package main import ( "context" "errors" "flag" "fmt" "log" "mxl-player/internal/source" "runtime" "time" "unsafe" vk "github.com/christerso/vulkan-go/vk" "github.com/ebitengine/purego" ) const ( APP_NAME = "MXL Player" APP_VER = "0.0.1" WIN_WIDTH int32 = 1280 WIN_HEIGHT int32 = 720 ) const ( initVideo uint32 = 0x00000020 windowVulkan uint64 = 0x0000000010000000 windowResizable uint64 = 0x0000000000000020 sdlEventQuit uint32 = 0x100 sdlEventWindowResized uint32 = 0x206 sdlEventPixelSizeChanged uint32 = 0x207 sdlEventKeyDown uint32 = 0x300 keyEscape uint32 = 0x1B ) var ( sdlInit func(flags uint32) bool sdlQuit func() sdlGetError func() uintptr sdlCreateWindow func(title *byte, w, h int32, flags uint64) uintptr sdlDestroyWindow func(window uintptr) sdlVulkanGetInstanceExtensions func(count *uint32) uintptr sdlVulkanCreateSurface func(window, instance, allocator uintptr, surface *uint64) bool sdlPollEvent func(event unsafe.Pointer) bool sdlGetWindowSizeInPixels func(window uintptr, w, h *int32) bool sdlSetWindowFullscreen func(windows uintptr, fullscreen bool) bool ) func sdlError() string { return cstr(sdlGetError()) } var sdlLoaded = false func loadSDLMissing() error { if sdlLoaded { return nil } h, err := purego.Dlopen("libSDL3.so.0", purego.RTLD_NOW|purego.RTLD_GLOBAL) if err != nil { return fmt.Errorf("win: load SDL3: %w", err) } purego.RegisterLibFunc(&sdlInit, h, "SDL_Init") purego.RegisterLibFunc(&sdlQuit, h, "SDL_Quit") purego.RegisterLibFunc(&sdlGetError, h, "SDL_GetError") purego.RegisterLibFunc(&sdlCreateWindow, h, "SDL_CreateWindow") purego.RegisterLibFunc(&sdlDestroyWindow, h, "SDL_DestroyWindow") purego.RegisterLibFunc(&sdlVulkanGetInstanceExtensions, h, "SDL_Vulkan_GetInstanceExtensions") purego.RegisterLibFunc(&sdlVulkanCreateSurface, h, "SDL_Vulkan_CreateSurface") purego.RegisterLibFunc(&sdlPollEvent, h, "SDL_PollEvent") purego.RegisterLibFunc(&sdlGetWindowSizeInPixels, h, "SDL_GetWindowSizeInPixels") purego.RegisterLibFunc(&sdlSetWindowFullscreen, h, "SDL_SetWindowFullscreen") sdlLoaded = true return nil } // cstr reads a NUL-terminated C string at p. func cstr(p uintptr) string { if p == 0 { return "" } var n int for *(*byte)(unsafe.Pointer(p + uintptr(n))) != 0 { n++ } return string(unsafe.Slice((*byte)(unsafe.Pointer(p)), n)) } // cbytes returns a NUL-terminated copy of s as *byte, kept alive by the caller. func cbytes(s string) *byte { b := make([]byte, len(s)+1) copy(b, s) runtime.KeepAlive(b) return &b[0] } func main() { // TODO: remove flags defaults mxlDomain := flag.String("d", "/dev/shm/mxl", "MXL domain") mxlVideoFlowID := flag.String("v", "5fbec3b1-1b0f-417d-9059-8b94a47197ed", "MXL video flow UUID") flag.Parse() runtime.LockOSThread() if err := loadSDLMissing(); err != nil { panic(err) } if !sdlInit(initVideo) { log.Fatalf("SDL_Init: %s", sdlError()) return } windowHandler := sdlCreateWindow(cbytes(fmt.Sprintf("%s %s", APP_NAME, APP_VER)), WIN_WIDTH, WIN_HEIGHT, windowVulkan|windowResizable) if windowHandler == 0 { sdlQuit() log.Fatalf("SDL_CreateWindow: %s", sdlError()) return } // Vulkan init if err := vk.Load(); err != nil { panic(err) } // Vulkan instance extensions SDL needs. var sdlExtCount uint32 arr := sdlVulkanGetInstanceExtensions(&sdlExtCount) if arr == 0 { log.Fatal("sdlVulkanGetInstanceExtensions is 0") return } sdlExtensions := make([]string, sdlExtCount) for i := uint32(0); i < sdlExtCount; i++ { p := *(*uintptr)(unsafe.Pointer(arr + uintptr(i)*unsafe.Sizeof(uintptr(0)))) sdlExtensions[i] = cstr(p) } // Vulkan Instance creation var vkLayers []string vkExtensions := append(sdlExtensions, vk.ExtDebugUtils) vkInstance, err := vk.CreateInstance(vk.InstanceConfig{ ApplicationName: APP_NAME, EngineName: "no engine", Extensions: vkExtensions, Layers: vkLayers, }) if err != nil { log.Fatalf(err.Error()) panic(err) } defer vkInstance.Destroy() // Vulkan surface var vkSurface uint64 if !sdlVulkanCreateSurface(windowHandler, uintptr(vkInstance), 0, &vkSurface) { log.Fatalf("SDL_Vulkan_CreateSurface: %s", sdlError()) return } vkSurf := vk.SurfaceKHR(vkSurface) defer vkInstance.DestroySurface(vkSurf) // Vulkan Physical Device devices, err := vkInstance.EnumeratePhysicalDevices() if err != nil || len(devices) == 0 { panic("No Vulkan devices") } for _, pd := range devices { info := pd.Info() fmt.Printf("%s (%s)\n", info.Name, info.Type) } vkPhysDevice := devices[0] // Vulkan Surface & Graphics Queue gfx, err := vkPhysDevice.GraphicsFamily() if err != nil { panic(err.Error()) } if !vkPhysDevice.SurfaceSupport(gfx, vkSurf) { log.Fatalf("graphics queue cannot present") return } vkDevice, vkQueue, err := vkPhysDevice.CreateDevice(vk.DeviceConfig{ GraphicsFamily: gfx, Extensions: []string{"VK_KHR_swapchain"}, }) if err != nil { panic(err) } defer vkDevice.Destroy() // TODO: colorspace & render pass var vkFormat vk.Format = vk.FormatUndefined var vkColorSpace uint32 formats, _ := vkPhysDevice.SurfaceFormats(vkSurf) for _, f := range formats { // Prefer 10-bit RGB (A2B10G10R10_UNORM = 64) to preserve V210's 10 bits if f.Format == vk.Format(64) && f.ColorSpace == vk.ColorSpaceSRGBNonlinear { vkFormat = f.Format vkColorSpace = f.ColorSpace } } if vkFormat == vk.FormatUndefined { vkFormat = formats[0].Format vkColorSpace = formats[0].ColorSpace } var vkPresentMode vk.PresentMode = vk.PresentModeFIFO modes, _ := vkPhysDevice.SurfacePresentModes(vkSurf) for _, m := range modes { if m == vk.PresentModeMailbox { vkPresentMode = m } } vkRenderPass, err := vkDevice.CreateColorDepthRenderPass(vkFormat, vk.FormatD32Sfloat) if err != nil { panic(err.Error()) } defer vkDevice.DestroyRenderPass(vkRenderPass) // Vulkan Command Pool vkCommandPool, err := vkDevice.CreateCommandPool(gfx) if err != nil { panic(err.Error()) } defer vkDevice.DestroyCommandPool(vkCommandPool) vkCommands, err := vkDevice.AllocateCommandBuffers(vkCommandPool, 1) if err != nil { panic(err.Error()) } // MXL Source mxlSrc, err := source.Open(*mxlDomain, *mxlVideoFlowID) if err != nil { log.Fatalf("source: %v\n", err) } defer mxlSrc.Close() frameSize := vk.DeviceSize(mxlSrc.Stride()) * vk.DeviceSize(mxlSrc.Height()) fmt.Printf("source: %dx%d stride=%d frameSize=%d\n", mxlSrc.Width(), mxlSrc.Height(), mxlSrc.Stride(), frameSize) // Staging buffer: host-visible, persistently mapped. The reader goroutine // writes V210 bytes here; the GPU copies from it. staging, err := vkDevice.CreateBuffer(vkPhysDevice, vk.BufferConfig{ Size: frameSize, Usage: vk.BufferUsageTransferSrc, Properties: vk.MemoryHostVisible | vk.MemoryHostCoherent, Map: true, }) if err != nil { panic(err) } defer vkDevice.DestroyBuffer(staging) // Device-local V210 buffer: fast for