Merge pull request #410 from wheremyfoodat/dsp

WIP: DSP support

include/audio/dsp_core.hpp

@@ -0,0 +1,55 @@
+#pragma once
+#include <array>
+#include <functional>
+#include <memory>
+#include <optional>
+#include <string>
+#include <vector>
+
+#include "helpers.hpp"
+#include "logger.hpp"
+#include "scheduler.hpp"
+
+// The DSP core must have access to the DSP service to be able to trigger interrupts properly
+class DSPService;
+class Memory;
+
+namespace Audio {
+	// There are 160 stereo samples in 1 audio frame, so 320 samples total
+	static constexpr u64 samplesInFrame = 160;
+	// 1 frame = 4096 DSP cycles = 8192 ARM11 cycles
+	static constexpr u64 cyclesPerFrame = samplesInFrame * 8192;
+	// For LLE DSP cores, we run the DSP for N cycles at a time, every N*2 arm11 cycles since the ARM11 runs twice as fast
+	static constexpr u64 lleSlice = 16384;
+
+	class DSPCore {
+	  protected:
+		Memory& mem;
+		Scheduler& scheduler;
+		DSPService& dspService;
+
+		MAKE_LOG_FUNCTION(log, dspLogger)
+
+	  public:
+		enum class Type { Null, Teakra };
+		DSPCore(Memory& mem, Scheduler& scheduler, DSPService& dspService) : mem(mem), scheduler(scheduler), dspService(dspService) {}
+
+		virtual void reset() = 0;
+		virtual void runAudioFrame() = 0;
+		virtual u8* getDspMemory() = 0;
+
+		virtual u16 recvData(u32 regId) = 0;
+		virtual bool recvDataIsReady(u32 regId) = 0;
+		virtual void setSemaphore(u16 value) = 0;
+		virtual void writeProcessPipe(u32 channel, u32 size, u32 buffer) = 0;
+		virtual std::vector<u8> readPipe(u32 channel, u32 peer, u32 size, u32 buffer) = 0;
+		virtual void loadComponent(std::vector<u8>& data, u32 programMask, u32 dataMask) = 0;
+		virtual void unloadComponent() = 0;
+		virtual void setSemaphoreMask(u16 value) = 0;
+
+		static Audio::DSPCore::Type typeFromString(std::string inString);
+		static const char* typeToString(Audio::DSPCore::Type type);
+	};
+
+	std::unique_ptr<DSPCore> makeDSPCore(DSPCore::Type type, Memory& mem, Scheduler& scheduler, DSPService& dspService);
+}  // namespace Audio

include/audio/null_core.hpp

@@ -0,0 +1,45 @@
+#pragma once
+#include <array>
+
+#include "audio/dsp_core.hpp"
+#include "memory.hpp"
+
+namespace Audio {
+	class NullDSP : public DSPCore {
+		enum class DSPState : u32 {
+			Off,
+			On,
+			Slep,
+		};
+
+		// Number of DSP pipes
+		static constexpr size_t pipeCount = 8;
+		DSPState dspState;
+
+		std::array<std::vector<u8>, pipeCount> pipeData;  // The data of each pipe
+		std::array<u8, Memory::DSP_RAM_SIZE> dspRam;
+
+		void resetAudioPipe();
+		bool loaded = false; // Have we loaded a component?
+
+	  public:
+		NullDSP(Memory& mem, Scheduler& scheduler, DSPService& dspService) : DSPCore(mem, scheduler, dspService) {}
+
+		void reset() override;
+		void runAudioFrame() override;
+
+		u8* getDspMemory() override { return dspRam.data(); }
+
+		u16 recvData(u32 regId) override;
+		bool recvDataIsReady(u32 regId) override { return true; }  // Treat data as always ready
+		void writeProcessPipe(u32 channel, u32 size, u32 buffer) override;
+		std::vector<u8> readPipe(u32 channel, u32 peer, u32 size, u32 buffer) override;
+
+		// NOPs for null DSP core
+		void loadComponent(std::vector<u8>& data, u32 programMask, u32 dataMask) override;
+		void unloadComponent() override;
+		void setSemaphore(u16 value) override {}
+		void setSemaphoreMask(u16 value) override {}
+	};
+
+}  // namespace Audio

