324 lines
11 KiB
Zig
Executable file
324 lines
11 KiB
Zig
Executable file
const std = @import("std");
|
|
const mach = @import("mach");
|
|
const gpu = @import("gpu");
|
|
const glfw = @import("glfw");
|
|
|
|
const App = @This();
|
|
|
|
const Vertex = struct {
|
|
pos: @Vector(4, f32),
|
|
uv: @Vector(2, f32),
|
|
};
|
|
|
|
const vertices = [_]Vertex{
|
|
.{ .pos = .{ -1, -1, 0, 1 }, .uv = .{ 0, 0 } },
|
|
.{ .pos = .{ 1, -1, 0, 1 }, .uv = .{ 1, 0 } },
|
|
.{ .pos = .{ 1, 1, 0, 1 }, .uv = .{ 1, 1 } },
|
|
.{ .pos = .{ -1, 1, 0, 1 }, .uv = .{ 0, 1 } },
|
|
};
|
|
const indices = [_]u16{ 0, 1, 2, 2, 3, 0 };
|
|
|
|
const UniformBufferObject = struct {
|
|
resolution: @Vector(2, f32),
|
|
time: f32,
|
|
};
|
|
|
|
var timer: std.time.Timer = undefined;
|
|
|
|
pipeline: gpu.RenderPipeline,
|
|
queue: gpu.Queue,
|
|
vertex_buffer: gpu.Buffer,
|
|
index_buffer: gpu.Buffer,
|
|
uniform_buffer: gpu.Buffer,
|
|
bind_group: gpu.BindGroup,
|
|
|
|
fragment_shader_file: std.fs.File,
|
|
fragment_shader_code: [:0]const u8,
|
|
last_mtime: i128,
|
|
|
|
pub fn init(app: *App, engine: *mach.Engine) !void {
|
|
timer = try std.time.Timer.start();
|
|
|
|
// On linux if we don't set a minimum size, you can squish the window to 0 pixels of width and height,
|
|
// this makes some strange effects when that happens, so it's better to leave a minimum size to avoid that,
|
|
// this doesn't prevent you from minimizing the window.
|
|
try engine.setOptions(.{
|
|
.size_min = .{ .width = 20, .height = 20 },
|
|
});
|
|
|
|
var fragment_file: std.fs.File = undefined;
|
|
var last_mtime: i128 = undefined;
|
|
if (std.fs.cwd().openFile("shaderexp/frag.wgsl", .{ .mode = .read_only })) |file| {
|
|
fragment_file = file;
|
|
if (file.stat()) |stat| {
|
|
last_mtime = stat.mtime;
|
|
} else |err| {
|
|
std.debug.print("Something went wrong when attempting to stat file: {}\n", .{err});
|
|
return;
|
|
}
|
|
} else |e| {
|
|
std.debug.print("Something went wrong when attempting to open file: {}\n", .{e});
|
|
return;
|
|
}
|
|
var code = try fragment_file.readToEndAllocOptions(engine.allocator, std.math.maxInt(u16), null, 1, 0);
|
|
|
|
const queue = engine.device.getQueue();
|
|
|
|
const vertex_buffer = engine.device.createBuffer(&.{
|
|
.usage = .{ .vertex = true },
|
|
.size = @sizeOf(Vertex) * vertices.len,
|
|
.mapped_at_creation = true,
|
|
});
|
|
var vertex_mapped = vertex_buffer.getMappedRange(Vertex, 0, vertices.len);
|
|
std.mem.copy(Vertex, vertex_mapped, vertices[0..]);
|
|
vertex_buffer.unmap();
|
|
|
|
const index_buffer = engine.device.createBuffer(&.{
|
|
.usage = .{ .index = true },
|
|
.size = @sizeOf(u16) * indices.len,
|
|
.mapped_at_creation = true,
|
|
});
|
|
var index_mapped = index_buffer.getMappedRange(@TypeOf(indices[0]), 0, indices.len);
|
|
std.mem.copy(u16, index_mapped, indices[0..]);
|
|
index_buffer.unmap();
|
|
|
|
// We need a bgl to bind the UniformBufferObject, but it is also needed for creating
|
|
// the RenderPipeline, so we pass it to recreatePipeline as a pointer
|
|
var bgl: gpu.BindGroupLayout = undefined;
|
|
const pipeline = recreatePipeline(engine, code, &bgl);
|
|
|
|
const uniform_buffer = engine.device.createBuffer(&.{
|
|
.usage = .{ .copy_dst = true, .uniform = true },
|
|
.size = @sizeOf(UniformBufferObject),
|
|
.mapped_at_creation = false,
|
|
});
|
|
const bind_group = engine.device.createBindGroup(
|
|
&gpu.BindGroup.Descriptor{
|
|
.layout = bgl,
|
|
.entries = &.{
|
|
gpu.BindGroup.Entry.buffer(0, uniform_buffer, 0, @sizeOf(UniformBufferObject)),
|
|
},
|
|
},
|
|
);
|
|
|
|
app.pipeline = pipeline;
|
|
app.queue = queue;
|
|
app.vertex_buffer = vertex_buffer;
|
|
app.index_buffer = index_buffer;
|
|
app.uniform_buffer = uniform_buffer;
|
|
app.bind_group = bind_group;
|
|
|
|
app.fragment_shader_file = fragment_file;
|
|
app.fragment_shader_code = code;
|
|
app.last_mtime = last_mtime;
|
|
|
|
bgl.release();
|
|
}