the GPU to read (M3 compute), CPU can't // write it. Filled each frame by a CopyBuffer from staging. v210Buf, err := vkDevice.CreateBuffer(vkPhysDevice, vk.BufferConfig{ Size: frameSize, Usage: vk.BufferUsageTransferDst | vk.BufferUsageStorageBuffer, Properties: vk.MemoryDeviceLocal, Map: false, }) if err != nil { panic(err) } defer vkDevice.DestroyBuffer(v210Buf) // Decode pipeline: fullscreen triangle, fragment reads V210 from the // staging buffer and writes 10-bit RGB to the swapchain color attachment. vertModule, err := vkDevice.CreateShaderModule(vertSPV) if err != nil { panic(err) } defer vkDevice.DestroyShaderModule(vertModule) fragModule, err := vkDevice.CreateShaderModule(fragSPV) if err != nil { panic(err) } defer vkDevice.DestroyShaderModule(fragModule) // Descriptor set layout: binding 0 = storage buffer (V210), fragment stage. decodeDSL, err := vkDevice.CreateDescriptorSetLayout([]vk.DescriptorBinding{ {Binding: 0, Type: vk.DescriptorStorageBuffer, Count: 1, Stages: vk.ShaderStageFragment}, }) if err != nil { panic(err) } defer vkDevice.DestroyDescriptorSetLayout(decodeDSL) // Pipeline layout: the set layout + push constants {width,height,strideBytes}. decodeLayout, err := vkDevice.CreatePipelineLayout([]vk.DescriptorSetLayout{decodeDSL}, vk.ShaderStageFragment, 12) if err != nil { panic(err) } defer vkDevice.DestroyPipelineLayout(decodeLayout) decodePipeline, err := vkDevice.CreateGraphicsPipeline(vk.GraphicsPipelineConfig{ Layout: decodeLayout, RenderPass: vkRenderPass, VertexShader: vertModule, FragShader: fragModule, Topology: vk.TopologyTriangleList, PolygonMode: vk.PolygonFill, CullMode: vk.CullNone, FrontFace: vk.FrontFaceCounterClockwise, }) if err != nil { panic(err) } defer vkDevice.DestroyPipeline(decodePipeline) // Descriptor pool + set, bound once to the staging buffer descPool, err := vkDevice.CreateDescriptorPool(1, map[vk.DescriptorType]uint32{ vk.DescriptorStorageBuffer: 1, }) if err != nil { panic(err) } defer vkDevice.DestroyDescriptorPool(descPool) decodeSet, err := vkDevice.AllocateDescriptorSet(descPool, decodeDSL) if err != nil { panic(err) } vkDevice.UpdateBufferDescriptor(decodeSet, 0, vk.DescriptorStorageBuffer, staging.Buffer, 0, vk.WholeSize) // Swapchain var ( vkExtent vk.Extent2D vkSwapchain vk.SwapchainKHR swapImages []vk.Image swapViews []vk.ImageView depthImg vk.AllocImage depthView vk.ImageView fbs []vk.Framebuffer ) destroySwapChain := func() { for _, fb := range fbs { vkDevice.DestroyFramebuffer(fb) } vkDevice.DestroyImageView(depthView) vkDevice.DestroyImage(depthImg) for _, v := range swapViews { vkDevice.DestroyImageView(v) } vkDevice.DestroySwapchain(vkSwapchain) fbs = nil swapViews = nil swapImages = nil depthView = 0 depthImg = vk.AllocImage{} vkSwapchain = 0 vkExtent = vk.Extent2D{} } createSwapChain := func() (err error) { defer func() { if err != nil { destroySwapChain() } }() caps, err := vkPhysDevice.SurfaceCapabilities(vkSurf) if err != nil { return err } //compute swapchain extent var w, h int32 = WIN_WIDTH, WIN_HEIGHT if !sdlGetWindowSizeInPixels(windowHandler, &w, &h) { return fmt.Errorf("SDL_GetWindowSizeInPixels: %s", sdlError()) } if w < int32(caps.MinImageExtent.Width) { w = int32(caps.MinImageExtent.Width) } if h < int32(caps.MinImageExtent.Height) { h = int32(caps.MinImageExtent.Height) } if