void VKTexture::CopyRectangleFromTexture(const AbstractTexture* src,
const MathUtil::Rectangle<int>& src_rect, u32 src_layer,
u32 src_level, const MathUtil::Rectangle<int>& dst_rect,
u32 dst_layer, u32 dst_level)
{
Texture2D* src_texture = static_cast<const VKTexture*>(src)->GetRawTexIdentifier();
_assert_msg_(VIDEO, static_cast<u32>(src_rect.GetWidth()) <= src_texture->GetWidth() &&
static_cast<u32>(src_rect.GetHeight()) <= src_texture->GetHeight(),
"Source rect is too large for CopyRectangleFromTexture");
_assert_msg_(VIDEO, static_cast<u32>(dst_rect.GetWidth()) <= m_config.width &&
static_cast<u32>(dst_rect.GetHeight()) <= m_config.height,
"Dest rect is too large for CopyRectangleFromTexture");
VkImageCopy image_copy = {
{VK_IMAGE_ASPECT_COLOR_BIT, src_level, src_layer, src_texture->GetLayers()},
{src_rect.left, src_rect.top, 0},
{VK_IMAGE_ASPECT_COLOR_BIT, dst_level, dst_layer, m_config.layers},
{dst_rect.left, dst_rect.top, 0},
{static_cast<uint32_t>(src_rect.GetWidth()), static_cast<uint32_t>(src_rect.GetHeight()), 1}};
// Must be called outside of a render pass.
StateTracker::GetInstance()->EndRenderPass();
src_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
m_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
vkCmdCopyImage(g_command_buffer_mgr->GetCurrentCommandBuffer(), src_texture->GetImage(),
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, m_texture->GetImage(),
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &image_copy);
// Ensure both textures remain in the SHADER_READ_ONLY layout so they can be bound.
src_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
m_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
}
void VulkanRenderManager::Submit(int frame, bool triggerFence) {
FrameData &frameData = frameData_[frame];
if (frameData.hasInitCommands) {
VkResult res = vkEndCommandBuffer(frameData.initCmd);
_assert_msg_(G3D, res == VK_SUCCESS, "vkEndCommandBuffer failed (init)! result=%s", VulkanResultToString(res));
}
VkResult res = vkEndCommandBuffer(frameData.mainCmd);
_assert_msg_(G3D, res == VK_SUCCESS, "vkEndCommandBuffer failed (main)! result=%s", VulkanResultToString(res));
VkCommandBuffer cmdBufs[2];
int numCmdBufs = 0;
if (frameData.hasInitCommands) {
cmdBufs[numCmdBufs++] = frameData.initCmd;
frameData.hasInitCommands = false;
if (splitSubmit_) {
// Send the init commands off separately. Used this once to confirm that the cause of a device loss was in the init cmdbuf.
VkSubmitInfo submit_info{ VK_STRUCTURE_TYPE_SUBMIT_INFO };
submit_info.commandBufferCount = (uint32_t)numCmdBufs;
submit_info.pCommandBuffers = cmdBufs;
res = vkQueueSubmit(vulkan_->GetGraphicsQueue(), 1, &submit_info, VK_NULL_HANDLE);
if (res == VK_ERROR_DEVICE_LOST) {
_assert_msg_(G3D, false, "Lost the Vulkan device!");
} else {
_assert_msg_(G3D, res == VK_SUCCESS, "vkQueueSubmit failed (init)! result=%s", VulkanResultToString(res));
}
numCmdBufs = 0;
}
}
cmdBufs[numCmdBufs++] = frameData.mainCmd;
VkSubmitInfo submit_info{ VK_STRUCTURE_TYPE_SUBMIT_INFO };
if (triggerFence && !frameData.skipSwap) {
submit_info.waitSemaphoreCount = 1;
submit_info.pWaitSemaphores = &acquireSemaphore_;
VkPipelineStageFlags waitStage[1]{ VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT };
submit_info.pWaitDstStageMask = waitStage;
}
submit_info.commandBufferCount = (uint32_t)numCmdBufs;
submit_info.pCommandBuffers = cmdBufs;
if (triggerFence && !frameData.skipSwap) {
submit_info.signalSemaphoreCount = 1;
submit_info.pSignalSemaphores = &renderingCompleteSemaphore_;
}
res = vkQueueSubmit(vulkan_->GetGraphicsQueue(), 1, &submit_info, triggerFence ? frameData.fence : VK_NULL_HANDLE);
if (res == VK_ERROR_DEVICE_LOST) {
_assert_msg_(G3D, false, "Lost the Vulkan device!");
} else {
_assert_msg_(G3D, res == VK_SUCCESS, "vkQueueSubmit failed (main, split=%d)! result=%s", (int)splitSubmit_, VulkanResultToString(res));
}
// When !triggerFence, we notify after syncing with Vulkan.
