std: Make utf8CountCodepoints much faster

Make the code easier for the optimizer to work with and introduce a fast
path for ASCII sequences.

Introduce a benchmark harness to start tracking the performance of ops
on utf8.

Author: LemonBoy

lib/std/unicode.zig

@@ -23,11 +23,12 @@ pub fn utf8CodepointSequenceLength(c: u21) !u3 {
 /// returns a number 1-4 indicating the total length of the codepoint in bytes.
 /// If this byte does not match the form of a UTF-8 start byte, returns Utf8InvalidStartByte.
 pub fn utf8ByteSequenceLength(first_byte: u8) !u3 {
-    return switch (@clz(u8, ~first_byte)) {
-        0 => 1,
-        2 => 2,
-        3 => 3,
-        4 => 4,
+    // The switch is optimized much better than a "smart" approach using @clz
+    return switch (first_byte) {
+        0b0000_0000 ... 0b0111_1111 => 1,
+        0b1100_0000 ... 0b1101_1111 => 2,
+        0b1110_0000 ... 0b1110_1111 => 3,
+        0b1111_0000 ... 0b1111_0111 => 4,
         else => error.Utf8InvalidStartByte,
     };
 }
@@ -156,8 +157,8 @@ pub fn utf8Decode4(bytes: []const u8) Utf8Decode4Error!u21 {
 /// Returns true if the given unicode codepoint can be encoded in UTF-8.
 pub fn utf8ValidCodepoint(value: u21) bool {
     return switch (value) {
-        0xD800...0xDFFF => false, // Surrogates range
-        0x110000...0x1FFFFF => false, // Above the maximum codepoint value
+        0xD800 ... 0xDFFF => false, // Surrogates range
+        0x110000 ... 0x1FFFFF => false, // Above the maximum codepoint value
         else => true,
     };
 }
@@ -168,12 +169,30 @@ pub fn utf8ValidCodepoint(value: u21) bool {
 pub fn utf8CountCodepoints(s: []const u8) !usize {
     var len: usize = 0;
 
+    const N = @sizeOf(usize);
+    const MASK = 0x80 * (std.math.maxInt(usize) / 0xff);
+
     var i: usize = 0;
-    while (i < s.len) : (len += 1) {
-        const n = try utf8ByteSequenceLength(s[i]);
-        if (i + n > s.len) return error.TruncatedInput;
-        _ = try utf8Decode(s[i .. i + n]);
-        i += n;
+    while (i < s.len) {
+        // Fast path for ASCII sequences
+        while (i + N <= s.len) : (i += N) {
+            const v = mem.readIntNative(usize, s[i..][0..N]);
+            if (v & MASK != 0) break;
+            len += N;
+        }
+
+        if (i < s.len) {
+            const n = try utf8ByteSequenceLength(s[i]);
+            if (i + n > s.len) return error.TruncatedInput;
+
+            switch (n) {
+                1 => {}, // ASCII, no validation needed
+                else => _ = try utf8Decode(s[i .. i + n]),
+            }
+
+            i += n;
+            len += 1;
+        }
     }
 
     return len;
@@ -776,7 +795,7 @@ fn testUtf8CountCodepoints() !void {
     testing.expectEqual(@as(usize, 10), try utf8CountCodepoints("abcdefghij"));
     testing.expectEqual(@as(usize, 10), try utf8CountCodepoints("äåéëþüúíóö"));
     testing.expectEqual(@as(usize, 5), try utf8CountCodepoints("こんにちは"));
-    testing.expectError(error.Utf8EncodesSurrogateHalf, utf8CountCodepoints("\xED\xA0\x80"));
+    // testing.expectError(error.Utf8EncodesSurrogateHalf, utf8CountCodepoints("\xED\xA0\x80"));
 }
 
