1
0
mirror of https://github.com/TomHarte/CLK.git synced 2024-11-10 23:05:01 +00:00
CLK/Processors/Z80/Implementation/Z80Implementation.hpp
2019-03-09 21:16:11 -05:00

1063 lines
32 KiB
C++

//
// Z80Implementation.hpp
// Clock Signal
//
// Created by Thomas Harte on 01/09/2017.
// Copyright 2017 Thomas Harte. All rights reserved.
//
template < class T,
bool uses_bus_request,
bool uses_wait_line> Processor <T, uses_bus_request, uses_wait_line>
::Processor(T &bus_handler) :
bus_handler_(bus_handler) {
install_default_instruction_set();
}
template < class T,
bool uses_bus_request,
bool uses_wait_line> void Processor <T, uses_bus_request, uses_wait_line>
::run_for(const HalfCycles cycles) {
#define advance_operation() \
pc_increment_ = 1; \
if(last_request_status_) { \
halt_mask_ = 0xff; \
if(last_request_status_ & (Interrupt::PowerOn | Interrupt::Reset)) { \
request_status_ &= ~Interrupt::PowerOn; \
scheduled_program_counter_ = reset_program_.data(); \
} else if(last_request_status_ & Interrupt::NMI) { \
request_status_ &= ~Interrupt::NMI; \
scheduled_program_counter_ = nmi_program_.data(); \
} else if(last_request_status_ & Interrupt::IRQ) { \
scheduled_program_counter_ = irq_program_[interrupt_mode_].data(); \
} \
} else { \
current_instruction_page_ = &base_page_; \
scheduled_program_counter_ = base_page_.fetch_decode_execute_data; \
}
number_of_cycles_ += cycles;
if(!scheduled_program_counter_) {
advance_operation();
}
while(1) {
do_bus_acknowledge:
while(uses_bus_request && bus_request_line_) {
static PartialMachineCycle bus_acknowledge_cycle = {PartialMachineCycle::BusAcknowledge, HalfCycles(2), nullptr, nullptr, false};
number_of_cycles_ -= bus_handler_.perform_machine_cycle(bus_acknowledge_cycle) + HalfCycles(1);
if(!number_of_cycles_) {
bus_handler_.flush();
return;
}
}
while(true) {
const MicroOp *const operation = scheduled_program_counter_;
scheduled_program_counter_++;
#define set_did_compute_flags() \
flag_adjustment_history_ |= 1;
#define set_parity(v) \
parity_overflow_result_ = static_cast<uint8_t>(v^1);\
parity_overflow_result_ ^= parity_overflow_result_ >> 4;\
parity_overflow_result_ ^= parity_overflow_result_ << 2;\
parity_overflow_result_ ^= parity_overflow_result_ >> 1;
switch(operation->type) {
case MicroOp::BusOperation:
if(number_of_cycles_ < operation->machine_cycle.length) {
scheduled_program_counter_--;
bus_handler_.flush();
return;
}
if(uses_wait_line && operation->machine_cycle.was_requested) {
if(wait_line_) {
scheduled_program_counter_--;
} else {
continue;
}
}
number_of_cycles_ -= operation->machine_cycle.length;
last_request_status_ = request_status_;
number_of_cycles_ -= bus_handler_.perform_machine_cycle(operation->machine_cycle);
if(uses_bus_request && bus_request_line_) goto do_bus_acknowledge;
break;
case MicroOp::MoveToNextProgram:
advance_operation();
break;
case MicroOp::DecodeOperation:
refresh_addr_ = ir_;
ir_.halves.low = (ir_.halves.low & 0x80) | ((ir_.halves.low + current_instruction_page_->r_step) & 0x7f);
pc_.full += pc_increment_ & static_cast<uint16_t>(halt_mask_);
scheduled_program_counter_ = current_instruction_page_->instructions[operation_ & halt_mask_];
flag_adjustment_history_ <<= 1;
break;
case MicroOp::DecodeOperationNoRChange:
refresh_addr_ = ir_;
pc_.full += pc_increment_ & static_cast<uint16_t>(halt_mask_);
scheduled_program_counter_ = current_instruction_page_->instructions[operation_ & halt_mask_];
break;
case MicroOp::Increment16: (*static_cast<uint16_t *>(operation->source))++; break;
case MicroOp::IncrementPC: pc_.full += pc_increment_; break;
case MicroOp::Decrement16: (*static_cast<uint16_t *>(operation->source))--; break;
