void SetBufferSwap(u32 screen_id, const FrameBufferInfo& info) {
u32 base_address = 0x400000;
PAddr phys_address_left = Memory::VirtualToPhysicalAddress(info.address_left);
PAddr phys_address_right = Memory::VirtualToPhysicalAddress(info.address_right);
if (info.active_fb == 0) {
WriteHWRegs(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].address_left1)), 4,
&phys_address_left);
WriteHWRegs(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].address_right1)), 4,
&phys_address_right);
} else {
WriteHWRegs(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].address_left2)), 4,
&phys_address_left);
WriteHWRegs(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].address_right2)), 4,
&phys_address_right);
}
WriteHWRegs(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].stride)), 4,
&info.stride);
WriteHWRegs(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].color_format)), 4,
&info.format);
WriteHWRegs(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].active_fb)), 4,
&info.shown_fb);
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::BufferSwapped, nullptr);
if (screen_id == 0) {
MicroProfileFlip();
}
}
static void SetBufferSwap(u32 screen_id, const FrameBufferInfo& info) {
u32 base_address = 0x400000;
if (info.active_fb == 0) {
WriteHWRegs(base_address + 4 * GPU_REG_INDEX(framebuffer_config[screen_id].address_left1), 4, &info.address_left);
WriteHWRegs(base_address + 4 * GPU_REG_INDEX(framebuffer_config[screen_id].address_right1), 4, &info.address_right);
} else {
WriteHWRegs(base_address + 4 * GPU_REG_INDEX(framebuffer_config[screen_id].address_left2), 4, &info.address_left);
WriteHWRegs(base_address + 4 * GPU_REG_INDEX(framebuffer_config[screen_id].address_right2), 4, &info.address_right);
}
WriteHWRegs(base_address + 4 * GPU_REG_INDEX(framebuffer_config[screen_id].stride), 4, &info.stride);
WriteHWRegs(base_address + 4 * GPU_REG_INDEX(framebuffer_config[screen_id].color_format), 4, &info.format);
WriteHWRegs(base_address + 4 * GPU_REG_INDEX(framebuffer_config[screen_id].active_fb), 4, &info.shown_fb);
}
ResultCode SetBufferSwap(u32 screen_id, const FrameBufferInfo& info) {
u32 base_address = 0x400000;
PAddr phys_address_left = VirtualToPhysicalAddress(info.address_left);
PAddr phys_address_right = VirtualToPhysicalAddress(info.address_right);
if (info.active_fb == 0) {
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(
framebuffer_config[screen_id].address_left1)),
phys_address_left);
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(
framebuffer_config[screen_id].address_right1)),
phys_address_right);
} else {
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(
framebuffer_config[screen_id].address_left2)),
phys_address_left);
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(
framebuffer_config[screen_id].address_right2)),
phys_address_right);
}
WriteSingleHWReg(base_address +
4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].stride)),
info.stride);
WriteSingleHWReg(base_address + 4 * static_cast<u32>(GPU_REG_INDEX(
framebuffer_config[screen_id].color_format)),
info.format);
WriteSingleHWReg(
base_address + 4 * static_cast<u32>(GPU_REG_INDEX(framebuffer_config[screen_id].active_fb)),
info.shown_fb);
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::BufferSwapped, nullptr);
if (screen_id == 0) {
MicroProfileFlip();
Core::System::GetInstance().perf_stats.EndGameFrame();
}
return RESULT_SUCCESS;
}
inline void Write(u32 addr, const T data) {
addr -= 0x1EF00000;
int index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T,u32>::value) {
ERROR_LOG(GPU, "unknown Write%d 0x%08X @ 0x%08X", sizeof(data) * 8, data, addr);
return;
}
g_regs[index] = data;
switch (index) {
// Memory fills are triggered once the fill value is written.
// NOTE: This is not verified.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].value, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].value, 0x00008 + 0x3):
{
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].value));
const auto& config = g_regs.memory_fill_config[is_second_filler];
// TODO: Not sure if this check should be done at GSP level instead
if (config.address_start) {
// TODO: Not sure if this algorithm is correct, particularly because it doesn't use the size member at all
u32* start = (u32*)Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetStartAddress()));
u32* end = (u32*)Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetEndAddress()));
for (u32* ptr = start; ptr < end; ++ptr)
*ptr = bswap32(config.value); // TODO: This is just a workaround to missing framebuffer format emulation
DEBUG_LOG(GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(), config.GetEndAddress());
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger):
{
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
u8* source_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalInputAddress()));
u8* dest_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalOutputAddress()));
for (int y = 0; y < config.output_height; ++y) {
// TODO: Why does the register seem to hold twice the framebuffer width?
for (int x = 0; x < config.output_width; ++x) {
struct {
int r, g, b, a;
} source_color = { 0, 0, 0, 0 };
switch (config.input_format) {
case Regs::FramebufferFormat::RGBA8:
{
// TODO: Most likely got the component order messed up.
