void ARM_Dynarmic::ExecuteInstructions(int num_instructions) {
MICROPROFILE_SCOPE(ARM_Jit);
unsigned ticks_executed = jit->Run(static_cast<unsigned>(num_instructions));
AddTicks(ticks_executed);
}
int main(int argc, char* argv[])
{
MicroProfileOnThreadCreate("Main");
printf("press ctrl-c to quit\n");
//turn on profiling
MicroProfileSetForceEnable(true);
MicroProfileSetEnableAllGroups(true);
MicroProfileSetForceMetaCounters(true);
MicroProfileStartContextSwitchTrace();
StartFakeWork();
while(!g_nQuit)
{
MICROPROFILE_SCOPE(MAIN);
{
usleep(16000);
}
MicroProfileFlip();
static bool once = false;
if(!once)
{
once = 1;
printf("open localhost:%d in chrome to capture profile data\n", MicroProfileWebServerPort());
}
}
StopFakeWork();
MicroProfileShutdown();
return 0;
}
void ShaderSetup::Run(UnitState& state, const InputVertex& input, int num_attributes) {
auto& config = g_state.regs.vs;
auto& setup = g_state.vs;
MICROPROFILE_SCOPE(GPU_Shader);
// Setup input register table
const auto& attribute_register_map = config.input_register_map;
for (unsigned i = 0; i < num_attributes; i++)
state.registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i];
state.conditional_code[0] = false;
state.conditional_code[1] = false;
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled) {
jit_shader->Run(setup, state, config.main_offset);
} else {
DebugData<false> dummy_debug_data;
RunInterpreter(setup, state, dummy_debug_data, config.main_offset);
}
#else
DebugData<false> dummy_debug_data;
RunInterpreter(setup, state, dummy_debug_data, config.main_offset);
#endif // ARCHITECTURE_x86_64
}
void CompositorBase::SubmitFovLayer(ovrRecti viewport[ovrEye_Count], ovrFovPort fov[ovrEye_Count], ovrTextureSwapChain swapChain[ovrEye_Count], unsigned int flags)
{
MICROPROFILE_SCOPE(SubmitFovLayer);
// If the right eye isn't set used the left eye for both
if (!swapChain[ovrEye_Right])
swapChain[ovrEye_Right] = swapChain[ovrEye_Left];
MICROPROFILE_META_CPU("SwapChain Right", swapChain[ovrEye_Right]->Identifier);
MICROPROFILE_META_CPU("SwapChain Left", swapChain[ovrEye_Left]->Identifier);
// Render the scene layer
for (int i = 0; i < ovrEye_Count; i++)
{
// Get the scene fov
ovrFovPort sceneFov;
if (m_SceneLayer->Type == ovrLayerType_EyeFov)
sceneFov = ((ovrLayerEyeFov*)m_SceneLayer)->Fov[i];
else if (m_SceneLayer->Type == ovrLayerType_EyeMatrix)
sceneFov = MatrixToFovPort(((ovrLayerEyeMatrix*)m_SceneLayer)->Matrix[i]);
// Calculate the fov quad
vr::HmdVector4_t quad;
quad.v[0] = fov[i].LeftTan / -sceneFov.LeftTan;
quad.v[1] = fov[i].RightTan / sceneFov.RightTan;
quad.v[2] = fov[i].UpTan / sceneFov.UpTan;
quad.v[3] = fov[i].DownTan / -sceneFov.DownTan;
// Calculate the texture bounds
vr::VRTextureBounds_t bounds = ViewportToTextureBounds(viewport[i], swapChain[i], flags);
// Composit the layer
if (m_SceneLayer->Type == ovrLayerType_EyeFov)
{
ovrLayerEyeFov* layer = (ovrLayerEyeFov*)m_SceneLayer;
RenderTextureSwapChain((vr::EVREye)i, swapChain[i], layer->ColorTexture[i], layer->Viewport[i], bounds, quad);
}
else if (m_SceneLayer->Type == ovrLayerType_EyeMatrix)
{
ovrLayerEyeMatrix* layer = (ovrLayerEyeMatrix*)m_SceneLayer;
RenderTextureSwapChain((vr::EVREye)i, swapChain[i], layer->ColorTexture[i], layer->Viewport[i], bounds, quad);
}
}
swapChain[ovrEye_Left]->Submit();
if (swapChain[ovrEye_Left] != swapChain[ovrEye_Right])
swapChain[ovrEye_Right]->Submit();
}
vr::VRCompositorError CompositorBase::SubmitSceneLayer(ovrRecti viewport[ovrEye_Count], ovrFovPort fov[ovrEye_Count], ovrTextureSwapChain swapChain[ovrEye_Count], unsigned int flags)
{
MICROPROFILE_SCOPE(SubmitSceneLayer);
// If the right eye isn't set used the left eye for both
if (!swapChain[ovrEye_Right])
swapChain[ovrEye_Right] = swapChain[ovrEye_Left];
MICROPROFILE_META_CPU("SwapChain Right", swapChain[ovrEye_Right]->Identifier);
MICROPROFILE_META_CPU("SwapChain Left", swapChain[ovrEye_Left]->Identifier);
// Submit the scene layer.