if (useThread_ && triggerFence) {
VLOG("PULL: Frame %d.readyForFence = true", frame);
std::unique_lock<std::mutex> lock(frameData.push_mutex);
frameData.readyForFence = true;
frameData.push_condVar.notify_all();
}
}
void PPCXEmitter::SetJumpTarget(FixupBranch const &branch)
{
s32 distance = s32(code) - (s32)branch.ptr;
_assert_msg_(DYNA_REC, distance > -32767
&& distance <= 32767,
"SetJumpTarget out of range (%p calls %p)", code,
branch.ptr);
switch(branch.type) {
case _B:
*(u32*)branch.ptr = (0x48000000 | ((s32)((distance) & 0x3fffffc)));
break;
case _BL:
*(u32*)branch.ptr = (0x48000001 | ((s32)((distance) & 0x3fffffc)));
break;
case _BEQ:
*(u32*)branch.ptr = (0x41820000 | ((s16)(((distance)+1)) & 0xfffc));
break;
case _BNE:
*(u32*)branch.ptr = (0x40820000 | ((s16)(((distance)+1)) & 0xfffc));
break;
case _BLT:
*(u32*)branch.ptr = (0x41800000 | ((s16)(((distance)+1)) & 0xfffc));
break;
case _BLE:
*(u32*)branch.ptr = (0x40810000 | ((s16)(((distance)+1)) & 0xfffc));
break;
case _BGT:
*(u32*)branch.ptr = (0x41810000 | ((s16)(((distance)+1)) & 0xfffc));
break;
case _BGE:
*(u32*)branch.ptr = (0x40800000 | ((s16)(((distance)+1)) & 0xfffc));
break;
default:
// Error !!!
_assert_msg_(DYNA_REC, 0, "SetJumpTarget unknow branch type: %d", branch.type);
break;
}
}
/**
* GSP_GPU::RegisterInterruptRelayQueue service function
* Inputs:
* 1 : "Flags" field, purpose is unknown
* 3 : Handle to GSP synchronization event
* Outputs:
* 0 : Result of function, 0 on success, otherwise error code
* 2 : Thread index into GSP command buffer
* 4 : Handle to GSP shared memory
*/
void RegisterInterruptRelayQueue(Service::Interface* self) {
u32* cmd_buff = Service::GetCommandBuffer();
u32 flags = cmd_buff[1];
g_interrupt_event = cmd_buff[3];
g_shared_memory = Kernel::CreateSharedMemory("GSPSharedMem");
_assert_msg_(GSP, (g_interrupt_event != 0), "handle is not valid!");
cmd_buff[2] = g_thread_id++; // ThreadID
cmd_buff[4] = g_shared_memory; // GSP shared memory
Kernel::SignalEvent(g_interrupt_event); // TODO(bunnei): Is this correct?
}
CEXIMemoryCard::CEXIMemoryCard(const int index, bool gciFolder) : card_index(index)
{
_assert_msg_(EXPANSIONINTERFACE, static_cast<std::size_t>(index) < s_et_cmd_done.size(),
"Trying to create invalid memory card index %d.", index);
// NOTE: When loading a save state, DMA completion callbacks (s_et_transfer_complete) and such
// may have been restored, we need to anticipate those arriving.
