ghostty/src/renderer/Metal.zig

//! Graphics API wrapper for Metal.
pub const Metal = @This();

const std = @import("std");
const assert = std.debug.assert;
const Allocator = std.mem.Allocator;
const builtin = @import("builtin");
const objc = @import("objc");
const macos = @import("macos");
const graphics = macos.graphics;
const apprt = @import("../apprt.zig");
const font = @import("../font/main.zig");
const configpkg = @import("../config.zig");
const rendererpkg = @import("../renderer.zig");
const Renderer = rendererpkg.GenericRenderer(Metal);
const shadertoy = @import("shadertoy.zig");

const mtl = @import("metal/api.zig");
const IOSurfaceLayer = @import("metal/IOSurfaceLayer.zig");

pub const GraphicsAPI = Metal;
pub const Target = @import("metal/Target.zig");
pub const Frame = @import("metal/Frame.zig");
pub const RenderPass = @import("metal/RenderPass.zig");
pub const Pipeline = @import("metal/Pipeline.zig");
const bufferpkg = @import("metal/buffer.zig");
pub const Buffer = bufferpkg.Buffer;
pub const Texture = @import("metal/Texture.zig");
pub const shaders = @import("metal/shaders.zig");

pub const custom_shader_target: shadertoy.Target = .msl;
// The fragCoord for Metal shaders is +Y = down.
pub const custom_shader_y_is_down = true;

/// Triple buffering.
pub const swap_chain_count = 3;

const log = std.log.scoped(.metal);

layer: IOSurfaceLayer,

/// MTLDevice
device: objc.Object,
/// MTLCommandQueue
queue: objc.Object,

/// Alpha blending mode
blending: configpkg.Config.AlphaBlending,

/// The default storage mode to use for resources created with our device.
///
/// This is based on whether the device is a discrete GPU or not, since
/// discrete GPUs do not have unified memory and therefore do not support
/// the "shared" storage mode, instead we have to use the "managed" mode.
default_storage_mode: mtl.MTLResourceOptions.StorageMode,

/// We start an AutoreleasePool before `drawFrame` and end it afterwards.
autorelease_pool: ?*objc.AutoreleasePool = null,

pub fn init(alloc: Allocator, opts: rendererpkg.Options) !Metal {
    comptime switch (builtin.os.tag) {
        .macos, .ios => {},
        else => @compileError("unsupported platform for Metal"),
    };

    _ = alloc;

    // Choose our MTLDevice and create a MTLCommandQueue for that device.
    const device = try chooseDevice();
    errdefer device.release();
    const queue = device.msgSend(objc.Object, objc.sel("newCommandQueue"), .{});
    errdefer queue.release();

    const default_storage_mode: mtl.MTLResourceOptions.StorageMode =
        if (device.getProperty(bool, "hasUnifiedMemory")) .shared else .managed;

    const ViewInfo = struct {
        view: objc.Object,
        scaleFactor: f64,
    };

    // Get the metadata about our underlying view that we'll be rendering to.
    const info: ViewInfo = switch (apprt.runtime) {
        apprt.embedded => .{
            .scaleFactor = @floatCast(opts.rt_surface.content_scale.x),
            .view = switch (opts.rt_surface.platform) {
                .macos => |v| v.nsview,
                .ios => |v| v.uiview,
            },
        },

        else => @compileError("unsupported apprt for metal"),
    };

    // Create an IOSurfaceLayer which we can assign to the view to make
    // it in to a "layer-hosting view", so that we can manually control
    // the layer contents.
    var layer = try IOSurfaceLayer.init();
    errdefer layer.release();

    // Add our layer to the view.
    //
    // On macOS we do this by making the view "layer-hosting"
    // by assigning it to the view's `layer` property BEFORE
    // setting `wantsLayer` to `true`.
    //
    // On iOS, views are always layer-backed, and `layer`
    // is readonly, so instead we add it as a sublayer.
    switch (comptime builtin.os.tag) {
        .macos => {
            info.view.setProperty("layer", layer.layer.value);
            info.view.setProperty("wantsLayer", true);
        },

        .ios => {
            info.view.msgSend(void, objc.sel("addSublayer"), .{layer.layer.value});
        },

        else => @compileError("unsupported target for Metal"),
    }

    // Ensure that if our layer is oversized it
    // does not overflow the bounds of the view.
    info.view.setProperty("clipsToBounds", true);