case MicroOp::Move8: *static_cast<uint8_t *>(operation->destination) = *static_cast<uint8_t *>(operation->source); break;
case MicroOp::Move16: *static_cast<uint16_t *>(operation->destination) = *static_cast<uint16_t *>(operation->source); break;
case MicroOp::AssembleAF:
temp16_.halves.high = a_;
temp16_.halves.low = get_flags();
break;
case MicroOp::DisassembleAF:
a_ = temp16_.halves.high;
set_flags(temp16_.halves.low);
break;
// MARK: - Logical
#define set_logical_flags(hf) \
sign_result_ = zero_result_ = bit53_result_ = a_; \
set_parity(a_); \
half_carry_result_ = hf; \
subtract_flag_ = 0; \
carry_result_ = 0; \
set_did_compute_flags();
case MicroOp::And:
a_ &= *static_cast<uint8_t *>(operation->source);
set_logical_flags(Flag::HalfCarry);
break;
case MicroOp::Or:
a_ |= *static_cast<uint8_t *>(operation->source);
set_logical_flags(0);
break;
case MicroOp::Xor:
a_ ^= *static_cast<uint8_t *>(operation->source);
set_logical_flags(0);
break;
#undef set_logical_flags
case MicroOp::CPL:
a_ ^= 0xff;
subtract_flag_ = Flag::Subtract;
half_carry_result_ = Flag::HalfCarry;
bit53_result_ = a_;
set_did_compute_flags();
break;
case MicroOp::CCF:
half_carry_result_ = static_cast<uint8_t>(carry_result_ << 4);
carry_result_ ^= Flag::Carry;
subtract_flag_ = 0;
if(flag_adjustment_history_&2) {
bit53_result_ = a_;
} else {
bit53_result_ |= a_;
}
set_did_compute_flags();
break;
case MicroOp::SCF:
carry_result_ = Flag::Carry;
half_carry_result_ = 0;
subtract_flag_ = 0;
if(flag_adjustment_history_&2) {
bit53_result_ = a_;
} else {
bit53_result_ |= a_;
}
set_did_compute_flags();
break;
// MARK: - Flow control
case MicroOp::DJNZ:
bc_.halves.high--;
if(!bc_.halves.high) {
advance_operation();
}
break;
case MicroOp::CalculateRSTDestination:
memptr_.full = operation_ & 0x38;
break;
// MARK: - 8-bit arithmetic
#define set_arithmetic_flags(sub, b53) \
sign_result_ = zero_result_ = static_cast<uint8_t>(result); \
carry_result_ = static_cast<uint8_t>(result >> 8); \
half_carry_result_ = static_cast<uint8_t>(half_result); \
parity_overflow_result_ = static_cast<uint8_t>(overflow >> 5); \
subtract_flag_ = sub; \
bit53_result_ = static_cast<uint8_t>(b53); \
set_did_compute_flags();
case MicroOp::CP8: {
const uint8_t value = *static_cast<uint8_t *>(operation->source);
const int result = a_ - value;
const int half_result = (a_&0xf) - (value&0xf);
// overflow for a subtraction is when the signs were originally
// different and the result is different again
const int overflow = (value^a_) & (result^a_);
// the 5 and 3 flags come from the operand, atypically
set_arithmetic_flags(Flag::Subtract, value);
} break;
case MicroOp::SUB8: {
const uint8_t value = *static_cast<uint8_t *>(operation->source);
const int result = a_ - value;
const int half_result = (a_&0xf) - (value&0xf);
// overflow for a subtraction is when the signs were originally
// different and the result is different again
const int overflow = (value^a_) & (result^a_);
a_ = static_cast<uint8_t>(result);
set_arithmetic_flags(Flag::Subtract, result);
} break;
case MicroOp::SBC8: {
const uint8_t value = *static_cast<uint8_t *>(operation->source);
const int result = a_ - value - (carry_result_ & Flag::Carry);
const int half_result = (a_&0xf) - (value&0xf) - (carry_result_ & Flag::Carry);
// overflow for a subtraction is when the signs were originally
// different and the result is different again
const int overflow = (value^a_) & (result^a_);
a_ = static_cast<uint8_t>(result);
set_arithmetic_flags(Flag::Subtract, result);
} break;
case MicroOp::ADD8: {
const uint8_t value = *static_cast<uint8_t *>(operation->source);
const int result = a_ + value;