vr::VRCompositorError err;
for (int i = 0; i < ovrEye_Count; i++)
{
ovrTextureSwapChain chain = swapChain[i];
vr::VRTextureBounds_t bounds = ViewportToTextureBounds(viewport[i], swapChain[i], flags);
// Shrink the bounds to account for the overlapping fov
vr::VRTextureBounds_t fovBounds = FovPortToTextureBounds((ovrEyeType)i, fov[i]);
// Combine the fov bounds with the viewport bounds
bounds.uMin += fovBounds.uMin * bounds.uMax;
bounds.uMax *= fovBounds.uMax;
bounds.vMin += fovBounds.vMin * bounds.vMax;
bounds.vMax *= fovBounds.vMax;
vr::Texture_t texture = chain->Textures[chain->SubmitIndex]->ToVRTexture();
err = vr::VRCompositor()->Submit((vr::EVREye)i, &texture, &bounds);
if (err != vr::VRCompositorError_None)
break;
}
swapChain[ovrEye_Left]->Submit();
if (swapChain[ovrEye_Left] != swapChain[ovrEye_Right])
swapChain[ovrEye_Right]->Submit();
return err;
}
void RasterizerCacheOpenGL::LoadAndBindTexture(OpenGLState &state, unsigned texture_unit, const Pica::DebugUtils::TextureInfo& info) {
const auto cached_texture = texture_cache.find(info.physical_address);
if (cached_texture != texture_cache.end()) {
state.texture_units[texture_unit].texture_2d = cached_texture->second->texture.handle;
state.Apply();
} else {
MICROPROFILE_SCOPE(OpenGL_TextureUpload);
std::unique_ptr<CachedTexture> new_texture = Common::make_unique<CachedTexture>();
new_texture->texture.Create();
state.texture_units[texture_unit].texture_2d = new_texture->texture.handle;
state.Apply();
glActiveTexture(GL_TEXTURE0 + texture_unit);
u8* texture_src_data = Memory::GetPhysicalPointer(info.physical_address);
new_texture->width = info.width;
new_texture->height = info.height;
new_texture->size = info.stride * info.height;
new_texture->addr = info.physical_address;
new_texture->hash = Common::ComputeHash64(texture_src_data, new_texture->size);
std::unique_ptr<Math::Vec4<u8>[]> temp_texture_buffer_rgba(new Math::Vec4<u8>[info.width * info.height]);
for (int y = 0; y < info.height; ++y) {
for (int x = 0; x < info.width; ++x) {
temp_texture_buffer_rgba[x + info.width * y] = Pica::DebugUtils::LookupTexture(texture_src_data, x, info.height - 1 - y, info);
}
}
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, info.width, info.height, 0, GL_RGBA, GL_UNSIGNED_BYTE, temp_texture_buffer_rgba.get());
texture_cache.emplace(info.physical_address, std::move(new_texture));
}
}
void ARM_Dynarmic::Run() {
ASSERT(Memory::GetCurrentPageTable() == current_page_table);
MICROPROFILE_SCOPE(ARM_Jit);
jit->Run();
}
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);
}
}
int main(int argc, char* argv[])
{
MicroProfileOnThreadCreate("AA_Main");
if(SDL_Init(SDL_INIT_VIDEO) < 0) {
return 1;
}
SDL_GL_SetAttribute(SDL_GL_RED_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_GREEN_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_BLUE_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_ALPHA_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_DEPTH_SIZE, 24);
SDL_GL_SetAttribute(SDL_GL_STENCIL_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_BUFFER_SIZE, 32);
SDL_GL_SetAttribute(SDL_GL_DOUBLEBUFFER, 1);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MAJOR_VERSION, 3);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MINOR_VERSION, 2);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_CORE);
SDL_GL_SetSwapInterval(1);
SDL_Window * pWindow = SDL_CreateWindow("microprofiledemo", 10, 10, WIDTH, HEIGHT, SDL_WINDOW_OPENGL);
if(!pWindow)
return 1;
SDL_GLContext glcontext = SDL_GL_CreateContext(pWindow);
glewExperimental=1;
GLenum err=glewInit();
printf("ERROR IS %d\n",err);
if(err!=GLEW_OK)
{
__BREAK();
}
glGetError(); //glew generates an error
InitGLBuffers();
#if MICROPROFILE_ENABLED
MicroProfileGpuInitGL();
#endif
SDL_GL_SetSwapInterval(1);
while(!g_nQuit)
{
MICROPROFILE_SCOPE(MAIN);
MICROPROFILE_COUNTER_ADD("engine/frames", 1);
SDL_Event Evt;
while(SDL_PollEvent(&Evt))
{
MICROPROFILE_COUNTER_LOCAL_ADD_ATOMIC(SDLFrameEvents, 1);
HandleEvent(&Evt);
}
MICROPROFILE_COUNTER_LOCAL_UPDATE_SET_ATOMIC(SDLFrameEvents);
glClearColor(0.3f,0.4f,0.6f,0.f);
glViewport(0, 0, WIDTH, HEIGHT);
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
DrawGLStuff();
MicroProfileFlip(0);
MICROPROFILE_SCOPEI("MAIN", "Flip", 0xffee00);
SDL_GL_SwapWindow(pWindow);
static bool bOnce = false;
if(!bOnce)
{
bOnce = true;
printf("open localhost:%d in chrome to capture profile data\n", MicroProfileWebServerPort());
}
}
MicroProfileShutdown();
SDL_GL_DeleteContext(glcontext);
SDL_DestroyWindow(pWindow);
SDL_Quit();
return 0;
}
vr::EVRCompositorError CompositorBase::SubmitFrame(ovrLayerHeader const * const * layerPtrList, unsigned int layerCount)
{
MICROPROFILE_SCOPE(SubmitFrame);
// Other layers are interpreted as overlays.