interruptSwitch = 0;
m_bInterruptSet = 0;
command = 0;
status = MC_STATUS_BUSY | MC_STATUS_UNLOCKED | MC_STATUS_READY;
m_uPosition = 0;
memset(programming_buffer, 0, sizeof(programming_buffer));
// Nintendo Memory Card EXI IDs
// 0x00000004 Memory Card 59 4Mbit
// 0x00000008 Memory Card 123 8Mb
// 0x00000010 Memory Card 251 16Mb
// 0x00000020 Memory Card 507 32Mb
// 0x00000040 Memory Card 1019 64Mb
// 0x00000080 Memory Card 2043 128Mb
// 0x00000510 16Mb "bigben" card
// card_id = 0xc243;
card_id = 0xc221; // It's a Nintendo brand memcard
// The following games have issues with memory cards bigger than 16Mb
// Darkened Skye GDQE6S GDQP6S
// WTA Tour Tennis GWTEA4 GWTJA4 GWTPA4
// Disney Sports : Skate Boarding GDXEA4 GDXPA4 GDXJA4
// Disney Sports : Soccer GDKEA4
// Use a 16Mb (251 block) memory card for these games
bool useMC251;
IniFile gameIni = SConfig::GetInstance().LoadGameIni();
gameIni.GetOrCreateSection("Core")->Get("MemoryCard251", &useMC251, false);
u16 sizeMb = useMC251 ? MemCard251Mb : MemCard2043Mb;
if (gciFolder)
{
SetupGciFolder(sizeMb);
}
else
{
SetupRawMemcard(sizeMb);
}
memory_card_size = memorycard->GetCardId() * SIZE_TO_Mb;
u8 header[20] = {0};
memorycard->Read(0, static_cast<s32>(ArraySize(header)), header);
SetCardFlashID(header, card_index);
}
void Interpreter::unknown_instruction(UGeckoInstruction _inst)
{
std::string disasm = GekkoDisassembler::Disassemble(PowerPC::HostRead_U32(last_pc), last_pc);
NOTICE_LOG(POWERPC, "Last PC = %08x : %s", last_pc, disasm.c_str());
Dolphin_Debugger::PrintCallstack();
NOTICE_LOG(POWERPC, "\nIntCPU: Unknown instruction %08x at PC = %08x last_PC = %08x LR = %08x\n", _inst.hex, PC, last_pc, LR);
for (int i = 0; i < 32; i += 4)
NOTICE_LOG(POWERPC, "r%d: 0x%08x r%d: 0x%08x r%d:0x%08x r%d: 0x%08x",
i, rGPR[i],
i + 1, rGPR[i + 1],
i + 2, rGPR[i + 2],
i + 3, rGPR[i + 3]);
_assert_msg_(POWERPC, 0, "\nIntCPU: Unknown instruction %08x at PC = %08x last_PC = %08x LR = %08x\n", _inst.hex, PC, last_pc, LR);
}
u32 sceCtrlReadBufferPositive(u32 ctrlDataPtr, u32 nBufs)
{
DEBUG_LOG(HLE,"sceCtrlReadBufferPositive(%08x, %i)", ctrlDataPtr, nBufs);
_assert_msg_(HLE, nBufs > 0, "sceCtrlReadBufferPositive: trying to read nothing?");
std::lock_guard<std::recursive_mutex> guard(ctrlMutex);
// Let's just ignore if ctrl is inited or not; some games don't init it (Super Fruit Fall)
//if (ctrlInited)
//{
SampleControls();
memcpy(Memory::GetPointer(ctrlDataPtr), &ctrl, sizeof(_ctrl_data));
//}
return 1;
}
u32 sceCtrlSetSamplingMode(u32 mode)
{
u32 retVal = 0;
DEBUG_LOG(HLE,"sceCtrlSetSamplingMode(%i)", mode);
_assert_msg_(HLE, mode >= 0 && mode <= 1, "sceCtrlSetSamplingMode: mode outside expected range.");
if (ctrlInited)
{
retVal = analogEnabled == true ? CTRL_MODE_ANALOG : CTRL_MODE_DIGITAL;
analogEnabled = mode == CTRL_MODE_ANALOG ? true : false;
}
return retVal;
}
/// Synchronize to an OS service
static Result SendSyncRequest(Handle handle) {
Kernel::Object* object = Kernel::g_object_pool.GetFast<Kernel::Object>(handle);
_assert_msg_(KERNEL, (object != nullptr), "called, but kernel object is nullptr!");
DEBUG_LOG(SVC, "called handle=0x%08X(%s)", handle, object->GetTypeName().c_str());
bool wait = false;
Result res = object->SyncRequest(&wait);
if (wait) {
Kernel::WaitCurrentThread(WAITTYPE_SYNCH); // TODO(bunnei): Is this correct?
