add higher-level app library

Fixes hexops/mach#190

Signed-off-by: Stephen Gutekanst <stephen@hexops.com>

src/c.zig

@@ -0,0 +1,5 @@
+pub const c = @cImport({
+    @cInclude("dawn/webgpu.h");
+    @cInclude("dawn/dawn_proc.h");
+    @cInclude("dawn_native_mach.h");
+});

src/main.zig

@@ -2,7 +2,233 @@ const std = @import("std");
 const testing = std.testing;
 
 const glfw = @import("glfw");
+const gpu = @import("gpu");
+const util = @import("util.zig");
+const c = @import("c.zig").c;
+
+/// For now, this contains nothing. In the future, this will include application configuration that
+/// can only be specified at compile-time.
+pub const AppConfig = struct {};
+
+/// Application options that can be configured at init time.
+pub const Options = struct {
+    /// The title of the window.
+    title: [*:0]const u8 = "Mach engine",
+
+    /// The width of the window.
+    width: u32 = 640,
+
+    /// The height of the window.
+    height: u32 = 480,
+
+    /// GPU features required by the application.
+    required_features: ?[]gpu.Feature = null,
+
+    /// GPU limits required by the application.
+    required_limits: ?gpu.Limits = null,
+
+    /// Whether the application has a preference for low power or high performance GPU.
+    power_preference: gpu.PowerPreference = .none,
+};
+
+/// A Mach application.
+///
+/// The Context type is your own data type which can later be accessed via app.context from within
+/// the frame function you pass to run().
+pub fn App(comptime Context: type, comptime config: AppConfig) type {
+    _ = config;
+    return struct {
+        context: Context,
+        device: gpu.Device,
+        window: glfw.Window,
+        backend_type: gpu.Adapter.BackendType,
+        allocator: std.mem.Allocator,
+        swap_chain: ?gpu.SwapChain,
+        swap_chain_format: gpu.Texture.Format,
+
+        // Internals
+        native_instance: gpu.NativeInstance,
+        surface: ?gpu.Surface,
+        current_desc: gpu.SwapChain.Descriptor,
+        target_desc: gpu.SwapChain.Descriptor,
+
+        const Self = @This();
+
+        pub fn init(allocator: std.mem.Allocator, context: Context, options: Options) !Self {
+            const backend_type = try util.detectBackendType(allocator);
+
+            try glfw.init(.{});
+
+            // Create the test window and discover adapters using it (esp. for OpenGL)
+            var hints = util.glfwWindowHintsForBackend(backend_type);
+            hints.cocoa_retina_framebuffer = true;
+            const window = try glfw.Window.create(
+                options.width,
+                options.height,
+                options.title,
+                null,
+                null,
+                hints,
+            );
+
+            const backend_procs = c.machDawnNativeGetProcs();
+            c.dawnProcSetProcs(backend_procs);
+
+            const instance = c.machDawnNativeInstance_init();
+            var native_instance = gpu.NativeInstance.wrap(c.machDawnNativeInstance_get(instance).?);
+
+            // Discover e.g. OpenGL adapters.
+            try util.discoverAdapters(instance, window, backend_type);
+
+            // Request an adapter.
+            //
+            // TODO: It would be nice if we could use gpu_interface.waitForAdapter here, however the webgpu.h
+            // API does not yet have a way to specify what type of backend you want (vulkan, opengl, etc.)
+            // In theory, I suppose we shouldn't need to and Dawn should just pick the best adapter - but in
+            // practice if Vulkan is not supported today waitForAdapter/requestAdapter merely generates an error.
+            //
+            // const gpu_interface = native_instance.interface();
+            // const backend_adapter = switch (gpu_interface.waitForAdapter(&.{
+            //     .power_preference = .high_performance,
+            // })) {
+            //     .adapter => |v| v,
+            //     .err => |err| {
+            //         std.debug.print("mach: failed to get adapter: error={} {s}\n", .{ err.code, err.message });
+            //         std.process.exit(1);
+            //     },
+            // };
+            const adapters = c.machDawnNativeInstance_getAdapters(instance);
+            var dawn_adapter: ?c.MachDawnNativeAdapter = null;
+            var i: usize = 0;
+            while (i < c.machDawnNativeAdapters_length(adapters)) : (i += 1) {
+                const adapter = c.machDawnNativeAdapters_index(adapters, i);
+                const properties = c.machDawnNativeAdapter_getProperties(adapter);
+                const found_backend_type = @intToEnum(gpu.Adapter.BackendType, c.machDawnNativeAdapterProperties_getBackendType(properties));
+                if (found_backend_type == backend_type) {
+                    dawn_adapter = adapter;
+                }
+            }
+            if (dawn_adapter == null) {
+                std.debug.print("mach: no matching adapter found for {s}", .{@tagName(backend_type)});
+                std.debug.print("-> maybe try GPU_BACKEND=opengl ?\n", .{});
+                std.process.exit(1);
+            }
+            std.debug.assert(dawn_adapter != null);
