benchmark: add screen-clone for benchmarking the Screen.clone method (#9624)

This adds a benchmark and some test coverage for a `screen-clone`
benchmark. This benchmarks the screen cloning which is a hot spot for
lock contention for the renderer + IO threads. I wasn't able to
meaningfully speed this up, but still want to commit this benchmark.
pull/9626/head
Mitchell Hashimoto 2025-11-17 08:51:27 -08:00 committed by GitHub
commit 3ff0cddee8
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 236 additions and 0 deletions

View File

@ -0,0 +1,155 @@
//! This benchmark tests the performance of the Screen.clone
//! function. This is useful because it is one of the primary lock
//! holders that impact IO performance when the renderer is active.
//! We do this very frequently.
const ScreenClone = @This();
const std = @import("std");
const assert = std.debug.assert;
const Allocator = std.mem.Allocator;
const terminalpkg = @import("../terminal/main.zig");
const Benchmark = @import("Benchmark.zig");
const options = @import("options.zig");
const Terminal = terminalpkg.Terminal;
const log = std.log.scoped(.@"terminal-stream-bench");
opts: Options,
terminal: Terminal,
pub const Options = struct {
/// The type of codepoint width calculation to use.
mode: Mode = .clone,
/// The size of the terminal. This affects benchmarking when
/// dealing with soft line wrapping and the memory impact
/// of page sizes.
@"terminal-rows": u16 = 80,
@"terminal-cols": u16 = 120,
/// The data to read as a filepath. If this is "-" then
/// we will read stdin. If this is unset, then we will
/// do nothing (benchmark is a noop). It'd be more unixy to
/// use stdin by default but I find that a hanging CLI command
/// with no interaction is a bit annoying.
///
/// This will be used to initialize the terminal screen state before
/// cloning. This data can switch to alt screen if it wants. The time
/// to read this is not part of the benchmark.
data: ?[]const u8 = null,
};
pub const Mode = enum {
/// The baseline mode copies the screen by value.
noop,
/// Full clone
clone,
};
pub fn create(
alloc: Allocator,
opts: Options,
) !*ScreenClone {
const ptr = try alloc.create(ScreenClone);
errdefer alloc.destroy(ptr);
ptr.* = .{
.opts = opts,
.terminal = try .init(alloc, .{
.rows = opts.@"terminal-rows",
.cols = opts.@"terminal-cols",
}),
};
return ptr;
}
pub fn destroy(self: *ScreenClone, alloc: Allocator) void {
self.terminal.deinit(alloc);
alloc.destroy(self);
}
pub fn benchmark(self: *ScreenClone) Benchmark {
return .init(self, .{
.stepFn = switch (self.opts.mode) {
.noop => stepNoop,
.clone => stepClone,
},
.setupFn = setup,
.teardownFn = teardown,
});
}
fn setup(ptr: *anyopaque) Benchmark.Error!void {
const self: *ScreenClone = @ptrCast(@alignCast(ptr));
// Always reset our terminal state
self.terminal.fullReset();
// Setup our terminal state
const data_f: std.fs.File = (options.dataFile(
self.opts.data,
) catch |err| {
log.warn("error opening data file err={}", .{err});
return error.BenchmarkFailed;
}) orelse return;
var stream = self.terminal.vtStream();
defer stream.deinit();
var read_buf: [4096]u8 = undefined;
var f_reader = data_f.reader(&read_buf);
const r = &f_reader.interface;
var buf: [4096]u8 = undefined;
while (true) {
const n = r.readSliceShort(&buf) catch {
log.warn("error reading data file err={?}", .{f_reader.err});
return error.BenchmarkFailed;
};
if (n == 0) break; // EOF reached
stream.nextSlice(buf[0..n]) catch |err| {
log.warn("error processing data file chunk err={}", .{err});
return error.BenchmarkFailed;
};
}
}
fn teardown(ptr: *anyopaque) void {
const self: *ScreenClone = @ptrCast(@alignCast(ptr));
_ = self;
}
fn stepNoop(ptr: *anyopaque) Benchmark.Error!void {
const self: *ScreenClone = @ptrCast(@alignCast(ptr));
// We loop because its so fast that a single benchmark run doesn't
// properly capture our speeds.
for (0..1000) |_| {
const s: terminalpkg.Screen = self.terminal.screens.active.*;
std.mem.doNotOptimizeAway(s);
}
}
fn stepClone(ptr: *anyopaque) Benchmark.Error!void {
const self: *ScreenClone = @ptrCast(@alignCast(ptr));
// We loop because its so fast that a single benchmark run doesn't
// properly capture our speeds.
for (0..1000) |_| {
const s: *terminalpkg.Screen = self.terminal.screens.active;
const copy = s.clone(
s.alloc,
.{ .viewport = .{} },
null,
) catch |err| {
log.warn("error cloning screen err={}", .{err});
return error.BenchmarkFailed;
};
std.mem.doNotOptimizeAway(copy);
// Note: we purposely do not free memory because we don't want
// to benchmark that. We'll free when the benchmark exits.
}
}