|
|
|
|
pub fn deinit(app: *App, engine: *mach.Engine) void {
|
|
app.fragment_shader_file.close();
|
|
engine.allocator.free(app.fragment_shader_code);
|
|
|
|
app.vertex_buffer.release();
|
|
app.index_buffer.release();
|
|
app.uniform_buffer.release();
|
|
app.bind_group.release();
|
|
}
|
|
|
|
pub fn update(app: *App, engine: *mach.Engine) !bool {
|
|
while (engine.pollEvent()) |event| {
|
|
switch (event) {
|
|
.key_press => |ev| {
|
|
if (ev.key == .space)
|
|
engine.setShouldClose(true);
|
|
},
|
|
else => {},
|
|
}
|
|
}
|
|
|
|
if (app.fragment_shader_file.stat()) |stat| {
|
|
if (app.last_mtime < stat.mtime) {
|
|
std.log.info("The fragment shader has been changed", .{});
|
|
app.last_mtime = stat.mtime;
|
|
app.fragment_shader_file.seekTo(0) catch unreachable;
|
|
app.fragment_shader_code = app.fragment_shader_file.readToEndAllocOptions(engine.allocator, std.math.maxInt(u32), null, 1, 0) catch |err| {
|
|
std.log.err("Err: {}", .{err});
|
|
return false;
|
|
};
|
|
app.pipeline = recreatePipeline(engine, app.fragment_shader_code, null);
|
|
}
|
|
} else |err| {
|
|
std.log.err("Something went wrong when attempting to stat file: {}\n", .{err});
|
|
}
|
|
|
|
const back_buffer_view = engine.swap_chain.?.getCurrentTextureView();
|
|
const color_attachment = gpu.RenderPassColorAttachment{
|
|
.view = back_buffer_view,
|
|
.resolve_target = null,
|
|
.clear_value = std.mem.zeroes(gpu.Color),
|
|
.load_op = .clear,
|
|
.store_op = .store,
|
|
};
|
|
|
|
const encoder = engine.device.createCommandEncoder(null);
|
|
const render_pass_info = gpu.RenderPassEncoder.Descriptor{
|
|
.color_attachments = &.{color_attachment},
|
|
.depth_stencil_attachment = null,
|
|
};
|
|
|
|
const time = @intToFloat(f32, timer.read()) / @as(f32, std.time.ns_per_s);
|
|
const ubo = UniformBufferObject{
|
|
.resolution = .{ @intToFloat(f32, engine.current_desc.width), @intToFloat(f32, engine.current_desc.height) },
|
|
.time = time,
|
|
};
|
|
encoder.writeBuffer(app.uniform_buffer, 0, UniformBufferObject, &.{ubo});
|
|
|
|
const pass = encoder.beginRenderPass(&render_pass_info);
|
|
pass.setVertexBuffer(0, app.vertex_buffer, 0, @sizeOf(Vertex) * vertices.len);
|
|
pass.setIndexBuffer(app.index_buffer, .uint16, 0, @sizeOf(u16) * indices.len);
|
|
pass.setPipeline(app.pipeline);
|
|
pass.setBindGroup(0, app.bind_group, &.{0});
|
|
pass.drawIndexed(indices.len, 1, 0, 0, 0);
|
|
pass.end();
|
|
pass.release();
|
|
|
|
var command = encoder.finish(null);
|
|
encoder.release();
|
|
|
|
app.queue.submit(&.{command});
|
|
command.release();
|
|
engine.swap_chain.?.present();
|
|
back_buffer_view.release();
|
|
|
|
return true;
|
|
}
|
|
|
|
fn recreatePipeline(engine: *mach.Engine, fragment_shader_code: [:0]const u8, bgl: ?*gpu.BindGroupLayout) gpu.RenderPipeline {
|
|
const vs_module = engine.device.createShaderModule(&.{
|
|
.label = "my vertex shader",
|
|
.code = .{ .wgsl = @embedFile("vert.wgsl") },
|
|
});
|
|
defer vs_module.release();
|
|
const vertex_attributes = [_]gpu.VertexAttribute{
|
|
.{ .format = .float32x4, .offset = @offsetOf(Vertex, "pos"), .shader_location = 0 },
|
|
.{ .format = .float32x2, .offset = @offsetOf(Vertex, "uv"), .shader_location = 1 },
|
|
};
|
|
const vertex_buffer_layout = gpu.VertexBufferLayout{
|
|
.array_stride = @sizeOf(Vertex),
|
|
.step_mode = .vertex,
|
|
.attribute_count = vertex_attributes.len,
|
|
.attributes = &vertex_attributes,
|
|
};
|
|
|
|
// Check wether the fragment shader code compiled successfully, if not
|
|
// print the validation layer error and show a black screen
|
|
engine.device.pushErrorScope(.validation);
|
|
var fs_module = engine.device.createShaderModule(&gpu.ShaderModule.Descriptor{
|
|
.label = "my fragment shader",
|
|
.code = .{ .wgsl = fragment_shader_code },
|
|
});
|
|
var error_occurred: bool = false;
|
|
// popErrorScope() returns always true, (unless maybe it fails to capture the error scope?)