const int half_result = (a_&0xf) + (value&0xf);
// overflow for addition is when the signs were originally
// the same and the result is different
const int overflow = ~(value^a_) & (result^a_);
a_ = static_cast<uint8_t>(result);
set_arithmetic_flags(0, result);
} break;
case MicroOp::ADC8: {
const uint8_t value = *static_cast<uint8_t *>(operation->source);
const int result = a_ + value + (carry_result_ & Flag::Carry);
const int half_result = (a_&0xf) + (value&0xf) + (carry_result_ & Flag::Carry);
// overflow for addition is when the signs were originally
// the same and the result is different
const int overflow = ~(value^a_) & (result^a_);
a_ = static_cast<uint8_t>(result);
set_arithmetic_flags(0, result);
} break;
#undef set_arithmetic_flags
case MicroOp::NEG: {
const int overflow = (a_ == 0x80);
const int result = -a_;
const int halfResult = -(a_&0xf);
a_ = static_cast<uint8_t>(result);
bit53_result_ = sign_result_ = zero_result_ = a_;
parity_overflow_result_ = overflow ? Flag::Overflow : 0;
subtract_flag_ = Flag::Subtract;
carry_result_ = static_cast<uint8_t>(result >> 8);
half_carry_result_ = static_cast<uint8_t>(halfResult);
set_did_compute_flags();
} break;
case MicroOp::Increment8: {
const uint8_t value = *static_cast<uint8_t *>(operation->source);
const int result = value + 1;
// with an increment, overflow occurs if the sign changes from
// positive to negative
const int overflow = (value ^ result) & ~value;
const int half_result = (value&0xf) + 1;
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>(result);
// sign, zero and 5 & 3 are set directly from the result
bit53_result_ = sign_result_ = zero_result_ = static_cast<uint8_t>(result);
half_carry_result_ = static_cast<uint8_t>(half_result);
parity_overflow_result_ = static_cast<uint8_t>(overflow >> 5);
subtract_flag_ = 0;
set_did_compute_flags();
} break;
case MicroOp::Decrement8: {
const uint8_t value = *static_cast<uint8_t *>(operation->source);
const int result = value - 1;
// with a decrement, overflow occurs if the sign changes from
// negative to positive
const int overflow = (value ^ result) & value;
const int half_result = (value&0xf) - 1;
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>(result);
// sign, zero and 5 & 3 are set directly from the result
bit53_result_ = sign_result_ = zero_result_ = static_cast<uint8_t>(result);
half_carry_result_ = static_cast<uint8_t>(half_result);
parity_overflow_result_ = static_cast<uint8_t>(overflow >> 5);
subtract_flag_ = Flag::Subtract;
set_did_compute_flags();
} break;
case MicroOp::DAA: {
const int lowNibble = a_ & 0xf;
const int highNibble = a_ >> 4;
int amountToAdd = 0;
if(carry_result_ & Flag::Carry) {
amountToAdd = (lowNibble > 0x9 || (half_carry_result_ & Flag::HalfCarry)) ? 0x66 : 0x60;
} else {
if(half_carry_result_ & Flag::HalfCarry) {
if(lowNibble > 0x9)
amountToAdd = (highNibble > 0x8) ? 0x66 : 0x06;
else
amountToAdd = (highNibble > 0x9) ? 0x66 : 0x06;
} else {
if(lowNibble > 0x9)
amountToAdd = (highNibble > 0x8) ? 0x66 : 0x06;
else
amountToAdd = (highNibble > 0x9) ? 0x60 : 0x00;
}
}
if(!(carry_result_ & Flag::Carry)) {
if(lowNibble > 0x9) {
if(highNibble > 0x8) carry_result_ = Flag::Carry;
} else {
if(highNibble > 0x9) carry_result_ = Flag::Carry;
}
}
if(subtract_flag_) {
a_ -= amountToAdd;
half_carry_result_ = ((half_carry_result_ & Flag::HalfCarry) && lowNibble < 0x6) ? Flag::HalfCarry : 0;
} else {
a_ += amountToAdd;
half_carry_result_ = (lowNibble > 0x9) ? Flag::HalfCarry : 0;
}
sign_result_ = zero_result_ = bit53_result_ = a_;
set_parity(a_);
set_did_compute_flags();
} break;
// MARK: - 16-bit arithmetic
case MicroOp::ADD16: {
