Duckstation/src/core/dma.cpp

639 lines
19 KiB
C++
Raw Normal View History

2019-09-09 07:01:26 +00:00
#include "dma.h"
#include "bus.h"
2019-09-21 15:12:16 +00:00
#include "cdrom.h"
#include "common/log.h"
2019-09-14 10:28:47 +00:00
#include "common/state_wrapper.h"
#include "common/string_util.h"
2020-10-18 04:43:09 +00:00
#include "cpu_code_cache.h"
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
#include "cpu_core.h"
2019-09-11 04:59:41 +00:00
#include "gpu.h"
2019-09-24 09:43:10 +00:00
#include "interrupt_controller.h"
2019-09-29 02:51:34 +00:00
#include "mdec.h"
#include "spu.h"
2019-09-24 11:38:58 +00:00
#include "system.h"
2020-09-26 09:33:10 +00:00
#ifdef WITH_IMGUI
#include "imgui.h"
#endif
2019-09-09 07:01:26 +00:00
Log_SetChannel(DMA);
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
DMA g_dma;
2019-09-09 07:01:26 +00:00
DMA::DMA() = default;
DMA::~DMA() = default;
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
void DMA::Initialize()
2019-09-09 07:01:26 +00:00
{
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
m_max_slice_ticks = g_settings.dma_max_slice_ticks;
m_halt_ticks = g_settings.dma_halt_ticks;
m_transfer_buffer.resize(32);
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
m_unhalt_event = TimingEvents::CreateTimingEvent("DMA Transfer Unhalt", 1, m_max_slice_ticks,
std::bind(&DMA::UnhaltTransfer, this, std::placeholders::_1), false);
Reset();
}
void DMA::Shutdown()
{
m_unhalt_event.reset();
2019-09-09 07:01:26 +00:00
}
void DMA::Reset()
{
for (u32 i = 0; i < NUM_CHANNELS; i++)
{
ChannelState& cs = m_state[i];
cs.base_address = 0;
cs.block_control.bits = 0;
cs.channel_control.bits = 0;
cs.request = false;
}
m_DPCR.bits = 0x07654321;
m_DICR.bits = 0;
m_halt_ticks_remaining = 0;
m_unhalt_event->Deactivate();
2019-09-09 07:01:26 +00:00
}
2019-09-14 10:28:47 +00:00
bool DMA::DoState(StateWrapper& sw)
{
sw.Do(&m_halt_ticks_remaining);
2019-09-14 10:28:47 +00:00
for (u32 i = 0; i < NUM_CHANNELS; i++)
{
ChannelState& cs = m_state[i];
sw.Do(&cs.base_address);
sw.Do(&cs.block_control.bits);
sw.Do(&cs.channel_control.bits);
sw.Do(&cs.request);
}
sw.Do(&m_DPCR.bits);
2019-09-24 09:43:10 +00:00
sw.Do(&m_DICR.bits);
if (sw.IsReading())
{
if (m_halt_ticks_remaining > 0)
m_unhalt_event->SetIntervalAndSchedule(m_halt_ticks_remaining);
else
m_unhalt_event->Deactivate();
}
2019-09-14 10:28:47 +00:00
return !sw.HasError();
}
2019-09-09 07:01:26 +00:00
u32 DMA::ReadRegister(u32 offset)
{
const u32 channel_index = offset >> 4;
if (channel_index < 7)
{
switch (offset & UINT32_C(0x0F))
{
case 0x00:
{
Log_TracePrintf("DMA%u base address -> 0x%08X", channel_index, m_state[channel_index].base_address);
2019-09-09 07:01:26 +00:00
return m_state[channel_index].base_address;
}
2019-09-09 07:01:26 +00:00
case 0x04:
{
Log_TracePrintf("DMA%u block control -> 0x%08X", channel_index, m_state[channel_index].block_control.bits);
2019-09-09 07:01:26 +00:00
return m_state[channel_index].block_control.bits;
}
2019-09-09 07:01:26 +00:00
case 0x08:
{
Log_TracePrintf("DMA%u channel control -> 0x%08X", channel_index, m_state[channel_index].channel_control.bits);
2019-09-09 07:01:26 +00:00
return m_state[channel_index].channel_control.bits;
}
2019-09-09 07:01:26 +00:00
default:
break;
}
}
else
{
if (offset == 0x70)
{
Log_TracePrintf("DPCR -> 0x%08X", m_DPCR.bits);
2019-09-09 07:01:26 +00:00
return m_DPCR.bits;
}
2019-09-09 07:01:26 +00:00
else if (offset == 0x74)
{
Log_TracePrintf("DPCR -> 0x%08X", m_DPCR.bits);
2019-09-24 09:43:10 +00:00
return m_DICR.bits;
}
2019-09-09 07:01:26 +00:00
}
Log_ErrorPrintf("Unhandled register read: %02X", offset);
return UINT32_C(0xFFFFFFFF);
}
void DMA::WriteRegister(u32 offset, u32 value)
{
const u32 channel_index = offset >> 4;
if (channel_index < 7)
{
ChannelState& state = m_state[channel_index];