w > int32(caps.MaxImageExtent.Width) { w = int32(caps.MaxImageExtent.Width) } if h > int32(caps.MaxImageExtent.Height) { h = int32(caps.MaxImageExtent.Height) } vkExtent = vk.Extent2D{Width: uint32(w), Height: uint32(h)} vkSwapchain, err = vkDevice.CreateSwapchain(vk.SwapchainConfig{ Surface: vkSurf, MinImageCount: caps.MinImageCount, Format: vkFormat, ColorSpace: vkColorSpace, Extent: vkExtent, PresentMode: vkPresentMode, PreTransform: caps.CurrentTransform, }) if err != nil { return err } swapImages, err = vkDevice.SwapchainImages(vkSwapchain) if err != nil { return err } swapViews = make([]vk.ImageView, len(swapImages)) for i, img := range swapImages { v, err := vkDevice.CreateImageView(img, vkFormat, vk.AspectColor) if err != nil { return err } swapViews[i] = v } depthImg, err = vkDevice.CreateImage2D( vkPhysDevice, vk.FormatD32Sfloat, vkExtent, vk.ImageUsageDepthStencilAttachment) if err != nil { return err } depthView, err = vkDevice.CreateImageView(depthImg.Image, vk.FormatD32Sfloat, vk.AspectDepth) if err != nil { return err } fbs = make([]vk.Framebuffer, len(swapViews)) for i, cv := range swapViews { fb, err := vkDevice.CreateFramebuffer(vkRenderPass, []vk.ImageView{cv, depthView}, vkExtent) if err != nil { return err } fbs[i] = fb } return nil } recreateSwapChain := func() error { if err := vkDevice.WaitIdle(); err != nil { return err } destroySwapChain() return createSwapChain() } if err := createSwapChain(); err != nil { panic(err) } defer destroySwapChain() // syncs // Semaphores sync GPU work to GPU work (you never wait on these from the CPU). You need two: // - imageAvailable — AcquireNextImage signals it when a swapchain image is ready to render into. // Your draw submission waits on it before touching color. // - renderFinished — your draw submission signals it when rendering is done. // Present waits on it before showing the frame. // Fences sync GPU work to CPU. You need one: // - inFlight — your submission signals it when the GPU is done with the command buffer. // Your CPU waits on it before re-recording that same command buffer. // It's created signaled so the very first frame's wait returns immediately // (otherwise you'd deadlock waiting for work that hasn't been submitted yet). imageAvailable, err := vkDevice.CreateSemaphore() if err != nil { panic(err.Error()) } defer vkDevice.DestroySemaphore(imageAvailable) renderFinished, err := vkDevice.CreateSemaphore() if err != nil { panic(err.Error()) } defer vkDevice.DestroySemaphore(renderFinished) inFlight, err := vkDevice.CreateFence(true) if err != nil { panic(err.Error()) } defer vkDevice.DestroyFence(inFlight) defer vkDevice.WaitIdle() // Reader goroutine: stages V210 bytes into the staging buffer. // Single-flight handshake: it only writes staging after the render thread // grants permission (post-WaitFence), so the GPU is never reading it ctx, cancel := context.WithCancel(context.Background()) defer cancel() grant := make(chan struct{}) staged := make(chan struct{}) go func() { for { select { case <-grant: case <-ctx.Done(): return } f, err := mxlSrc.NextCtx(ctx, 200*time.Millisecond) if err != nil { if errors.Is(err, context.Canceled) { return } log.Printf("source: %v", err) cancel() return } // Copy the borrowed payload into staging BEFORE the next read // invalidates it. Within the