|
|
_ = engine.device.popErrorScope(&gpu.ErrorCallback.init(*bool, &error_occurred, struct {
|
|
fn callback(ctx: *bool, typ: gpu.ErrorType, message: [*:0]const u8) void {
|
|
if (typ != .noError) {
|
|
std.debug.print("🔴🔴🔴🔴:\n{s}\n", .{message});
|
|
ctx.* = true;
|
|
}
|
|
}
|
|
}.callback));
|
|
if (error_occurred) {
|
|
fs_module = engine.device.createShaderModule(&gpu.ShaderModule.Descriptor{
|
|
.label = "my fragment shader",
|
|
.code = .{ .wgsl = @embedFile("black_screen_frag.wgsl") },
|
|
});
|
|
}
|
|
defer fs_module.release();
|
|
|
|
const blend = gpu.BlendState{
|
|
.color = .{
|
|
.operation = .add,
|
|
.src_factor = .one,
|
|
.dst_factor = .zero,
|
|
},
|
|
.alpha = .{
|
|
.operation = .add,
|
|
.src_factor = .one,
|
|
.dst_factor = .zero,
|
|
},
|
|
};
|
|
const color_target = gpu.ColorTargetState{
|
|
.format = engine.swap_chain_format,
|
|
.blend = &blend,
|
|
.write_mask = gpu.ColorWriteMask.all,
|
|
};
|
|
const fragment = gpu.FragmentState{
|
|
.module = fs_module,
|
|
.entry_point = "main",
|
|
.targets = &.{color_target},
|
|
.constants = null,
|
|
};
|
|
|
|
const bgle = gpu.BindGroupLayout.Entry.buffer(0, .{ .fragment = true }, .uniform, true, 0);
|
|
// bgl is needed outside, for the creation of the uniform_buffer in main
|
|
const bgl_tmp = engine.device.createBindGroupLayout(
|
|
&gpu.BindGroupLayout.Descriptor{
|
|
.entries = &.{bgle},
|
|
},
|
|
);
|
|
defer {
|
|
// In frame we don't need to use bgl, so we can release it inside this function, else we pass bgl
|
|
if (bgl == null) {
|
|
bgl_tmp.release();
|
|
} else {
|
|
bgl.?.* = bgl_tmp;
|
|
}
|
|
}
|
|
|
|
const bind_group_layouts = [_]gpu.BindGroupLayout{bgl_tmp};
|
|
const pipeline_layout = engine.device.createPipelineLayout(&.{
|
|
.bind_group_layouts = &bind_group_layouts,
|
|
});
|
|
defer pipeline_layout.release();
|
|
|
|
const pipeline_descriptor = gpu.RenderPipeline.Descriptor{
|
|
.fragment = &fragment,
|
|
.layout = pipeline_layout,
|
|
.depth_stencil = null,
|
|
.vertex = .{
|
|
.module = vs_module,
|
|
.entry_point = "main",
|
|
.buffers = &.{vertex_buffer_layout},
|
|
},
|
|
.multisample = .{
|
|
.count = 1,
|
|
.mask = 0xFFFFFFFF,
|
|
.alpha_to_coverage_enabled = false,
|
|
},
|
|
.primitive = .{
|
|
.front_face = .ccw,
|
|
.cull_mode = .none,
|
|
.topology = .triangle_list,
|
|
.strip_index_format = .none,
|
|
},
|
|
};
|
|
|
|
// Create the render pipeline. Even if the shader compilation succeeded, this could fail if the
|
|
// shader is missing a `main` entrypoint.
|
|
engine.device.pushErrorScope(.validation);
|
|
const pipeline = engine.device.createRenderPipeline(&pipeline_descriptor);
|
|
// popErrorScope() returns always true, (unless maybe it fails to capture the error scope?)
|
|
_ = engine.device.popErrorScope(&gpu.ErrorCallback.init(*bool, &error_occurred, struct {
|
|
fn callback(ctx: *bool, typ: gpu.ErrorType, message: [*:0]const u8) void {
|
|
if (typ != .noError) {
|
|
std.debug.print("🔴🔴🔴🔴:\n{s}\n", .{message});
|
|
ctx.* = true;
|
|
}
|
|
}
|
|
}.callback));
|
|
if (error_occurred) {
|
|
// Retry with black_screen_frag which we know will work.
|
|
return recreatePipeline(engine, @embedFile("black_screen_frag.wgsl"), bgl);
|
|
}
|
|
return pipeline;
|
|
}
|