u8* srcptr = source_pointer + x * 4 + y * config.input_width * 4;
source_color.r = srcptr[0]; // blue
source_color.g = srcptr[1]; // green
source_color.b = srcptr[2]; // red
source_color.a = srcptr[3]; // alpha
break;
}
default:
ERROR_LOG(GPU, "Unknown source framebuffer format %x", config.input_format.Value());
break;
}
switch (config.output_format) {
/*case Regs::FramebufferFormat::RGBA8:
{
// TODO: Untested
u8* dstptr = (u32*)(dest_pointer + x * 4 + y * config.output_width * 4);
dstptr[0] = source_color.r;
dstptr[1] = source_color.g;
dstptr[2] = source_color.b;
dstptr[3] = source_color.a;
break;
}*/
case Regs::FramebufferFormat::RGB8:
{
// TODO: Most likely got the component order messed up.
u8* dstptr = dest_pointer + x * 3 + y * config.output_width * 3;
dstptr[0] = source_color.r; // blue
dstptr[1] = source_color.g; // green
dstptr[2] = source_color.b; // red
break;
}
default:
ERROR_LOG(GPU, "Unknown destination framebuffer format %x", config.output_format.Value());
break;
}
}
}
DEBUG_LOG(GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%dx%d)-> 0x%08x(%dx%d), dst format %x",
config.output_height * config.output_width * 4,
config.GetPhysicalInputAddress(), (int)config.input_width, (int)config.input_height,
config.GetPhysicalOutputAddress(), (int)config.output_width, (int)config.output_height,
config.output_format.Value());
}
break;
}
// Seems like writing to this register triggers processing
case GPU_REG_INDEX(command_processor_config.trigger):
{
const auto& config = g_regs.command_processor_config;
if (config.trigger & 1)
{
u32* buffer = (u32*)Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalAddress()));
u32 size = config.size << 3;
Pica::CommandProcessor::ProcessCommandList(buffer, size);
}
break;
}
default:
break;
}
}
/// Executes the next GSP command
void ExecuteCommand(const Command& command) {
// Utility function to convert register ID to address
auto WriteGPURegister = [](u32 id, u32 data) {
GPU::Write<u32>(0x1EF00000 + 4 * id, data);
};
switch (command.id) {
// GX request DMA - typically used for copying memory from GSP heap to VRAM
case CommandId::REQUEST_DMA:
memcpy(Memory::GetPointer(command.dma_request.dest_address),
Memory::GetPointer(command.dma_request.source_address),
command.dma_request.size);
break;
// ctrulib homebrew sends all relevant command list data with this command,
// hence we do all "interesting" stuff here and do nothing in SET_COMMAND_LIST_FIRST.
// TODO: This will need some rework in the future.
case CommandId::SET_COMMAND_LIST_LAST:
{
auto& params = command.set_command_list_last;
WriteGPURegister(GPU_REG_INDEX(command_processor_config.address), Memory::VirtualToPhysicalAddress(params.address) >> 3);
WriteGPURegister(GPU_REG_INDEX(command_processor_config.size), params.size >> 3);
// TODO: Not sure if we are supposed to always write this .. seems to trigger processing though
WriteGPURegister(GPU_REG_INDEX(command_processor_config.trigger), 1);
// TODO: Move this to GPU
// TODO: Not sure what units the size is measured in
g_debugger.CommandListCalled(params.address,
(u32*)Memory::GetPointer(params.address),
params.size);
SignalInterrupt(InterruptId::P3D);
break;
}
// It's assumed that the two "blocks" behave equivalently.
// Presumably this is done simply to allow two memory fills to run in parallel.
case CommandId::SET_MEMORY_FILL:
{
auto& params = command.memory_fill;
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].address_start), Memory::VirtualToPhysicalAddress(params.start1) >> 3);
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].address_end), Memory::VirtualToPhysicalAddress(params.end1) >> 3);
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].size), params.end1 - params.start1);
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].value), params.value1);
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].address_start), Memory::VirtualToPhysicalAddress(params.start2) >> 3);
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].address_end), Memory::VirtualToPhysicalAddress(params.end2) >> 3);
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].size), params.end2 - params.start2);
WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].value), params.value2);
break;
}
case CommandId::SET_DISPLAY_TRANSFER:
{
auto& params = command.image_copy;
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.input_address), Memory::VirtualToPhysicalAddress(params.in_buffer_address) >> 3);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.output_address), Memory::VirtualToPhysicalAddress(params.out_buffer_address) >> 3);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.input_size), params.in_buffer_size);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.output_size), params.out_buffer_size);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.flags), params.flags);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.trigger), 1);
// TODO(bunnei): Signalling all of these interrupts here is totally wrong, but it seems to
// work well enough for running demos. Need to figure out how these all work and trigger
// them correctly.