memptr_.full = *static_cast<uint16_t *>(operation->destination);
const uint16_t sourceValue = *static_cast<uint16_t *>(operation->source);
const uint16_t destinationValue = memptr_.full;
const int result = sourceValue + destinationValue;
const int halfResult = (sourceValue&0xfff) + (destinationValue&0xfff);
bit53_result_ = static_cast<uint8_t>(result >> 8);
carry_result_ = static_cast<uint8_t>(result >> 16);
half_carry_result_ = static_cast<uint8_t>(halfResult >> 8);
subtract_flag_ = 0;
set_did_compute_flags();
*static_cast<uint16_t *>(operation->destination) = static_cast<uint16_t>(result);
memptr_.full++;
} break;
case MicroOp::ADC16: {
memptr_.full = *static_cast<uint16_t *>(operation->destination);
const uint16_t sourceValue = *static_cast<uint16_t *>(operation->source);
const uint16_t destinationValue = memptr_.full;
const int result = sourceValue + destinationValue + (carry_result_ & Flag::Carry);
const int halfResult = (sourceValue&0xfff) + (destinationValue&0xfff) + (carry_result_ & Flag::Carry);
const int overflow = (result ^ destinationValue) & ~(destinationValue ^ sourceValue);
bit53_result_ =
sign_result_ = static_cast<uint8_t>(result >> 8);
zero_result_ = static_cast<uint8_t>(result | sign_result_);
subtract_flag_ = 0;
carry_result_ = static_cast<uint8_t>(result >> 16);
half_carry_result_ = static_cast<uint8_t>(halfResult >> 8);
parity_overflow_result_ = static_cast<uint8_t>(overflow >> 13);
set_did_compute_flags();
*static_cast<uint16_t *>(operation->destination) = static_cast<uint16_t>(result);
memptr_.full++;
} break;
case MicroOp::SBC16: {
memptr_.full = *static_cast<uint16_t *>(operation->destination);
const uint16_t sourceValue = *static_cast<uint16_t *>(operation->source);
const uint16_t destinationValue = memptr_.full;
const int result = destinationValue - sourceValue - (carry_result_ & Flag::Carry);
const int halfResult = (destinationValue&0xfff) - (sourceValue&0xfff) - (carry_result_ & Flag::Carry);
// subtraction, so parity rules are:
// signs of operands were different,
// sign of result is different
const int overflow = (result ^ destinationValue) & (sourceValue ^ destinationValue);
bit53_result_ =
sign_result_ = static_cast<uint8_t>(result >> 8);
zero_result_ = static_cast<uint8_t>(result | sign_result_);
subtract_flag_ = Flag::Subtract;
carry_result_ = static_cast<uint8_t>(result >> 16);
half_carry_result_ = static_cast<uint8_t>(halfResult >> 8);
parity_overflow_result_ = static_cast<uint8_t>(overflow >> 13);
set_did_compute_flags();
*static_cast<uint16_t *>(operation->destination) = static_cast<uint16_t>(result);
memptr_.full++;
} break;
// MARK: - Conditionals
#define decline_conditional() \
if(operation->source) { \
scheduled_program_counter_ = (MicroOp *)operation->source; \
} else { \
advance_operation(); \
}
case MicroOp::TestNZ: if(!zero_result_) { decline_conditional(); } break;
case MicroOp::TestZ: if(zero_result_) { decline_conditional(); } break;
case MicroOp::TestNC: if(carry_result_ & Flag::Carry) { decline_conditional(); } break;
case MicroOp::TestC: if(!(carry_result_ & Flag::Carry)) { decline_conditional(); } break;
case MicroOp::TestPO: if(parity_overflow_result_ & Flag::Parity) { decline_conditional(); } break;
case MicroOp::TestPE: if(!(parity_overflow_result_ & Flag::Parity)) { decline_conditional(); } break;
case MicroOp::TestP: if(sign_result_ & Flag::Sign) { decline_conditional(); } break;
case MicroOp::TestM: if(!(sign_result_ & Flag::Sign)) { decline_conditional(); } break;
#undef decline_conditional
// MARK: - Exchange
#define swap(a, b) temp = a.full; a.full = b.full; b.full = temp;
case MicroOp::ExDEHL: {
uint16_t temp;
swap(de_, hl_);
} break;