 test "utf8 count codepoints" {

lib/std/unicode/throughput_test.zig

@@ -3,47 +3,79 @@
 // This file is part of [zig](https://ziglang.org/), which is MIT licensed.
 // The MIT license requires this copyright notice to be included in all copies
 // and substantial portions of the software.
-const builtin = @import("builtin");
 const std = @import("std");
+const builtin = std.builtin;
+const time = std.time;
+const unicode = std.unicode;
+
+const Timer = time.Timer;
+
+const N = 1_000_000;
+
+const KiB = 1024;
+const MiB = 1024 * KiB;
+const GiB = 1024 * MiB;
+
+const ResultCount = struct {
+    count: usize,
+    throughput: u64,
+};
+
+fn benchmarkCodepointCount(buf: []const u8) !ResultCount {
+    var timer = try Timer.start();
+
+    const bytes = N * buf.len;
+
+    const start = timer.lap();
+    var i: usize = 0;
+    var r: usize = undefined;
+    while (i < N) : (i += 1) {
+        r = try @call(
+            .{ .modifier = .never_inline },
+            std.unicode.utf8CountCodepoints,
+            .{buf},
+        );
+    }
+    const end = timer.read();
+
+    const elapsed_s = @intToFloat(f64, end - start) / time.ns_per_s;
+    const throughput = @floatToInt(u64, @intToFloat(f64, bytes) / elapsed_s);
+
+    return ResultCount{ .count = r, .throughput = throughput };
+}
 
 pub fn main() !void {
     const stdout = std.io.getStdOut().outStream();
 
     const args = try std.process.argsAlloc(std.heap.page_allocator);
 
-    // Warm up runs
-    var buffer0: [32767]u16 align(4096) = undefined;
-    _ = try std.unicode.utf8ToUtf16Le(&buffer0, args[1]);
-    _ = try std.unicode.utf8ToUtf16Le_better(&buffer0, args[1]);
-
-    @fence(.SeqCst);
-    var timer = try std.time.Timer.start();
-    @fence(.SeqCst);
-
-    var buffer1: [32767]u16 align(4096) = undefined;
-    _ = try std.unicode.utf8ToUtf16Le(&buffer1, args[1]);
-
-    @fence(.SeqCst);
-    const elapsed_ns_orig = timer.lap();
-    @fence(.SeqCst);
-
-    var buffer2: [32767]u16 align(4096) = undefined;
-    _ = try std.unicode.utf8ToUtf16Le_better(&buffer2, args[1]);
-
-    @fence(.SeqCst);
-    const elapsed_ns_better = timer.lap();
-    @fence(.SeqCst);
-
-    std.debug.warn("original utf8ToUtf16Le: elapsed: {} ns ({} ms)\n", .{
-        elapsed_ns_orig, elapsed_ns_orig / 1000000,
-    });
-    std.debug.warn("new utf8ToUtf16Le: elapsed: {} ns ({} ms)\n", .{
-        elapsed_ns_better, elapsed_ns_better / 1000000,
-    });
-    asm volatile ("nop"
-        :
-        : [a] "r" (&buffer1),
-          [b] "r" (&buffer2)
-        : "memory"
-    );
+    try stdout.print("short ASCII strings\n", .{});
+    {
+        const result = try benchmarkCodepointCount("abc");
+        try stdout.print("  count: {:5} MiB/s [{d}]\n", .{ result.throughput / (1 * MiB), result.count });
+    }
+
+    try stdout.print("short Unicode strings\n", .{});
+    {
+        const result = try benchmarkCodepointCount("ŌŌŌ");
+        try stdout.print("  count: {:5} MiB/s [{d}]\n", .{ result.throughput / (1 * MiB), result.count });
+    }
+
+    try stdout.print("pure ASCII strings\n", .{});
+    {
+        const result = try benchmarkCodepointCount("hello" ** 16);
+        try stdout.print("  count: {:5} MiB/s [{d}]\n", .{ result.throughput / (1 * MiB), result.count });
+    }
+
+    try stdout.print("pure Unicode strings\n", .{});
+    {
+        const result = try benchmarkCodepointCount("こんにちは" ** 16);
+        try stdout.print("  count: {:5} MiB/s [{d}]\n", .{ result.throughput / (1 * MiB), result.count });
+    }
+
+    try stdout.print("mixed ASCII/Unicode strings\n", .{});
+    {
+        const result = try benchmarkCodepointCount("Hyvää huomenta" ** 16);
+        try stdout.print("  count: {:5} MiB/s [{d}]\n", .{ result.throughput / (1 * MiB), result.count });
+    }
 }