mirror of
https://github.com/TomHarte/CLK.git
synced 2024-11-25 16:31:42 +00:00
340 lines
10 KiB
C++
340 lines
10 KiB
C++
//
|
|
// Dave.cpp
|
|
// Clock Signal
|
|
//
|
|
// Created by Thomas Harte on 22/06/2021.
|
|
// Copyright © 2021 Thomas Harte. All rights reserved.
|
|
//
|
|
|
|
#include "Dave.hpp"
|
|
|
|
using namespace Enterprise::Dave;
|
|
|
|
// MARK: - Audio generator
|
|
|
|
Audio::Audio(Concurrency::AsyncTaskQueue<false> &audio_queue) :
|
|
audio_queue_(audio_queue) {}
|
|
|
|
void Audio::write(uint16_t address, uint8_t value) {
|
|
address &= 0x1f;
|
|
audio_queue_.enqueue([address, value, this] {
|
|
switch(address) {
|
|
case 0: case 2: case 4:
|
|
channels_[address >> 1].reload = (channels_[address >> 1].reload & 0xff00) | value;
|
|
break;
|
|
case 1: case 3: case 5:
|
|
channels_[address >> 1].reload = uint16_t((channels_[address >> 1].reload & 0x00ff) | ((value & 0xf) << 8));
|
|
channels_[address >> 1].distortion = Channel::Distortion((value >> 4)&3);
|
|
channels_[address >> 1].high_pass = value & 0x40;
|
|
channels_[address >> 1].ring_modulate = value & 0x80;
|
|
break;
|
|
case 6:
|
|
noise_.frequency = Noise::Frequency(value&3);
|
|
noise_.polynomial = Noise::Polynomial((value >> 2)&3);
|
|
noise_.swap_polynomial = value & 0x10;
|
|
noise_.low_pass = value & 0x20;
|
|
noise_.high_pass = value & 0x40;
|
|
noise_.ring_modulate = value & 0x80;
|
|
break;
|
|
|
|
case 7:
|
|
channels_[0].sync = value & 0x01;
|
|
channels_[1].sync = value & 0x02;
|
|
channels_[2].sync = value & 0x04;
|
|
use_direct_output_[0] = value & 0x08;
|
|
use_direct_output_[1] = value & 0x10;
|
|
// Interrupt bits are handled separately.
|
|
break;
|
|
|
|
case 8: case 9: case 10:
|
|
channels_[address - 8].amplitude[0] = value & 0x3f;
|
|
break;
|
|
case 12: case 13: case 14:
|
|
channels_[address - 12].amplitude[1] = value & 0x3f;
|
|
break;
|
|
case 11: noise_.amplitude[0] = value & 0x3f; break;
|
|
case 15: noise_.amplitude[1] = value & 0x3f; break;
|
|
|
|
case 31:
|
|
global_divider_reload_ = 2 + ((value >> 1)&1);
|
|
break;
|
|
}
|
|
});
|
|
}
|
|
|
|
void Audio::set_sample_volume_range(int16_t range) {
|
|
audio_queue_.enqueue([range, this] {
|
|
volume_ = range / (63*4);
|
|
});
|
|
}
|
|
|
|
void Audio::update_channel(int c) {
|
|
auto output = channels_[c].output & 1;
|
|
channels_[c].output <<= 1;
|
|
if(channels_[c].sync) {
|
|
channels_[c].count = channels_[c].reload;
|
|
output = 0;
|
|
} else {
|
|
if(!channels_[c].count) {
|
|
channels_[c].count = channels_[c].reload;
|
|
|
|
if(channels_[c].distortion == Channel::Distortion::None)
|
|
output ^= 1;
|
|
else
|
|
output = poly_state_[int(channels_[c].distortion)];
|
|
} else {
|
|
--channels_[c].count;
|
|
}
|
|
|
|
if(channels_[c].high_pass && (channels_[(c+1)%3].output&3) == 2) {
|
|
output = 0;
|
|
}
|
|
}
|
|
|
|
// Ring modulation applies even when sync is enabled, per SIDBasic.
|
|
if(channels_[c].ring_modulate) {
|
|
output = ~(output ^ channels_[(c+2)%3].output) & 1;
|
|
}
|
|
|
|
channels_[c].output |= output;
|
|
}
|
|
|
|
template <Outputs::Speaker::Action action>
|
|
void Audio::apply_samples(std::size_t number_of_samples, Outputs::Speaker::StereoSample *target) {
|
|
Outputs::Speaker::StereoSample output_level;
|
|
|
|
size_t c = 0;
|
|
while(c < number_of_samples) {
|
|
// I'm unclear on the details of the time division multiplexing so,
|
|
// for now, just sum the outputs.
|
|
output_level.left =
|
|
volume_ *
|
|
(use_direct_output_[0] ?
|
|
channels_[0].amplitude[0]
|
|
: (
|
|
channels_[0].amplitude[0] * (channels_[0].output & 1) +
|
|
channels_[1].amplitude[0] * (channels_[1].output & 1) +
|
|
channels_[2].amplitude[0] * (channels_[2].output & 1) +
|
|
noise_.amplitude[0] * noise_.final_output
|
|
));
|
|
|
|
output_level.right =
|
|
volume_ *
|
|
(use_direct_output_[1] ?