switch (offset & UINT32_C(0x0F))
{
case 0x00:
{
2019-11-11 10:34:41 +00:00
state.base_address = value & BASE_ADDRESS_MASK;
2019-09-30 10:01:41 +00:00
Log_TracePrintf("DMA channel %u base address <- 0x%08X", channel_index, state.base_address);
return;
}
case 0x04:
{
2019-09-30 10:01:41 +00:00
Log_TracePrintf("DMA channel %u block control <- 0x%08X", channel_index, value);
state.block_control.bits = value;
return;
}
case 0x08:
{
state.channel_control.bits = (state.channel_control.bits & ~ChannelState::ChannelControl::WRITE_MASK) |
(value & ChannelState::ChannelControl::WRITE_MASK);
2019-09-30 10:01:41 +00:00
Log_TracePrintf("DMA channel %u channel control <- 0x%08X", channel_index, state.channel_control.bits);
// start/trigger bit must be enabled for OTC
if (static_cast<Channel>(channel_index) == Channel::OTC)
SetRequest(static_cast<Channel>(channel_index), state.channel_control.start_trigger);
if (CanTransferChannel(static_cast<Channel>(channel_index)))
TransferChannel(static_cast<Channel>(channel_index));
return;
}
default:
break;
}
}
else
{
switch (offset)
{
case 0x70:
{
2019-09-30 10:01:41 +00:00
Log_TracePrintf("DPCR <- 0x%08X", value);
m_DPCR.bits = value;
for (u32 i = 0; i < NUM_CHANNELS; i++)
{
if (CanTransferChannel(static_cast<Channel>(i)))
{
if (!TransferChannel(static_cast<Channel>(i)))
break;
}
}
return;
}
case 0x74:
{
2019-09-30 10:01:41 +00:00
Log_TracePrintf("DCIR <- 0x%08X", value);
2019-09-24 09:43:10 +00:00
m_DICR.bits = (m_DICR.bits & ~DICR_WRITE_MASK) | (value & DICR_WRITE_MASK);
m_DICR.bits = m_DICR.bits & ~(value & DICR_RESET_MASK);
2019-09-24 09:43:10 +00:00
m_DICR.UpdateMasterFlag();
return;
}
default:
break;
}
}
2019-09-09 07:01:26 +00:00
Log_ErrorPrintf("Unhandled register write: %02X <- %08X", offset, value);
}
void DMA::SetRequest(Channel channel, bool request)
{
ChannelState& cs = m_state[static_cast<u32>(channel)];
if (cs.request == request)
return;
cs.request = request;
if (CanTransferChannel(channel))
TransferChannel(channel);
}
bool DMA::CanTransferChannel(Channel channel) const
{
if (!m_DPCR.GetMasterEnable(channel))
return false;
const ChannelState& cs = m_state[static_cast<u32>(channel)];
if (!cs.channel_control.enable_busy)
return false;
if (cs.channel_control.sync_mode != SyncMode::Manual && IsTransferHalted())
return false;
return cs.request;
}
bool DMA::IsTransferHalted() const
{
return m_unhalt_event->IsActive();
}
void DMA::UpdateIRQ()
{
m_DICR.UpdateMasterFlag();
if (m_DICR.master_flag)
{
Log_TracePrintf("Firing DMA master interrupt");
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
g_interrupt_controller.InterruptRequest(InterruptController::IRQ::DMA);
}
}
bool DMA::TransferChannel(Channel channel)
{
ChannelState& cs = m_state[static_cast<u32>(channel)];
const bool copy_to_device = cs.channel_control.copy_to_device;
// start/trigger bit is cleared on beginning of transfer
cs.channel_control.start_trigger = false;
2019-11-11 10:34:41 +00:00
PhysicalMemoryAddress current_address = cs.base_address;
2019-09-11 04:59:41 +00:00
const PhysicalMemoryAddress increment = cs.channel_control.address_step_reverse ? static_cast<u32>(-4) : UINT32_C(4);
switch (cs.channel_control.sync_mode)
{
case SyncMode::Manual:
{
const u32 word_count = cs.block_control.manual.GetWordCount();
2019-09-30 10:01:41 +00:00
Log_DebugPrintf("DMA%u: Copying %u words %s 0x%08X", static_cast<u32>(channel), word_count,
copy_to_device ? "from" : "to", current_address & ADDRESS_MASK);
TickCount used_ticks;
if (copy_to_device)
used_ticks = TransferMemoryToDevice(channel, current_address & ADDRESS_MASK, increment, word_count);
else
used_ticks = TransferDeviceToMemory(channel, current_address & ADDRESS_MASK, increment, word_count);
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
CPU::AddPendingTicks(used_ticks);
