allemang.d/zig-experiments
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

remove unused swapchain.zig

Browse Source
This commit is contained in:
David Allemang
2024-04-03 16:12:59 -04:00
parent 002761d7f8
commit 5809b64f1c
1 changed files with 0 additions and 336 deletions
Download Patch File Download Diff File Expand all files Collapse all files

336
src/swapchain.zig
View File

@@ -1,336 +0,0 @@
const std = @import("std");
const vk = @import("vk");
const gfx = @import("gfx.zig");
const Allocator = std.mem.Allocator;
pub const Context = struct {
vki: gfx.InstanceDispatch,
vkd: gfx.DeviceDispatch,
pdev: vk.PhysicalDevice,
dev: vk.Device,
surface: vk.SurfaceKHR,
queue: vk.Queue,
family: u32,
};
pub const Swapchain = struct {
pub const PresentState = enum {
optimal,
suboptimal,
};
gc: *const Context,
allocator: Allocator,
present_mode: vk.PresentModeKHR,
extent: vk.Extent2D,
handle: vk.SwapchainKHR,
swap_images: []SwapImage,
image_index: u32,
next_image_acquired: vk.Semaphore,
pub fn init(
gc: *const Context,
allocator: Allocator,
extent: vk.Extent2D,
format: vk.SurfaceFormatKHR,
) !Swapchain {
return try initRecycle(gc, allocator, extent, format, .null_handle);
}
pub fn initRecycle(
gc: *const Context,
allocator: Allocator,
extent: vk.Extent2D,
format: vk.SurfaceFormatKHR,
old_handle: vk.SwapchainKHR,
) !Swapchain {
const caps = try gc.vki.getPhysicalDeviceSurfaceCapabilitiesKHR(gc.pdev, gc.surface);
const actual_extent = findActualExtent(caps, extent);
if (actual_extent.width == 0 or actual_extent.height == 0) {
return error.InvalidSurfaceDimensions;
}
const present_mode = try findPresentMode(gc, allocator);
var image_count = caps.min_image_count + 1;
if (caps.max_image_count > 0) {
image_count = @min(image_count, caps.max_image_count);
}
const handle = try gc.vkd.createSwapchainKHR(gc.dev, &.{
.surface = gc.surface,
.min_image_count = image_count,
.image_format = format.format,
.image_color_space = format.color_space,
.image_extent = actual_extent,
.image_array_layers = 1,
.image_usage = .{ .color_attachment_bit = true, .transfer_dst_bit = true },
.image_sharing_mode = .exclusive,
.pre_transform = caps.current_transform,
.composite_alpha = .{ .opaque_bit_khr = true },
.present_mode = present_mode,
.clipped = vk.TRUE,
.old_swapchain = old_handle,
}, null);
errdefer gc.vkd.destroySwapchainKHR(gc.dev, handle, null);
if (old_handle != .null_handle) {
// Apparently, the old swapchain handle still needs to be destroyed after recreating.
gc.vkd.destroySwapchainKHR(gc.dev, old_handle, null);
}
const swap_images = try initSwapchainImages(gc, handle, format.format, allocator);
errdefer {
for (swap_images) |si| si.deinit(gc);
allocator.free(swap_images);
}
var next_image_acquired = try gc.vkd.createSemaphore(gc.dev, &.{}, null);
errdefer gc.vkd.destroySemaphore(gc.dev, next_image_acquired, null);
const result = try gc.vkd.acquireNextImageKHR(
gc.dev,
handle,
std.math.maxInt(u64),
next_image_acquired,
.null_handle,
);
switch (result.result) {
vk.Result.success, vk.Result.suboptimal_khr => {},
vk.Result.timeout => return error.Timeout,
vk.Result.not_ready => return error.NotReady,
else => unreachable,
}
std.mem.swap(vk.Semaphore, &swap_images[result.image_index].image_acquired, &next_image_acquired);
return Swapchain{
.gc = gc,
.allocator = allocator,
.present_mode = present_mode,
.extent = actual_extent,
.handle = handle,
.swap_images = swap_images,
.image_index = result.image_index,
.next_image_acquired = next_image_acquired,
};
}
fn deinitExceptSwapchain(self: Swapchain) void {
for (self.swap_images) |si| si.deinit(self.gc);
self.allocator.free(self.swap_images);
self.gc.vkd.destroySemaphore(self.gc.dev, self.next_image_acquired, null);
}
pub fn waitForAllFences(self: Swapchain) !void {
for (self.swap_images) |si| si.waitForFence(self.gc) catch {};
}
pub fn deinit(self: Swapchain) void {
self.deinitExceptSwapchain();
self.gc.vkd.destroySwapchainKHR(self.gc.dev, self.handle, null);
}
pub fn recreate(
self: *Swapchain,
new_extent: vk.Extent2D,
format: vk.SurfaceFormatKHR,
) !void {
const gc = self.gc;
const allocator = self.allocator;
const old_handle = self.handle;
self.deinitExceptSwapchain();