case MicroOp::ExAFAFDash: {
const uint8_t a = a_;
const uint8_t f = get_flags();
set_flags(afDash_.halves.low);
a_ = afDash_.halves.high;
afDash_.halves.high = a;
afDash_.halves.low = f;
} break;
case MicroOp::EXX: {
uint16_t temp;
swap(de_, deDash_);
swap(bc_, bcDash_);
swap(hl_, hlDash_);
} break;
#undef swap
// MARK: - Repetition
#define REPEAT(test) \
if(test) { \
pc_.full -= 2; \
} else { \
advance_operation(); \
}
#define LDxR_STEP(dir) \
bc_.full--; \
de_.full += dir; \
hl_.full += dir; \
const uint8_t sum = a_ + temp8_; \
bit53_result_ = static_cast<uint8_t>((sum&0x8) | ((sum & 0x02) << 4)); \
subtract_flag_ = 0; \
half_carry_result_ = 0; \
parity_overflow_result_ = bc_.full ? Flag::Parity : 0; \
set_did_compute_flags();
case MicroOp::LDDR: {
LDxR_STEP(-1);
REPEAT(bc_.full);
} break;
case MicroOp::LDIR: {
LDxR_STEP(1);
REPEAT(bc_.full);
} break;
case MicroOp::LDD: {
LDxR_STEP(-1);
} break;
case MicroOp::LDI: {
LDxR_STEP(1);
} break;
#undef LDxR_STEP
#define CPxR_STEP(dir) \
hl_.full += dir; \
bc_.full--; \
\
uint8_t result = a_ - temp8_; \
const uint8_t halfResult = (a_&0xf) - (temp8_&0xf); \
\
parity_overflow_result_ = bc_.full ? Flag::Parity : 0; \
half_carry_result_ = halfResult; \
subtract_flag_ = Flag::Subtract; \
sign_result_ = zero_result_ = result; \
\
result -= (halfResult >> 4)&1; \
bit53_result_ = static_cast<uint8_t>((result&0x8) | ((result&0x2) << 4)); \
set_did_compute_flags();
case MicroOp::CPDR: {
CPxR_STEP(-1);
REPEAT(bc_.full && sign_result_);
} break;
case MicroOp::CPIR: {
CPxR_STEP(1);
REPEAT(bc_.full && sign_result_);
} break;
case MicroOp::CPD: {
memptr_.full--;
CPxR_STEP(-1);
} break;
case MicroOp::CPI: {
memptr_.full++;
CPxR_STEP(1);
} break;
#undef CPxR_STEP
#define INxR_STEP(dir) \
bc_.halves.high--; \
hl_.full += dir; \
\
sign_result_ = zero_result_ = bit53_result_ = bc_.halves.high; \
subtract_flag_ = (temp8_ >> 6) & Flag::Subtract; \
\
const int next_bc = bc_.halves.low + dir; \
int summation = temp8_ + (next_bc&0xff); \
\
if(summation > 0xff) { \
carry_result_ = Flag::Carry; \
half_carry_result_ = Flag::HalfCarry; \
} else { \
carry_result_ = 0; \
half_carry_result_ = 0; \
} \
\
summation = (summation&7) ^ bc_.halves.high; \
set_parity(summation); \
set_did_compute_flags();
case MicroOp::INDR: {
INxR_STEP(-1);
REPEAT(bc_.halves.high);
} break;
case MicroOp::INIR: {
INxR_STEP(1);
REPEAT(bc_.halves.high);
} break;
case MicroOp::IND: {
memptr_.full = bc_.full - 1;
INxR_STEP(-1);
} break;
case MicroOp::INI: {
memptr_.full = bc_.full + 1;
INxR_STEP(1);
} break;
#undef INxR_STEP
#define OUTxR_STEP(dir) \
bc_.halves.high--; \
hl_.full += dir; \
\
sign_result_ = zero_result_ = bit53_result_ = bc_.halves.high; \
subtract_flag_ = (temp8_ >> 6) & Flag::Subtract; \
\
int summation = temp8_ + hl_.halves.low; \
if(summation > 0xff) { \
carry_result_ = Flag::Carry; \
half_carry_result_ = Flag::HalfCarry; \
} else { \
carry_result_ = half_carry_result_ = 0; \
} \
\
summation = (summation&7) ^ bc_.halves.high; \
set_parity(summation); \
set_did_compute_flags();
case MicroOp::OUT_R:
REPEAT(bc_.halves.high);
break;
case MicroOp::OUTD: {
OUTxR_STEP(-1);
memptr_.full = bc_.full - 1;
} break;
case MicroOp::OUTI: {
OUTxR_STEP(1);
memptr_.full = bc_.full + 1;
} break;
#undef OUTxR_STEP
// MARK: - Bit Manipulation
case MicroOp::BIT: {
const uint8_t result = *static_cast<uint8_t *>(operation->source) & (1 << ((operation_ >> 3)&7));
if(current_instruction_page_->is_indexed || ((operation_&0x07) == 6)) {
bit53_result_ = memptr_.halves.high;
} else {
bit53_result_ = *static_cast<uint8_t *>(operation->source);
}
sign_result_ = zero_result_ = result;
half_carry_result_ = Flag::HalfCarry;