SignalInterrupt(InterruptId::PSC0);
SignalInterrupt(InterruptId::PSC1);
SignalInterrupt(InterruptId::PPF);
SignalInterrupt(InterruptId::P3D);
SignalInterrupt(InterruptId::DMA);
break;
}
// TODO: Check if texture copies are implemented correctly..
case CommandId::SET_TEXTURE_COPY:
{
auto& params = command.image_copy;
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.input_address), Memory::VirtualToPhysicalAddress(params.in_buffer_address) >> 3);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.output_address), Memory::VirtualToPhysicalAddress(params.out_buffer_address) >> 3);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.input_size), params.in_buffer_size);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.output_size), params.out_buffer_size);
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.flags), params.flags);
// TODO: Should this register be set to 1 or should instead its value be OR-ed with 1?
WriteGPURegister(GPU_REG_INDEX(display_transfer_config.trigger), 1);
break;
}
// TODO: Figure out what exactly SET_COMMAND_LIST_FIRST and SET_COMMAND_LIST_LAST
// are supposed to do.
case CommandId::SET_COMMAND_LIST_FIRST:
{
break;
}
default:
ERROR_LOG(GSP, "unknown command 0x%08X", (int)command.id.Value());
}
}
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write%lu 0x%08X @ 0x%08X", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
switch (index) {
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].trigger, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].trigger, 0x00008 + 0x3): {
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) {
MemoryFill(config);
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(),
config.GetEndAddress());
// It seems that it won't signal interrupt if "address_start" is zero.
// TODO: hwtest this
if (config.GetStartAddress() != 0) {
if (!is_second_filler) {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
} else {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
}
}
// Reset "trigger" flag and set the "finish" flag
// NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
config.trigger.Assign(0);
config.finished.Assign(1);
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger): {
MICROPROFILE_SCOPE(GPU_DisplayTransfer);
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer,
nullptr);
if (config.is_texture_copy) {
TextureCopy(config);
LOG_TRACE(HW_GPU, "TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> "
"0x%08X(%u+%u), flags 0x%08X",
config.texture_copy.size, config.GetPhysicalInputAddress(),
config.texture_copy.input_width * 16, config.texture_copy.input_gap * 16,
config.GetPhysicalOutputAddress(), config.texture_copy.output_width * 16,
config.texture_copy.output_gap * 16, config.flags);
} else {
DisplayTransfer(config);
LOG_TRACE(HW_GPU, "DisplayTransfer: 0x%08x(%ux%u)-> "
"0x%08x(%ux%u), dst format %x, flags 0x%08X",
config.GetPhysicalInputAddress(), config.input_width.Value(),
config.input_height.Value(), config.GetPhysicalOutputAddress(),
config.output_width.Value(), config.output_height.Value(),
config.output_format.Value(), config.flags);
}
g_regs.display_transfer_config.trigger = 0;
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
}
break;
}
// Seems like writing to this register triggers processing
case GPU_REG_INDEX(command_processor_config.trigger): {
const auto& config = g_regs.command_processor_config;
if (config.trigger & 1) {
MICROPROFILE_SCOPE(GPU_CmdlistProcessing);
u32* buffer = (u32*)Memory::GetPhysicalPointer(config.GetPhysicalAddress());
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
Pica::g_debug_context->recorder->MemoryAccessed(
(u8*)buffer, config.size * sizeof(u32), config.GetPhysicalAddress());
}
Pica::CommandProcessor::ProcessCommandList(buffer, config.size);
g_regs.command_processor_config.trigger = 0;
}
break;
}
default:
break;
}
// Notify tracer about the register write
// This is happening *after* handling the write to make sure we properly catch all memory reads.
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
// addr + GPU VBase - IO VBase + IO PBase
Pica::g_debug_context->recorder->RegisterWritten<T>(
addr + 0x1EF00000 - 0x1EC00000 + 0x10100000, data);
}
}
/// Executes the next GSP command
static void ExecuteCommand(const Command& command, u32 thread_id) {
// Utility function to convert register ID to address
static auto WriteGPURegister = [](u32 id, u32 data) {
GPU::Write<u32>(0x1EF00000 + 4 * id, data);
};
switch (command.id) {
// GX request DMA - typically used for copying memory from GSP heap to VRAM
case CommandId::REQUEST_DMA:
VideoCore::g_renderer->hw_rasterizer->NotifyPreRead(Memory::VirtualToPhysicalAddress(command.dma_request.source_address),
command.dma_request.size);
memcpy(Memory::GetPointer(command.dma_request.dest_address),
Memory::GetPointer(command.dma_request.source_address),
command.dma_request.size);
SignalInterrupt(InterruptId::DMA);
VideoCore::g_renderer->hw_rasterizer->NotifyFlush(Memory::VirtualToPhysicalAddress(command.dma_request.dest_address),
command.dma_request.size);
break;
// ctrulib homebrew sends all relevant command list data with this command,
// hence we do all "interesting" stuff here and do nothing in SET_COMMAND_LIST_FIRST.
