Re: How to add pseudo vector types

[Yuan Fu <casouri@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1080 bytes --]


> On Jul 17, 2021, at 1:30 PM, Stefan Monnier <monnier@iro.umontreal.ca> wrote:
> 
> In your benchmark , you give numbers for:
> - initial full-text parse (a bit above 1MB/s)
> - cost of update-without-reparse
> 
> but I think it would be nice to see the cost of the reparse after
> those updates (should be much faster than the initial parse).

I have done some more benchmark. Initially I thought tree-sitter doesn’t scale, because re-parsing my JSON file is unexpectedly slow, but then I retired with xdisp.c with tree-sitter's C parser, and that is really fast and matches my expectation of tree-sitter. So from now on I’ll use xdispf.c and the C parser for benchmarking. I guess the json parser is simply bad-written?

I benchmarked with a simple C program. The programs are in main-c.c and main-json.c, and the shell output of the measurements is in benchmark.3.txt.

JSON: Initial parse takes 1.2s, re-parse (with no change) takes 0.7s, uses 307MB memory
C: Initial parse takes 0.14s, re-parse (with no change) takes 0.009s, uses 20MB memory

Yuan


[-- Attachment #2: benchmark.3.txt --]
[-- Type: text/plain, Size: 2875 bytes --]

On benchmark.2.json (1.6M)

One full parse: 1.2s
________________________________________________________
Executed in    1.30 secs   fish           external
   usr time  1210.81 millis  142.00 micros  1210.67 millis
   sys time   87.40 millis  756.00 micros   86.65 millis


One full parse and a re-parse:
________________________________________________________
Executed in    2.40 secs   fish           external
   usr time    1.95 secs  154.00 micros    1.95 secs
   sys time    0.15 secs  763.00 micros    0.15 secs

Re-parse takes 1.95 - 1.21 = 0.74s

Memory usage of full-parse + re-parse:

        2.17 real         2.00 user         0.16 sys
           307269632  maximum resident set size
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
               75035  page reclaims
                   0  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                   0  signals received
                   0  voluntary context switches
                 463  involuntary context switches
         14674957821  instructions retired
          7838514409  cycles elapsed
           306745344  peak memory footprint

307MB for two trees that "shares internal structure".

\f

On xdisp.c (1.2M)

One full paese: 0.139s
________________________________________________________
Executed in  478.23 millis    fish           external
   usr time  139.69 millis  134.00 micros  139.55 millis
   sys time    8.05 millis  829.00 micros    7.22 millis

Full parse and re-parse:
________________________________________________________
Executed in  456.58 millis    fish           external
   usr time  148.23 millis  153.00 micros  148.08 millis
   sys time    9.08 millis  791.00 micros    8.29 millis

148 - 139 = 0.009s

Memory usage of full-parse + re-parse:

        0.16 real         0.15 user         0.00 sys
            20131840  maximum resident set size
                   0  average shared memory size
                   0  average unshared data size
                   0  average unshared stack size
                4932  page reclaims
                   0  page faults
                   0  swaps
                   0  block input operations
                   0  block output operations
                   0  messages sent
                   0  messages received
                   0  signals received
                   0  voluntary context switches
                  28  involuntary context switches
          1070525817  instructions retired
           581557699  cycles elapsed
            19271680  peak memory footprint

20MB

[-- Attachment #3: main-c.c --]
[-- Type: application/octet-stream, Size: 1180 bytes --]

#include <string.h>
#include <stdio.h>
#include <tree_sitter/api.h>

TSLanguage *tree_sitter_c();

struct buffer {
  char *buf;
  long len;
};

const char *read_file(void *payload, uint32_t byte_index,
                TSPoint position, uint32_t *bytes_read) {
  long len = ((struct buffer *) payload)->len;
  if (byte_index >= len) {
    *bytes_read = 0;
    return (char *) "";
  } else {
    *bytes_read = len - byte_index;
    return (char *) (((struct buffer *) payload)->buf) + byte_index;
  }
}

int main() {
  TSParser *parser = ts_parser_new();
  ts_parser_set_language(parser, tree_sitter_c());

  /* Copy the file into BUFFER. */
  FILE *file = fopen("xdisp.c", "rb");
  fseek(file, 0, SEEK_END);
  long length = ftell (file);
  fseek(file, 0, SEEK_SET);
  char *buffer = malloc (length);
  fread(buffer, 1, length, file);
  fclose (file);

  struct buffer buf = {buffer, length};
  TSInput input = {&buf, read_file, TSInputEncodingUTF8};
  
  TSTree *tree = ts_parser_parse(parser, NULL, input);
  TSTree *new_tree = ts_parser_parse(parser, tree, input);
  
  free(buffer);
  ts_tree_delete(tree);
  ts_tree_delete(new_tree);
  ts_parser_delete(parser);

  return 0;
}

[-- Attachment #4: main-json.c --]
[-- Type: application/octet-stream, Size: 1195 bytes --]

#include <string.h>
#include <stdio.h>
#include <tree_sitter/api.h>

TSLanguage *tree_sitter_json();

struct buffer {
  char *buf;
  long len;
};

const char *read_file(void *payload, uint32_t byte_index,
                TSPoint position, uint32_t *bytes_read) {
  long len = ((struct buffer *) payload)->len;
  if (byte_index >= len) {
    *bytes_read = 0;
    return (char *) "";
  } else {
    *bytes_read = len - byte_index;
    return (char *) (((struct buffer *) payload)->buf) + byte_index;
  }
}

int main() {
  TSParser *parser = ts_parser_new();
  ts_parser_set_language(parser, tree_sitter_json());

  /* Copy the file into BUFFER. */
  FILE *file = fopen("benchmark.3.json", "rb");
  fseek(file, 0, SEEK_END);
  long length = ftell (file);
  fseek(file, 0, SEEK_SET);
  char *buffer = malloc (length);
  fread(buffer, 1, length, file);
  fclose (file);

  struct buffer buf = {buffer, length};
  TSInput input = {&buf, read_file, TSInputEncodingUTF8};
  
  TSTree *tree = ts_parser_parse(parser, NULL, input);
  TSTree *new_tree = ts_parser_parse(parser, tree, input);
  
  free(buffer);
  ts_tree_delete(tree);
  ts_tree_delete(new_tree);
  ts_parser_delete(parser);

  return 0;
}