DS>(instruction, destination_w(), context);
context.segments.did_update(Source::DS);
}
return;
case Operation::LES:
if constexpr (data_size == DataSize::Word) {
Primitive::ld<Source::ES>(instruction, destination_w(), context);
context.segments.did_update(Source::ES);
}
return;
case Operation::LEA: Primitive::lea<IntT>(instruction, destination_w(), context); return;
case Operation::MOV:
Primitive::mov<IntT>(destination_w(), source_r());
if constexpr (std::is_same_v<IntT, uint16_t>) {
context.segments.did_update(instruction.destination().source());
}
break;
case Operation::JO: jcc(context.flags.template condition<Condition::Overflow>()); return;
case Operation::JNO: jcc(!context.flags.template condition<Condition::Overflow>()); return;
case Operation::JB: jcc(context.flags.template condition<Condition::Below>()); return;
case Operation::JNB: jcc(!context.flags.template condition<Condition::Below>()); return;
case Operation::JZ: jcc(context.flags.template condition<Condition::Zero>()); return;
case Operation::JNZ: jcc(!context.flags.template condition<Condition::Zero>()); return;
case Operation::JBE: jcc(context.flags.template condition<Condition::BelowOrEqual>()); return;
case Operation::JNBE: jcc(!context.flags.template condition<Condition::BelowOrEqual>()); return;
case Operation::JS: jcc(context.flags.template condition<Condition::Sign>()); return;
case Operation::JNS: jcc(!context.flags.template condition<Condition::Sign>()); return;
case Operation::JP: jcc(!context.flags.template condition<Condition::ParityOdd>()); return;
case Operation::JNP: jcc(context.flags.template condition<Condition::ParityOdd>()); return;
case Operation::JL: jcc(context.flags.template condition<Condition::Less>()); return;
case Operation::JNL: jcc(!context.flags.template condition<Condition::Less>()); return;
case Operation::JLE: jcc(context.flags.template condition<Condition::LessOrEqual>()); return;
case Operation::JNLE: jcc(!context.flags.template condition<Condition::LessOrEqual>()); return;
case Operation::RCL: Primitive::rcl<IntT>(destination_rmw(), shift_count(), context); break;
case Operation::RCR: Primitive::rcr<IntT>(destination_rmw(), shift_count(), context); break;
case Operation::ROL: Primitive::rol<IntT>(destination_rmw(), shift_count(), context); break;
case Operation::ROR: Primitive::ror<IntT>(destination_rmw(), shift_count(), context); break;
case Operation::SAL: Primitive::sal<IntT>(destination_rmw(), shift_count(), context); break;
case Operation::SAR: Primitive::sar<IntT>(destination_rmw(), shift_count(), context); break;
case Operation::SHR: Primitive::shr<IntT>(destination_rmw(), shift_count(), context); break;
case Operation::CLC: Primitive::clc(context); return;
case Operation::CLD: Primitive::cld(context); return;
case Operation::CLI: Primitive::cli(context); return;
case Operation::STC: Primitive::stc(context); return;
case Operation::STD: Primitive::std(context); return;
case Operation::STI: Primitive::sti(context); return;
case Operation::CMC: Primitive::cmc(context); return;
case Operation::XCHG: Primitive::xchg<IntT>(destination_rmw(), source_rmw()); break;
case Operation::SALC: Primitive::salc(context.registers.al(), context); return;
case Operation::SETMO:
if constexpr (ContextT::model == Model::i8086) {
Primitive::setmo<IntT>(destination_w(), context);
break;
} else {
static_assert(int(Operation::SETMO) == int(Operation::ENTER));
Primitive::enter<IntT>(instruction, context);
}
return;
case Operation::SETMOC:
if constexpr (ContextT::model == Model::i8086) {
// Test CL out here to avoid taking a reference to memory if
// no write is going to occur.
if(context.registers.cl()) {
Primitive::setmo<IntT>(destination_w(), context);
}
break;
} else {
static_assert(int(Operation::SETMOC) == int(Operation::BOUND));
Primitive::bound<IntT>(instruction, destination_r(), source_r(), context);
}
return;
case Operation::OUT: Primitive::out<IntT>(port(instruction.destination().source()), pair_low(), context); return;
case Operation::IN: Primitive::in<IntT>(port(instruction.source().source()), pair_low(), context); return;
case Operation::XLAT: Primitive::xlat<AddressT>(instruction, context); return;
case Operation::POP:
destination_w() = Primitive::pop<IntT, false>(context);
if constexpr (std::is_same_v<IntT, uint16_t>) {
context.segments.did_update(instruction.destination().source());
}
break;
case Operation::PUSH:
Primitive::push<IntT, false>(source_rmw(), context); // PUSH SP modifies SP before pushing it;
// hence PUSH is sometimes read-modify-write.