// TODO: This will need some rework in the future.
case CommandId::SET_COMMAND_LIST_LAST:
{
auto& params = command.set_command_list_last;
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.address)),
Memory::VirtualToPhysicalAddress(params.address) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.size)), params.size);
// TODO: Not sure if we are supposed to always write this .. seems to trigger processing though
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.trigger)), 1);
break;
}
// It's assumed that the two "blocks" behave equivalently.
// Presumably this is done simply to allow two memory fills to run in parallel.
case CommandId::SET_MEMORY_FILL:
{
auto& params = command.memory_fill;
if (params.start1 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_start)),
Memory::VirtualToPhysicalAddress(params.start1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_end)),
Memory::VirtualToPhysicalAddress(params.end1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].value_32bit)), params.value1);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].control)), params.control1);
}
if (params.start2 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_start)),
Memory::VirtualToPhysicalAddress(params.start2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_end)),
Memory::VirtualToPhysicalAddress(params.end2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].value_32bit)), params.value2);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].control)), params.control2);
}
break;
}
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write%lu 0x%08X @ 0x%08X", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
switch (index) {
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].trigger, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].trigger, 0x00008 + 0x3):
{
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) {
if (config.address_start) { // Some games pass invalid values here
u8* start = Memory::GetPhysicalPointer(config.GetStartAddress());
u8* end = Memory::GetPhysicalPointer(config.GetEndAddress());
if (config.fill_24bit) {
// fill with 24-bit values
for (u8* ptr = start; ptr < end; ptr += 3) {
ptr[0] = config.value_24bit_r;
ptr[1] = config.value_24bit_g;
ptr[2] = config.value_24bit_b;
}
} else if (config.fill_32bit) {
// fill with 32-bit values
for (u32* ptr = (u32*)start; ptr < (u32*)end; ++ptr)
*ptr = config.value_32bit;
} else {
// fill with 16-bit values
for (u16* ptr = (u16*)start; ptr < (u16*)end; ++ptr)
*ptr = config.value_16bit;
}
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(), config.GetEndAddress());
if (!is_second_filler) {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
} else {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
}
VideoCore::g_renderer->rasterizer->InvalidateRegion(config.GetStartAddress(), config.GetEndAddress() - config.GetStartAddress());
}
// Reset "trigger" flag and set the "finish" flag
// NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
config.trigger = 0;
config.finished = 1;
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger):
{
MICROPROFILE_SCOPE(GPU_DisplayTransfer);
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
if (Pica::g_debug_context)
Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer, nullptr);
u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress());
u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress());
if (config.is_texture_copy) {
u32 input_width = config.texture_copy.input_width * 16;
u32 input_gap = config.texture_copy.input_gap * 16;
u32 output_width = config.texture_copy.output_width * 16;
u32 output_gap = config.texture_copy.output_gap * 16;
size_t contiguous_input_size = config.texture_copy.size / input_width * (input_width + input_gap);
VideoCore::g_renderer->rasterizer->FlushRegion(config.GetPhysicalInputAddress(), contiguous_input_size);
u32 remaining_size = config.texture_copy.size;
u32 remaining_input = input_width;
u32 remaining_output = output_width;
while (remaining_size > 0) {
u32 copy_size = std::min({ remaining_input, remaining_output, remaining_size });
std::memcpy(dst_pointer, src_pointer, copy_size);
src_pointer += copy_size;
dst_pointer += copy_size;
remaining_input -= copy_size;
remaining_output -= copy_size;
remaining_size -= copy_size;
if (remaining_input == 0) {
remaining_input = input_width;
src_pointer += input_gap;
}
if (remaining_output == 0) {
remaining_output = output_width;
dst_pointer += output_gap;
}
}
LOG_TRACE(HW_GPU, "TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> 0x%08X(%u+%u), flags 0x%08X",
config.texture_copy.size,
config.GetPhysicalInputAddress(), input_width, input_gap,
config.GetPhysicalOutputAddress(), output_width, output_gap,
config.flags);
size_t contiguous_output_size = config.texture_copy.size / output_width * (output_width + output_gap);
VideoCore::g_renderer->rasterizer->InvalidateRegion(config.GetPhysicalOutputAddress(), contiguous_output_size);
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
break;
}
if (config.scaling > config.ScaleXY) {
LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u", config.scaling.Value());
UNIMPLEMENTED();
break;
}
if (config.input_linear && config.scaling != config.NoScale) {
LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
UNIMPLEMENTED();
break;
}
bool horizontal_scale = config.scaling != config.NoScale;
bool vertical_scale = config.scaling == config.ScaleXY;
u32 output_width = config.output_width >> horizontal_scale;
u32 output_height = config.output_height >> vertical_scale;
u32 input_size = config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
VideoCore::g_renderer->rasterizer->FlushRegion(config.GetPhysicalInputAddress(), input_size);
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
Math::Vec4<u8> src_color;
// Calculate the [x,y] position of the input image
// based on the current output position and the scale
u32 input_x = x << horizontal_scale;
u32 input_y = y << vertical_scale;
if (config.flip_vertically) {
// Flip the y value of the output data,
// we do this after calculating the [x,y] position of the input image
// to account for the scaling options.
