renderer refactoring done

This commit is contained in:
Dmitry Sergeev
2026-08-23 10:42:06 +03:00
parent c932777cc3
commit ef92dbe7fa
7 changed files with 500 additions and 391 deletions
+32 -390
View File
@@ -6,6 +6,7 @@ import (
"flag"
"fmt"
"log"
"mxl-player/internal/renderer"
"mxl-player/internal/sdl"
"mxl-player/internal/source"
"runtime"
@@ -19,13 +20,11 @@ import (
const (
APP_NAME = "MXL Player"
APP_VER = "0.0.1"
WIN_WIDTH int32 = 1280
WIN_HEIGHT int32 = 720
WIN_WIDTH int32 = 1920
WIN_HEIGHT int32 = 1080
)
func main() {
var errMinimized = errors.New("Window minimized")
// 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()
@@ -46,18 +45,15 @@ func main() {
return
}
// Vulkan init
if err := vk.Load(); err != nil {
panic(err)
}
// Vulkan instance extensions SDL needs.
sdlExtensions := sdl.VulkanGetInstanceExtensions()
if len(sdlExtensions) == 0 {
log.Fatal("sdlVulkanGetInstanceExtensions is 0")
return
}
// Vulkan Instance creation
var vkLayers []string
vkExtensions := append(sdlExtensions, vk.ExtDebugUtils)
vkInstance, err := vk.CreateInstance(vk.InstanceConfig{
@@ -71,7 +67,7 @@ func main() {
panic(err)
}
defer vkInstance.Destroy()
// Vulkan surface
var vkSurface uint64
if !sdl.VulkanCreateSurface(windowHandler, uintptr(vkInstance), 0, &vkSurface) {
log.Fatalf("SDL_Vulkan_CreateSurface: %s", sdl.GetError())
@@ -80,7 +76,6 @@ func main() {
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")
@@ -90,7 +85,7 @@ func main() {
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())
@@ -99,6 +94,7 @@ func main() {
log.Fatalf("graphics queue cannot present")
return
}
vkDevice, vkQueue, err := vkPhysDevice.CreateDevice(vk.DeviceConfig{
GraphicsFamily: gfx,
Extensions: []string{"VK_KHR_swapchain"},
@@ -107,286 +103,31 @@ func main() {
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)
log.Fatalf("source: %v", err)
}
defer func() { _ = 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,
fmt.Printf("source: %dx%d stride=%d\n", mxlSrc.Width(), mxlSrc.Height(), mxlSrc.Stride())
r, err := renderer.New(renderer.Config{
PhysDevice: vkPhysDevice,
Device: vkDevice,
Queue: vkQueue,
Surface: vkSurf,
Window: windowHandler,
GraphicsFamily: gfx,
VideoWidth: mxlSrc.Width(),
VideoHeight: mxlSrc.Height(),
VideoStride: mxlSrc.Stride(),
})
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 !sdl.GetWindowSizeInPixels(windowHandler, &w, &h) {
return fmt.Errorf("SDL_GetWindowSizeInPixels: %s", sdl.GetError())
}
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)
}
if w == 0 || h == 0 {
return errMinimized
}
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 r.Destroy()
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{}, 1)
@@ -406,28 +147,10 @@ func main() {
s, err := source.Open(*mxlDomain, *mxlVideoFlowID)
if err == nil {
newSize := vk.DeviceSize(s.Stride()) * vk.DeviceSize(s.Height())
if newSize != frameSize {
if err := vkDevice.WaitIdle(); err != nil {
if newSize != r.FrameSize() {
if err := r.RecreateBuffers(newSize); err != nil {
return err
}
vkDevice.DestroyBuffer(v210Buf)
vkDevice.DestroyBuffer(staging)
staging, err = vkDevice.CreateBuffer(vkPhysDevice, vk.BufferConfig{
Size: newSize, Usage: vk.BufferUsageTransferSrc,
Properties: vk.MemoryHostVisible | vk.MemoryHostCoherent, Map: true,
})
if err != nil {
return err
}
v210Buf, err = vkDevice.CreateBuffer(vkPhysDevice, vk.BufferConfig{
Size: newSize, Usage: vk.BufferUsageTransferDst | vk.BufferUsageStorageBuffer,
Properties: vk.MemoryDeviceLocal, Map: false,
})
if err != nil {
return err
}
frameSize = newSize
vkDevice.UpdateBufferDescriptor(decodeSet, 0, vk.DescriptorStorageBuffer, staging.Buffer, 0, vk.WholeSize)
fmt.Printf("source: resolution changed, frameSize=%d\n", newSize)
}
mxlSrc = s
@@ -468,10 +191,7 @@ func main() {
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)
vk.CopyToMapped(r.StagingMapped(), f.Payload)
select {
case staged <- f.Index:
case <-ctx.Done():
@@ -480,11 +200,6 @@ func main() {
}
}()
type decodePushConstants struct {
Width, Height, StrideBytes, WinW, WinH uint32
}
// Core Loop
running := true
resized := false
granted := false
@@ -521,8 +236,8 @@ func main() {
break
}
if resized {