2019-09-11 04:59:41 +00:00
}
break;
case SyncMode::LinkedList:
{
TickCount used_ticks = 0;
2019-09-11 04:59:41 +00:00
if (!copy_to_device)
{
Panic("Linked list not implemented for DMA reads");
return true;
2019-09-11 04:59:41 +00:00
}
2019-09-30 10:01:41 +00:00
Log_DebugPrintf("DMA%u: Copying linked list starting at 0x%08X to device", static_cast<u32>(channel),
current_address & ADDRESS_MASK);
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
u8* ram_pointer = Bus::g_ram;
bool halt_transfer = false;
while (cs.request)
{
u32 header;
std::memcpy(&header, &ram_pointer[current_address & ADDRESS_MASK], sizeof(header));
used_ticks += 10;
const u32 word_count = header >> 24;
const u32 next_address = header & UINT32_C(0x00FFFFFF);
Log_TracePrintf(" .. linked list entry at 0x%08X size=%u(%u words) next=0x%08X", current_address & ADDRESS_MASK,
word_count * UINT32_C(4), word_count, next_address);
if (word_count > 0)
{
used_ticks += 5;
used_ticks +=
TransferMemoryToDevice(channel, (current_address + sizeof(header)) & ADDRESS_MASK, 4, word_count);
}
else if ((current_address & ADDRESS_MASK) == (next_address & ADDRESS_MASK))
{
2019-11-11 10:34:41 +00:00
current_address = next_address;
halt_transfer = true;
break;
}
current_address = next_address;
if (current_address & UINT32_C(0x800000))
break;
if (used_ticks >= m_max_slice_ticks)
{
halt_transfer = true;
break;
}
}
2019-11-11 10:34:41 +00:00
cs.base_address = current_address;
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
CPU::AddPendingTicks(used_ticks);
if (current_address & UINT32_C(0x800000))
break;
if (halt_transfer)
{
// stall the transfer for a bit if we ran for too long
HaltTransfer(m_halt_ticks);
return false;
}
else
{
// linked list not yet complete
return true;
}
}
break;
case SyncMode::Request:
2019-09-12 15:09:44 +00:00
{
Log_DebugPrintf("DMA%u: Copying %u blocks of size %u (%u total words) %s 0x%08X", static_cast<u32>(channel),
cs.block_control.request.GetBlockCount(), cs.block_control.request.GetBlockSize(),
cs.block_control.request.GetBlockCount() * cs.block_control.request.GetBlockSize(),
copy_to_device ? "from" : "to", current_address & ADDRESS_MASK);
const u32 block_size = cs.block_control.request.GetBlockSize();
u32 blocks_remaining = cs.block_control.request.GetBlockCount();
TickCount used_ticks = 0;
2019-09-12 15:09:44 +00:00
if (copy_to_device)
{
do
{
blocks_remaining--;
used_ticks += TransferMemoryToDevice(channel, current_address & ADDRESS_MASK, increment, block_size);
2019-11-11 10:34:41 +00:00
current_address = (current_address + (increment * block_size));
} while (cs.request && blocks_remaining > 0);
2019-09-12 15:09:44 +00:00
}
else
{
do
{
blocks_remaining--;
used_ticks += TransferDeviceToMemory(channel, current_address & ADDRESS_MASK, increment, block_size);
2019-11-11 10:34:41 +00:00
current_address = (current_address + (increment * block_size));
} while (cs.request && blocks_remaining > 0);
2019-09-12 15:09:44 +00:00
}
2019-11-11 10:34:41 +00:00
cs.base_address = current_address & BASE_ADDRESS_MASK;
cs.block_control.request.block_count = blocks_remaining;
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
CPU::AddPendingTicks(used_ticks);
// finish transfer later if the request was cleared
if (blocks_remaining > 0)
return true;
2019-09-12 15:09:44 +00:00
}
break;
default:
Panic("Unimplemented sync mode");
break;
}
// start/busy bit is cleared on end of transfer
cs.channel_control.enable_busy = false;
2019-09-24 09:43:10 +00:00
if (m_DICR.IsIRQEnabled(channel))
{
Log_DebugPrintf("Set DMA interrupt for channel %u", static_cast<u32>(channel));
2019-09-24 09:43:10 +00:00
m_DICR.SetIRQFlag(channel);
UpdateIRQ();
2019-09-24 09:43:10 +00:00
}
return true;
}
void DMA::HaltTransfer(TickCount duration)
{
m_halt_ticks_remaining += duration;