std::vector<vr::VROverlayHandle_t> activeOverlays;
for (uint32_t i = 0; i < layerCount; i++)
{
if (layerPtrList[i] == nullptr)
continue;
if (layerPtrList[i]->Type == ovrLayerType_Quad)
{
ovrLayerQuad* layer = (ovrLayerQuad*)layerPtrList[i];
ovrTextureSwapChain chain = layer->ColorTexture;
// Every overlay is associated with a swapchain.
// This is necessary because the position of the layer may change in the array,
// which would otherwise cause flickering between overlays.
// TODO: Support multiple overlays using the same texture.
vr::VROverlayHandle_t overlay = layer->ColorTexture->Overlay;
if (overlay == vr::k_ulOverlayHandleInvalid)
{
overlay = CreateOverlay();
layer->ColorTexture->Overlay = overlay;
}
activeOverlays.push_back(overlay);
// Set the layer rendering order.
vr::VROverlay()->SetOverlaySortOrder(overlay, i);
// Transform the overlay.
vr::HmdMatrix34_t transform = REV::Matrix4f(layer->QuadPoseCenter);
vr::VROverlay()->SetOverlayWidthInMeters(overlay, layer->QuadSize.x);
if (layer->Header.Flags & ovrLayerFlag_HeadLocked)
vr::VROverlay()->SetOverlayTransformTrackedDeviceRelative(overlay, vr::k_unTrackedDeviceIndex_Hmd, &transform);
else
vr::VROverlay()->SetOverlayTransformAbsolute(overlay, vr::VRCompositor()->GetTrackingSpace(), &transform);
// Set the texture and show the overlay.
vr::VRTextureBounds_t bounds = ViewportToTextureBounds(layer->Viewport, layer->ColorTexture, layer->Header.Flags);
vr::Texture_t texture = chain->Textures[chain->SubmitIndex]->ToVRTexture();
vr::VROverlay()->SetOverlayTextureBounds(overlay, &bounds);
vr::VROverlay()->SetOverlayTexture(overlay, &texture);
chain->Submit();
// Show the overlay, unfortunately we have no control over the order in which
// overlays are drawn.
// TODO: Support ovrLayerFlag_HighQuality for overlays with anisotropic sampling.
// TODO: Handle overlay errors.
vr::VROverlay()->ShowOverlay(overlay);
}
else if (layerPtrList[i]->Type == ovrLayerType_EyeFov)
{
ovrLayerEyeFov* sceneLayer = (ovrLayerEyeFov*)layerPtrList[i];
if (!m_SceneLayer)
m_SceneLayer = layerPtrList[i];
else
SubmitFovLayer(sceneLayer->Viewport, sceneLayer->Fov, sceneLayer->ColorTexture, sceneLayer->Header.Flags);
}
else if (layerPtrList[i]->Type == ovrLayerType_EyeMatrix)
{
ovrLayerEyeMatrix* sceneLayer = (ovrLayerEyeMatrix*)layerPtrList[i];
ovrFovPort fov[ovrEye_Count] = {
MatrixToFovPort(sceneLayer->Matrix[ovrEye_Left]),
MatrixToFovPort(sceneLayer->Matrix[ovrEye_Right])
};
if (!m_SceneLayer)
m_SceneLayer = layerPtrList[i];
else
SubmitFovLayer(sceneLayer->Viewport, fov, sceneLayer->ColorTexture, sceneLayer->Header.Flags);
}
}
// Hide previous overlays that are not part of the current layers.
for (vr::VROverlayHandle_t overlay : m_ActiveOverlays)
{
// Find the overlay in the current active overlays, if it was not found then hide it.