|
|
channels_[0].amplitude[1]
|
|
: (
|
|
channels_[0].amplitude[1] * (channels_[0].output & 1) +
|
|
channels_[1].amplitude[1] * (channels_[1].output & 1) +
|
|
channels_[2].amplitude[1] * (channels_[2].output & 1) +
|
|
noise_.amplitude[1] * noise_.final_output
|
|
));
|
|
|
|
while(global_divider_ && c < number_of_samples) {
|
|
--global_divider_;
|
|
Outputs::Speaker::apply<action>(target[c], output_level);
|
|
++c;
|
|
}
|
|
global_divider_ = global_divider_reload_;
|
|
|
|
poly_state_[int(Channel::Distortion::FourBit)] = poly4_.next();
|
|
poly_state_[int(Channel::Distortion::FiveBit)] = poly5_.next();
|
|
poly_state_[int(Channel::Distortion::SevenBit)] = poly7_.next();
|
|
if(noise_.swap_polynomial) {
|
|
poly_state_[int(Channel::Distortion::SevenBit)] = poly_state_[int(Channel::Distortion::None)];
|
|
}
|
|
|
|
// Update tone channels.
|
|
update_channel(0);
|
|
update_channel(1);
|
|
update_channel(2);
|
|
|
|
// Update noise channel.
|
|
|
|
// Step 1: decide whether there is a tick to apply.
|
|
bool noise_tick = false;
|
|
if(noise_.frequency == Noise::Frequency::DivideByFour) {
|
|
if(!noise_.count) {
|
|
noise_tick = true;
|
|
noise_.count = 3;
|
|
} else {
|
|
--noise_.count;
|
|
}
|
|
} else {
|
|
noise_tick = (channels_[int(noise_.frequency) - 1].output&3) == 2;
|
|
}
|
|
|
|
// Step 2: tick if necessary.
|
|
int noise_output = noise_.output & 1;
|
|
noise_.output <<= 1;
|
|
if(noise_tick) {
|
|
switch(noise_.polynomial) {
|
|
case Noise::Polynomial::SeventeenBit:
|
|
poly_state_[int(Channel::Distortion::None)] = uint8_t(poly17_.next());
|
|
break;
|
|
case Noise::Polynomial::FifteenBit:
|
|
poly_state_[int(Channel::Distortion::None)] = uint8_t(poly15_.next());
|
|
break;
|
|
case Noise::Polynomial::ElevenBit:
|
|
poly_state_[int(Channel::Distortion::None)] = uint8_t(poly11_.next());
|
|
break;
|
|
case Noise::Polynomial::NineBit:
|
|
poly_state_[int(Channel::Distortion::None)] = uint8_t(poly9_.next());
|
|
break;
|
|
}
|
|
|
|
noise_output = poly_state_[int(Channel::Distortion::None)];
|
|
}
|
|
noise_.output |= noise_output;
|
|
|
|
// Low pass: sample channel 2 on downward transitions of the prima facie output.
|
|
if(noise_.low_pass && (noise_.output & 3) == 2) {
|
|
noise_.output = (noise_.output & ~1) | (channels_[2].output & 1);
|
|
}
|
|
|
|
// Apply noise high-pass.
|
|
if(noise_.high_pass && (channels_[0].output & 3) == 2) {
|
|
noise_.output &= ~1;
|
|
}
|
|
|
|
// Update noise ring modulation, if any.
|
|
if(noise_.ring_modulate) {
|
|
noise_.final_output = !((noise_.output ^ channels_[1].output) & 1);
|
|
} else {
|
|
noise_.final_output = noise_.output & 1;
|
|
}
|
|
}
|
|
}
|
|
template void Audio::apply_samples<Outputs::Speaker::Action::Mix>(std::size_t, Outputs::Speaker::StereoSample *);
|
|
template void Audio::apply_samples<Outputs::Speaker::Action::Store>(std::size_t, Outputs::Speaker::StereoSample *);
|
|
template void Audio::apply_samples<Outputs::Speaker::Action::Ignore>(std::size_t, Outputs::Speaker::StereoSample *);
|
|
|
|
// MARK: - Interrupt source
|
|
|
|
uint8_t TimedInterruptSource::get_new_interrupts() {
|
|
const uint8_t result = interrupts_;
|
|
interrupts_ = 0;
|
|
return result;
|
|
}
|
|
|
|
void TimedInterruptSource::write(uint16_t address, uint8_t value) {
|
|
address &= 0x1f;
|
|
switch(address) {
|
|
default: break;
|
|
|
|
case 0: case 2:
|
|
channels_[address >> 1].reload = (channels_[address >> 1].reload & 0xff00) | value;
|
|
break;
|
|
case 1: case 3:
|
|
channels_[address >> 1].reload = uint16_t((channels_[address >> 1].reload & 0x00ff) | ((value & 0xf) << 8));
|
|
break;
|
|
|
|
case 7:
|
|
channels_[0].sync = value & 0x01;
|
|
channels_[1].sync = value & 0x02;
|
|
rate_ = InterruptRate((value >> 5) & 3);
|
|
break;
|
|
|
|
case 31:
|
|
global_divider_ = Cycles(2 + ((value >> 1)&1));
|
|
break;
|
|
}
|
|
}
|
|
|
|
void TimedInterruptSource::update_channel(int c, bool is_linked, int decrement) {
|
|
if(channels_[c].sync) {
|
|
channels_[c].value = channels_[c].reload;
|
|
} else {
|
|
if(decrement <= channels_[c].value) {
|
|
channels_[c].value -= decrement;
|
|
} else {
|
|
// The decrement is greater than the current value, therefore
|
|
// there'll be at least one flip.