include/audio/teakra_core.hpp

@@ -0,0 +1,99 @@
+#pragma once
+#include "audio/dsp_core.hpp"
+#include "memory.hpp"
+#include "swap.hpp"
+#include "teakra/teakra.h"
+
+namespace Audio {
+	class TeakraDSP : public DSPCore {
+		Teakra::Teakra teakra;
+		u32 pipeBaseAddr;
+		bool running;  // Is the DSP running?
+		bool loaded;   // Have we finished loading a binary with LoadComponent?
+
+		// Get a pointer to a data memory address
+		u8* getDataPointer(u32 address) { return getDspMemory() + Memory::DSP_DATA_MEMORY_OFFSET + address; }
+
+		enum class PipeDirection {
+			DSPtoCPU = 0,
+			CPUtoDSP = 1,
+		};
+
+		// A lot of Teakra integration code, especially pipe stuff is based on Citra's integration here:
+		// https://github.com/citra-emu/citra/blob/master/src/audio_core/lle/lle.cpp
+		struct PipeStatus {
+			// All addresses and sizes here refer to byte values, NOT 16-bit values.
+			u16_le address;
+			u16_le byteSize;
+			u16_le readPointer;
+			u16_le writePointer;
+			u8 slot;
+			u8 flags;
+
+			static constexpr u16 wrapBit = 0x8000;
+			static constexpr u16 pointerMask = 0x7FFF;
+
+			bool isFull() const { return (readPointer ^ writePointer) == wrapBit; }
+			bool isEmpty() const { return (readPointer ^ writePointer) == 0; }
+
+			// isWrapped: Are read and write pointers in different memory passes.
+			// true:   xxxx]----[xxxx (data is wrapping around the end of memory)
+			// false:  ----[xxxx]----
+			bool isWrapped() const { return (readPointer ^ writePointer) >= wrapBit; }
+		};
+		static_assert(sizeof(PipeStatus) == 10, "Teakra: Pipe Status size is wrong");
+		static constexpr u8 pipeToSlotIndex(u8 pipe, PipeDirection direction) { return (pipe * 2) + u8(direction); }
+
+		PipeStatus getPipeStatus(u8 pipe, PipeDirection direction) {
+			PipeStatus ret;
+			const u8 index = pipeToSlotIndex(pipe, direction);
+
+			std::memcpy(&ret, getDataPointer(pipeBaseAddr * 2 + index * sizeof(PipeStatus)), sizeof(PipeStatus));
+			return ret;
+		}
+
+		void updatePipeStatus(const PipeStatus& status) {
+			u8 slot = status.slot;
+			u8* statusAddress = getDataPointer(pipeBaseAddr * 2 + slot * sizeof(PipeStatus));
+
+			if (slot % 2 == 0) {
+				std::memcpy(statusAddress + 4, &status.readPointer, sizeof(u16));
+			} else {
+				std::memcpy(statusAddress + 6, &status.writePointer, sizeof(u16));
+			}
+		}
+
+		bool signalledData;
+		bool signalledSemaphore;
+
+		// Run 1 slice of DSP instructions
+		void runSlice() {
+			if (running) {
+				teakra.Run(Audio::lleSlice);
+			}
+		}
+
+	  public:
+		TeakraDSP(Memory& mem, Scheduler& scheduler, DSPService& dspService);
+
+		void reset() override;
+
+		// Run 1 slice of DSP instructions and schedule the next audio frame
+		void runAudioFrame() override {
+			runSlice();
+			scheduler.addEvent(Scheduler::EventType::RunDSP, scheduler.currentTimestamp + Audio::lleSlice * 2);
+		}
+
+		u8* getDspMemory() override { return teakra.GetDspMemory().data(); }
+
+		u16 recvData(u32 regId) override { return teakra.RecvData(regId); }
+		bool recvDataIsReady(u32 regId) override { return teakra.RecvDataIsReady(regId); }
+		void setSemaphore(u16 value) override { teakra.SetSemaphore(value); }
+		void setSemaphoreMask(u16 value) override { teakra.MaskSemaphore(value); }
+
+		void writeProcessPipe(u32 channel, u32 size, u32 buffer) override;
+		std::vector<u8> readPipe(u32 channel, u32 peer, u32 size, u32 buffer) override;
+		void loadComponent(std::vector<u8>& data, u32 programMask, u32 dataMask) override;
+		void unloadComponent() override;
+	};
+}  // namespace Audio

include/config.hpp

@@ -1,6 +1,7 @@
 #pragma once
 #include <filesystem>
 
+#include "audio/dsp_core.hpp"
 #include "renderer.hpp"
 