}
return res;
}
void ArmRegCache::Unlock(ARMReg R0, ARMReg R1, ARMReg R2, ARMReg R3)
{
for (u8 RegNum = 0; RegNum < NUMARMREG; ++RegNum)
{
if (ArmRegs[RegNum].Reg == R0)
{
_assert_msg_(_DYNA_REC, !ArmRegs[RegNum].free, "This register is already unlocked");
ArmRegs[RegNum].free = true;
}
if ( R1 != INVALID_REG && ArmRegs[RegNum].Reg == R1) ArmRegs[RegNum].free = true;
if ( R2 != INVALID_REG && ArmRegs[RegNum].Reg == R2) ArmRegs[RegNum].free = true;
if ( R3 != INVALID_REG && ArmRegs[RegNum].Reg == R3) ArmRegs[RegNum].free = true;
}
}
void FPURegCache::BindToRegister(const int i, bool doLoad, bool makeDirty) {
_assert_msg_(DYNA_REC, !regs[i].location.IsImm(), "WTF - load - imm");
if (!regs[i].away) {
// Reg is at home in the memory register file. Let's pull it out.
X64Reg xr = GetFreeXReg();
_assert_msg_(DYNA_REC, xr < NUM_X_FPREGS, "WTF - load - invalid reg");
xregs[xr].mipsReg = i;
xregs[xr].dirty = makeDirty;
OpArg newloc = ::Gen::R(xr);
if (doLoad) {
if (!regs[i].location.IsImm() && (regs[i].location.offset & 0x3)) {
PanicAlert("WARNING - misaligned fp register location %i", i);
}
emit->MOVSS(xr, regs[i].location);
}
regs[i].location = newloc;
regs[i].away = true;
} else {
// There are no immediates in the FPR reg file, so we already had this in a register. Make dirty as necessary.
xregs[RX(i)].dirty |= makeDirty;
_assert_msg_(DYNA_REC, regs[i].location.IsSimpleReg(), "not loaded and not simple.");
}
}
u16 DSPCore_ReadRegister(int reg)
{
switch (reg)
{
case DSP_REG_AR0:
case DSP_REG_AR1:
case DSP_REG_AR2:
case DSP_REG_AR3:
return g_dsp.r.ar[reg - DSP_REG_AR0];
case DSP_REG_IX0:
case DSP_REG_IX1:
case DSP_REG_IX2:
case DSP_REG_IX3:
return g_dsp.r.ix[reg - DSP_REG_IX0];
case DSP_REG_WR0:
case DSP_REG_WR1:
case DSP_REG_WR2:
case DSP_REG_WR3:
return g_dsp.r.wr[reg - DSP_REG_WR0];
case DSP_REG_ST0:
case DSP_REG_ST1:
case DSP_REG_ST2:
case DSP_REG_ST3:
return g_dsp.r.st[reg - DSP_REG_ST0];
case DSP_REG_ACH0:
case DSP_REG_ACH1:
return g_dsp.r.ac[reg - DSP_REG_ACH0].h;
case DSP_REG_CR: return g_dsp.r.cr;
case DSP_REG_SR: return g_dsp.r.sr;
case DSP_REG_PRODL: return g_dsp.r.prod.l;
case DSP_REG_PRODM: return g_dsp.r.prod.m;
case DSP_REG_PRODH: return g_dsp.r.prod.h;
case DSP_REG_PRODM2: return g_dsp.r.prod.m2;
case DSP_REG_AXL0:
case DSP_REG_AXL1:
return g_dsp.r.ax[reg - DSP_REG_AXL0].l;
case DSP_REG_AXH0:
case DSP_REG_AXH1:
return g_dsp.r.ax[reg - DSP_REG_AXH0].h;
case DSP_REG_ACL0:
case DSP_REG_ACL1:
return g_dsp.r.ac[reg - DSP_REG_ACL0].l;
case DSP_REG_ACM0:
case DSP_REG_ACM1:
return g_dsp.r.ac[reg - DSP_REG_ACM0].m;
default:
_assert_msg_(DSP_CORE, 0, "cannot happen");
return 0;
}
}
void FPURegCache::DiscardVS(int vreg) {
_assert_msg_(JIT, !vregs[vreg].location.IsImm(), "FPU can't handle imm yet.");
if (vregs[vreg].away) {
_assert_msg_(JIT, vregs[vreg].lane != 0, "VS expects a SIMD reg.");
X64Reg xr = vregs[vreg].location.GetSimpleReg();
_assert_msg_(JIT, xr >= 0 && xr < NUM_X_FPREGS, "DiscardR: MipsReg had bad X64Reg");
// Note that we DO NOT write it back here. That's the whole point of Discard.
for (int i = 0; i < 4; ++i) {
int mr = xregs[xr].mipsRegs[i];
if (mr != -1) {
regs[mr].location = GetDefaultLocation(mr);
regs[mr].away = false;
regs[mr].tempLocked = false;
}
xregs[xr].mipsRegs[i] = -1;
}
xregs[xr].dirty = false;
} else {
vregs[vreg].tempLocked = false;
}
Invariant();
}
EventType* RegisterEvent(const std::string& name, TimedCallback callback)
{
// check for existing type with same name.
