code: Better screen support

include/PICA/gpu.hpp

@@ -29,6 +29,7 @@ class GPU {
 	static constexpr u32 vramSize = u32(6_MB);
 	Registers regs;                           // GPU internal registers
 	std::array<vec4f, 16> currentAttributes;  // Vertex attributes before being passed to the shader
+	std::array<u32, 0x1000> external_regs; // GPU external registers
 
 	std::array<vec4f, 16> immediateModeAttributes;  // Vertex attributes uploaded via immediate mode submission
 	std::array<PICA::Vertex, 3> immediateModeVertices;
@@ -97,6 +98,9 @@ class GPU {
 	u32 readReg(u32 address);
 	void writeReg(u32 address, u32 value);
 
+	u32 readExternalReg(u32 index);
+	void writeExternalReg(u32 index, u32 value);
+
 	// Used when processing GPU command lists
 	u32 readInternalReg(u32 index);
 	void writeInternalReg(u32 index, u32 value, u32 mask);

include/PICA/regs.hpp

@@ -178,6 +178,54 @@ namespace PICA {
 		};
 	}
 
+	namespace ExternalRegs {
+		enum : u32 {
+			MemFill1BufferStartPaddr = 0x3,
+			MemFill1BufferEndPAddr = 0x4,
+			MemFill1Value = 0x5,
+			MemFill1Control = 0x6,
+			MemFill2BufferStartPaddr = 0x7,
+			MemFill2BufferEndPAddr = 0x8,
+			MemFill2Value = 0x9,
+			MemFill2Control = 0xA,
+			VramBankControl = 0xB,
+			GPUBusy = 0xC,
+			BacklightControl = 0xBC,
+			// TODO: Framebuffer regs
+			Framebuffer0Size = 0x2F,
+			Framebuffer0AFirstAddr = 0x119,
+			Framebuffer0ASecondAddr = 0x11A,
+			Framebuffer0Config = 0x11B,
+			Framebuffer0Select = 0x11D,
+			Framebuffer0Stride = 0x123,
+			Framebuffer0BFirstAddr = 0x124,
+			Framebuffer0BSecondAddr = 0x125,
+			Framebuffer1Size = 0x156,
+			Framebuffer1AFirstAddr = 0x159,
+			Framebuffer1ASecondAddr = 0x15A,
+			Framebuffer1Config = 0x15B,
+			Framebuffer1Select = 0x15D,
+			Framebuffer1Stride = 0x163,
+			Framebuffer1BFirstAddr = 0x164,
+			Framebuffer1BSecondAddr = 0x165,
+			TransferInputPAddr = 0x2FF,
+			TransferOutputPAddr = 0x300,
+			DisplayTransferOutputDim = 0x301,
+			DisplayTransferInputDim = 0x302,
+			TransferFlags = 0x303,
+			TransferTrigger = 0x305,
+			TextureCopyTotalBytes = 0x307,
+			TextureCopyInputLineGap = 0x308,
+			TextureCopyOutputLineGap = 0x309,
+		};
+	}
+
+	enum class Scaling : u32 {
+		None = 0,
+		X = 1,
+		XY = 2,
+	};
+
 	namespace Lights {
 		enum : u32 {
 			LUT_D0 = 0,

include/memory.hpp

@@ -248,4 +248,4 @@ public:
 
 	void setVRAM(u8* pointer) { vram = pointer; }
 	bool allocateMainThreadStack(u32 size);
-};
+};

include/renderer.hpp

@@ -65,4 +65,4 @@ class Renderer {
 
 	void setColourBufferLoc(u32 loc) { colourBufferLoc = loc; }
 	void setDepthBufferLoc(u32 loc) { depthBufferLoc = loc; }
-};
+};

include/renderer_gl/gl_state.hpp

@@ -213,4 +213,4 @@ struct GLStateManager {
 };
 
 static_assert(std::is_trivially_constructible<GLStateManager>(), "OpenGL State Manager class is not trivially constructible!");
-static_assert(std::is_trivially_destructible<GLStateManager>(), "OpenGL State Manager class is not trivially destructible!");
+static_assert(std::is_trivially_destructible<GLStateManager>(), "OpenGL State Manager class is not trivially destructible!");

include/renderer_gl/renderer_gl.hpp

@@ -44,8 +44,8 @@ class RendererGL final : public Renderer {
 	float oldDepthOffset = 0.0;
 	bool oldDepthmapEnable = false;
 
-	SurfaceCache<DepthBuffer, 10, true> depthBufferCache;
-	SurfaceCache<ColourBuffer, 10, true> colourBufferCache;
+	SurfaceCache<DepthBuffer, 64, true> depthBufferCache;
+	SurfaceCache<ColourBuffer, 64, true> colourBufferCache;
 	SurfaceCache<Texture, 256, true> textureCache;
 