    // Ensure that our layer has a content scale set to
    // match the scale factor of the window. This avoids
    // magnification issues leading to blurry rendering.
    layer.layer.setProperty("contentsScale", info.scaleFactor);

    // This makes it so that our display callback will actually be called.
    layer.layer.setProperty("needsDisplayOnBoundsChange", true);

    return .{
        .layer = layer,
        .device = device,
        .queue = queue,
        .blending = opts.config.blending,
        .default_storage_mode = default_storage_mode,
    };
}

pub fn deinit(self: *Metal) void {
    self.queue.release();
    self.device.release();
    self.layer.release();
}

pub fn loopEnter(self: *Metal) void {
    const renderer: *align(1) Renderer = @fieldParentPtr("api", self);
    self.layer.setDisplayCallback(
        @ptrCast(&displayCallback),
        @ptrCast(renderer),
    );
}

fn displayCallback(renderer: *Renderer) align(8) void {
    renderer.drawFrame(true) catch |err| {
        log.warn("Error drawing frame in display callback, err={}", .{err});
    };
}

/// Actions taken before doing anything in `drawFrame`.
///
/// Right now we use this to start an AutoreleasePool.
pub fn drawFrameStart(self: *Metal) void {
    assert(self.autorelease_pool == null);
    self.autorelease_pool = .init();
}

/// Actions taken after `drawFrame` is done.
///
/// Right now we use this to end our AutoreleasePool.
pub fn drawFrameEnd(self: *Metal) void {
    assert(self.autorelease_pool != null);
    self.autorelease_pool.?.deinit();
    self.autorelease_pool = null;
}

pub fn initShaders(
    self: *const Metal,
    alloc: Allocator,
    custom_shaders: []const [:0]const u8,
) !shaders.Shaders {
    return try shaders.Shaders.init(
        alloc,
        self.device,
        custom_shaders,
        // Using an `*_srgb` pixel format makes Metal gamma encode
        // the pixels written to it *after* blending, which means
        // we get linear alpha blending rather than gamma-incorrect
        // blending.
        if (self.blending.isLinear())
            mtl.MTLPixelFormat.bgra8unorm_srgb
        else
            mtl.MTLPixelFormat.bgra8unorm,
    );
}

/// Get the current size of the runtime surface.
pub fn surfaceSize(self: *const Metal) !struct { width: u32, height: u32 } {
    const bounds = self.layer.layer.getProperty(graphics.Rect, "bounds");
    const scale = self.layer.layer.getProperty(f64, "contentsScale");
    return .{
        .width = @intFromFloat(bounds.size.width * scale),
        .height = @intFromFloat(bounds.size.height * scale),
    };
}

/// Initialize a new render target which can be presented by this API.
pub fn initTarget(self: *const Metal, width: usize, height: usize) !Target {
    return Target.init(.{
        .device = self.device,
        // Using an `*_srgb` pixel format makes Metal gamma encode the pixels
        // written to it *after* blending, which means we get linear alpha
        // blending rather than gamma-incorrect blending.
        .pixel_format = if (self.blending.isLinear())
            .bgra8unorm_srgb
        else
            .bgra8unorm,
        .storage_mode = self.default_storage_mode,
        .width = width,
        .height = height,
    });
}

/// Present the provided target.
pub inline fn present(self: *Metal, target: Target, sync: bool) !void {
    if (sync) {
        self.layer.setSurfaceSync(target.surface);
    } else {
        try self.layer.setSurface(target.surface);
    }
}

/// Present the last presented target again. (noop for Metal)
pub inline fn presentLastTarget(self: *Metal) !void {
    _ = self;
}

/// Returns the options to use when constructing buffers.
pub inline fn bufferOptions(self: Metal) bufferpkg.Options {
    return .{
        .device = self.device,
        .resource_options = .{
            // Indicate that the CPU writes to this resource but never reads it.
            .cpu_cache_mode = .write_combined,
            .storage_mode = self.default_storage_mode,
        },
    };
}

pub const instanceBufferOptions = bufferOptions;
pub const uniformBufferOptions = bufferOptions;
pub const fgBufferOptions = bufferOptions;
pub const bgBufferOptions = bufferOptions;
pub const imageBufferOptions = bufferOptions;
pub const bgImageBufferOptions = bufferOptions;