// we want event type names to remain unique so that we can use them for serialization.
_assert_msg_(POWERPC, s_event_types.find(name) == s_event_types.end(),
"CoreTiming Event \"%s\" is already registered. Events should only be registered "
"during Init to avoid breaking save states.",
name.c_str());
auto info = s_event_types.emplace(name, EventType{callback, nullptr});
EventType* event_type = &info.first->second;
event_type->name = &info.first->first;
return event_type;
}
ARMReg ArmRegCache::GetReg(bool AutoLock)
{
for (u8 a = 0; a < NUMARMREG; ++a)
if (ArmRegs[a].free)
{
// Alright, this one is free
if (AutoLock)
ArmRegs[a].free = false;
return ArmRegs[a].Reg;
}
// Uh Oh, we have all them locked....
_assert_msg_(_DYNA_REC_, false, "All available registers are locked dumb dumb");
return R0;
}
/**
* GSP_GPU::RegisterInterruptRelayQueue service function
* Inputs:
* 1 : "Flags" field, purpose is unknown
* 3 : Handle to GSP synchronization event
* Outputs:
* 0 : Result of function, 0 on success, otherwise error code
* 2 : Thread index into GSP command buffer
* 4 : Handle to GSP shared memory
*/
static void RegisterInterruptRelayQueue(Service::Interface* self) {
u32* cmd_buff = Service::GetCommandBuffer();
u32 flags = cmd_buff[1];
g_interrupt_event = cmd_buff[3];
g_shared_memory = Kernel::CreateSharedMemory("GSPSharedMem");
_assert_msg_(GSP, (g_interrupt_event != 0), "handle is not valid!");
cmd_buff[1] = 0x2A07; // Value verified by 3dmoo team, purpose unknown, but needed for GSP init
cmd_buff[2] = g_thread_id++; // Thread ID
cmd_buff[4] = g_shared_memory; // GSP shared memory
Kernel::SignalEvent(g_interrupt_event); // TODO(bunnei): Is this correct?
}
void TextureCache::CopyTextureRectangle(TCacheEntry* dst_texture,
const MathUtil::Rectangle<int>& dst_rect,
Texture2D* src_texture,
const MathUtil::Rectangle<int>& src_rect)
{
_assert_msg_(VIDEO, static_cast<u32>(src_rect.GetWidth()) <= src_texture->GetWidth() &&
static_cast<u32>(src_rect.GetHeight()) <= src_texture->GetHeight(),
"Source rect is too large for CopyRectangleFromTexture");
_assert_msg_(VIDEO, static_cast<u32>(dst_rect.GetWidth()) <= dst_texture->config.width &&
static_cast<u32>(dst_rect.GetHeight()) <= dst_texture->config.height,
"Dest rect is too large for CopyRectangleFromTexture");
VkImageCopy image_copy = {
{ VK_IMAGE_ASPECT_COLOR_BIT, 0, 0,
src_texture->GetLayers() }, // VkImageSubresourceLayers srcSubresource
{ src_rect.left, src_rect.top, 0 }, // VkOffset3D srcOffset
{ VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, // VkImageSubresourceLayers dstSubresource
dst_texture->config.layers },
{ dst_rect.left, dst_rect.top, 0 }, // VkOffset3D dstOffset
{ static_cast<uint32_t>(src_rect.GetWidth()), static_cast<uint32_t>(src_rect.GetHeight()),
1 } // VkExtent3D extent
};
// Must be called outside of a render pass.
StateTracker::GetInstance()->EndRenderPass();
src_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
dst_texture->GetTexture()->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
vkCmdCopyImage(g_command_buffer_mgr->GetCurrentCommandBuffer(), src_texture->GetImage(),
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_texture->GetTexture()->GetImage(),
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &image_copy);
}
/// Duplicates a kernel handle
static Result DuplicateHandle(Handle* out, Handle handle) {
DEBUG_LOG(SVC, "called handle=0x%08X", handle);
// Translate kernel handles -> real handles
if (handle == Kernel::CurrentThread) {
handle = Kernel::GetCurrentThreadHandle();
}
_assert_msg_(KERNEL, (handle != Kernel::CurrentProcess),
"(UNIMPLEMENTED) process handle duplication!");
// TODO(bunnei): FixMe - This is a hack to return the handle that we were asked to duplicate.