y = output_height - y - 1;
}
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
u32 src_offset;
u32 dst_offset;
if (config.input_linear) {
if (!config.dont_swizzle) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = y & ~7;
u32 stride = output_width * dst_bytes_per_pixel;
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + coarse_y * stride;
} else {
// Both input and output are linear
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
}
} else {
if (!config.dont_swizzle) {
// Interpret the input as tiled and the output as linear
u32 coarse_y = input_y & ~7;
u32 stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) + coarse_y * stride;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
} else {
// Both input and output are tiled
u32 out_coarse_y = y & ~7;
u32 out_stride = output_width * dst_bytes_per_pixel;
u32 in_coarse_y = input_y & ~7;
u32 in_stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) + in_coarse_y * in_stride;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + out_coarse_y * out_stride;
}
}
const u8* src_pixel = src_pointer + src_offset;
src_color = DecodePixel(config.input_format, src_pixel);
if (config.scaling == config.ScaleX) {
Math::Vec4<u8> pixel = DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
src_color = ((src_color + pixel) / 2).Cast<u8>();
} else if (config.scaling == config.ScaleXY) {
Math::Vec4<u8> pixel1 = DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
Math::Vec4<u8> pixel2 = DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
Math::Vec4<u8> pixel3 = DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
}
u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
case Regs::PixelFormat::RGBA8:
Color::EncodeRGBA8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB8:
Color::EncodeRGB8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB565:
Color::EncodeRGB565(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB5A1:
Color::EncodeRGB5A1(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGBA4:
Color::EncodeRGBA4(src_color, dst_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x", config.output_format.Value());
break;
}
}
}
LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), dst format %x, flags 0x%08X",
config.output_height * output_width * GPU::Regs::BytesPerPixel(config.output_format),
config.GetPhysicalInputAddress(), config.input_width.Value(), config.input_height.Value(),
config.GetPhysicalOutputAddress(), output_width, output_height,
config.output_format.Value(), config.flags);
g_regs.display_transfer_config.trigger = 0;
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
VideoCore::g_renderer->rasterizer->InvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
}
break;
}
// Seems like writing to this register triggers processing
case GPU_REG_INDEX(command_processor_config.trigger):
{
const auto& config = g_regs.command_processor_config;
if (config.trigger & 1)
{
MICROPROFILE_SCOPE(GPU_CmdlistProcessing);
u32* buffer = (u32*)Memory::GetPhysicalPointer(config.GetPhysicalAddress());
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
Pica::g_debug_context->recorder->MemoryAccessed((u8*)buffer, config.size * sizeof(u32), config.GetPhysicalAddress());
}
Pica::CommandProcessor::ProcessCommandList(buffer, config.size);
g_regs.command_processor_config.trigger = 0;
}
break;
}
default:
break;
}
// Notify tracer about the register write
// This is happening *after* handling the write to make sure we properly catch all memory reads.
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
// addr + GPU VBase - IO VBase + IO PBase
Pica::g_debug_context->recorder->RegisterWritten<T>(addr + 0x1EF00000 - 0x1EC00000 + 0x10100000, data);
}
}
inline void Write(u32 addr, const T data) {
addr -= 0x1EF00000;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write%lu 0x%08X @ 0x%08X", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
switch (index) {
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].trigger, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].trigger, 0x00008 + 0x3):
{
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.address_start && config.trigger) {
u8* start = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetStartAddress()));
u8* end = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetEndAddress()));
if (config.fill_24bit) {
// fill with 24-bit values
for (u8* ptr = start; ptr < end; ptr += 3) {
ptr[0] = config.value_24bit_b;
ptr[1] = config.value_24bit_g;
ptr[2] = config.value_24bit_r;
}
} else if (config.fill_32bit) {
// fill with 32-bit values
for (u32* ptr = (u32*)start; ptr < (u32*)end; ++ptr)
*ptr = config.value_32bit;
} else {
// fill with 16-bit values
for (u16* ptr = (u16*)start; ptr < (u16*)end; ++ptr)
*ptr = config.value_16bit;
}
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(), config.GetEndAddress());
config.trigger = 0;
config.finished = 1;
if (!is_second_filler) {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
} else {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
}
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger):
{
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
u8* src_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalInputAddress()));
u8* dst_pointer = Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalOutputAddress()));
unsigned horizontal_scale = (config.scale_horizontally != 0) ? 2 : 1;
unsigned vertical_scale = (config.scale_vertically != 0) ? 2 : 1;
u32 output_width = config.output_width / horizontal_scale;
u32 output_height = config.output_height / vertical_scale;
if (config.raw_copy) {
// Raw copies do not perform color conversion nor tiled->linear / linear->tiled conversions
// TODO(Subv): Verify if raw copies perform scaling
memcpy(dst_pointer, src_pointer, config.output_width * config.output_height *
GPU::Regs::BytesPerPixel(config.output_format));
LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), flags 0x%08X, Raw copy",
config.output_height * output_width * GPU::Regs::BytesPerPixel(config.output_format),
config.GetPhysicalInputAddress(), config.input_width.Value(), config.input_height.Value(),
config.GetPhysicalOutputAddress(), config.output_width.Value(), config.output_height.Value(),
config.output_format.Value(), config.flags);
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
break;
}
// TODO(Subv): Blend the pixels when horizontal / vertical scaling is enabled,
// right now we're just skipping the extra pixels.