Log_DebugPrintf("Halting DMA for %d ticks", m_halt_ticks_remaining);
DebugAssert(!m_unhalt_event->IsActive());
m_unhalt_event->SetIntervalAndSchedule(m_halt_ticks_remaining);
}
void DMA::UnhaltTransfer(TickCount ticks)
{
Log_DebugPrintf("Resuming DMA after %d ticks, %d ticks late", ticks, -(m_halt_ticks_remaining - ticks));
m_halt_ticks_remaining -= ticks;
m_unhalt_event->Deactivate();
// TODO: Use channel priority. But doing it in ascending order is probably good enough.
// Main thing is that OTC happens after GPU, because otherwise it'll wipe out the LL.
for (u32 i = 0; i < NUM_CHANNELS; i++)
{
if (CanTransferChannel(static_cast<Channel>(i)))
{
if (!TransferChannel(static_cast<Channel>(i)))
return;
}
}
// We didn't run too long, so reset timer.
m_halt_ticks_remaining = 0;
}
TickCount DMA::TransferMemoryToDevice(Channel channel, u32 address, u32 increment, u32 word_count)
{
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
const u32* src_pointer = reinterpret_cast<u32*>(Bus::g_ram + address);
2020-08-01 14:25:07 +00:00
if (channel != Channel::GPU &&
(static_cast<s32>(increment) < 0 || ((address + (increment * word_count)) & ADDRESS_MASK) <= address))
{
// Use temp buffer if it's wrapping around
if (m_transfer_buffer.size() < word_count)
m_transfer_buffer.resize(word_count);
src_pointer = m_transfer_buffer.data();
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
u8* ram_pointer = Bus::g_ram;
for (u32 i = 0; i < word_count; i++)
{
std::memcpy(&m_transfer_buffer[i], &ram_pointer[address], sizeof(u32));
address = (address + increment) & ADDRESS_MASK;
}
}
2019-09-11 04:59:41 +00:00
switch (channel)
{
2019-09-11 04:59:41 +00:00
case Channel::GPU:
2020-08-01 14:25:07 +00:00
{
if (g_gpu->BeginDMAWrite())
{
u8* ram_pointer = Bus::g_ram;
for (u32 i = 0; i < word_count; i++)
{
u32 value;
std::memcpy(&value, &ram_pointer[address], sizeof(u32));
g_gpu->DMAWrite(address, value);
address = (address + increment) & ADDRESS_MASK;
}
g_gpu->EndDMAWrite();
}
}
break;
2019-09-21 15:12:16 +00:00
case Channel::SPU:
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
g_spu.DMAWrite(src_pointer, word_count);
break;
2019-09-29 02:51:34 +00:00
case Channel::MDECin:
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
g_mdec.DMAWrite(src_pointer, word_count);
break;
case Channel::CDROM:
case Channel::MDECout:
case Channel::PIO:
default:
Log_ErrorPrintf("Unhandled DMA channel %u for device write", static_cast<u32>(channel));
break;
}
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
return Bus::GetDMARAMTickCount(word_count);
}
TickCount DMA::TransferDeviceToMemory(Channel channel, u32 address, u32 increment, u32 word_count)
{
if (channel == Channel::OTC)
2019-09-11 04:59:41 +00:00
{
// clear ordering table
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
u8* ram_pointer = Bus::g_ram;
const u32 word_count_less_1 = word_count - 1;
for (u32 i = 0; i < word_count_less_1; i++)
{
u32 value = ((address - 4) & ADDRESS_MASK);
std::memcpy(&ram_pointer[address], &value, sizeof(value));
address = (address - 4) & ADDRESS_MASK;
}
const u32 terminator = UINT32_C(0xFFFFFF);
std::memcpy(&ram_pointer[address], &terminator, sizeof(terminator));
2020-10-18 04:43:09 +00:00
CPU::CodeCache::InvalidateCodePages(address, word_count);
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
return Bus::GetDMARAMTickCount(word_count);
}
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
u32* dest_pointer = reinterpret_cast<u32*>(&Bus::g_ram[address]);
if (static_cast<s32>(increment) < 0 || ((address + (increment * word_count)) & ADDRESS_MASK) <= address)
{
// Use temp buffer if it's wrapping around
if (m_transfer_buffer.size() < word_count)
m_transfer_buffer.resize(word_count);
dest_pointer = m_transfer_buffer.data();
}
2019-09-11 04:59:41 +00:00
// Read from device.