*out = handle;
return 0;
}
// Can be called multiple times with no bad side effects. This is so that we can either begin a frame the normal way,
// or stop it in the middle for a synchronous readback, then start over again mostly normally but without repeating
// the backbuffer image acquisition.
void VulkanRenderManager::BeginSubmitFrame(int frame) {
FrameData &frameData = frameData_[frame];
if (!frameData.hasBegun) {
// Get the index of the next available swapchain image, and a semaphore to block command buffer execution on.
// Now, I wonder if we should do this early in the frame or late? Right now we do it early, which should be fine.
VkResult res = vkAcquireNextImageKHR(vulkan_->GetDevice(), vulkan_->GetSwapchain(), UINT64_MAX, acquireSemaphore_, (VkFence)VK_NULL_HANDLE, &frameData.curSwapchainImage);
if (res == VK_SUBOPTIMAL_KHR) {
// Hopefully the resize will happen shortly. Ignore - one frame might look bad or something.
} else if (res == VK_ERROR_OUT_OF_DATE_KHR) {
frameData.skipSwap = true;
} else {
_assert_msg_(G3D, res == VK_SUCCESS, "vkAcquireNextImageKHR failed! result=%s", VulkanResultToString(res));
}
VkCommandBufferBeginInfo begin{ VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO };
begin.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
res = vkBeginCommandBuffer(frameData.mainCmd, &begin);
_assert_msg_(G3D, res == VK_SUCCESS, "vkBeginCommandBuffer failed! result=%s", VulkanResultToString(res));
queueRunner_.SetBackbuffer(framebuffers_[frameData.curSwapchainImage], swapchainImages_[frameData.curSwapchainImage].image);
frameData.hasBegun = true;
}
}
void Jit64::ps_arith(UGeckoInstruction inst)
{
INSTRUCTION_START
JITDISABLE(bJITPairedOff);
FALLBACK_IF(inst.Rc);
switch (inst.SUBOP5)
{
case 18: tri_op(inst.FD, inst.FA, inst.FB, false, &XEmitter::DIVPD); break; //div
case 20: tri_op(inst.FD, inst.FA, inst.FB, false, &XEmitter::SUBPD); break; //sub
case 21: tri_op(inst.FD, inst.FA, inst.FB, true, &XEmitter::ADDPD); break; //add
case 25: tri_op(inst.FD, inst.FA, inst.FC, true, &XEmitter::MULPD); break; //mul
default:
_assert_msg_(DYNA_REC, 0, "ps_arith WTF!!!");
}
}
// This is to be called when outside threads, such as the graphics thread, wants to
// schedule things to be executed on the main thread.
void ScheduleEvent_Threadsafe(s64 cyclesIntoFuture, int event_type, u64 userdata)
{
_assert_msg_(POWERPC, !Core::IsCPUThread(), "ScheduleEvent_Threadsafe from wrong thread");
if (Core::g_want_determinism)
{
ERROR_LOG(POWERPC, "Someone scheduled an off-thread \"%s\" event while netplay or movie play/record "
"was active. This is likely to cause a desync.",
event_types[event_type].name.c_str());
}
std::lock_guard<std::mutex> lk(tsWriteLock);
Event ne;
ne.time = g_globalTimer + cyclesIntoFuture;
ne.type = event_type;
ne.userdata = userdata;
tsQueue.Push(ne);
}
void Init()
{
s_devnode_name_map.clear();
// During initialization we use udev to iterate over all /dev/input/event* devices.
// Note: the Linux kernel is currently limited to just 32 event devices. If this ever
// changes, hopefully udev will take care of this.
udev* udev = udev_new();
_assert_msg_(PAD, udev != nullptr, "Couldn't initialize libudev.");
// List all input devices
udev_enumerate* enumerate = udev_enumerate_new(udev);
udev_enumerate_add_match_subsystem(enumerate, "input");
udev_enumerate_scan_devices(enumerate);
udev_list_entry* devices = udev_enumerate_get_list_entry(enumerate);
// Iterate over all input devices
udev_list_entry* dev_list_entry;
udev_list_entry_foreach(dev_list_entry, devices)
{
const char* path = udev_list_entry_get_name(dev_list_entry);
udev_device* dev = udev_device_new_from_syspath(udev, path);
const char* devnode = udev_device_get_devnode(dev);
// We only care about devices which we have read/write access to.
if (devnode && access(devnode, W_OK) == 0)
{
// Unfortunately udev gives us no way to filter out the non event device interfaces.