break;
case Operation::POPF:
if constexpr (std::is_same_v<IntT, uint16_t> || std::is_same_v<IntT, uint32_t>) {
Primitive::popf(context);
}
return;
case Operation::PUSHF:
if constexpr (std::is_same_v<IntT, uint16_t> || std::is_same_v<IntT, uint32_t>) {
Primitive::pushf(context);
}
return;
case Operation::POPA:
if constexpr (std::is_same_v<IntT, uint16_t> || std::is_same_v<IntT, uint32_t>) {
Primitive::popa<IntT>(context);
}
return;
case Operation::PUSHA:
if constexpr (std::is_same_v<IntT, uint16_t> || std::is_same_v<IntT, uint32_t>) {
Primitive::pusha<IntT>(context);
}
return;
case Operation::CMPS:
Primitive::cmps<IntT, AddressT, Repetition::None>(instruction, eCX(), eSI(), eDI(), context);
return;
case Operation::CMPS_REPE:
Primitive::cmps<IntT, AddressT, Repetition::RepE>(instruction, eCX(), eSI(), eDI(), context);
return;
case Operation::CMPS_REPNE:
Primitive::cmps<IntT, AddressT, Repetition::RepNE>(instruction, eCX(), eSI(), eDI(), context);
return;
case Operation::SCAS:
Primitive::scas<IntT, AddressT, Repetition::None>(eCX(), eDI(), pair_low(), context);
return;
case Operation::SCAS_REPE:
Primitive::scas<IntT, AddressT, Repetition::RepE>(eCX(), eDI(), pair_low(), context);
return;
case Operation::SCAS_REPNE:
Primitive::scas<IntT, AddressT, Repetition::RepNE>(eCX(), eDI(), pair_low(), context);
return;
case Operation::LODS:
Primitive::lods<IntT, AddressT, Repetition::None>(instruction, eCX(), eSI(), pair_low(), context);
return;
case Operation::LODS_REP:
Primitive::lods<IntT, AddressT, Repetition::Rep>(instruction, eCX(), eSI(), pair_low(), context);
return;
case Operation::MOVS:
Primitive::movs<IntT, AddressT, Repetition::None>(instruction, eCX(), eSI(), eDI(), context);
break;
case Operation::MOVS_REP:
Primitive::movs<IntT, AddressT, Repetition::Rep>(instruction, eCX(), eSI(), eDI(), context);
break;
case Operation::STOS:
Primitive::stos<IntT, AddressT, Repetition::None>(eCX(), eDI(), pair_low(), context);
break;
case Operation::STOS_REP:
Primitive::stos<IntT, AddressT, Repetition::Rep>(eCX(), eDI(), pair_low(), context);
break;
case Operation::OUTS:
Primitive::outs<IntT, AddressT, Repetition::None>(instruction, eCX(), context.registers.dx(), eSI(), context);
return;
case Operation::OUTS_REP:
Primitive::outs<IntT, AddressT, Repetition::Rep>(instruction, eCX(), context.registers.dx(), eSI(), context);
return;
case Operation::INS:
Primitive::ins<IntT, AddressT, Repetition::None>(eCX(), context.registers.dx(), eDI(), context);
break;
case Operation::INS_REP:
Primitive::ins<IntT, AddressT, Repetition::Rep>(eCX(), context.registers.dx(), eDI(), context);
break;
}
// Write to memory if required to complete this operation.
//
// This is not currently handled via RAII because of the amount of context that would need to place onto the stack;
// instead code has been set up to make sure there is only at most one writeable target on loan for potential
// write back. I might flip-flop on this, especially if I can verify whether extra stack context is easily
// optimised out.
context.memory.template write_back<IntT>();
}
//
// Public function; just a trampoline into a version of perform templated on data and address size.
//
// Which, yes, means there's an outer switch leading to an inner switch, which could be reduced to one big switch.
// It'd be a substantial effort to find the most neat expression of that, I think, so it is not currently done.
//
template <
typename InstructionT,
typename ContextT
> void perform(
const InstructionT &instruction,
ContextT &context
) {
auto size = [](DataSize operation_size, AddressSize address_size) constexpr -> int {
return int(operation_size) + (int(address_size) << 2);
};
// Dispatch to a function specialised on data and address size.
switch(size(instruction.operation_size(), instruction.address_size())) {
// 16-bit combinations.
case size(DataSize::Byte, AddressSize::b16):
perform<DataSize::Byte, AddressSize::b16>(instruction, context);
return;
case size(DataSize::Word, AddressSize::b16):
perform<DataSize::Word, AddressSize::b16>(instruction, context);
return;
// 32-bit combinations.
//
// The if constexprs below ensure that `perform` isn't compiled for incompatible data or address size and
// model combinations. So if a caller nominates a 16-bit model it can supply registers and memory objects
// that don't implement 32-bit registers or accesses.
case size(DataSize::Byte, AddressSize::b32):
if constexpr (is_32bit(ContextT::model)) {
perform<DataSize::Byte, AddressSize::b32>(instruction, context);
return;
}
break;
case size(DataSize::Word, AddressSize::b32):
if constexpr (is_32bit(ContextT::model)) {
perform<DataSize::Word, AddressSize::b32>(instruction, context);
return;
}
break;
case size(DataSize::DWord, AddressSize::b16):
if constexpr (is_32bit(ContextT::model)) {
perform<DataSize::DWord, AddressSize::b16>(instruction, context);
return;
}
break;
case size(DataSize::DWord, AddressSize::b32):
if constexpr (is_32bit(ContextT::model)) {
perform<DataSize::DWord, AddressSize::b32>(instruction, context);
return;
}
break;
default: break;
}
// This is reachable only if the data and address size combination in use isn't available
// on the processor model nominated.
assert(false);
}
template <
typename ContextT
> void interrupt(
int index,
ContextT &context
) {
const uint32_t address = static_cast<uint32_t>(index) << 2;
context.memory.preauthorise_read(address, sizeof(uint16_t) * 2);
context.memory.preauthorise_stack_write(sizeof(uint16_t) * 3);
const uint16_t ip = context.memory.template access<uint16_t, AccessType::PreauthorisedRead>(address);
const uint16_t cs = context.memory.template access<uint16_t, AccessType::PreauthorisedRead>(address + 2);
auto flags = context.flags.get();
Primitive::push<uint16_t, true>(flags, context);
context.flags.template set_from<Flag::Interrupt, Flag::Trap>(0);
// Push CS and IP.
Primitive::push<uint16_t, true>(context.registers.cs(), context);
Primitive::push<uint16_t, true>(context.registers.ip(), context);
// Set new destination.
context.flow_controller.jump(cs, ip);
}
}
#endif /* PerformImplementation_h */