switch (channel)
{
case Channel::GPU:
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
g_gpu->DMARead(dest_pointer, word_count);
break;
case Channel::CDROM:
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
g_cdrom.DMARead(dest_pointer, word_count);
break;
case Channel::SPU:
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
g_spu.DMARead(dest_pointer, word_count);
2019-09-29 02:51:34 +00:00
break;
2019-09-11 04:59:41 +00:00
case Channel::MDECout:
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
g_mdec.DMARead(dest_pointer, word_count);
break;
2019-09-11 04:59:41 +00:00
default:
Log_ErrorPrintf("Unhandled DMA channel %u for device read", static_cast<u32>(channel));
std::fill_n(dest_pointer, word_count, UINT32_C(0xFFFFFFFF));
2019-09-11 04:59:41 +00:00
break;
}
if (dest_pointer == m_transfer_buffer.data())
{
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
u8* ram_pointer = Bus::g_ram;
for (u32 i = 0; i < word_count; i++)
{
std::memcpy(&ram_pointer[address], &m_transfer_buffer[i], sizeof(u32));
address = (address + increment) & ADDRESS_MASK;
}
}
2020-10-18 04:43:09 +00:00
CPU::CodeCache::InvalidateCodePages(address, word_count);
JIT optimizations and refactoring (#675) * CPU/Recompiler: Use rel32 call where possible for no-args * JitCodeBuffer: Support using preallocated buffer * CPU/Recompiler/AArch64: Use bl instead of blr for short branches * CPU/CodeCache: Allocate recompiler buffer in program space This means we don't need 64-bit moves for every call out of the recompiler. * GTE: Don't store as u16 and load as u32 * CPU/Recompiler: Add methods to emit global load/stores * GTE: Convert class to namespace * CPU/Recompiler: Call GTE functions directly * Settings: Turn into a global variable * GPU: Replace local pointers with global * InterruptController: Turn into a global pointer * System: Replace local pointers with global * Timers: Turn into a global instance * DMA: Turn into a global instance * SPU: Turn into a global instance * CDROM: Turn into a global instance * MDEC: Turn into a global instance * Pad: Turn into a global instance * SIO: Turn into a global instance * CDROM: Move audio FIFO to the heap * CPU/Recompiler: Drop ASMFunctions No longer needed since we have code in the same 4GB window. * CPUCodeCache: Turn class into namespace * Bus: Local pointer -> global pointers * CPU: Turn class into namespace * Bus: Turn into namespace * GTE: Store registers in CPU state struct Allows relative addressing on ARM. * CPU/Recompiler: Align code storage to page size * CPU/Recompiler: Fix relative branches on A64 * HostInterface: Local references to global * System: Turn into a namespace, move events out * Add guard pages * Android: Fix build
2020-07-31 07:09:18 +00:00
return Bus::GetDMARAMTickCount(word_count);
}
2020-09-26 09:33:10 +00:00
void DMA::DrawDebugStateWindow()
{
#ifdef WITH_IMGUI
static constexpr u32 NUM_COLUMNS = 10;
static constexpr std::array<const char*, NUM_COLUMNS> column_names = {
{"#", "Req", "Direction", "Chopping", "Mode", "Busy", "Enable", "Priority", "IRQ", "Flag"}};