|
|
//
|
|
// After decreasing the decrement by the current value + 1,
|
|
// it'll be clear how many decrements are left after reload.
|
|
//
|
|
// Dividing that by the number of decrements necessary for a
|
|
// flip will provide the total number of flips.
|
|
const int decrements_after_flip = decrement - (channels_[c].value + 1);
|
|
const int num_flips = 1 + decrements_after_flip / (channels_[c].reload + 1);
|
|
|
|
// If this is a linked channel, set the interrupt mask if a transition
|
|
// from high to low is amongst the included flips.
|
|
if(is_linked && num_flips + channels_[c].level >= 2) {
|
|
interrupts_ |= uint8_t(Interrupt::VariableFrequency);
|
|
programmable_level_ ^= true;
|
|
}
|
|
channels_[c].level ^= (num_flips & 1);
|
|
|
|
// Apply the modulo number of decrements to the reload value to
|
|
// figure out where things stand now.
|
|
channels_[c].value = channels_[c].reload - decrements_after_flip % (channels_[c].reload + 1);
|
|
}
|
|
}
|
|
}
|
|
|
|
void TimedInterruptSource::run_for(Cycles duration) {
|
|
// Determine total number of ticks.
|
|
run_length_ += duration;
|
|
const Cycles cycles = run_length_.divide(global_divider_);
|
|
if(cycles == Cycles(0)) {
|
|
return;
|
|
}
|
|
|
|
// Update the two-second counter, from which the 1Hz, 50Hz and 1000Hz signals
|
|
// are derived.
|
|
const int previous_counter = two_second_counter_;
|
|
two_second_counter_ = (two_second_counter_ + cycles.as<int>()) % 500'000;
|
|
|
|
// Check for a 1Hz rollover.
|
|
if(previous_counter / 250'000 != two_second_counter_ / 250'000) {
|
|
interrupts_ |= uint8_t(Interrupt::OneHz);
|
|
}
|
|
|
|
// Check for 1kHz or 50Hz rollover;
|
|
switch(rate_) {
|
|
default: break;
|
|
case InterruptRate::OnekHz:
|
|
if(previous_counter / 250 != two_second_counter_ / 250) {
|
|
interrupts_ |= uint8_t(Interrupt::VariableFrequency);
|
|
programmable_level_ ^= true;
|
|
}
|
|
break;
|
|
case InterruptRate::FiftyHz:
|
|
if(previous_counter / 5'000 != two_second_counter_ / 5'000) {
|
|
interrupts_ |= uint8_t(Interrupt::VariableFrequency);
|
|
programmable_level_ ^= true;
|
|
}
|
|
break;
|
|
}
|
|
|
|
// Update the two tone channels.
|
|
update_channel(0, rate_ == InterruptRate::ToneGenerator0, cycles.as<int>());
|
|
update_channel(1, rate_ == InterruptRate::ToneGenerator1, cycles.as<int>());
|
|
}
|
|
|
|
Cycles TimedInterruptSource::next_sequence_point() const {
|
|
// Since both the 1kHz and 50Hz timers are integer dividers of the 1Hz timer, there's no need
|
|
// to factor that one in when determining the next sequence point for either of those.
|
|
switch(rate_) {
|
|
default:
|
|
case InterruptRate::OnekHz: return Cycles(250 - (two_second_counter_ % 250));
|
|
case InterruptRate::FiftyHz: return Cycles(5000 - (two_second_counter_ % 5000));
|
|
|
|
case InterruptRate::ToneGenerator0:
|
|
case InterruptRate::ToneGenerator1: {
|
|
const auto &channel = channels_[int(rate_) - int(InterruptRate::ToneGenerator0)];
|
|
const int cycles_until_interrupt = channel.value + 1 + (!channel.level) * (channel.reload + 1);
|
|
|
|
return Cycles(std::min(
|
|
250'000 - (two_second_counter_ % 250'000),
|
|
cycles_until_interrupt
|
|
));
|
|
}
|
|
}
|
|
}
|
|
|
|
uint8_t TimedInterruptSource::get_divider_state() {
|
|
return uint8_t((two_second_counter_ / 250'000) * 4 | programmable_level_);
|
|
}
|