// TODO: Handle overlay errors.
if (std::find(activeOverlays.begin(), activeOverlays.end(), overlay) == activeOverlays.end())
vr::VROverlay()->HideOverlay(overlay);
}
m_ActiveOverlays = activeOverlays;
vr::EVRCompositorError error = vr::VRCompositorError_None;
if (m_SceneLayer && m_SceneLayer->Type == ovrLayerType_EyeFov)
{
ovrLayerEyeFov* sceneLayer = (ovrLayerEyeFov*)m_SceneLayer;
error = SubmitSceneLayer(sceneLayer->Viewport, sceneLayer->Fov, sceneLayer->ColorTexture, sceneLayer->Header.Flags);
if (m_MirrorTexture && error == vr::VRCompositorError_None)
RenderMirrorTexture(m_MirrorTexture, sceneLayer->ColorTexture);
}
else if (m_SceneLayer && m_SceneLayer->Type == ovrLayerType_EyeMatrix)
{
ovrLayerEyeMatrix* sceneLayer = (ovrLayerEyeMatrix*)m_SceneLayer;
ovrFovPort fov[ovrEye_Count] = {
MatrixToFovPort(sceneLayer->Matrix[ovrEye_Left]),
MatrixToFovPort(sceneLayer->Matrix[ovrEye_Right])
};
error = SubmitSceneLayer(sceneLayer->Viewport, fov, sceneLayer->ColorTexture, sceneLayer->Header.Flags);
if (m_MirrorTexture && error == vr::VRCompositorError_None)
RenderMirrorTexture(m_MirrorTexture, sceneLayer->ColorTexture);
}
m_SceneLayer = nullptr;
return error;
}
OutputVertex Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
auto& config = g_state.regs.vs;
Common::Profiling::ScopeTimer timer(shader_category);
MICROPROFILE_SCOPE(GPU_VertexShader);
state.program_counter = config.main_offset;
state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0;
// Setup input register table
const auto& attribute_register_map = config.input_register_map;
// TODO: Instead of this cumbersome logic, just load the input data directly like
// for (int attr = 0; attr < num_attributes; ++attr) { input_attr[0] = state.registers.input[attribute_register_map.attribute0_register]; }
if (num_attributes > 0) state.registers.input[attribute_register_map.attribute0_register] = input.attr[0];
if (num_attributes > 1) state.registers.input[attribute_register_map.attribute1_register] = input.attr[1];
if (num_attributes > 2) state.registers.input[attribute_register_map.attribute2_register] = input.attr[2];
if (num_attributes > 3) state.registers.input[attribute_register_map.attribute3_register] = input.attr[3];
if (num_attributes > 4) state.registers.input[attribute_register_map.attribute4_register] = input.attr[4];
if (num_attributes > 5) state.registers.input[attribute_register_map.attribute5_register] = input.attr[5];
if (num_attributes > 6) state.registers.input[attribute_register_map.attribute6_register] = input.attr[6];
if (num_attributes > 7) state.registers.input[attribute_register_map.attribute7_register] = input.attr[7];
if (num_attributes > 8) state.registers.input[attribute_register_map.attribute8_register] = input.attr[8];
if (num_attributes > 9) state.registers.input[attribute_register_map.attribute9_register] = input.attr[9];
if (num_attributes > 10) state.registers.input[attribute_register_map.attribute10_register] = input.attr[10];
if (num_attributes > 11) state.registers.input[attribute_register_map.attribute11_register] = input.attr[11];
if (num_attributes > 12) state.registers.input[attribute_register_map.attribute12_register] = input.attr[12];
if (num_attributes > 13) state.registers.input[attribute_register_map.attribute13_register] = input.attr[13];
if (num_attributes > 14) state.registers.input[attribute_register_map.attribute14_register] = input.attr[14];
if (num_attributes > 15) state.registers.input[attribute_register_map.attribute15_register] = input.attr[15];
state.conditional_code[0] = false;
state.conditional_code[1] = false;
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled)
jit_shader(&state.registers);
else
RunInterpreter(state);
#else
RunInterpreter(state);
#endif // ARCHITECTURE_x86_64
// Setup output data
OutputVertex ret;
// TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to
// figure out what those circumstances are and enable the remaining outputs then.
for (int i = 0; i < 7; ++i) {
const auto& output_register_map = g_state.regs.vs_output_attributes[i]; // TODO: Don't hardcode VS here
u32 semantics[4] = {
output_register_map.map_x, output_register_map.map_y,
output_register_map.map_z, output_register_map.map_w
};
for (int comp = 0; comp < 4; ++comp) {
float24* out = ((float24*)&ret) + semantics[comp];
if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
*out = state.registers.output[i][comp];
} else {
// Zero output so that attributes which aren't output won't have denormals in them,
// which would slow us down later.
memset(out, 0, sizeof(*out));
}
}
}
// The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
for (int i = 0; i < 4; ++i) {
ret.color[i] = float24::FromFloat32(
std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
}
LOG_TRACE(Render_Software, "Output vertex: pos(%.2f, %.2f, %.2f, %.2f), quat(%.2f, %.2f, %.2f, %.2f), "
"col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f), view(%.2f, %.2f, %.2f)",
ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),
ret.quat.x.ToFloat32(), ret.quat.y.ToFloat32(), ret.quat.z.ToFloat32(), ret.quat.w.ToFloat32(),
ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32(),
ret.view.x.ToFloat32(), ret.view.y.ToFloat32(), ret.view.z.ToFloat32());
return ret;
}
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);
}
}
/// 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;
}
OutputVertex ShaderSetup::Run(UnitState<false>& state, const InputVertex& input, int num_attributes) {
auto& config = g_state.regs.vs;
MICROPROFILE_SCOPE(GPU_Shader);
state.program_counter = config.main_offset;
state.debug.max_offset = 0;
state.debug.max_opdesc_id = 0;
// Setup input register table
const auto& attribute_register_map = config.input_register_map;
for (unsigned i = 0; i < num_attributes; i++)
state.registers.input[attribute_register_map.GetRegisterForAttribute(i)] = input.attr[i];
state.conditional_code[0] = false;
state.conditional_code[1] = false;
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled)
jit_shader->Run(&state.registers, g_state.regs.vs.main_offset);
else
RunInterpreter(state);
#else
RunInterpreter(state);
#endif // ARCHITECTURE_x86_64
// Setup output data
OutputVertex ret;
// TODO(neobrain): Under some circumstances, up to 16 attributes may be output. We need to
// figure out what those circumstances are and enable the remaining outputs then.