if err := recreateSwapChain(); err != nil {
if errors.Is(err, errMinimized) {
if err := r.RecreateSwapchain(); err != nil {
if errors.Is(err, renderer.ErrMinimized) {
resized = true
continue
}
@@ -545,6 +260,7 @@ func main() {
granted = false
case <-failed:
granted = false
continue
case <-ctx.Done():
running = false
continue
@@ -552,26 +268,18 @@ func main() {
continue
}
// Acquire the next swapchain image. Right after we have a frame
imageIndex, r := vkDevice.AcquireNextImage(vkSwapchain, imageAvailable, ^uint64(0))
if r == vk.ErrorOutOfDateKHR || r == vk.SuboptimalKHR {
if err := recreateSwapChain(); err != nil {
if errors.Is(err, errMinimized) {
err := r.DrawFrame(mxlSrc.Width(), mxlSrc.Height(), mxlSrc.Stride())
if errors.Is(err, renderer.ErrOutOfDate) {
if rerr := r.RecreateSwapchain(); rerr != nil {
if errors.Is(rerr, renderer.ErrMinimized) {
resized = true
continue
}
panic(err)
panic(rerr)
}
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 {
if err != nil {
panic(err)
}
@@ -591,71 +299,5 @@ func main() {
dropped = 0
lastReport = now
}
// 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 {
if errors.Is(err, errMinimized) {
resized = true
continue
}
panic(err)
}
continue
}
if !r.Ok() {
panic(fmt.Errorf("present: %v", r))
}
}
}
-9
View File
@@ -1,9 +0,0 @@
package main
import _ "embed"
//go:embed shaders/triangle.vert.spv
var vertSPV []byte
//go:embed shaders/decode.frag.spv
var fragSPV []byte
-76
View File
@@ -1,76 +0,0 @@
#version 450
layout(set = 0, binding = 0, std430) readonly buffer V210 {
uint words[];
};
layout(push_constant) uniform PC {
uint width;
uint height;
uint strideBytes;
uint winW;
uint winH;
} pc;
layout(location = 0) out vec4 fragColor;
void main() {
// Letterbox: fit video into the window, preserving aspect ratio.
float sx = float(pc.winW) / float(pc.width);
float sy = float(pc.winH) / float(pc.height);
float scale = min(sx, sy);
float dispW = float(pc.width) * scale;
float dispH = float(pc.height) * scale;
float offX = (float(pc.winW) - dispW) * 0.5;
float offY = (float(pc.winH) - dispH) * 0.5;
float fbx = gl_FragCoord.x - offX;
float fby = gl_FragCoord.y - offY;
if (fbx < 0.0 || fbx >= dispW || fby < 0.0 || fby >= dispH) {
fragColor = vec4(0.0, 0.0, 0.0, 1.0);
return;
}
// Framebuffer y is bottom-origin; flip to video top-origin.
uint x = uint(fbx / scale);
uint y = uint(fby / scale);
// V210: 6 pixels per group of 4 words; 3 ten-bit components per word
// (bits 0-9 / 10-19 / 20-29). Stream order: Cb Y Cr Y Cb Y Cr Y ...
uint group = x / 6u;
uint sub = x % 6u;
uint wordsPerLine = pc.strideBytes / 4u;
uint base = y * wordsPerLine + group * 4u;
uint w0 = words[base + 0u];
uint w1 = words[base + 1u];
uint w2 = words[base + 2u];
uint w3 = words[base + 3u];
uint cb0 = (w0 ) & 0x3FFu;
uint y0 = (w0 >> 10u) & 0x3FFu;
uint cr0 = (w0 >> 20u) & 0x3FFu;
uint y1 = (w1 ) & 0x3FFu;
uint cb1 = (w1 >> 10u) & 0x3FFu;
uint y2 = (w1 >> 20u) & 0x3FFu;
uint cr1 = (w2 ) & 0x3FFu;
uint y3 = (w2 >> 10u) & 0x3FFu;
uint cb2 = (w2 >> 20u) & 0x3FFu;
uint y4 = (w3 ) & 0x3FFu;
uint cr2 = (w3 >> 10u) & 0x3FFu;
uint y5 = (w3 >> 20u) & 0x3FFu;
float Y, Cb, Cr;
if (sub == 0u) { Y = float(y0); Cb = float(cb0); Cr = float(cr0); }
else if (sub == 1u) { Y = float(y1); Cb = float(cb0); Cr = float(cr0); }
else if (sub == 2u) { Y = float(y2); Cb = float(cb1); Cr = float(cr1); }
else if (sub == 3u) { Y = float(y3); Cb = float(cb1); Cr = float(cr1); }
else if (sub == 4u) { Y = float(y4); Cb = float(cb2); Cr = float(cr2); }
else { Y = float(y5); Cb = float(cb2); Cr = float(cr2); }
float yf = (Y - 64.0) / 876.0;
float uf = (Cb - 512.0) / 896.0;
float vf = (Cr - 512.0) / 896.0;
float r = yf + 1.5748 * vf;
float g = yf - 0.1873 * uf - 0.4681 * vf;
float b = yf + 1.8556 * uf;
fragColor = vec4(clamp(r, 0.0, 1.0), clamp(g, 0.0, 1.0), clamp(b, 0.0, 1.0), 1.0);
}
Binary file not shown.
-12
View File
@@ -1,12 +0,0 @@
#version 450
// Fullscreen triangle, no vertex buffer. One triangle covers the viewport.
vec2 positions[3] = vec2[](
vec2(-1.0, -1.0),
vec2( 3.0, -1.0),
vec2(-1.0, 3.0)
);
void main() {
gl_Position = vec4(positions[gl_VertexIndex], 0.0, 1.0);
}
Binary file not shown.