// So we open it and see if it works with evdev ioctls or not.
std::string name = GetName(devnode);
auto input = std::make_shared<evdevDevice>(devnode);
if (input->IsInteresting())
{
g_controller_interface.AddDevice(std::move(input));
s_devnode_name_map.insert(std::pair<std::string, std::string>(devnode, name));
}
}
udev_device_unref(dev);
}
udev_enumerate_unref(enumerate);
udev_unref(udev);
StartHotplugThread();
}
// This must be run ONLY from within the CPU thread
// cyclesIntoFuture may be VERY inaccurate if called from anything else
// than Advance
void ScheduleEvent(s64 cyclesIntoFuture, int event_type, u64 userdata)
{
_assert_msg_(POWERPC, Core::IsCPUThread() || Core::GetState() == Core::CORE_PAUSE,
"ScheduleEvent from wrong thread");
Event *ne = GetNewEvent();
ne->userdata = userdata;
ne->type = event_type;
ne->time = GetTicks() + cyclesIntoFuture;
// If this event needs to be scheduled before the next advance(), force one early
if (!globalTimerIsSane)
ForceExceptionCheck(cyclesIntoFuture);
AddEventToQueue(ne);
}
void VKTexture::ScaleRectangleFromTexture(const AbstractTexture* source,
const MathUtil::Rectangle<int>& src_rect,
const MathUtil::Rectangle<int>& dst_rect)
{
Texture2D* src_texture = static_cast<const VKTexture*>(source)->GetRawTexIdentifier();
// Can't do this within a game render pass.
StateTracker::GetInstance()->EndRenderPass();
StateTracker::GetInstance()->SetPendingRebind();
// Can't render to a non-rendertarget (no framebuffer).
_assert_msg_(VIDEO, m_config.rendertarget,
"Destination texture for partial copy is not a rendertarget");
// Render pass expects dst_texture to be in COLOR_ATTACHMENT_OPTIMAL state.
// src_texture should already be in SHADER_READ_ONLY state, but transition in case (XFB).
src_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
m_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
VkRenderPass render_pass = g_object_cache->GetRenderPass(
m_texture->GetFormat(), VK_FORMAT_UNDEFINED, 1, VK_ATTACHMENT_LOAD_OP_DONT_CARE);
UtilityShaderDraw draw(g_command_buffer_mgr->GetCurrentCommandBuffer(),
g_object_cache->GetPipelineLayout(PIPELINE_LAYOUT_STANDARD), render_pass,
g_shader_cache->GetPassthroughVertexShader(),
g_shader_cache->GetPassthroughGeometryShader(),
TextureCache::GetInstance()->GetCopyShader());
VkRect2D region = {
{dst_rect.left, dst_rect.top},
{static_cast<u32>(dst_rect.GetWidth()), static_cast<u32>(dst_rect.GetHeight())}};
draw.BeginRenderPass(m_framebuffer, region);
draw.SetPSSampler(0, src_texture->GetView(), g_object_cache->GetLinearSampler());
draw.DrawQuad(dst_rect.left, dst_rect.top, dst_rect.GetWidth(), dst_rect.GetHeight(),
src_rect.left, src_rect.top, 0, src_rect.GetWidth(), src_rect.GetHeight(),
static_cast<int>(src_texture->GetWidth()),
static_cast<int>(src_texture->GetHeight()));
draw.EndRenderPass();
// Ensure both textures remain in the SHADER_READ_ONLY layout so they can be bound.