 // Remember to initialize every field here to its default value otherwise bad things will happen
@@ -15,6 +16,7 @@ struct EmulatorConfig {
 	bool shaderJitEnabled = shaderJitDefault;
 	bool discordRpcEnabled = false;
 	RendererType rendererType = RendererType::OpenGL;
+	Audio::DSPCore::Type dspType = Audio::DSPCore::Type::Null;
 
 	bool sdCardInserted = true;
 	bool sdWriteProtected = false;

include/emulator.hpp

@@ -2,10 +2,12 @@
 
 #include <filesystem>
 #include <fstream>
+#include <memory>
 #include <optional>
 #include <span>
 
 #include "PICA/gpu.hpp"
+#include "audio/dsp_core.hpp"
 #include "cheats.hpp"
 #include "config.hpp"
 #include "cpu.hpp"
@@ -41,9 +43,11 @@ class Emulator {
 	GPU gpu;
 	Memory memory;
 	Kernel kernel;
+	std::unique_ptr<Audio::DSPCore> dsp;
+	Scheduler scheduler;
+
 	Crypto::AESEngine aesEngine;
 	Cheats cheats;
-	Scheduler scheduler;
 
 	// Variables to keep track of whether the user is controlling the 3DS analog stick with their keyboard
 	// This is done so when a gamepad is connected, we won't automatically override the 3DS analog stick settings with the gamepad's state

include/kernel/kernel.hpp

@@ -66,15 +66,20 @@ class Kernel {
 	Handle makeMemoryBlock(u32 addr, u32 size, u32 myPermission, u32 otherPermission);
 
 public:
-	Handle makeEvent(ResetType resetType); // Needs to be public to be accessible to the APT/HID services
-	Handle makeMutex(bool locked = false); // Needs to be public to be accessible to the APT/DSP services
-	Handle makeSemaphore(u32 initialCount, u32 maximumCount); // Needs to be public to be accessible to the service manager port
+	// Needs to be public to be accessible to the APT/HID services
+	Handle makeEvent(ResetType resetType, Event::CallbackType callback = Event::CallbackType::None);
+	// Needs to be public to be accessible to the APT/DSP services
+	Handle makeMutex(bool locked = false);
+	// Needs to be public to be accessible to the service manager port
+	Handle makeSemaphore(u32 initialCount, u32 maximumCount);
 	Handle makeTimer(ResetType resetType);
 	void pollTimers();
 
 	// Signals an event, returns true on success or false if the event does not exist
 	bool signalEvent(Handle e);
-	
+	// Run the callback for "special" events that have callbacks
+	void runEventCallback(Event::CallbackType callback);
+
 	void clearEvent(Handle e) {
 		KernelObject* object = getObject(e, KernelObjectType::Event);
 		if (object != nullptr) {
@@ -240,6 +245,5 @@ public:
 	ServiceManager& getServiceManager() { return serviceManager; }
 
 	void sendGPUInterrupt(GPUInterrupt type) { serviceManager.sendGPUInterrupt(type); }
-	void signalDSPEvents() { serviceManager.signalDSPEvents(); }
 	void clearInstructionCache();
 };

include/kernel/kernel_types.hpp

@@ -62,11 +62,19 @@ struct Process {
 };
 
 struct Event {
+    // Some events (for now, only the DSP semaphore events) need to execute a callback when signalled
+    // This enum stores what kind of callback they should execute
+    enum class CallbackType : u32 {
+        None, DSPSemaphore,
+    };
+
     u64 waitlist; // A bitfield where each bit symbolizes if the thread with thread with the corresponding index is waiting on the event
     ResetType resetType = ResetType::OneShot;
+    CallbackType callback = CallbackType::None;
     bool fired = false;
 
     Event(ResetType resetType) : resetType(resetType), waitlist(0) {}
+    Event(ResetType resetType, CallbackType cb) : resetType(resetType), waitlist(0), callback(cb) {}
 };
 
 struct Port {

include/logger.hpp

@@ -36,6 +36,7 @@ namespace Log {
 	static Logger<false> gpuLogger;
 	static Logger<false> rendererLogger;
 	static Logger<false> shaderJITLogger;
+	static Logger<false> dspLogger;
 