static constexpr std::array<const char*, NUM_CHANNELS> channel_names = {
{"MDECin", "MDECout", "GPU", "CDROM", "SPU", "PIO", "OTC"}};
static constexpr std::array<const char*, 4> sync_mode_names = {{"Manual", "Request", "LinkedList", "Reserved"}};
const float framebuffer_scale = ImGui::GetIO().DisplayFramebufferScale.x;
ImGui::SetNextWindowSize(ImVec2(850.0f * framebuffer_scale, 250.0f * framebuffer_scale), ImGuiCond_FirstUseEver);
if (!ImGui::Begin("DMA State", &g_settings.debugging.show_dma_state))
{
ImGui::End();
return;
}
ImGui::Columns(NUM_COLUMNS);
ImGui::SetColumnWidth(0, 100.0f * framebuffer_scale);
ImGui::SetColumnWidth(1, 50.0f * framebuffer_scale);
ImGui::SetColumnWidth(2, 100.0f * framebuffer_scale);
ImGui::SetColumnWidth(3, 150.0f * framebuffer_scale);
ImGui::SetColumnWidth(4, 80.0f * framebuffer_scale);
ImGui::SetColumnWidth(5, 80.0f * framebuffer_scale);
ImGui::SetColumnWidth(6, 80.0f * framebuffer_scale);
ImGui::SetColumnWidth(7, 80.0f * framebuffer_scale);
ImGui::SetColumnWidth(8, 80.0f * framebuffer_scale);
ImGui::SetColumnWidth(9, 80.0f * framebuffer_scale);
for (const char* title : column_names)
{
ImGui::TextUnformatted(title);
ImGui::NextColumn();
}
const ImVec4 active(1.0f, 1.0f, 1.0f, 1.0f);
const ImVec4 inactive(0.5f, 0.5f, 0.5f, 1.0f);
for (u32 i = 0; i < NUM_CHANNELS; i++)
{
const ChannelState& cs = m_state[i];
ImGui::TextColored(cs.channel_control.enable_busy ? active : inactive, "%u[%s]", i, channel_names[i]);
ImGui::NextColumn();
ImGui::TextColored(cs.request ? active : inactive, cs.request ? "Yes" : "No");
ImGui::NextColumn();
ImGui::Text("%s%s", cs.channel_control.copy_to_device ? "FromRAM" : "ToRAM",
cs.channel_control.address_step_reverse ? " Addr+" : " Addr-");
ImGui::NextColumn();
ImGui::TextColored(cs.channel_control.chopping_enable ? active : inactive, "%s/%u/%u",
cs.channel_control.chopping_enable ? "Yes" : "No",
cs.channel_control.chopping_cpu_window_size.GetValue(),
cs.channel_control.chopping_dma_window_size.GetValue());
ImGui::NextColumn();
ImGui::Text("%s", sync_mode_names[static_cast<u8>(cs.channel_control.sync_mode.GetValue())]);
ImGui::NextColumn();
ImGui::TextColored(cs.channel_control.enable_busy ? active : inactive, "%s%s",
cs.channel_control.enable_busy ? "Busy" : "Idle",
cs.channel_control.start_trigger ? " (Trigger)" : "");
ImGui::NextColumn();
ImGui::TextColored(m_DPCR.GetMasterEnable(static_cast<Channel>(i)) ? active : inactive,
m_DPCR.GetMasterEnable(static_cast<Channel>(i)) ? "Enabled" : "Disabled");
ImGui::NextColumn();
ImGui::TextColored(m_DPCR.GetMasterEnable(static_cast<Channel>(i)) ? active : inactive, "%u",
m_DPCR.GetPriority(static_cast<Channel>(i)));
ImGui::NextColumn();
ImGui::TextColored(m_DICR.IsIRQEnabled(static_cast<Channel>(i)) ? active : inactive,
m_DICR.IsIRQEnabled(static_cast<Channel>(i)) ? "Enabled" : "Disabled");
ImGui::NextColumn();
ImGui::TextColored(m_DICR.GetIRQFlag(static_cast<Channel>(i)) ? active : inactive,
m_DICR.GetIRQFlag(static_cast<Channel>(i)) ? "IRQ" : "");
ImGui::NextColumn();
}
ImGui::Columns(1);
ImGui::End();
#endif
}