Call cjieba using PHP's FFI

Recommended: "PHP Video Tutorial"

phpjieba_ffi

Use the FFI test of PHP 7.4 to directly call cjieba Word segmentation dynamic library

The reason for choosing CJieba is that FFI uses the C calling convention. If you use Cpp, you have to package it yourself, and then extern C to let the compiler generate a standard C dynamic library.

Problems encountered

Segmentation fault

C variable is not initialized

C is called directly The function is not called by the C object initialized by FFI

Non-null judgment needs to use FFI::isNull($x)

Arrays in the form of pointers cannot use foreach

Looping of arrays in pointer form

Looking at the C code, we found that the Cut part is as follows:

CJiebaWord* Cut(Jieba handle, const char* sentence, size_t len) {
  cppjieba::Jieba* x = (cppjieba::Jieba*)handle;
  vector words;
  string s(sentence, len);
  x->Cut(s, words);
  
  CJiebaWord* res = (CJiebaWord*)malloc(sizeof(CJiebaWord) * (words.size() + 1));
  size_t offset = 0;
  for (size_t i = 0; i < words.size(); i++) {
    res[i].word = sentence + offset;
    res[i].len = words[i].size();
    offset += res[i].len;
  }
  if (offset != len) {
    free(res);
    return NULL;
  }
  res[words.size()].word = NULL;
  res[words.size()].len = 0;
  return res;
}

What is returned is a structure pointer. In C language, the array name is actually The pointer address of the first variable in the array, so it can be traversed through the operation of the pointer address. What about in FFI?

For this array, I used a foreach loop at first and directly reported a segmentation error. Later, I used pointers directly like C and found that it was feasible. I give FFI a thumbs up because it can also directly operate C pointers. .

Get the word segmentation result

As shown in the above code, for a single word segmentation CJiebaWord, it is not the saved word segmentation, but the sentence offset, which means that the result of the first word segmentation is certain is a raw string.

In the C demo, it is printf formatting (. indicates field width and alignment), but there is no similar method in PHP. You need to intercept the string substr($x->word, 0, $x- >len)

  for (x = words; x->word; x++) {
    printf("%*.*s\n", x->len, x->len, x->word);
  }

Usage example

Compile dynamic library

make libjieba.so

Run

time php demo.php

Run c demo

make demo
time ./demo

Result

PHP
load: 0.00025701522827148
real    1m59.619s
user    1m56.093s
sys     0m3.517s
C
real    1m54.738s
user    1m50.382s
sys     0m4.323s
CPU 占用 基本都是 12%

It can be found that using FFI, the speed of PHP is basically the same as C. If you have a business that takes up a lot of CPU, you can try to use other languages ​​​​(C/C, golang, Rust, etc.) writes and then exports a standard C dynamic library.

Use of FFI

Before there was FFI, where system calls or sdk calls were required, PHP needed to develop extensions, but developing extensions not only required understanding of the C language , you still have to understand the PHP kernel, which is more difficult. Now it is much more convenient, you can directly use FFI to call the dynamic library.

Extended macro expansion

For example, there are a large number of macros in Hikvision's sdk gcc -E -P HCNetSDK.h -o HCNetSDK_unfold.h supports type define, feel free to use it

Original address: https://github.com/dwdcth/phpjieba_ffi

The above is the detailed content of Call cjieba using PHP's FFI. For more information, please follow other related articles on the PHP Chinese website!