我如何优化这个 for 循环以在 PHP 上快速运行。在 javascript 上运行得非常快

Question

所以我一直在将一些代码从 Javascript 移植到现在可以运行的 PHP 中，但是，执行时间很短。在 PHP 中大约需要 26 秒，而在 javascript 中它非常快。我已经能够将其分解为导致问题的 for 循环。这是循环

for ($R0d = 0; $R0d < $F0d; $R0d = $R0d + 16) {

                $f0d = [bitwise_and(JS_charCodeAt($C0d,($R0d + 4)), 0xff) | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($R0d + 5)), 0xff)), 8) | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($R0d + 6)), 255)), 16) | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($R0d + 7)), 255)), 24), bitwise_and(JS_charCodeAt($C0d, ($R0d)), 255) | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($R0d + 1)), 255)), 8) | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($R0d + 2)), 255)), 16) | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($R0d + 3)), 0xff)), 24)];

                $f0d = x64Multiply($f0d, $M0d);

                $f0d = x64Rotl($f0d, 31);

                $f0d = x64Multiply($f0d, $I0d);

                $P0d = x64Xor($P0d, $f0d);

                $P0d = x64Rotl($P0d, 27);

                $P0d = x64Add($P0d, $o0d);

                $P0d = x64Add(x64Multiply($P0d, [0, 5]), [0, 1390208809]);

                $H0d = x64Multiply($H0d, $I0d);

                $H0d = x64Rotl($H0d, 33);

                $H0d = x64Multiply($H0d, $M0d);

                $o0d = x64Xor($o0d, $H0d);

                $o0d = x64Rotl($o0d, 31);

                $o0d = x64Add($o0d, $P0d);

                $o0d = x64Add(x64Multiply($o0d, [0, 5]), [0, 944331445]);       

            }

`

每个单独的函数调用都很好，我已经看到每次迭代大约需要 0.01 秒，而对于我在这段代码中使用的测试字符串，大约需要 24-27 秒。这次的循环运行了 2444 次。想象$F0d = 39104

无论如何，我是否可以像在使用 JavaScript 的浏览器中那样将执行时间缩短到不到一秒？

Answer 1

如果这是 MurmurHash，那么去看看https://github.com/lastguest/murmurhash-php。否则，您可以摆脱大量不必要的函数调用，如下所示：for ($i = 0; $i < $F0d; $i += 16) {&nbsp; &nbsp; // Let's start with this block. What does it do? It clearly takes a block of 8 bytes&nbsp; &nbsp; // four at a time, and builds a pair of 32-bit words.&nbsp; &nbsp; // Each byte is AND-ed bitwise with 0xFF to ensure it is a 8-bit value.&nbsp; &nbsp; $f0d = [&nbsp; &nbsp; &nbsp; &nbsp; bitwise_and(JS_charCodeAt($C0d, ($i + 4)), 0xff)&nbsp; &nbsp; &nbsp; &nbsp; | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($i + 5)), 0xff)), 8)&nbsp; &nbsp; &nbsp; &nbsp; | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($i + 6)), 255)), 16)&nbsp; &nbsp; &nbsp; &nbsp; | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($i + 7)), 255)), 24),&nbsp; &nbsp; &nbsp; &nbsp; bitwise_and(JS_charCodeAt($C0d, ($i)), 255)&nbsp; &nbsp; &nbsp; &nbsp; | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($i + 1)), 255)), 8)&nbsp; &nbsp; &nbsp; &nbsp; | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($i + 2)), 255)), 16)&nbsp; &nbsp; &nbsp; &nbsp; | shift_left_32((bitwise_and(JS_charCodeAt($C0d, ($i + 3)), 0xff)), 24),&nbsp; &nbsp; ];&nbsp; &nbsp; // We can do the same thing in PHP using the unpack() function and type V on an appropriate substr() of $C0d.&nbsp; &nbsp; // But due to the vagaries of unpack() and the fact that the elements are reversed,&nbsp; &nbsp; // we cannot use unpack alone (well, we COULD, if we changed the algorithm below, but it would be too awkward).&nbsp; &nbsp; $x0d = array_reverse(array_values(unpack('V2', substr($C0d, $i, 8))));&nbsp; &nbsp; // If you now dump $f0d and $x0d, they should be identical.&nbsp; &nbsp; // You can do the same with $H0d.&nbsp;&nbsp; &nbsp; // Actually you can do both together, avoiding the array manipulation, BUT I HAVEN'T CHECKED (just do a simple test run and verify $x0d's contents comparing it to $f0d and $H0d's):&nbsp; &nbsp; $x0d = unpack('V4', substr($C0d, $i, 16));&nbsp; &nbsp; $f0d = [ $x0d[2], $x0d[1] ];&nbsp; &nbsp; $H0d = [ $x0d[4], $x0d[3] ];&nbsp; &nbsp; // and get all 16 bytes neatly converted in two two-32-bit word arrays.

Answer 2

您的主要问题在于错误编写的 JS_charCodeAt 函数。通过增加输入大小，性能呈指数级下降，因为每次调用单个字符时，此函数都会从 UTF8 - UTF16 转换整个输入。添加这个简单的缓存将工作时间减少了 100 倍。只需替换此功能：function JS_charCodeAt($str, $index) {&nbsp; &nbsp; static $utf16s = [];&nbsp; &nbsp; if (isset($utf16s[$str]))&nbsp; &nbsp; &nbsp; &nbsp; $utf16 = $utf16s[$str];&nbsp; &nbsp; else&nbsp; &nbsp; &nbsp; &nbsp; $utf16s[$str] = $utf16 = mb_convert_encoding($str, 'UTF-16LE', 'UTF-8');&nbsp; &nbsp; return ord($utf16[$index*2]) + (ord($utf16[$index*2+1]) << 8);}当然，有很多事情是错误的，但这是非常容易和快速修复的，并且会产生最大的不同。到目前为止，如果您需要更高的性能，我可以进一步研究它。但该代码真的应该完全重写，这是一团糟。但现在它至少有效。