View File

@ -8,6 +8,7 @@ const cli = @import("../cli.zig");
pub const Action = enum {
@"codepoint-width",
@"grapheme-break",
@"screen-clone",
@"terminal-parser",
@"terminal-stream",
@"is-symbol",
@ -22,6 +23,7 @@ pub const Action = enum {
/// See TerminalStream for an example.
pub fn Struct(comptime action: Action) type {
return switch (action) {
.@"screen-clone" => @import("ScreenClone.zig"),
.@"terminal-stream" => @import("TerminalStream.zig"),
.@"codepoint-width" => @import("CodepointWidth.zig"),
.@"grapheme-break" => @import("GraphemeBreak.zig"),

View File

@ -4,6 +4,7 @@ pub const CApi = @import("CApi.zig");
pub const TerminalStream = @import("TerminalStream.zig");
pub const CodepointWidth = @import("CodepointWidth.zig");
pub const GraphemeBreak = @import("GraphemeBreak.zig");
pub const ScreenClone = @import("ScreenClone.zig");
pub const TerminalParser = @import("TerminalParser.zig");
pub const IsSymbol = @import("IsSymbol.zig");

View File

@ -2233,6 +2233,84 @@ test "Page clone" {
}
}
test "Page clone graphemes" {
var page = try Page.init(.{
.cols = 10,
.rows = 10,
.styles = 8,
});
defer page.deinit();
// Append some graphemes
{
const rac = page.getRowAndCell(0, 0);
rac.cell.* = .init(0x09);
try page.appendGrapheme(rac.row, rac.cell, 0x0A);
try page.appendGrapheme(rac.row, rac.cell, 0x0B);
}
// Clone it
var page2 = try page.clone();
defer page2.deinit();
{
const rac = page2.getRowAndCell(0, 0);
try testing.expect(rac.row.grapheme);
try testing.expect(rac.cell.hasGrapheme());
try testing.expectEqualSlices(u21, &.{ 0x0A, 0x0B }, page2.lookupGrapheme(rac.cell).?);
}
}
test "Page clone styles" {
var page = try Page.init(.{
.cols = 10,
.rows = 10,
.styles = 8,
});
defer page.deinit();
// Write with some styles
{
const id = try page.styles.add(page.memory, .{ .flags = .{
.bold = true,
} });
for (0..page.size.cols) |x| {
const rac = page.getRowAndCell(x, 0);
rac.row.styled = true;
rac.cell.* = .{
.content_tag = .codepoint,
.content = .{ .codepoint = @intCast(x + 1) },
.style_id = id,
};
page.styles.use(page.memory, id);
}
}
// Clone it
var page2 = try page.clone();
defer page2.deinit();
{
const id: u16 = style: {
const rac = page2.getRowAndCell(0, 0);
break :style rac.cell.style_id;
};
for (0..page.size.cols) |x| {
const rac = page.getRowAndCell(x, 0);
try testing.expect(rac.row.styled);
try testing.expectEqual(id, rac.cell.style_id);
}
const style = page.styles.get(
page.memory,
id,
);
try testing.expect((Style{ .flags = .{
.bold = true,
} }).eql(style.*));
}
}
test "Page cloneFrom" {
var page = try Page.init(.{
.cols = 10,