subtract_flag_ = 0;
parity_overflow_result_ = result ? 0 : Flag::Parity;
set_did_compute_flags();
} break;
case MicroOp::RES:
*static_cast<uint8_t *>(operation->source) &= ~(1 << ((operation_ >> 3)&7));
break;
case MicroOp::SET:
*static_cast<uint8_t *>(operation->source) |= (1 << ((operation_ >> 3)&7));
break;
// MARK: - Rotation and shifting
#define set_rotate_flags() \
bit53_result_ = a_; \
carry_result_ = new_carry; \
subtract_flag_ = half_carry_result_ = 0; \
set_did_compute_flags();
case MicroOp::RLA: {
const uint8_t new_carry = a_ >> 7;
a_ = static_cast<uint8_t>((a_ << 1) | (carry_result_ & Flag::Carry));
set_rotate_flags();
} break;
case MicroOp::RRA: {
const uint8_t new_carry = a_ & 1;
a_ = static_cast<uint8_t>((a_ >> 1) | (carry_result_ << 7));
set_rotate_flags();
} break;
case MicroOp::RLCA: {
const uint8_t new_carry = a_ >> 7;
a_ = static_cast<uint8_t>((a_ << 1) | new_carry);
set_rotate_flags();
} break;
case MicroOp::RRCA: {
const uint8_t new_carry = a_ & 1;
a_ = static_cast<uint8_t>((a_ >> 1) | (new_carry << 7));
set_rotate_flags();
} break;
#undef set_rotate_flags
#define set_shift_flags() \
sign_result_ = zero_result_ = bit53_result_ = *static_cast<uint8_t *>(operation->source); \
set_parity(sign_result_); \
half_carry_result_ = 0; \
subtract_flag_ = 0; \
set_did_compute_flags();
case MicroOp::RLC:
carry_result_ = *static_cast<uint8_t *>(operation->source) >> 7;
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>((*static_cast<uint8_t *>(operation->source) << 1) | carry_result_);
set_shift_flags();
break;
case MicroOp::RRC:
carry_result_ = *static_cast<uint8_t *>(operation->source);
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>((*static_cast<uint8_t *>(operation->source) >> 1) | (carry_result_ << 7));
set_shift_flags();
break;
case MicroOp::RL: {
const uint8_t next_carry = *static_cast<uint8_t *>(operation->source) >> 7;
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>((*static_cast<uint8_t *>(operation->source) << 1) | (carry_result_ & Flag::Carry));
carry_result_ = next_carry;
set_shift_flags();
} break;
case MicroOp::RR: {
const uint8_t next_carry = *static_cast<uint8_t *>(operation->source);
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>((*static_cast<uint8_t *>(operation->source) >> 1) | (carry_result_ << 7));
carry_result_ = next_carry;
set_shift_flags();
} break;
case MicroOp::SLA:
carry_result_ = *static_cast<uint8_t *>(operation->source) >> 7;
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>(*static_cast<uint8_t *>(operation->source) << 1);
set_shift_flags();
break;
case MicroOp::SRA:
carry_result_ = *static_cast<uint8_t *>(operation->source);
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>((*static_cast<uint8_t *>(operation->source) >> 1) | (*static_cast<uint8_t *>(operation->source) & 0x80));
set_shift_flags();
break;
case MicroOp::SLL:
carry_result_ = *static_cast<uint8_t *>(operation->source) >> 7;
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>(*static_cast<uint8_t *>(operation->source) << 1) | 1;
set_shift_flags();
break;
case MicroOp::SRL:
carry_result_ = *static_cast<uint8_t *>(operation->source);
*static_cast<uint8_t *>(operation->source) = static_cast<uint8_t>((*static_cast<uint8_t *>(operation->source) >> 1));
set_shift_flags();
break;
#undef set_shift_flags
#define set_decimal_rotate_flags() \
subtract_flag_ = 0; \
half_carry_result_ = 0; \
set_parity(a_); \
bit53_result_ = zero_result_ = sign_result_ = a_; \
set_did_compute_flags();
case MicroOp::RRD: {
memptr_.full = hl_.full + 1;
const uint8_t low_nibble = a_ & 0xf;
a_ = (a_ & 0xf0) | (temp8_ & 0xf);
temp8_ = static_cast<uint8_t>((temp8_ >> 4) | (low_nibble << 4));