+            const backend_adapter = gpu.NativeInstance.fromWGPUAdapter(c.machDawnNativeAdapter_get(dawn_adapter.?).?);
+
+            // Print which adapter we are going to use.
+            const props = backend_adapter.properties;
+            std.debug.print("mach: found {s} backend on {s} adapter: {s}, {s}\n", .{
+                gpu.Adapter.backendTypeName(props.backend_type),
+                gpu.Adapter.typeName(props.adapter_type),
+                props.name,
+                props.driver_description,
+            });
+
+            const device = switch (backend_adapter.waitForDevice(&.{
+                .required_features = options.required_features,
+                .required_limits = options.required_limits,
+            })) {
+                .device => |v| v,
+                .err => |err| {
+                    // TODO: return a proper error type
+                    std.debug.print("mach: failed to get device: error={} {s}\n", .{ err.code, err.message });
+                    std.process.exit(1);
+                },
+            };
+
+            var framebuffer_size = try window.getFramebufferSize();
+
+            // If targeting OpenGL, we can't use the newer WGPUSurface API. Instead, we need to use the
+            // older Dawn-specific API. https://bugs.chromium.org/p/dawn/issues/detail?id=269&q=surface&can=2
+            const use_legacy_api = backend_type == .opengl or backend_type == .opengles;
+            var descriptor: gpu.SwapChain.Descriptor = undefined;
+            var swap_chain: ?gpu.SwapChain = null;
+            var swap_chain_format: gpu.Texture.Format = undefined;
+            var surface: ?gpu.Surface = null;
+            if (!use_legacy_api) {
+                swap_chain_format = .bgra8_unorm;
+                descriptor = .{
+                    .label = "basic swap chain",
+                    .usage = .render_attachment,
+                    .format = swap_chain_format,
+                    .width = framebuffer_size.width,
+                    .height = framebuffer_size.height,
+                    .present_mode = .fifo,
+                    .implementation = 0,
+                };
+                surface = util.createSurfaceForWindow(
+                    &native_instance,
+                    window,
+                    comptime util.detectGLFWOptions(),
+                );
+            } else {
+                const binding = c.machUtilsCreateBinding(@enumToInt(backend_type), @ptrCast(*c.GLFWwindow, window.handle), @ptrCast(c.WGPUDevice, device.ptr));
+                if (binding == null) {
+                    @panic("failed to create Dawn backend binding");
+                }
+                descriptor = std.mem.zeroes(gpu.SwapChain.Descriptor);
+                descriptor.implementation = c.machUtilsBackendBinding_getSwapChainImplementation(binding);
+                swap_chain = device.nativeCreateSwapChain(null, &descriptor);
+
+                swap_chain_format = @intToEnum(gpu.Texture.Format, @intCast(u32, c.machUtilsBackendBinding_getPreferredSwapChainTextureFormat(binding)));
+                swap_chain.?.configure(
+                    swap_chain_format,
+                    .render_attachment,
+                    framebuffer_size.width,
+                    framebuffer_size.height,
+                );
+            }
+
+            device.setUncapturedErrorCallback(&util.printUnhandledErrorCallback);
+            return Self{
+                .context = context,
+                .device = device,
+                .window = window,
+                .backend_type = backend_type,
+                .allocator = allocator,
+
+                .native_instance = native_instance,
+                .surface = surface,
+                .swap_chain = swap_chain,
+                .swap_chain_format = swap_chain_format,
+                .current_desc = descriptor,
+                .target_desc = descriptor,
+            };
+        }
+
+        const FrameFunc = fn (app: *Self, ctx: Context) error{OutOfMemory}!void;
+
+        pub fn run(app: *Self, frame: FrameFunc) !void {
+            while (!app.window.shouldClose()) {
+                try glfw.pollEvents();
+
+                var framebuffer_size = try app.window.getFramebufferSize();
+                app.target_desc.width = framebuffer_size.width;
+                app.target_desc.height = framebuffer_size.height;
+
+                if (app.swap_chain == null or !app.current_desc.equal(&app.target_desc)) {
+                    const use_legacy_api = app.surface == null;
+                    if (!use_legacy_api) {
+                        app.swap_chain = app.device.nativeCreateSwapChain(app.surface, &app.target_desc);
+                    } else app.swap_chain.?.configure(
+                        app.swap_chain_format,
+                        .render_attachment,
+                        app.target_desc.width,
+                        app.target_desc.height,
+                    );
+                    app.current_desc = app.target_desc;
+                }
+
+                try frame(app, app.context);
+                std.time.sleep(16 * std.time.ns_per_ms); // TODO: this is very naive
+            }
+        }
+    };
+}
 
