case Regs::PixelFormat::RGB565:
return Color::DecodeRGB565(src_pixel);
case Regs::PixelFormat::RGB5A1:
return Color::DecodeRGB5A1(src_pixel);
case Regs::PixelFormat::RGBA4:
return Color::DecodeRGBA4(src_pixel);
default:
LOG_ERROR(HW_GPU, "Unknown source framebuffer format %x", input_format);
return {0, 0, 0, 0};
}
}
MICROPROFILE_DEFINE(GPU_DisplayTransfer, "GPU", "DisplayTransfer", MP_RGB(100, 100, 255));
MICROPROFILE_DEFINE(GPU_CmdlistProcessing, "GPU", "Cmdlist Processing", MP_RGB(100, 255, 100));
static void MemoryFill(const Regs::MemoryFillConfig& config) {
const PAddr start_addr = config.GetStartAddress();
const PAddr end_addr = config.GetEndAddress();
// TODO: do hwtest with these cases
if (!Memory::IsValidPhysicalAddress(start_addr)) {
LOG_CRITICAL(HW_GPU, "invalid start address 0x%08X", start_addr);
return;
}
if (!Memory::IsValidPhysicalAddress(end_addr)) {
LOG_CRITICAL(HW_GPU, "invalid end address 0x%08X", end_addr);
return;
s64 ticks = CoreTiming::GetDowncount();
return static_cast<u64>(ticks <= 0 ? 0 : ticks);
}
ARM_Dynarmic& parent;
};
ARM_Dynarmic::ARM_Dynarmic(PrivilegeMode initial_mode)
: cb(std::make_unique<DynarmicUserCallbacks>(*this)) {
interpreter_state = std::make_shared<ARMul_State>(initial_mode);
PageTableChanged();
}
ARM_Dynarmic::~ARM_Dynarmic() = default;
MICROPROFILE_DEFINE(ARM_Jit, "ARM JIT", "ARM JIT", MP_RGB(255, 64, 64));
void ARM_Dynarmic::Run() {
ASSERT(Memory::GetCurrentPageTable() == current_page_table);
MICROPROFILE_SCOPE(ARM_Jit);
jit->Run();
}
void ARM_Dynarmic::Step() {
cb->InterpreterFallback(jit->Regs()[15], 1);
}
void ARM_Dynarmic::SetPC(u32 pc) {
jit->Regs()[15] = pc;
}
}
}
}
GSUnitState::GSUnitState() : UnitState(&emitter) {}
void GSUnitState::SetVertexHandler(VertexHandler vertex_handler, WindingSetter winding_setter) {
emitter.handlers->vertex_handler = std::move(vertex_handler);
emitter.handlers->winding_setter = std::move(winding_setter);
}
void GSUnitState::ConfigOutput(const ShaderRegs& config) {
emitter.output_mask = config.output_mask;
}
MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240));
#ifdef ARCHITECTURE_x86_64
static std::unique_ptr<JitX64Engine> jit_engine;
#endif // ARCHITECTURE_x86_64
static InterpreterEngine interpreter_engine;
ShaderEngine* GetEngine() {
#ifdef ARCHITECTURE_x86_64
// TODO(yuriks): Re-initialize on each change rather than being persistent
if (VideoCore::g_shader_jit_enabled) {
if (jit_engine == nullptr) {
jit_engine = std::make_unique<JitX64Engine>();
}
return jit_engine.get();
}
#include "common/make_unique.h"
#include "common/math_util.h"
#include "common/microprofile.h"
#include "common/vector_math.h"
#include "core/memory.h"
#include "video_core/debug_utils/debug_utils.h"
#include "video_core/renderer_opengl/gl_rasterizer_cache.h"
#include "video_core/renderer_opengl/pica_to_gl.h"
RasterizerCacheOpenGL::~RasterizerCacheOpenGL() {
InvalidateAll();
}
MICROPROFILE_DEFINE(OpenGL_TextureUpload, "OpenGL", "Texture Upload", MP_RGB(128, 64, 192));
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();
// right), but the PDC0/1 interrupts are signaled for every registered thread.
if (interrupt_id == InterruptId::PDC0 || interrupt_id == InterruptId::PDC1) {
for (u32 thread_id = 0; thread_id < MaxGSPThreads; ++thread_id) {
SignalInterruptForThread(interrupt_id, thread_id);
}
return;
}
// For normal interrupts, don't do anything if no process has acquired the GPU right.