 	// Dummy VAO/VBO for blitting the final output
@@ -78,6 +78,8 @@ class RendererGL final : public Renderer {
 	void displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32 outputSize, u32 flags) override;  // Perform display transfer
 	void drawVertices(PICA::PrimType primType, std::span<const PICA::Vertex> vertices) override;             // Draw the given vertices
 
+	ColourBuffer getColourBuffer(u32 addr, PICA::ColorFmt format, u32 width, u32 height);
+
 	// Take a screenshot of the screen and store it in a file
 	void screenshot(const std::string& name) override;
 };

include/renderer_gl/surfaces.hpp

@@ -19,6 +19,10 @@ struct ColourBuffer {
     OpenGL::Texture texture;
     OpenGL::Framebuffer fbo;
 
+    GLenum internalFormat;
+    GLenum fmt;
+    GLenum type;
+
     ColourBuffer() : valid(false) {}
 
     ColourBuffer(u32 loc, PICA::ColorFmt format, u32 x, u32 y, bool valid = true)
@@ -29,12 +33,32 @@ struct ColourBuffer {
         range = Interval<u32>(loc, (u32)endLoc);
     }
 
-    void allocate() {
+	void allocate() {
+		// Internal formats for the texture based on format
+		static constexpr std::array<GLenum, 5> internalFormats = {
+			GL_RGBA8, GL_RGB8, GL_RGB5_A1, GL_RGB565, GL_RGBA4
+		};
+
+		// Format of the texture
+		static constexpr std::array<GLenum, 5> formats = {
+			GL_RGBA, GL_BGR, GL_RGBA, GL_RGB, GL_RGBA,
+		};
+
+		static constexpr std::array<GLenum, 5> types = {
+			GL_UNSIGNED_INT_8_8_8_8, GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT_5_5_5_1,
+			GL_UNSIGNED_SHORT_5_6_5, GL_UNSIGNED_SHORT_4_4_4_4,
+		};
+
+		internalFormat = internalFormats[(int)format];
+		fmt = formats[(int)format];
+		type = types[(int)format];
+
+
         // Create texture for the FBO, setting up filters and the like
         // Reading back the current texture is slow, but allocate calls should be few and far between.
         // If this becomes a bottleneck, we can fix it semi-easily
         auto prevTexture = OpenGL::getTex2D();
-        texture.create(size.x(), size.y(), GL_RGBA8);
+        texture.create(size.x(), size.y(), internalFormat);
         texture.bind();
         texture.setMinFilter(OpenGL::Linear);
         texture.setMagFilter(OpenGL::Linear);

include/renderer_gl/textures.hpp

@@ -53,7 +53,7 @@ struct Texture {
     static u32 getSwizzledOffset_4bpp(u32 u, u32 v, u32 width);
 
     // Returns the format of this texture as a string
-    std::string formatToString() {
+    std::string_view formatToString() {
         return PICA::textureFormatToString(format);
     }
 

include/renderer_null/renderer_null.hpp

@@ -14,4 +14,4 @@ class RendererNull final : public Renderer {
 	void displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32 outputSize, u32 flags) override;
 	void drawVertices(PICA::PrimType primType, std::span<const PICA::Vertex> vertices) override;
 	void screenshot(const std::string& name) override;
-};
+};

include/services/gsp_gpu.hpp

@@ -18,6 +18,24 @@ enum class GPUInterrupt : u8 {
 	DMA = 6
 };
 
+struct FramebufferInfo {
+	u32 activeFb;
+	u32 leftFramebufferVaddr;
+	u32 rightFramebufferVaddr;
+	u32 stride;
+	u32 format;
+	u32 displayFb;
+	u32 attribute;
+};
+
+struct FrameBufferUpdate {
+	u8 index;
+	u8 dirtyFlag;
+	u16 pad0;
+	std::array<FramebufferInfo, 2> framebufferInfo;
+	u32 pad1;
+};
+
 // More circular dependencies
 class Kernel;
 
@@ -45,6 +63,7 @@ class GPUService {
 	void flushDataCache(u32 messagePointer);
 	void registerInterruptRelayQueue(u32 messagePointer);
 	void setAxiConfigQoSMode(u32 messagePointer);
+	void setBufferSwap(u32 messagePointer);
 	void setInternalPriorities(u32 messagePointer);
 	void setLCDForceBlack(u32 messagePointer);
 	void storeDataCache(u32 messagePointer);
@@ -60,6 +79,8 @@ class GPUService {
 	void triggerTextureCopy(u32* cmd);
 	void flushCacheRegions(u32* cmd);
 
+	void setBufferSwapImpl(u32 screen_id, const FramebufferInfo& info);
+
 public:
 	GPUService(Memory& mem, GPU& gpu, Kernel& kernel, u32& currentPID) : mem(mem), gpu(gpu),
 		kernel(kernel), currentPID(currentPID) {}
@@ -72,4 +93,4 @@ public:
 			std::memset(ptr, 0, 0x1000);
 		}
 	}
-};
+};