self.* = try initRecycle(gc, allocator, new_extent, format, old_handle);
}
pub fn currentImage(self: Swapchain) vk.Image {
return self.swap_images[self.image_index].image;
}
pub fn currentSwapImage(self: Swapchain) *const SwapImage {
return &self.swap_images[self.image_index];
}
pub fn present(self: *Swapchain, cmdbuf: vk.CommandBuffer) !PresentState {
// Simple method:
// 1) Acquire next image
// 2) Wait for and reset fence of the acquired image
// 3) Submit command buffer with fence of acquired image,
// dependendent on the semaphore signalled by the first step.
// 4) Present current frame, dependent on semaphore signalled by previous step
// Problem: This way we can't reference the current image while rendering.
// Better method: Shuffle the steps around such that acquire next image is the last step,
// leaving the swapchain in a state with the current image.
// 1) Wait for and reset fence of current image
// 2) Submit command buffer, signalling fence of current image and dependent on
// the semaphore signalled by step 4.
// 3) Present current frame, dependent on semaphore signalled by the submit
// 4) Acquire next image, signalling its semaphore
// One problem that arises is that we can't know beforehand which semaphore to signal,
// so we keep an extra auxilery semaphore that is swapped around
// Step 1: Make sure the current frame has finished rendering
const current = self.currentSwapImage();
try current.waitForFence(self.gc);
try self.gc.vkd.resetFences(self.gc.dev, 1, @ptrCast(&current.frame_fence));
// Step 2: Submit the command buffer
const wait_stage = [_]vk.PipelineStageFlags{.{ .top_of_pipe_bit = true }};
try self.gc.vkd.queueSubmit(self.gc.queue, 1, &[_]vk.SubmitInfo{.{
.wait_semaphore_count = 1,
.p_wait_semaphores = @ptrCast(&current.image_acquired),
.p_wait_dst_stage_mask = &wait_stage,
.command_buffer_count = 1,
.p_command_buffers = @ptrCast(&cmdbuf),
.signal_semaphore_count = 1,
.p_signal_semaphores = @ptrCast(&current.render_finished),
}}, current.frame_fence);
// Step 3: Present the current frame
_ = try self.gc.vkd.queuePresentKHR(self.gc.queue, &.{
.wait_semaphore_count = 1,
.p_wait_semaphores = @as([*]const vk.Semaphore, @ptrCast(&current.render_finished)),
.swapchain_count = 1,
.p_swapchains = @as([*]const vk.SwapchainKHR, @ptrCast(&self.handle)),
.p_image_indices = @as([*]const u32, @ptrCast(&self.image_index)),
});
// Step 4: Acquire next frame
const result = try self.gc.vkd.acquireNextImageKHR(
self.gc.dev,
self.handle,
std.math.maxInt(u64),
self.next_image_acquired,
.null_handle,
);
std.mem.swap(vk.Semaphore, &self.swap_images[result.image_index].image_acquired, &self.next_image_acquired);
self.image_index = result.image_index;
return switch (result.result) {
.success => .optimal,
.suboptimal_khr => .suboptimal,
else => unreachable,
};
}
};
const SwapImage = struct {
image: vk.Image,
view: vk.ImageView,
image_acquired: vk.Semaphore,
render_finished: vk.Semaphore,
frame_fence: vk.Fence,
fn init(gc: *const Context, image: vk.Image, format: vk.Format) !SwapImage {
const view = try gc.vkd.createImageView(gc.dev, &.{
.image = image,
.view_type = .@"2d",
.format = format,
.components = .{ .r = .identity, .g = .identity, .b = .identity, .a = .identity },
.subresource_range = .{
.aspect_mask = .{ .color_bit = true },
.base_mip_level = 0,
.level_count = 1,
.base_array_layer = 0,
.layer_count = 1,
},
}, null);
errdefer gc.vkd.destroyImageView(gc.dev, view, null);
const image_acquired = try gc.vkd.createSemaphore(gc.dev, &.{}, null);
errdefer gc.vkd.destroySemaphore(gc.dev, image_acquired, null);
const render_finished = try gc.vkd.createSemaphore(gc.dev, &.{}, null);
errdefer gc.vkd.destroySemaphore(gc.dev, render_finished, null);
const frame_fence = try gc.vkd.createFence(gc.dev, &.{ .flags = .{ .signaled_bit = true } }, null);
errdefer gc.vkd.destroyFence(gc.dev, frame_fence, null);
return SwapImage{
.image = image,
.view = view,
.image_acquired = image_acquired,
.render_finished = render_finished,
.frame_fence = frame_fence,
};
}
fn deinit(self: SwapImage, gc: *const Context) void {
// todo critical: this waitForFence deadlocks when recreating swapchain on nvidia.