 	// Service loggers
 	static Logger<false> acLogger;

include/memory.hpp

@@ -100,8 +100,8 @@ namespace KernelMemoryTypes {
 
 class Memory {
 	u8* fcram;
-	u8* dspRam;
-	u8* vram;  // Provided to the memory class by the GPU class
+	u8* dspRam;  // Provided to us by Audio
+	u8* vram;    // Provided to the memory class by the GPU class
 
 	u64& cpuTicks; // Reference to the CPU tick counter
 	using SharedMemoryBlock = KernelMemoryTypes::SharedMemoryBlock;
@@ -281,6 +281,8 @@ private:
 	u32 getUsedUserMem() { return usedUserMemory; }
 
 	void setVRAM(u8* pointer) { vram = pointer; }
+	void setDSPMem(u8* pointer) { dspRam = pointer; }
+
 	bool allocateMainThreadStack(u32 size);
 	Regions getConsoleRegion();
 	void copySharedFont(u8* ptr);

include/scheduler.hpp

@@ -10,7 +10,8 @@ struct Scheduler {
 	enum class EventType {
 		VBlank = 0,          // End of frame event
 		UpdateTimers = 1,    // Update kernel timer objects
-		Panic = 2,           // Dummy event that is always pending and should never be triggered (Timestamp = UINT64_MAX)
+		RunDSP = 2,          // Make the emulated DSP run for one audio frame
+		Panic = 3,           // Dummy event that is always pending and should never be triggered (Timestamp = UINT64_MAX)
 		TotalNumberOfEvents  // How many event types do we have in total?
 	};
 	static constexpr usize totalNumberOfEvents = static_cast<usize>(EventType::TotalNumberOfEvents);

include/services/dsp.hpp

@@ -1,17 +1,12 @@
 #pragma once
 #include <array>
 #include <optional>
+#include "audio/dsp_core.hpp"
 #include "helpers.hpp"
 #include "logger.hpp"
 #include "memory.hpp"
 #include "result/result.hpp"
 
-namespace DSPPipeType {
-	enum : u32 {
-		Debug = 0, DMA = 1, Audio = 2, Binary = 3
-	};
-}
-
 // Circular dependencies!
 class Kernel;
 
@@ -19,15 +14,11 @@ class DSPService {
 	Handle handle = KernelHandles::DSP;
 	Memory& mem;
 	Kernel& kernel;
+	Audio::DSPCore* dsp = nullptr;
 	MAKE_LOG_FUNCTION(log, dspServiceLogger)
 
-	enum class DSPState : u32 {
-		Off, On, Slep
-	};
-
 	// Number of DSP pipes
 	static constexpr size_t pipeCount = 8;
-	DSPState dspState;
 
 	// DSP service event handles
 	using DSPEvent = std::optional<Handle>;
@@ -36,10 +27,7 @@ class DSPService {
 	DSPEvent interrupt0;
 	DSPEvent interrupt1;
 	std::array<DSPEvent, pipeCount> pipeEvents;
-	std::array<std::vector<u8>, pipeCount> pipeData; // The data of each pipe
-
-	void resetAudioPipe();
-	std::vector<u8> readPipe(u32 pipe, u32 size);
+	u16 semaphoreMask = 0;
 
 	DSPEvent& getEventRef(u32 type, u32 pipe);
 	static constexpr size_t maxEventCount = 6;
@@ -67,6 +55,10 @@ public:
 	DSPService(Memory& mem, Kernel& kernel) : mem(mem), kernel(kernel) {}
 	void reset();
 	void handleSyncRequest(u32 messagePointer);
+	void setDSPCore(Audio::DSPCore* pointer) { dsp = pointer; }
+	
+	// Special callback that's ran when the semaphore event is signalled
+	void onSemaphoreEventSignal() { dsp->setSemaphore(semaphoreMask); }
 
 	enum class SoundOutputMode : u8 {
 		Mono = 0,
@@ -74,5 +66,8 @@ public:
 		Surround = 2
 	};
 
-	void signalEvents();
+	void triggerPipeEvent(int index);
+	void triggerSemaphoreEvent();
+	void triggerInterrupt0();
+	void triggerInterrupt1();
 };

include/services/service_manager.hpp

@@ -105,9 +105,8 @@ class ServiceManager {
 	void setHIDSharedMem(u8* ptr) { hid.setSharedMem(ptr); }
 	void setCSNDSharedMem(u8* ptr) { csnd.setSharedMemory(ptr); }
 
-	void signalDSPEvents() { dsp.signalEvents(); }
-
 	// Input function wrappers
 	HIDService& getHID() { return hid; }
 	NFCService& getNFC() { return nfc; }
+	DSPService& getDSP() { return dsp; }
 };