grain's valid lifetime vk.CopyToMapped(staging.Mapped, f.Payload) // fmt.Println("staged", f.Index) select { case staged <- struct{}{}: case <-ctx.Done(): return } } }() type decodePushConstants struct { Width, Height, StrideBytes, WinW, WinH uint32 } // Core Loop running := true resized := false fullscreen := false for running { var event [128]byte for sdlPollEvent(unsafe.Pointer(&event[0])) { eventType := *(*uint32)(unsafe.Pointer(&event[0])) switch eventType { case sdlEventQuit: running = false case sdlEventWindowResized, sdlEventPixelSizeChanged: resized = true case sdlEventKeyDown: key := *(*int32)(unsafe.Pointer(&event[28])) switch uint32(key) { case keyEscape: running = false // f case 0x66: fullscreen = !fullscreen sdlSetWindowFullscreen(windowHandler, fullscreen) resized = true } } } if !running { break } if resized { if err := recreateSwapChain(); err != nil { panic(err) } resized = false } // Acquire the next swapchain image imageIndex, r := vkDevice.AcquireNextImage(vkSwapchain, imageAvailable, ^uint64(0)) if r == vk.ErrorOutOfDateKHR || r == vk.SuboptimalKHR { if err := recreateSwapChain(); err != nil { panic(err) } continue } if !r.Ok() { panic(fmt.Errorf("acquire: %v", r)) } // Wait until GPU is done with our command buffer, then reset fence if err := vkDevice.WaitFence(inFlight, ^uint64(0)); err != nil { panic(err) } if err := vkDevice.ResetFence(inFlight); err != nil { panic(err) } // Grant the goroutine permission to write staging (GPU is idle now), // then wait for it to stage a new payload. select { case grant <- struct{}{}: case <-ctx.Done(): running = false continue } select { case <-staged: case <-ctx.Done(): running = false continue } // Record: copy staged V210 into device-local buffer, then clear cmd := vkCommands[0] if err := cmd.Reset(); err != nil { panic(err) } if err := cmd.Begin(vk.CommandBufferOneTimeSubmit); err != nil { panic(err) } cmd.CopyBuffer(staging.Buffer, v210Buf.Buffer, frameSize) cmd.BeginRenderPass( vkRenderPass, fbs[imageIndex], vk.Rect2D{Offset: vk.Offset2D{X: 0, Y: 0}, Extent: vkExtent}, []vk.ClearValue{ vk.ClearColor(0.0, 0.4, 0.7, 1.0), vk.ClearDepthStencil(1.0, 0), }, ) cmd.SetViewport(vk.Viewport{ X: 0, Y: 0, Width: float32(vkExtent.Width), Height: float32(vkExtent.Height), MinDepth: 0, MaxDepth: 1, }) cmd.SetScissor(vk.Rect2D{Offset: vk.Offset2D{X: 0, Y: 0}, Extent: vkExtent}) cmd.BindPipeline(decodePipeline) cmd.BindDescriptorSet(decodeLayout, 0, decodeSet) decodePC := decodePushConstants{ Width: mxlSrc.Width(), Height: mxlSrc.Height(), StrideBytes: mxlSrc.Stride(), WinW: vkExtent.Width, WinH: vkExtent.Height, } cmd.PushConstants(decodeLayout, vk.ShaderStageFragment, 0, unsafe.Pointer(&decodePC), 20) cmd.Draw(3, 1, 0, 0) cmd.EndRenderPass() if err := cmd.End(); err != nil { panic(err) } // Submit: wait on imageAvailable, signal RenderFinished + inFlight if err := vkQueue.Submit(vk.SubmitConfig{ Wait: imageAvailable, WaitStage: vk.StageColorAttachmentOutput, Command: cmd, Signal: renderFinished, Fence: inFlight, }); err != nil { panic(err) } // Present. Wait on renderFinished r = vkQueue.Present(vkSwapchain, imageIndex, renderFinished) if r == vk.ErrorOutOfDateKHR || r == vk.SuboptimalKHR { if err := recreateSwapChain(); err != nil { panic(err) } continue } if !r.Ok() { panic(fmt.Errorf("present: %v", r)) } } }