// Something about the main "present" loop is fucked. Can't just ignore the fence; validation layers show errors
// that you can't destroy a fence while a queue depends on it (details may be wrong... I don't fully understand)
// I suspect this is more an issue with the vulkan-zig example, so I probably need to revisit vulkan-tutorial or
// try to rebuild the swapchain infrastructure myself.
//
// I do think it's clunky how the swapchain is created and recreated; duplicate logic in creation and recreation
// that could maybe be avoided with .null_handle? Maybe there's some more straightforward way to handle it.
self.waitForFence(gc) catch return;
gc.vkd.destroyImageView(gc.dev, self.view, null);
gc.vkd.destroySemaphore(gc.dev, self.image_acquired, null);
gc.vkd.destroySemaphore(gc.dev, self.render_finished, null);
gc.vkd.destroyFence(gc.dev, self.frame_fence, null);
}
fn waitForFence(self: SwapImage, gc: *const Context) !void {
_ = try gc.vkd.waitForFences(gc.dev, 1, @ptrCast(&self.frame_fence), vk.TRUE, std.math.maxInt(u64));
}
};
fn initSwapchainImages(gc: *const Context, swapchain: vk.SwapchainKHR, format: vk.Format, allocator: Allocator) ![]SwapImage {
var count: u32 = undefined;
_ = try gc.vkd.getSwapchainImagesKHR(gc.dev, swapchain, &count, null);
const images = try allocator.alloc(vk.Image, count);
defer allocator.free(images);
_ = try gc.vkd.getSwapchainImagesKHR(gc.dev, swapchain, &count, images.ptr);
const swap_images = try allocator.alloc(SwapImage, count);
errdefer allocator.free(swap_images);
var i: usize = 0;
errdefer for (swap_images[0..i]) |si| si.deinit(gc);
for (images) |image| {
swap_images[i] = try SwapImage.init(gc, image, format);
i += 1;
}
return swap_images;
}
fn findPresentMode(gc: *const Context, allocator: Allocator) !vk.PresentModeKHR {
var count: u32 = undefined;
_ = try gc.vki.getPhysicalDeviceSurfacePresentModesKHR(gc.pdev, gc.surface, &count, null);
const present_modes = try allocator.alloc(vk.PresentModeKHR, count);
defer allocator.free(present_modes);
_ = try gc.vki.getPhysicalDeviceSurfacePresentModesKHR(gc.pdev, gc.surface, &count, present_modes.ptr);
const preferred = [_]vk.PresentModeKHR{
.mailbox_khr,
.immediate_khr,
};
for (preferred) |mode| {
if (std.mem.indexOfScalar(vk.PresentModeKHR, present_modes, mode) != null) {
return mode;
}
}
return .fifo_khr;
}
fn findActualExtent(caps: vk.SurfaceCapabilitiesKHR, extent: vk.Extent2D) vk.Extent2D {
if (caps.current_extent.width != 0xFFFF_FFFF) {
return caps.current_extent;
} else {
return .{
.width = std.math.clamp(extent.width, caps.min_image_extent.width, caps.max_image_extent.width),
.height = std.math.clamp(extent.height, caps.min_image_extent.height, caps.max_image_extent.height),
};
}
}