if (active_thread_id == -1)
return;
SignalInterruptForThread(interrupt_id, active_thread_id);
}
MICROPROFILE_DEFINE(GPU_GSP_DMA, "GPU", "GSP DMA", MP_RGB(100, 0, 255));
/// 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();
namespace Shader {
#ifdef ARCHITECTURE_x86_64
static std::unordered_map<u64, std::unique_ptr<JitShader>> shader_map;
static const JitShader* jit_shader;
#endif // ARCHITECTURE_x86_64
void ClearCache() {
#ifdef ARCHITECTURE_x86_64
shader_map.clear();
#endif // ARCHITECTURE_x86_64
}
void ShaderSetup::Setup() {
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled) {
u64 cache_key = (Common::ComputeHash64(&g_state.vs.program_code, sizeof(g_state.vs.program_code)) ^
Common::ComputeHash64(&g_state.vs.swizzle_data, sizeof(g_state.vs.swizzle_data)));
auto iter = shader_map.find(cache_key);
if (iter != shader_map.end()) {
jit_shader = iter->second.get();
} else {
auto shader = std::make_unique<JitShader>();
shader->Compile();
jit_shader = shader.get();
shader_map[cache_key] = std::move(shader);
}
}
#endif // ARCHITECTURE_x86_64
}
MICROPROFILE_DEFINE(GPU_Shader, "GPU", "Shader", MP_RGB(50, 50, 240));
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;
}
DebugData<true> ShaderSetup::ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup) {
UnitState<true> state;
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;
float24 dummy_register;
boost::fill(state.registers.input, &dummy_register);
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;
RunInterpreter(state);
return state.debug;
}
} // namespace Shader
namespace Shader {
#ifdef ARCHITECTURE_x86_64
static std::unordered_map<u64, CompiledShader*> shader_map;
static JitCompiler jit;
static CompiledShader* jit_shader;
#endif // ARCHITECTURE_x86_64
void Setup(UnitState<false>& state) {
#ifdef ARCHITECTURE_x86_64
if (VideoCore::g_shader_jit_enabled) {
u64 cache_key = (Common::ComputeHash64(&g_state.vs.program_code, sizeof(g_state.vs.program_code)) ^
Common::ComputeHash64(&g_state.vs.swizzle_data, sizeof(g_state.vs.swizzle_data)) ^
g_state.regs.vs.main_offset);
auto iter = shader_map.find(cache_key);
if (iter != shader_map.end()) {
jit_shader = iter->second;
} else {
jit_shader = jit.Compile();
shader_map.emplace(cache_key, jit_shader);
}
}
#endif // ARCHITECTURE_x86_64
}
void Shutdown() {
#ifdef ARCHITECTURE_x86_64
shader_map.clear();
#endif // ARCHITECTURE_x86_64
}
static Common::Profiling::TimingCategory shader_category("Vertex Shader");
MICROPROFILE_DEFINE(GPU_VertexShader, "GPU", "Vertex Shader", MP_RGB(50, 50, 240));
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;
}
DebugData<true> ProduceDebugInfo(const InputVertex& input, int num_attributes, const Regs::ShaderConfig& config, const ShaderSetup& setup) {
UnitState<true> state;
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;
float24 dummy_register;
boost::fill(state.registers.input, &dummy_register);
if (num_attributes > 0) state.registers.input[attribute_register_map.attribute0_register] = &input.attr[0].x;
if (num_attributes > 1) state.registers.input[attribute_register_map.attribute1_register] = &input.attr[1].x;
if (num_attributes > 2) state.registers.input[attribute_register_map.attribute2_register] = &input.attr[2].x;
if (num_attributes > 3) state.registers.input[attribute_register_map.attribute3_register] = &input.attr[3].x;
if (num_attributes > 4) state.registers.input[attribute_register_map.attribute4_register] = &input.attr[4].x;
if (num_attributes > 5) state.registers.input[attribute_register_map.attribute5_register] = &input.attr[5].x;
if (num_attributes > 6) state.registers.input[attribute_register_map.attribute6_register] = &input.attr[6].x;
if (num_attributes > 7) state.registers.input[attribute_register_map.attribute7_register] = &input.attr[7].x;
if (num_attributes > 8) state.registers.input[attribute_register_map.attribute8_register] = &input.attr[8].x;
if (num_attributes > 9) state.registers.input[attribute_register_map.attribute9_register] = &input.attr[9].x;
if (num_attributes > 10) state.registers.input[attribute_register_map.attribute10_register] = &input.attr[10].x;
if (num_attributes > 11) state.registers.input[attribute_register_map.attribute11_register] = &input.attr[11].x;
if (num_attributes > 12) state.registers.input[attribute_register_map.attribute12_register] = &input.attr[12].x;
if (num_attributes > 13) state.registers.input[attribute_register_map.attribute13_register] = &input.attr[13].x;
if (num_attributes > 14) state.registers.input[attribute_register_map.attribute14_register] = &input.attr[14].x;
if (num_attributes > 15) state.registers.input[attribute_register_map.attribute15_register] = &input.attr[15].x;
state.conditional_code[0] = false;
state.conditional_code[1] = false;
RunInterpreter(state);
return state.debug;
}
} // namespace Shader