src_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
m_texture->TransitionToLayout(g_command_buffer_mgr->GetCurrentCommandBuffer(),
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
}
/**
* Maps a shared memory block to an address in system memory
* @param handle Shared memory block handle
* @param address Address in system memory to map shared memory block to
* @param permissions Memory block map permissions (specified by SVC field)
* @param other_permissions Memory block map other permissions (specified by SVC field)
* @return Result of operation, 0 on success, otherwise error code
*/
Result MapSharedMemory(u32 handle, u32 address, MemoryPermission permissions,
MemoryPermission other_permissions) {
if (address < Memory::SHARED_MEMORY_VADDR || address >= Memory::SHARED_MEMORY_VADDR_END) {
ERROR_LOG(KERNEL, "cannot map handle=0x%08X, address=0x%08X outside of shared mem bounds!",
handle);
return -1;
}
SharedMemory* shared_memory = Kernel::g_object_pool.GetFast<SharedMemory>(handle);
_assert_msg_(KERNEL, (shared_memory != nullptr), "handle 0x%08X is not valid!", handle);
shared_memory->base_address = address;
shared_memory->permissions = permissions;
shared_memory->other_permissions = other_permissions;
return 0;
}
void GenerateDepalShader(char *buffer, GEBufferFormat pixelFormat, ShaderLanguage language) {
switch (language) {
case GLSL_140:
GenerateDepalShaderFloat(buffer, pixelFormat, language);
break;
case GLSL_300:
case GLSL_VULKAN:
case HLSL_D3D11:
GenerateDepalShader300(buffer, pixelFormat, language);
break;
case HLSL_DX9:
GenerateDepalShaderFloat(buffer, pixelFormat, language);
break;
case HLSL_D3D11_LEVEL9:
default:
_assert_msg_(G3D, false, "Depal shader language not supported: %d", (int)language);
}
}
/// Synchronize to an OS service
static Result SendSyncRequest(Handle handle) {
// TODO(yuriks): ObjectPool::Get tries to check the Object type, which fails since this is a generic base Object,
// so we are forced to use GetFast and manually verify the handle.
if (!Kernel::g_object_pool.IsValid(handle)) {
return InvalidHandle(ErrorModule::Kernel).raw;
}
Kernel::Object* object = Kernel::g_object_pool.GetFast<Kernel::Object>(handle);
_assert_msg_(KERNEL, (object != nullptr), "called, but kernel object is nullptr!");
DEBUG_LOG(SVC, "called handle=0x%08X(%s)", handle, object->GetTypeName().c_str());
ResultVal<bool> wait = object->SyncRequest();
if (wait.Succeeded() && *wait) {
Kernel::WaitCurrentThread(WAITTYPE_SYNCH); // TODO(bunnei): Is this correct?
}
return wait.Code().raw;
}
GekkoOPInfo *GetOpInfo(UGeckoInstruction _inst)
{
GekkoOPInfo *info = m_infoTable[_inst.OPCD];
if ((info->type & 0xFFFFFF) == OPTYPE_SUBTABLE)
{
int table = info->type>>24;
switch(table)
{
case 4: return m_infoTable4[_inst.SUBOP10];
case 19: return m_infoTable19[_inst.SUBOP10];
case 31: return m_infoTable31[_inst.SUBOP10];
case 59: return m_infoTable59[_inst.SUBOP5];
case 63: return m_infoTable63[_inst.SUBOP10];
default:
_assert_msg_(POWERPC,0,"GetOpInfo - invalid subtable op %08x @ %08x", _inst.hex, PC);
return 0;
}
}
/// Stops the current thread
void StopThread(Handle handle, const char* reason) {
Thread* thread = g_object_pool.GetFast<Thread>(handle);
_assert_msg_(KERNEL, (thread != nullptr), "called, but thread is nullptr!");
ChangeReadyState(thread, false);
thread->status = THREADSTATUS_DORMANT;
for (size_t i = 0; i < thread->waiting_threads.size(); ++i) {
const Handle waiting_thread = thread->waiting_threads[i];
if (VerifyWait(waiting_thread, WAITTYPE_THREADEND, handle)) {
ResumeThreadFromWait(waiting_thread);
}
}
thread->waiting_threads.clear();
// Stopped threads are never waiting.
thread->wait_type = WAITTYPE_NONE;
thread->wait_handle = 0;
}
void FPURegCache::Flush() {
for (int i = 0; i < NUM_MIPS_FPRS; i++) {
if (regs[i].locked) {
PanicAlert("Somebody forgot to unlock MIPS reg %i.", i);
}
if (regs[i].away) {
if (regs[i].location.IsSimpleReg()) {
X64Reg xr = RX(i);
StoreFromRegister(i);
xregs[xr].dirty = false;
} else if (regs[i].location.IsImm()) {
StoreFromRegister(i);
} else {
_assert_msg_(DYNA_REC,0,"Jit64 - Flush unhandled case, reg %i PC: %08x", i, mips->pc);
}
}
}
}