unsigned index = 0;
for (unsigned i = 0; i < 7; ++i) {
if (index >= g_state.regs.vs_output_total)
break;
if ((g_state.regs.vs.output_mask & (1 << i)) == 0)
continue;
const auto& output_register_map = g_state.regs.vs_output_attributes[index]; // TODO: Don't hardcode VS here
u32 semantics[4] = {
output_register_map.map_x, output_register_map.map_y,
output_register_map.map_z, output_register_map.map_w
};
for (unsigned comp = 0; comp < 4; ++comp) {
float24* out = ((float24*)&ret) + semantics[comp];
if (semantics[comp] != Regs::VSOutputAttributes::INVALID) {
*out = state.registers.output[i][comp];
} else {
// Zero output so that attributes which aren't output won't have denormals in them,
// which would slow us down later.
memset(out, 0, sizeof(*out));
}
}
index++;
}
// The hardware takes the absolute and saturates vertex colors like this, *before* doing interpolation
for (unsigned i = 0; i < 4; ++i) {
ret.color[i] = float24::FromFloat32(
std::fmin(std::fabs(ret.color[i].ToFloat32()), 1.0f));
}
LOG_TRACE(HW_GPU, "Output vertex: pos(%.2f, %.2f, %.2f, %.2f), quat(%.2f, %.2f, %.2f, %.2f), "
"col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f), view(%.2f, %.2f, %.2f)",
ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),
ret.quat.x.ToFloat32(), ret.quat.y.ToFloat32(), ret.quat.z.ToFloat32(), ret.quat.w.ToFloat32(),
ret.color.x.ToFloat32(), ret.color.y.ToFloat32(), ret.color.z.ToFloat32(), ret.color.w.ToFloat32(),
ret.tc0.u().ToFloat32(), ret.tc0.v().ToFloat32(),
ret.view.x.ToFloat32(), ret.view.y.ToFloat32(), ret.view.z.ToFloat32());
return ret;
}
int main(int argc, char* argv[])
{
MICROPROFILE_REGISTER_GROUP("Thread0", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("Thread1", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("Thread2", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("Thread2xx", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("MAIN", "main", 0x88fff00f);
g_QueueGraphics = MICROPROFILE_GPU_INIT_QUEUE("GPU-Graphics-Queue");
printf("press 'z' to toggle microprofile drawing\n");
printf("press 'right shift' to pause microprofile update\n");
printf("press 'x' to toggle profiling\n");
printf("press 'c' to toggle enable of all profiler groups\n");
MicroProfileOnThreadCreate("AA_Main");
if(SDL_Init(SDL_INIT_VIDEO) < 0) {
return 1;
}
SDL_GL_SetAttribute(SDL_GL_RED_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_GREEN_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_BLUE_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_ALPHA_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_DEPTH_SIZE, 24);
SDL_GL_SetAttribute(SDL_GL_STENCIL_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_BUFFER_SIZE, 32);
SDL_GL_SetAttribute(SDL_GL_DOUBLEBUFFER, 1);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MAJOR_VERSION, 3);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MINOR_VERSION, 2);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_CORE);
SDL_GL_SetSwapInterval(1);
SDL_Window * pWindow = SDL_CreateWindow("microprofiledemo", 10, 10, WIDTH, HEIGHT, SDL_WINDOW_OPENGL);
if(!pWindow)
return 1;
SDL_GLContext glcontext = SDL_GL_CreateContext(pWindow);
glewExperimental=1;
GLenum err=glewInit();
if(err!=GLEW_OK)
{
__BREAK();
}
glGetError(); //glew generates an error
#if MICROPROFILE_ENABLED
MicroProfileGpuInitGL();
MicroProfileDrawInit();
MP_ASSERT(glGetError() == 0);
MicroProfileToggleDisplayMode();
MicroProfileInitUI();
MicroProfileCustomGroup("Custom1", 2, 47, 2.f, MICROPROFILE_CUSTOM_BARS);
MicroProfileCustomGroupAddTimer("Custom1", "MicroProfile", "Draw");