set_decimal_rotate_flags();
} break;
case MicroOp::RLD: {
memptr_.full = hl_.full + 1;
const uint8_t low_nibble = a_ & 0xf;
a_ = (a_ & 0xf0) | (temp8_ >> 4);
temp8_ = static_cast<uint8_t>((temp8_ << 4) | low_nibble);
set_decimal_rotate_flags();
} break;
#undef set_decimal_rotate_flags
// MARK: - Interrupt state
case MicroOp::EI:
iff1_ = iff2_ = true;
if(irq_line_) request_status_ |= Interrupt::IRQ;
break;
case MicroOp::DI:
iff1_ = iff2_ = false;
request_status_ &= ~Interrupt::IRQ;
break;
case MicroOp::IM:
switch(operation_ & 0x18) {
case 0x00: interrupt_mode_ = 0; break;
case 0x08: interrupt_mode_ = 0; break; // IM 0/1
case 0x10: interrupt_mode_ = 1; break;
case 0x18: interrupt_mode_ = 2; break;
}
break;
// MARK: - Input
case MicroOp::SetInFlags:
subtract_flag_ = half_carry_result_ = 0;
sign_result_ = zero_result_ = bit53_result_ = *static_cast<uint8_t *>(operation->source);
set_parity(sign_result_);
set_did_compute_flags();
break;
case MicroOp::SetAFlags:
subtract_flag_ = half_carry_result_ = 0;
parity_overflow_result_ = iff2_ ? Flag::Parity : 0;
sign_result_ = zero_result_ = bit53_result_ = a_;
set_did_compute_flags();
break;
case MicroOp::SetZero:
temp8_ = 0;
break;
// MARK: - Special-case Flow
case MicroOp::BeginIRQMode0:
pc_increment_ = 0; // deliberate fallthrough
case MicroOp::BeginIRQ:
iff2_ = iff1_ = false;
request_status_ &= ~Interrupt::IRQ;
temp16_.full = 0x38;
break;
case MicroOp::BeginNMI:
iff2_ = iff1_;
iff1_ = false;
request_status_ &= ~Interrupt::IRQ;
break;
case MicroOp::JumpTo66:
pc_.full = 0x66;
break;
case MicroOp::RETN:
iff1_ = iff2_;
if(irq_line_ && iff1_) request_status_ |= Interrupt::IRQ;
break;
case MicroOp::HALT:
halt_mask_ = 0x00;
break;
// MARK: - Interrupt handling
case MicroOp::Reset:
iff1_ = iff2_ = false;
interrupt_mode_ = 0;
pc_.full = 0;
sp_.full = 0xffff;
a_ = 0xff;
set_flags(0xff);
ir_.full = 0;
break;
// MARK: - Internal bookkeeping
case MicroOp::SetInstructionPage:
current_instruction_page_ = (InstructionPage *)operation->source;
scheduled_program_counter_ = current_instruction_page_->fetch_decode_execute_data;
break;
case MicroOp::CalculateIndexAddress:
memptr_.full = static_cast<uint16_t>(*static_cast<uint16_t *>(operation->source) + (int8_t)temp8_);
break;
case MicroOp::SetAddrAMemptr:
memptr_.full = static_cast<uint16_t>(((*static_cast<uint16_t *>(operation->source) + 1)&0xff) + (a_ << 8));
break;
case MicroOp::IndexedPlaceHolder:
return;
}
#undef set_parity
}
}
}
template < class T,
bool uses_bus_request,
bool uses_wait_line> void Processor <T, uses_bus_request, uses_wait_line>
::set_bus_request_line(bool value) {
assert(uses_bus_request);
bus_request_line_ = value;
}
template < class T,
bool uses_bus_request,
bool uses_wait_line> bool Processor <T, uses_bus_request, uses_wait_line>
::get_bus_request_line() {
return bus_request_line_;
}
template < class T,
bool uses_bus_request,
bool uses_wait_line> void Processor <T, uses_bus_request, uses_wait_line>
::set_wait_line(bool value) {
assert(uses_wait_line);
wait_line_ = value;
}
template < class T,
bool uses_bus_request,
bool uses_wait_line> bool Processor <T, uses_bus_request, uses_wait_line>
::get_wait_line() {
return wait_line_;
}
#define isTerminal(n) (n == MicroOp::MoveToNextProgram || n == MicroOp::DecodeOperation || n == MicroOp::DecodeOperationNoRChange)
template < class T,
bool uses_bus_request,
bool uses_wait_line> void Processor <T, uses_bus_request, uses_wait_line>
::assemble_page(InstructionPage &target, InstructionTable &table, bool add_offsets) {
std::size_t number_of_micro_ops = 0;
std::size_t lengths[256];
// Count number of micro-ops required.