 test "glfw_basic" {
+    _ = Options;
+    _ = App;
     glfw.basicTest() catch unreachable;
 }

src/util.zig

@@ -0,0 +1,168 @@
+const std = @import("std");
+
+const glfw = @import("glfw");
+const gpu = @import("gpu");
+const c = @import("c.zig").c;
+const objc = @cImport({
+    @cInclude("objc/message.h");
+});
+
+fn printUnhandledError(_: void, typ: gpu.ErrorType, message: [*:0]const u8) void {
+    switch (typ) {
+        .validation => std.debug.print("gpu: validation error: {s}\n", .{message}),
+        .out_of_memory => std.debug.print("gpu: out of memory: {s}\n", .{message}),
+        .device_lost => std.debug.print("gpu: device lost: {s}\n", .{message}),
+        .unknown => std.debug.print("gpu: unknown error: {s}\n", .{message}),
+        else => unreachable,
+    }
+    std.os.exit(1);
+}
+pub var printUnhandledErrorCallback = gpu.ErrorCallback.init(void, {}, printUnhandledError);
+
+fn getEnvVarOwned(allocator: std.mem.Allocator, key: []const u8) error{ OutOfMemory, InvalidUtf8 }!?[]u8 {
+    return std.process.getEnvVarOwned(allocator, key) catch |err| switch (err) {
+        error.EnvironmentVariableNotFound => @as(?[]u8, null),
+        else => |e| e,
+    };
+}
+
+pub fn detectBackendType(allocator: std.mem.Allocator) !gpu.Adapter.BackendType {
+    const GPU_BACKEND = try getEnvVarOwned(allocator, "GPU_BACKEND");
+    if (GPU_BACKEND) |backend| {
+        defer allocator.free(backend);
+        if (std.ascii.eqlIgnoreCase(backend, "opengl")) return .opengl;
+        if (std.ascii.eqlIgnoreCase(backend, "opengles")) return .opengles;
+        if (std.ascii.eqlIgnoreCase(backend, "d3d11")) return .d3d11;
+        if (std.ascii.eqlIgnoreCase(backend, "d3d12")) return .d3d12;
+        if (std.ascii.eqlIgnoreCase(backend, "metal")) return .metal;
+        if (std.ascii.eqlIgnoreCase(backend, "null")) return .nul;
+        if (std.ascii.eqlIgnoreCase(backend, "vulkan")) return .vulkan;
+        @panic("unknown GPU_BACKEND type");
+    }
+
+    const target = @import("builtin").target;
+    if (target.isDarwin()) return .metal;
+    if (target.os.tag == .windows) return .d3d12;
+    return .vulkan;
+}
+
+pub fn glfwWindowHintsForBackend(backend: gpu.Adapter.BackendType) glfw.Window.Hints {
+    return switch (backend) {
+        .opengl => .{
+            // Ask for OpenGL 4.4 which is what the GL backend requires for compute shaders and
+            // texture views.
+            .context_version_major = 4,
+            .context_version_minor = 4,
+            .opengl_forward_compat = true,
+            .opengl_profile = .opengl_core_profile,
+        },
+        .opengles => .{
+            .context_version_major = 3,
+            .context_version_minor = 1,
+            .client_api = .opengl_es_api,
+            .context_creation_api = .egl_context_api,
+        },
+        else => .{
+            // Without this GLFW will initialize a GL context on the window, which prevents using
+            // the window with other APIs (by crashing in weird ways).
+            .client_api = .no_api,
+        },
+    };
+}
+
+pub fn discoverAdapters(instance: c.MachDawnNativeInstance, window: glfw.Window, typ: gpu.Adapter.BackendType) !void {
+    switch (typ) {
+        .opengl => {
+            try glfw.makeContextCurrent(window);
+            const adapter_options = c.MachDawnNativeAdapterDiscoveryOptions_OpenGL{
+                .getProc = @ptrCast(fn ([*c]const u8) callconv(.C) ?*anyopaque, glfw.getProcAddress),
+            };
+            _ = c.machDawnNativeInstance_discoverAdapters(instance, @enumToInt(typ), &adapter_options);
+        },
+        .opengles => {
+            try glfw.makeContextCurrent(window);
+            const adapter_options = c.MachDawnNativeAdapterDiscoveryOptions_OpenGLES{
+                .getProc = @ptrCast(fn ([*c]const u8) callconv(.C) ?*anyopaque, glfw.getProcAddress),
+            };
