Skip to content

perf: SIMD scan ASCII runs in input loop #288

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
dyxushuai wants to merge 7 commits into
base: main
Choose a base branch
from
Open

perf: SIMD scan ASCII runs in input loop #288

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
caa7637
perf: SIMD scan ASCII runs in input loop
dyxushuai Dec 28, 2025
72ae497
refactor: share ASCII run scanner
dyxushuai Dec 28, 2025
6da8289
fix: refine ASCII run boundary handling
dyxushuai Dec 28, 2025
109f291
bench: isolate parser instances
dyxushuai Dec 28, 2025
4a06cd6
fix: include buffered bytes in read loop
dyxushuai Dec 28, 2025
c222cf5
test: cover ASCII boundary and incomplete UTF-8 cases
dyxushuai Dec 28, 2025
d5d295b
fix: treat variation selectors as combining
dyxushuai Dec 28, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
96 changes: 96 additions & 0 deletions bench/bench.zig
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,93 @@ fn printResults(writer: anytype, label: []const u8, iterations: usize, elapsed_n
);
}

fn asciiPrintableRunLen(input: []const u8) usize {
const VecLenOpt = std.simd.suggestVectorLength(u8);
if (VecLenOpt) |VecLen| {
const Vec = @Vector(VecLen, u8);
const lo: Vec = @splat(0x20);
const hi: Vec = @splat(0x7E);
var i: usize = 0;
while (i + VecLen <= input.len) : (i += VecLen) {
const chunk = @as(*const [VecLen]u8, @ptrCast(input[i..].ptr)).*;
const vec: Vec = chunk;
const ok = (vec >= lo) & (vec <= hi);
if (!@reduce(.And, ok)) {
var j: usize = 0;
while (j < VecLen) : (j += 1) {
const b = input[i + j];
if (b < 0x20 or b > 0x7E) return i + j;
}
}
}
while (i < input.len) : (i += 1) {
const b = input[i];
if (b < 0x20 or b > 0x7E) return i;
}
return input.len;
}

var i: usize = 0;
while (i < input.len) : (i += 1) {
const b = input[i];
if (b < 0x20 or b > 0x7E) return i;
}
return input.len;
}

fn benchParseStreamBaseline(writer: anytype, label: []const u8, parser: *vaxis.Parser, input: []const u8, iterations: usize) !void {
var timer = try std.time.Timer.start();
var i: usize = 0;
while (i < iterations) : (i += 1) {
var idx: usize = 0;
while (idx < input.len) {
const result = try parser.parse(input[idx..], null);
if (result.n == 0) break;
idx += result.n;
std.mem.doNotOptimizeAway(result);
}
std.mem.doNotOptimizeAway(idx);
}
const elapsed_ns = timer.read();
try printResults(writer, label, iterations, elapsed_ns, input.len * iterations);
}

fn benchParseStreamSimd(writer: anytype, label: []const u8, parser: *vaxis.Parser, input: []const u8, iterations: usize) !void {
var timer = try std.time.Timer.start();
var i: usize = 0;
while (i < iterations) : (i += 1) {
var idx: usize = 0;
while (idx < input.len) {
const slice = input[idx..];
var ascii_len = asciiPrintableRunLen(slice);
if (ascii_len > 0 and ascii_len < slice.len and slice[ascii_len] >= 0x80) {
ascii_len -= 1;
}
if (ascii_len > 0) {
var j: usize = 0;
while (j < ascii_len) : (j += 1) {
const key: vaxis.Key = .{
.codepoint = slice[j],
.text = slice[j .. j + 1],
};
const event: vaxis.Event = .{ .key_press = key };
std.mem.doNotOptimizeAway(event);
}
idx += ascii_len;
continue;
}

const result = try parser.parse(slice, null);
if (result.n == 0) break;
idx += result.n;
std.mem.doNotOptimizeAway(result);
}
std.mem.doNotOptimizeAway(idx);
}
const elapsed_ns = timer.read();
try printResults(writer, label, iterations, elapsed_ns, input.len * iterations);
}

pub fn main() !void {
var gpa = std.heap.GeneralPurposeAllocator(.{}){};
defer _ = gpa.deinit();
Expand Down Expand Up @@ -59,4 +146,13 @@ pub fn main() !void {
const dirty_ns = timer.read();
const dirty_bytes: usize = dirty_writer.writer.end;
try printResults(stdout, "dirty", iterations, dirty_ns, dirty_bytes);

var parser: vaxis.Parser = .{};
const mixed_stream = "The quick brown fox jumps over the lazy dog " ++
"1234567890 !@#$%^&*() " ++
"\x1b[A" ++
"世界 1️⃣ 👩‍🚀!" ++
"\r";
try benchParseStreamBaseline(stdout, "parse_stream_loop_baseline", &parser, mixed_stream, iterations);
try benchParseStreamSimd(stdout, "parse_stream_loop_simd", &parser, mixed_stream, iterations);
}
55 changes: 55 additions & 0 deletions src/Loop.zig
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,40 @@ const Vaxis = @import("Vaxis.zig");

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

fn asciiPrintableRunLen(input: []const u8) usize {
const VecLenOpt = std.simd.suggestVectorLength(u8);
if (VecLenOpt) |VecLen| {
const Vec = @Vector(VecLen, u8);
const lo: Vec = @splat(0x20);
const hi: Vec = @splat(0x7E);
var i: usize = 0;
while (i + VecLen <= input.len) : (i += VecLen) {
const chunk = @as(*const [VecLen]u8, @ptrCast(input[i..].ptr)).*;
const vec: Vec = chunk;
const ok = (vec >= lo) & (vec <= hi);
if (!@reduce(.And, ok)) {
var j: usize = 0;
while (j < VecLen) : (j += 1) {
const b = input[i + j];
if (b < 0x20 or b > 0x7E) return i + j;
}
}
}
while (i < input.len) : (i += 1) {
const b = input[i];
if (b < 0x20 or b > 0x7E) return i;
}
return input.len;
}

var i: usize = 0;
while (i < input.len) : (i += 1) {
const b = input[i];
if (b < 0x20 or b > 0x7E) return i;
}
return input.len;
}

pub fn Loop(comptime T: type) type {
return struct {
const Self = @This();
Expand Down Expand Up @@ -137,6 +171,27 @@ pub fn Loop(comptime T: type) type {
const n = try self.tty.read(buf[read_start..]);
var seq_start: usize = 0;
while (seq_start < n) {
if (@hasField(Event, "key_press")) {
const input = buf[seq_start..n];
var ascii_len = asciiPrintableRunLen(input);
if (ascii_len > 0 and ascii_len < input.len and input[ascii_len] >= 0x80) {
ascii_len -= 1;
}
if (ascii_len > 0) {
var i: usize = 0;
while (i < ascii_len) : (i += 1) {
const key: vaxis.Key = .{
.codepoint = input[i],
.text = input[i .. i + 1],
};
const event: Event = .{ .key_press = key };
try handleEventGeneric(self, self.vaxis, &cache, Event, event, paste_allocator);
}
read_start = 0;
seq_start += ascii_len;
continue;
}
}
const result = try parser.parse(buf[seq_start..n], paste_allocator);
if (result.n == 0) {
// copy the read to the beginning. We don't use memcpy because
Expand Down