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
Math::Vec4<u8> src_color = { 0, 0, 0, 0 };
u32 scaled_x = x * horizontal_scale;
u32 scaled_y = y * vertical_scale;
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
u32 src_offset;
u32 dst_offset;
if (config.output_tiled) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = y & ~7;
u32 stride = output_width * dst_bytes_per_pixel;
src_offset = (scaled_x + scaled_y * config.input_width) * src_bytes_per_pixel;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + coarse_y * stride;
} else {
// Interpret the input as tiled and the output as linear
u32 coarse_y = scaled_y & ~7;
u32 stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(scaled_x, scaled_y, src_bytes_per_pixel) + coarse_y * stride;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
}
const u8* src_pixel = src_pointer + src_offset;
switch (config.input_format) {
case Regs::PixelFormat::RGBA8:
src_color = Color::DecodeRGBA8(src_pixel);
break;
case Regs::PixelFormat::RGB8:
src_color = Color::DecodeRGB8(src_pixel);
break;
case Regs::PixelFormat::RGB565:
src_color = Color::DecodeRGB565(src_pixel);
break;
case Regs::PixelFormat::RGB5A1:
src_color = Color::DecodeRGB5A1(src_pixel);
break;
case Regs::PixelFormat::RGBA4:
src_color = Color::DecodeRGBA4(src_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown source framebuffer format %x", config.input_format.Value());
break;
}
u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
case Regs::PixelFormat::RGBA8:
Color::EncodeRGBA8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB8:
Color::EncodeRGB8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB565:
Color::EncodeRGB565(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB5A1:
Color::EncodeRGB5A1(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGBA4:
Color::EncodeRGBA4(src_color, dst_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x", config.output_format.Value());
break;
}
}
}
LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), dst format %x, flags 0x%08X",
config.output_height * output_width * GPU::Regs::BytesPerPixel(config.output_format),
config.GetPhysicalInputAddress(), config.input_width.Value(), config.input_height.Value(),
config.GetPhysicalOutputAddress(), output_width, output_height,
config.output_format.Value(), config.flags);
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
}
break;
}
// Seems like writing to this register triggers processing
case GPU_REG_INDEX(command_processor_config.trigger):
{
const auto& config = g_regs.command_processor_config;
if (config.trigger & 1)
{
u32* buffer = (u32*)Memory::GetPointer(Memory::PhysicalToVirtualAddress(config.GetPhysicalAddress()));
Pica::CommandProcessor::ProcessCommandList(buffer, config.size);
}
break;
}
default:
break;
}
}
/// Executes the next GSP command
static void ExecuteCommand(const Command& command, u32 thread_id) {
// Utility function to convert register ID to address
static auto WriteGPURegister = [](u32 id, u32 data) {
GPU::Write<u32>(0x1EF00000 + 4 * id, data);
};
switch (command.id) {
// GX request DMA - typically used for copying memory from GSP heap to VRAM
case CommandId::REQUEST_DMA: {
MICROPROFILE_SCOPE(GPU_GSP_DMA);
Memory::MemorySystem& memory = Core::System::GetInstance().Memory();
// TODO: Consider attempting rasterizer-accelerated surface blit if that usage is ever
// possible/likely
Memory::RasterizerFlushVirtualRegion(command.dma_request.source_address,
command.dma_request.size, Memory::FlushMode::Flush);
Memory::RasterizerFlushVirtualRegion(command.dma_request.dest_address,
command.dma_request.size,
Memory::FlushMode::Invalidate);
// TODO(Subv): These memory accesses should not go through the application's memory mapping.
// They should go through the GSP module's memory mapping.
memory.CopyBlock(*Core::System::GetInstance().Kernel().GetCurrentProcess(),
command.dma_request.dest_address, command.dma_request.source_address,
command.dma_request.size);
SignalInterrupt(InterruptId::DMA);
break;
}
// TODO: This will need some rework in the future. (why?)
case CommandId::SUBMIT_GPU_CMDLIST: {
auto& params = command.submit_gpu_cmdlist;
if (params.do_flush) {
// This flag flushes the command list (params.address, params.size) from the cache.
// Command lists are not processed by the hardware renderer, so we don't need to
// actually flush them in Citra.
}
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.address)),
VirtualToPhysicalAddress(params.address) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.size)),
params.size);
// TODO: Not sure if we are supposed to always write this .. seems to trigger processing
// though
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.trigger)), 1);
// TODO(yuriks): Figure out the meaning of the `flags` field.
break;
}
// It's assumed that the two "blocks" behave equivalently.