MicroProfileCustomGroupAddTimer("Custom1", "MicroProfile", "Detailed View");
MicroProfileCustomGroup("Custom2", 2, 100, 20.f, MICROPROFILE_CUSTOM_BARS|MICROPROFILE_CUSTOM_BAR_SOURCE_MAX);
MicroProfileCustomGroupAddTimer("Custom2", "MicroProfile", "Draw");
MicroProfileCustomGroupAddTimer("Custom2", "MicroProfile", "Detailed View");
MicroProfileCustomGroup("Custom3", 2, 0, 5.f, MICROPROFILE_CUSTOM_STACK|MICROPROFILE_CUSTOM_STACK_SOURCE_MAX);
MicroProfileCustomGroupAddTimer("Custom3", "MicroProfile", "Draw");
MicroProfileCustomGroupAddTimer("Custom3", "MicroProfile", "Detailed View");
MicroProfileCustomGroup("ThreadSafe", 6, 10, 600.f, MICROPROFILE_CUSTOM_BARS | MICROPROFILE_CUSTOM_STACK);
MicroProfileCustomGroupAddTimer("ThreadSafe", "ThreadSafe", "main");
MicroProfileCustomGroupAddTimer("ThreadSafe", "ThreadSafe", "inner0");
MicroProfileCustomGroupAddTimer("ThreadSafe", "ThreadSafe", "inner1");
MicroProfileCustomGroupAddTimer("ThreadSafe", "ThreadSafe", "inner2");
MicroProfileCustomGroupAddTimer("ThreadSafe", "ThreadSafe", "inner3");
MicroProfileCustomGroupAddTimer("ThreadSafe", "ThreadSafe", "inner4");
#endif
MICROPROFILE_COUNTER_CONFIG("memory/main", MICROPROFILE_COUNTER_FORMAT_BYTES, 10ll<<30ll, 0);
MICROPROFILE_COUNTER_CONFIG("memory/gpu/indexbuffers", MICROPROFILE_COUNTER_FORMAT_BYTES, 0, 0);
MICROPROFILE_COUNTER_CONFIG("memory/gpu/vertexbuffers", MICROPROFILE_COUNTER_FORMAT_BYTES, 0, 0);
MICROPROFILE_COUNTER_CONFIG("memory/mainx", MICROPROFILE_COUNTER_FORMAT_BYTES, 10000, 0);
MICROPROFILE_COUNTER_ADD("memory/main", 1000);
MICROPROFILE_COUNTER_ADD("memory/gpu/vertexbuffers", 1000);
MICROPROFILE_COUNTER_ADD("memory/gpu/indexbuffers", 200);
MICROPROFILE_COUNTER_ADD("memory//", 10<<10);
MICROPROFILE_COUNTER_ADD("memory//main", (32ll<<30ll) + (1ll <<29ll));
MICROPROFILE_COUNTER_ADD("//memory//mainx/\\//", 1000);
MICROPROFILE_COUNTER_ADD("//memoryx//mainx/", 1000);
MICROPROFILE_COUNTER_ADD("//memoryy//main/", -1000000);
MICROPROFILE_COUNTER_ADD("//\\\\///lala////lelel", 1000);
MICROPROFILE_COUNTER_CONFIG("engine/frames", MICROPROFILE_COUNTER_FORMAT_DEFAULT, 1000, 0);
MICROPROFILE_COUNTER_SET("fisk/geder/", 42);
MICROPROFILE_COUNTER_SET("fisk/aborre/", -2002);
MICROPROFILE_COUNTER_SET_LIMIT("fisk/aborre/", 120);
static int Frames = 0;
static uint64_t FramesX = 0;
MICROPROFILE_COUNTER_SET_INT64_PTR("frames/int64", &FramesX);
MICROPROFILE_COUNTER_SET_INT32_PTR("frames/int32", &Frames);
MICROPROFILE_COUNTER_CONFIG("/test/sinus", MICROPROFILE_COUNTER_FORMAT_BYTES, 0, MICROPROFILE_COUNTER_FLAG_DETAILED);
MICROPROFILE_COUNTER_CONFIG("/test/cosinus", MICROPROFILE_COUNTER_FORMAT_DEFAULT, 0, MICROPROFILE_COUNTER_FLAG_DETAILED);
MICROPROFILE_COUNTER_CONFIG("/runtime/sdl_frame_events", MICROPROFILE_COUNTER_FORMAT_DEFAULT, 0, MICROPROFILE_COUNTER_FLAG_DETAILED);
StartFakeWork();
while(!g_nQuit)
{
Frames++;
FramesX += 1024*1024;
MICROPROFILE_SCOPE(MAIN);
MICROPROFILE_COUNTER_ADD("engine/frames", 1);
SDL_Event Evt;
while(SDL_PollEvent(&Evt))
{
MICROPROFILE_COUNTER_LOCAL_ADD(SDLFrameEvents, 1);
HandleEvent(&Evt);
}
MICROPROFILE_COUNTER_LOCAL_UPDATE_SET(SDLFrameEvents);
glClearColor(0.3f,0.4f,0.6f,0.f);
glViewport(0, 0, WIDTH, HEIGHT);
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
#if 1||FAKE_WORK
{
MICROPROFILE_SCOPEI("BMain", "Dummy", 0xff3399ff);
for(uint32_t i = 0; i < 14; ++i)
{
MICROPROFILE_SCOPEI("BMain", "1ms", 0xff3399ff);
MICROPROFILE_META_CPU("Sleep",1);
usleep(1000);
}
}
#endif
MicroProfileMouseButton(g_MouseDown0, g_MouseDown1);