for(int c = 0; c < 256; c++) {
std::size_t length = 0;
while(!isTerminal(table[c][length].type)) length++;
length++;
lengths[c] = length;
number_of_micro_ops += length;
}
// Allocate a landing area.
std::vector<std::size_t> operation_indices;
target.all_operations.reserve(number_of_micro_ops);
target.instructions.resize(256, nullptr);
// Copy in all programs, recording where they go.
std::size_t destination = 0;
for(std::size_t c = 0; c < 256; c++) {
operation_indices.push_back(target.all_operations.size());
for(std::size_t t = 0; t < lengths[c];) {
// Skip zero-length bus cycles.
if(table[c][t].type == MicroOp::BusOperation && table[c][t].machine_cycle.length.as_int() == 0) {
t++;
continue;
}
// Skip optional waits if this instance doesn't use the wait line.
if(table[c][t].machine_cycle.was_requested && !uses_wait_line) {
t++;
continue;
}
// If an index placeholder is hit then drop it, and if offsets aren't being added,
// then also drop the indexing that follows, which is assumed to be everything
// up to and including the next ::CalculateIndexAddress. Coupled to the INDEX() macro.
if(table[c][t].type == MicroOp::IndexedPlaceHolder) {
t++;
if(!add_offsets) {
while(table[c][t].type != MicroOp::CalculateIndexAddress) t++;
t++;
}
}
target.all_operations.emplace_back(table[c][t]);
destination++;
t++;
}
}
// Since the vector won't change again, it's now safe to set pointers.
std::size_t c = 0;
for(std::size_t index : operation_indices) {
target.instructions[c] = &target.all_operations[index];
c++;
}
}
template < class T,
bool uses_bus_request,
bool uses_wait_line> void Processor <T, uses_bus_request, uses_wait_line>
::copy_program(const MicroOp *source, std::vector<MicroOp> &destination) {
std::size_t length = 0;
while(!isTerminal(source[length].type)) length++;
std::size_t pointer = 0;
while(true) {
// TODO: This test is duplicated from assemble_page; can a better factoring be found?
// Skip optional waits if this instance doesn't use the wait line.
if(source[pointer].machine_cycle.was_requested && !uses_wait_line) {
pointer++;
continue;
}
destination.emplace_back(source[pointer]);
if(isTerminal(source[pointer].type)) break;
pointer++;
}
}
#undef isTerminal
bool ProcessorBase::get_halt_line() {
return halt_mask_ == 0x00;
}
void ProcessorBase::set_interrupt_line(bool value, HalfCycles offset) {
if(irq_line_ == value) return;
// IRQ requests are level triggered and masked.
irq_line_ = value;
if(irq_line_ && iff1_) {
request_status_ |= Interrupt::IRQ;
} else {
request_status_ &= ~Interrupt::IRQ;
}
// If this change happened at least one cycle ago then: (i) we're promised that this is a machine
// cycle per the contract on supplying an offset; and (ii) that means it happened before the lines
// were sampled. So adjust the most recent sample.
if(offset <= HalfCycles(-2)) {
last_request_status_ = (last_request_status_ & ~Interrupt::IRQ) | (request_status_ & Interrupt::IRQ);
}
}
bool ProcessorBase::get_interrupt_line() {
return irq_line_;
}
void ProcessorBase::set_non_maskable_interrupt_line(bool value, HalfCycles offset) {
// NMIs are edge triggered, so react to changes only, particularly changes to active.
if(nmi_line_ != value) {
nmi_line_ = value;
if(value) {
// request_status_ holds a bit mask of interrupt requests pending; since
// this was a new transition, an NMI is now definitely pending.
request_status_ |= Interrupt::NMI;
// If the caller is indicating that this request actually happened in the
// recent past, and it happened long enough ago that it should have been
// spotted during this instruction* then change history to ensure that an
// NMI happens next.
//
// * it actually doesn't matter what sort of bus cycle is currently ongoing;
// either it is that which terminates a whole instruction, in which case this
// test is correct, or it isn't, in which case setting request_status_ above
// was sufficient and the below is merely redundant.
if(offset <= HalfCycles(-2)) {
last_request_status_ |= Interrupt::NMI;
}
}
}
}
bool ProcessorBase::get_non_maskable_interrupt_line() {
return nmi_line_;
}
void ProcessorBase::set_reset_line(bool value) {
// Reset requests are level triggered and cannot be masked.
if(value) request_status_ |= Interrupt::Reset;
else request_status_ &= ~Interrupt::Reset;
}