// Presumably this is done simply to allow two memory fills to run in parallel.
case CommandId::SET_MEMORY_FILL: {
auto& params = command.memory_fill;
if (params.start1 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_start)),
VirtualToPhysicalAddress(params.start1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_end)),
VirtualToPhysicalAddress(params.end1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].value_32bit)),
params.value1);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].control)),
params.control1);
}
if (params.start2 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_start)),
VirtualToPhysicalAddress(params.start2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_end)),
VirtualToPhysicalAddress(params.end2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].value_32bit)),
params.value2);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].control)),
params.control2);
}
break;
}
/// Executes the next GSP command
static void ExecuteCommand(const Command& command, u32 thread_id) {
// Utility function to convert register ID to address
static auto WriteGPURegister = [](u32 id, u32 data) {
GPU::Write<u32>(0x1EF00000 + 4 * id, data);
};
switch (command.id) {
// GX request DMA - typically used for copying memory from GSP heap to VRAM
case CommandId::REQUEST_DMA:
VideoCore::g_renderer->hw_rasterizer->NotifyPreRead(Memory::VirtualToPhysicalAddress(command.dma_request.source_address),
command.dma_request.size);
memcpy(Memory::GetPointer(command.dma_request.dest_address),
Memory::GetPointer(command.dma_request.source_address),
command.dma_request.size);
SignalInterrupt(InterruptId::DMA);
VideoCore::g_renderer->hw_rasterizer->NotifyFlush(Memory::VirtualToPhysicalAddress(command.dma_request.dest_address),
command.dma_request.size);
break;
// TODO: This will need some rework in the future. (why?)
case CommandId::SUBMIT_GPU_CMDLIST:
{
auto& params = command.submit_gpu_cmdlist;
if (params.do_flush) {
// This flag flushes the command list (params.address, params.size) from the cache.
// Command lists are not processed by the hardware renderer, so we don't need to
// actually flush them in Citra.
}
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.address)),
Memory::VirtualToPhysicalAddress(params.address) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.size)), params.size);
// TODO: Not sure if we are supposed to always write this .. seems to trigger processing though
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(command_processor_config.trigger)), 1);
// TODO(yuriks): Figure out the meaning of the `flags` field.
break;
}
// It's assumed that the two "blocks" behave equivalently.
// Presumably this is done simply to allow two memory fills to run in parallel.
case CommandId::SET_MEMORY_FILL:
{
auto& params = command.memory_fill;
if (params.start1 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_start)),
Memory::VirtualToPhysicalAddress(params.start1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].address_end)),
Memory::VirtualToPhysicalAddress(params.end1) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].value_32bit)), params.value1);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[0].control)), params.control1);
}
if (params.start2 != 0) {
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_start)),
Memory::VirtualToPhysicalAddress(params.start2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].address_end)),
Memory::VirtualToPhysicalAddress(params.end2) >> 3);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].value_32bit)), params.value2);
WriteGPURegister(static_cast<u32>(GPU_REG_INDEX(memory_fill_config[1].control)), params.control2);
}
break;
}
inline void Write(u32 addr, const T data) {
addr -= HW::VADDR_GPU;
u32 index = addr / 4;
// Writes other than u32 are untested, so I'd rather have them abort than silently fail
if (index >= Regs::NumIds() || !std::is_same<T, u32>::value) {
LOG_ERROR(HW_GPU, "unknown Write%lu 0x%08X @ 0x%08X", sizeof(data) * 8, (u32)data, addr);
return;
}
g_regs[index] = static_cast<u32>(data);
switch (index) {
// Memory fills are triggered once the fill value is written.
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].trigger, 0x00004 + 0x3):
case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].trigger, 0x00008 + 0x3):
{
const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].trigger));
auto& config = g_regs.memory_fill_config[is_second_filler];
if (config.trigger) {
if (config.address_start) { // Some games pass invalid values here
u8* start = Memory::GetPhysicalPointer(config.GetStartAddress());
u8* end = Memory::GetPhysicalPointer(config.GetEndAddress());
if (config.fill_24bit) {
// fill with 24-bit values
for (u8* ptr = start; ptr < end; ptr += 3) {
ptr[0] = config.value_24bit_r;
ptr[1] = config.value_24bit_g;
ptr[2] = config.value_24bit_b;
}
} else if (config.fill_32bit) {
// fill with 32-bit values
for (u32* ptr = (u32*)start; ptr < (u32*)end; ++ptr)
*ptr = config.value_32bit;
} else {
// fill with 16-bit values
for (u16* ptr = (u16*)start; ptr < (u16*)end; ++ptr)
*ptr = config.value_16bit;
}
LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(), config.GetEndAddress());
if (!is_second_filler) {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC0);
} else {
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
}
VideoCore::g_renderer->hw_rasterizer->NotifyFlush(config.GetStartAddress(), config.GetEndAddress() - config.GetStartAddress());
}
// Reset "trigger" flag and set the "finish" flag
// NOTE: This was confirmed to happen on hardware even if "address_start" is zero.