MicroProfileMousePosition(g_MouseX, g_MouseY, g_MouseDelta);
g_MouseDelta = 0;
MicroProfileFlip(0);
static float f = 0;
f += 0.1f;
int sinus = (int)(10000000 * (sinf(f)));
int cosinus = int(cosf(f*1.3f) * 100000 + 50000);
MICROPROFILE_COUNTER_SET("/test/sinus", sinus);
MICROPROFILE_COUNTER_SET("/test/cosinus", cosinus);
{
MICROPROFILE_SCOPEGPUI("MicroProfileDraw", 0x88dd44);
float projection[16];
float left = 0.f;
float right = WIDTH;
float bottom = HEIGHT;
float top = 0.f;
float near = -1.f;
float far = 1.f;
memset(&projection[0], 0, sizeof(projection));
projection[0] = 2.0f / (right - left);
projection[5] = 2.0f / (top - bottom);
projection[10] = -2.0f / (far - near);
projection[12] = - (right + left) / (right - left);
projection[13] = - (top + bottom) / (top - bottom);
projection[14] = - (far + near) / (far - near);
projection[15] = 1.f;
glEnable(GL_BLEND);
glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
glDisable(GL_DEPTH_TEST);
#if MICROPROFILE_ENABLED
MicroProfileBeginDraw(WIDTH, HEIGHT, &projection[0]);
MicroProfileDraw(WIDTH, HEIGHT);
MicroProfileEndDraw();
#endif
glDisable(GL_BLEND);
}
MICROPROFILE_SCOPEI("MAIN", "Flip", 0xffee00);
SDL_GL_SwapWindow(pWindow);
}
StopFakeWork();
MicroProfileShutdown();
SDL_GL_DeleteContext(glcontext);
SDL_DestroyWindow(pWindow);
SDL_Quit();
return 0;
}
void WorkerThread(int threadId)
{
char name[100];
snprintf(name, 99, "Worker%d", threadId);
MicroProfileOnThreadCreate(&name[0]);
uint32_t c0 = 0xff3399ff;
uint32_t c1 = 0xffff99ff;
uint32_t c2 = 0xff33ff00;
uint32_t c3 = 0xff3399ff;
uint32_t c4 = 0xff33ff33;
while(!g_nQuit)
{
switch(threadId)
{
case 0:
{
usleep(100);
{
MICROPROFILE_SCOPEI("Thread0", "Work Thread0", c4);
MICROPROFILE_META_CPU("Sleep",10);
usleep(200);
{
MICROPROFILE_SCOPEI("Thread0", "Work Thread1", c3);
MICROPROFILE_META_CPU("DrawCalls", 1);
MICROPROFILE_META_CPU("DrawCalls", 1);
MICROPROFILE_META_CPU("DrawCalls", 1);
usleep(200);
{
MICROPROFILE_SCOPEI("Thread0", "Work Thread2", c2);
MICROPROFILE_META_CPU("DrawCalls", 1);
usleep(200);
{
MICROPROFILE_SCOPEI("Thread0", "Work Thread3", c1);
MICROPROFILE_META_CPU("DrawCalls", 4);
MICROPROFILE_META_CPU("Triangles",1000);
usleep(200);
}
}
}
}
}
break;
case 1:
{
usleep(100);
MICROPROFILE_SCOPEI("Thread1", "Work Thread 1", c1);
usleep(2000);
}
break;
case 2:
{
usleep(1000);
{
MICROPROFILE_SCOPEI("Thread2", "Worker2", c0);
spinsleep(100000);
{
MICROPROFILE_SCOPEI("Thread2", "InnerWork0", c1);
spinsleep(100);
{
MICROPROFILE_SCOPEI("Thread2", "InnerWork1", c2);
usleep(100);
{
MICROPROFILE_SCOPEI("Thread2", "InnerWork2", c3);
usleep(100);
{
// for(uint32_t i = 0; i < 1000; ++i)
// {
// MICROPROFILE_SCOPEI("Thread2", "InnerWork3", c4);
// spinsleep(10);
// }
}
}
}
}
}
}
break;
case 3:
{
MICROPROFILE_SCOPEI("ThreadWork", "MAIN", c0);
usleep(1000);;
for(uint32_t i = 0; i < 10; ++i)
{
MICROPROFILE_SCOPEI("ThreadWork", "Inner0", c1);
usleep(100);
for(uint32_t j = 0; j < 4; ++j)
{
MICROPROFILE_SCOPEI("ThreadWork", "Inner1", c4);
usleep(50);
MICROPROFILE_SCOPEI("ThreadWork", "Inner1", c4);
usleep(50);
MICROPROFILE_SCOPEI("ThreadWork", "Inner2", c2);
usleep(50);
MICROPROFILE_SCOPEI("ThreadWork", "Inner3", c3);
usleep(50);
MICROPROFILE_SCOPEI("ThreadWork", "Inner4", c3);
usleep(50);
}
}
}
break;
default:
MICROPROFILE_SCOPE(ThreadSafeMain);
usleep(1000);;
for(uint32_t i = 0; i < 5; ++i)
{
MICROPROFILE_SCOPE(ThreadSafeInner0);
usleep(1000);
for(uint32_t j = 0; j < 4; ++j)
{
MICROPROFILE_META_CPU("custom_very_long_meta", 1);
MICROPROFILE_SCOPE(ThreadSafeInner1);
usleep(500);
MICROPROFILE_SCOPE(ThreadSafeInner2);
