diff options
| -rw-r--r-- | src/string_immutable.zig | 80 |
1 files changed, 37 insertions, 43 deletions
diff --git a/src/string_immutable.zig b/src/string_immutable.zig index 98ec70646..dd81247a9 100644 --- a/src/string_immutable.zig +++ b/src/string_immutable.zig @@ -715,45 +715,33 @@ pub fn toUTF8Alloc(allocator: std.mem.Allocator, js: []const u16) !string { } pub inline fn appendUTF8MachineWordToUTF16MachineWord(output: *[@sizeOf(usize) / 2]u16, input: *const [@sizeOf(usize) / 2]u8) void { - comptime var i: usize = 0; - inline while (i < @sizeOf(usize) / 2) : (i += 1) { - output[i] = input[i]; - } + output[0 .. @sizeOf(usize) / 2].* = @bitCast( + [4]u16, + @as( + @Vector(4, u16), + @bitCast(@Vector(4, u8), input[0 .. @sizeOf(usize) / 2].*), + ), + ); } pub inline fn copyU8IntoU16(output_: []u16, input_: []const u8) void { var output = output_; var input = input_; - const word = @sizeOf(usize) / 2; if (comptime Environment.allow_assert) { std.debug.assert(input.len <= output.len); } // https://zig.godbolt.org/z/9rTn1orcY - const group = comptime if (Environment.isAarch64) - // on ARM64, 128 seems to be the best choice judging by lines of ASM - 128 - else - // on x64, 128 seems to be the best choice judging by lines of ASM - 16; - - while (input.len >= group) { - const input_vec: @Vector(group, u8) = input[0..group].*; - const output_vec: @Vector(group, u16) = @as(@Vector(group, u16), input_vec); - output[0..group].* = output_vec; - output = output[group..]; - input = input[group..]; - } + var input_ptr = input.ptr; + var output_ptr = output.ptr; - while (input.len >= word) { - appendUTF8MachineWordToUTF16MachineWord(output[0..word], input[0..word]); - output = output[word..]; - input = input[word..]; - } + const last_input_ptr = input_ptr + @minimum(input.len, output.len); - for (input) |c, i| { - output[i] = c; + while (last_input_ptr != input_ptr) { + output_ptr[0] = input_ptr[0]; + output_ptr += 1; + input_ptr += 1; } } @@ -814,29 +802,30 @@ pub inline fn copyU16IntoU8(output_: []u8, comptime InputType: type, input_: Inp if (comptime Environment.allow_assert) { std.debug.assert(input_.len <= output_.len); } - - if (comptime !JSC.is_bindgen) { + if (comptime !JSC.is_bindgen and Environment.isAarch64) { + // faster on aarch64 + // but it only uses SSE2 when it could use AVX2 + // so it's better to let llvm auto-vectorize it JSC.WTF.copyLCharsFromUCharSource(output_.ptr, InputType, input_); } else { var output = output_; var input = input_; + if (comptime Environment.allow_assert) { + std.debug.assert(input.len <= output.len); + } - // on X64, this is 4 - // on WASM, this is 2 - const machine_word_length = comptime @sizeOf(usize) / @sizeOf(u16); + // https://zig.godbolt.org/z/Y1qa9PTo1 + // https://github.com/ziglang/zig/issues/11830 + // this auto-vectorizes on x64 and aarch64 + var input_ptr = input.ptr; + var output_ptr = output.ptr; - while (input.len >= machine_word_length) { - comptime var machine_word_i: usize = 0; - inline while (machine_word_i < machine_word_length) : (machine_word_i += 1) { - output[machine_word_i] = @intCast(u8, input[machine_word_i]); - } + const last_input_ptr = input_ptr + @minimum(input.len, output.len); - output = output[machine_word_length..]; - input = input[machine_word_length..]; - } - - for (input) |c, i| { - output[i] = @intCast(u8, c); + while (last_input_ptr != input_ptr) { + output_ptr[0] = @truncate(u8, input_ptr[0]); + output_ptr += 1; + input_ptr += 1; } } } @@ -1281,7 +1270,12 @@ pub fn copyLatin1IntoUTF16(comptime Buffer: type, buf_: Buffer, comptime Type: t var latin1 = latin1_; while (buf.len > 0 and latin1.len > 0) { const to_write = strings.firstNonASCII(latin1) orelse @truncate(u32, latin1.len); - strings.copyU8IntoU16WithAlignment(std.meta.alignment(Buffer), buf, latin1[0..to_write]); + if (comptime std.meta.alignment(Buffer) != @alignOf(u16)) { + strings.copyU8IntoU16WithAlignment(std.meta.alignment(Buffer), buf, latin1[0..to_write]); + } else { + strings.copyU8IntoU16(buf, latin1[0..to_write]); + } + latin1 = latin1[to_write..]; buf = buf[to_write..]; if (latin1.len > 0 and buf.len >= 2) { |