/// Returns the options to use when constructing textures.
pub inline fn textureOptions(self: Metal) Texture.Options {
    return .{
        .device = self.device,
        // Using an `*_srgb` pixel format makes Metal gamma encode the pixels
        // written to it *after* blending, which means we get linear alpha
        // blending rather than gamma-incorrect blending.
        .pixel_format = if (self.blending.isLinear())
            .bgra8unorm_srgb
        else
            .bgra8unorm,
        .resource_options = .{
            // Indicate that the CPU writes to this resource but never reads it.
            .cpu_cache_mode = .write_combined,
            .storage_mode = self.default_storage_mode,
        },
    };
}

/// Pixel format for image texture options.
pub const ImageTextureFormat = enum {
    /// 1 byte per pixel grayscale.
    gray,
    /// 4 bytes per pixel RGBA.
    rgba,
    /// 4 bytes per pixel BGRA.
    bgra,

    fn toPixelFormat(
        self: ImageTextureFormat,
        srgb: bool,
    ) mtl.MTLPixelFormat {
        return switch (self) {
            .gray => if (srgb) .r8unorm_srgb else .r8unorm,
            .rgba => if (srgb) .rgba8unorm_srgb else .rgba8unorm,
            .bgra => if (srgb) .bgra8unorm_srgb else .bgra8unorm,
        };
    }
};

/// Returns the options to use when constructing textures for images.
pub inline fn imageTextureOptions(
    self: Metal,
    format: ImageTextureFormat,
    srgb: bool,
) Texture.Options {
    return .{
        .device = self.device,
        .pixel_format = format.toPixelFormat(srgb),
        .resource_options = .{
            // Indicate that the CPU writes to this resource but never reads it.
            .cpu_cache_mode = .write_combined,
            .storage_mode = self.default_storage_mode,
        },
    };
}

/// Initializes a Texture suitable for the provided font atlas.
pub fn initAtlasTexture(
    self: *const Metal,
    atlas: *const font.Atlas,
) Texture.Error!Texture {
    const pixel_format: mtl.MTLPixelFormat = switch (atlas.format) {
        .grayscale => .r8unorm,
        .bgra => .bgra8unorm_srgb,
        else => @panic("unsupported atlas format for Metal texture"),
    };

    return try Texture.init(
        .{
            .device = self.device,
            .pixel_format = pixel_format,
            .resource_options = .{
                // Indicate that the CPU writes to this resource but never reads it.
                .cpu_cache_mode = .write_combined,
                .storage_mode = self.default_storage_mode,
            },
        },
        atlas.size,
        atlas.size,
        null,
    );
}

/// Begin a frame.
pub inline fn beginFrame(
    self: *const Metal,
    /// Once the frame has been completed, the `frameCompleted` method
    /// on the renderer is called with the health status of the frame.
    renderer: *Renderer,
    /// The target is presented via the provided renderer's API when completed.
    target: *Target,
) !Frame {
    return try Frame.begin(.{ .queue = self.queue }, renderer, target);
}

fn chooseDevice() error{NoMetalDevice}!objc.Object {
    var chosen_device: ?objc.Object = null;

    switch (comptime builtin.os.tag) {
        .macos => {
            const devices = objc.Object.fromId(mtl.MTLCopyAllDevices());
            defer devices.release();

            var iter = devices.iterate();
            while (iter.next()) |device| {
                // We want a GPU that’s connected to a display.
                if (device.getProperty(bool, "isHeadless")) continue;
                chosen_device = device;
                // If the user has an eGPU plugged in, they probably want
                // to use it. Otherwise, integrated GPUs are better for
                // battery life and thermals.
                if (device.getProperty(bool, "isRemovable") or
                    device.getProperty(bool, "isLowPower")) break;
            }
        },
        .ios => {
            chosen_device = objc.Object.fromId(mtl.MTLCreateSystemDefaultDevice());
        },
        else => @compileError("unsupported target for Metal"),
    }

    const device = chosen_device orelse return error.NoMetalDevice;
    return device.retain();
}