usleep(150);
MICROPROFILE_SCOPE(ThreadSafeInner3);
usleep(150);
MICROPROFILE_SCOPE(ThreadSafeInner4);
usleep(150);
}
}
break;
}
}
}
int main(int argc, char* argv[])
{
MICROPROFILE_REGISTER_GROUP("Thread0", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("Thread1", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("Thread2", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("Thread2xx", "Threads", 0x88008800);
MICROPROFILE_REGISTER_GROUP("GPU", "main", 0x88fff00f);
MICROPROFILE_REGISTER_GROUP("MAIN", "main", 0x88fff00f);
printf("press 'z' to toggle microprofile drawing\n");
printf("press 'right shift' to pause microprofile update\n");
printf("press 'x' to toggle profiling\n");
printf("press 'c' to toggle enable of all profiler groups\n");
MicroProfileOnThreadCreate("AA_Main");
if(SDL_Init(SDL_INIT_VIDEO) < 0) {
return 1;
}
SDL_GL_SetAttribute(SDL_GL_RED_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_GREEN_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_BLUE_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_ALPHA_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_DEPTH_SIZE, 24);
SDL_GL_SetAttribute(SDL_GL_STENCIL_SIZE, 8);
SDL_GL_SetAttribute(SDL_GL_BUFFER_SIZE, 32);
SDL_GL_SetAttribute(SDL_GL_DOUBLEBUFFER, 1);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MAJOR_VERSION, 3);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MINOR_VERSION, 2);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_CORE);
SDL_GL_SetSwapInterval(1);
SDL_Window * pWindow = SDL_CreateWindow("microprofiledemo", 10, 10, WIDTH, HEIGHT, SDL_WINDOW_OPENGL);
if(!pWindow)
return 1;
SDL_GLContext glcontext = SDL_GL_CreateContext(pWindow);
glewExperimental=1;
GLenum err=glewInit();
if(err!=GLEW_OK)
{
__BREAK();
}
glGetError(); //glew generates an error
#if MICROPROFILE_ENABLED
MicroProfileGpuInitGL();
MicroProfileDrawInit();
MP_ASSERT(glGetError() == 0);
MicroProfileToggleDisplayMode();
#endif
StartFakeWork();
while(!g_nQuit)
{
MICROPROFILE_SCOPE(MAIN);
SDL_Event Evt;
while(SDL_PollEvent(&Evt))
{
HandleEvent(&Evt);
}
glClearColor(0.3f,0.4f,0.6f,0.f);
glViewport(0, 0, WIDTH, HEIGHT);
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
#if 1||FAKE_WORK
{
MICROPROFILE_SCOPEI("BMain", "Dummy", 0xff3399ff);
for(uint32_t i = 0; i < 14; ++i)
{
MICROPROFILE_SCOPEI("BMain", "1ms", 0xff3399ff);
MICROPROFILE_META_CPU("Sleep",1);
usleep(1000);
}
}
#endif
MicroProfileMouseButton(g_MouseDown0, g_MouseDown1);
MicroProfileMousePosition(g_MouseX, g_MouseY, g_MouseDelta);
g_MouseDelta = 0;
MicroProfileFlip();
{
MICROPROFILE_SCOPEGPUI("MicroProfileDraw", 0x88dd44);
float projection[16];
float left = 0.f;
float right = WIDTH;
float bottom = HEIGHT;
float top = 0.f;
float near = -1.f;
float far = 1.f;
memset(&projection[0], 0, sizeof(projection));
projection[0] = 2.0f / (right - left);
projection[5] = 2.0f / (top - bottom);
projection[10] = -2.0f / (far - near);
projection[12] = - (right + left) / (right - left);
projection[13] = - (top + bottom) / (top - bottom);
projection[14] = - (far + near) / (far - near);
projection[15] = 1.f;
glEnable(GL_BLEND);
glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
glDisable(GL_DEPTH_TEST);
#if MICROPROFILE_ENABLED
MicroProfileBeginDraw(WIDTH, HEIGHT, &projection[0]);
MicroProfileDraw(WIDTH, HEIGHT);
MicroProfileEndDraw();
#endif
glDisable(GL_BLEND);
}
MICROPROFILE_SCOPEI("MAIN", "Flip", 0xffee00);
SDL_GL_SwapWindow(pWindow);
}
StopFakeWork();
MicroProfileShutdown();
SDL_GL_DeleteContext(glcontext);
SDL_DestroyWindow(pWindow);
SDL_Quit();
return 0;
}