+            _ = c.machDawnNativeInstance_discoverAdapters(instance, @enumToInt(typ), &adapter_options);
+        },
+        else => {
+            c.machDawnNativeInstance_discoverDefaultAdapters(instance);
+        },
+    }
+}
+
+pub fn detectGLFWOptions() glfw.BackendOptions {
+    const target = @import("builtin").target;
+    if (target.isDarwin()) return .{ .cocoa = true };
+    return switch (target.os.tag) {
+        .windows => .{ .win32 = true },
+        .linux => .{ .x11 = true },
+        else => .{},
+    };
+}
+
+pub fn createSurfaceForWindow(
+    native_instance: *const gpu.NativeInstance,
+    window: glfw.Window,
+    comptime glfw_options: glfw.BackendOptions,
+) gpu.Surface {
+    const glfw_native = glfw.Native(glfw_options);
+    const descriptor = if (glfw_options.win32) gpu.Surface.Descriptor{
+        .windows_hwnd = .{
+            .label = "basic surface",
+            .hinstance = std.os.windows.kernel32.GetModuleHandleW(null).?,
+            .hwnd = glfw_native.getWin32Window(window),
+        },
+    } else if (glfw_options.x11) gpu.Surface.Descriptor{
+        .xlib = .{
+            .label = "basic surface",
+            .display = glfw_native.getX11Display(),
+            .window = glfw_native.getX11Window(window),
+        },
+    } else if (glfw_options.cocoa) blk: {
+        const ns_window = glfw_native.getCocoaWindow(window);
+        const ns_view = msgSend(ns_window, "contentView", .{}, *anyopaque); // [nsWindow contentView]
+
+        // Create a CAMetalLayer that covers the whole window that will be passed to CreateSurface.
+        msgSend(ns_view, "setWantsLayer:", .{true}, void); // [view setWantsLayer:YES]
+        const layer = msgSend(objc.objc_getClass("CAMetalLayer"), "layer", .{}, ?*anyopaque); // [CAMetalLayer layer]
+        if (layer == null) @panic("failed to create Metal layer");
+        msgSend(ns_view, "setLayer:", .{layer.?}, void); // [view setLayer:layer]
+
+        // Use retina if the window was created with retina support.
+        const scale_factor = msgSend(ns_window, "backingScaleFactor", .{}, f64); // [ns_window backingScaleFactor]
+        msgSend(layer.?, "setContentsScale:", .{scale_factor}, void); // [layer setContentsScale:scale_factor]
+
+        break :blk gpu.Surface.Descriptor{
+            .metal_layer = .{
+                .label = "basic surface",
+                .layer = layer.?,
+            },
+        };
+    } else if (glfw_options.wayland) {
+        @panic("Dawn does not yet have Wayland support, see https://bugs.chromium.org/p/dawn/issues/detail?id=1246&q=surface&can=2");
+    } else unreachable;
+
+    return native_instance.createSurface(&descriptor);
+}
+
+// Borrowed from https://github.com/hazeycode/zig-objcrt
+fn msgSend(obj: anytype, sel_name: [:0]const u8, args: anytype, comptime ReturnType: type) ReturnType {
+    const args_meta = @typeInfo(@TypeOf(args)).Struct.fields;
+
+    const FnType = switch (args_meta.len) {
+        0 => fn (@TypeOf(obj), objc.SEL) callconv(.C) ReturnType,
+        1 => fn (@TypeOf(obj), objc.SEL, args_meta[0].field_type) callconv(.C) ReturnType,
+        2 => fn (@TypeOf(obj), objc.SEL, args_meta[0].field_type, args_meta[1].field_type) callconv(.C) ReturnType,
+        3 => fn (@TypeOf(obj), objc.SEL, args_meta[0].field_type, args_meta[1].field_type, args_meta[2].field_type) callconv(.C) ReturnType,
+        4 => fn (@TypeOf(obj), objc.SEL, args_meta[0].field_type, args_meta[1].field_type, args_meta[2].field_type, args_meta[3].field_type) callconv(.C) ReturnType,
+        else => @compileError("Unsupported number of args"),
+    };
+
+    // NOTE: func is a var because making it const causes a compile error which I believe is a compiler bug
+    var func = @ptrCast(FnType, objc.objc_msgSend);
+    const sel = objc.sel_getUid(sel_name);
+
+    return @call(.{}, func, .{ obj, sel } ++ args);
+}