config.trigger = 0;
config.finished = 1;
}
break;
}
case GPU_REG_INDEX(display_transfer_config.trigger):
{
const auto& config = g_regs.display_transfer_config;
if (config.trigger & 1) {
u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress());
u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress());
if (config.scaling > config.ScaleXY) {
LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u", config.scaling.Value());
UNIMPLEMENTED();
break;
}
if (config.output_tiled &&
(config.scaling == config.ScaleXY || config.scaling == config.ScaleX)) {
LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
UNIMPLEMENTED();
break;
}
bool horizontal_scale = config.scaling != config.NoScale;
bool vertical_scale = config.scaling == config.ScaleXY;
u32 output_width = config.output_width >> horizontal_scale;
u32 output_height = config.output_height >> vertical_scale;
u32 input_size = config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
VideoCore::g_renderer->hw_rasterizer->NotifyPreRead(config.GetPhysicalInputAddress(), input_size);
if (config.raw_copy) {
// Raw copies do not perform color conversion nor tiled->linear / linear->tiled conversions
// TODO(Subv): Verify if raw copies perform scaling
memcpy(dst_pointer, src_pointer, output_size);
LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), output format: %x, flags 0x%08X, Raw copy",
output_size,
config.GetPhysicalInputAddress(), config.input_width.Value(), config.input_height.Value(),
config.GetPhysicalOutputAddress(), config.output_width.Value(), config.output_height.Value(),
config.output_format.Value(), config.flags);
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
VideoCore::g_renderer->hw_rasterizer->NotifyFlush(config.GetPhysicalOutputAddress(), output_size);
break;
}
for (u32 y = 0; y < output_height; ++y) {
for (u32 x = 0; x < output_width; ++x) {
Math::Vec4<u8> src_color;
// Calculate the [x,y] position of the input image
// based on the current output position and the scale
u32 input_x = x << horizontal_scale;
u32 input_y = y << vertical_scale;
if (config.flip_vertically) {
// Flip the y value of the output data,
// we do this after calculating the [x,y] position of the input image
// to account for the scaling options.
y = output_height - y - 1;
}
u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
u32 src_offset;
u32 dst_offset;
if (config.output_tiled) {
// Interpret the input as linear and the output as tiled
u32 coarse_y = y & ~7;
u32 stride = output_width * dst_bytes_per_pixel;
src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + coarse_y * stride;
} else {
// Interpret the input as tiled and the output as linear
u32 coarse_y = input_y & ~7;
u32 stride = config.input_width * src_bytes_per_pixel;
src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) + coarse_y * stride;
dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
}
const u8* src_pixel = src_pointer + src_offset;
src_color = DecodePixel(config.input_format, src_pixel);
if (config.scaling == config.ScaleX) {
Math::Vec4<u8> pixel = DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
src_color = ((src_color + pixel) / 2).Cast<u8>();
} else if (config.scaling == config.ScaleXY) {
Math::Vec4<u8> pixel1 = DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
Math::Vec4<u8> pixel2 = DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
Math::Vec4<u8> pixel3 = DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
}
u8* dst_pixel = dst_pointer + dst_offset;
switch (config.output_format) {
case Regs::PixelFormat::RGBA8:
Color::EncodeRGBA8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB8:
Color::EncodeRGB8(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB565:
Color::EncodeRGB565(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGB5A1:
Color::EncodeRGB5A1(src_color, dst_pixel);
break;
case Regs::PixelFormat::RGBA4:
Color::EncodeRGBA4(src_color, dst_pixel);
break;
default:
LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x", config.output_format.Value());
break;
}
}
}
LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), dst format %x, flags 0x%08X",
config.output_height * output_width * GPU::Regs::BytesPerPixel(config.output_format),
config.GetPhysicalInputAddress(), config.input_width.Value(), config.input_height.Value(),
config.GetPhysicalOutputAddress(), output_width, output_height,
config.output_format.Value(), config.flags);
g_regs.display_transfer_config.trigger = 0;
GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
VideoCore::g_renderer->hw_rasterizer->NotifyFlush(config.GetPhysicalOutputAddress(), output_size);
}
break;
}
// Seems like writing to this register triggers processing
case GPU_REG_INDEX(command_processor_config.trigger):
{
const auto& config = g_regs.command_processor_config;
if (config.trigger & 1)
{
u32* buffer = (u32*)Memory::GetPhysicalPointer(config.GetPhysicalAddress());
if (Pica::g_debug_context && Pica::g_debug_context->recorder) {
Pica::g_debug_context->recorder->MemoryAccessed((u8*)buffer, config.size * sizeof(u32), config.GetPhysicalAddress());
}
Pica::CommandProcessor::ProcessCommandList(buffer, config.size);
g_regs.command_processor_config.trigger = 0;
}
break;
}
default:
break;
}