Skip to content

Instantly share code, notes, and snippets.

@cellularmitosis
Last active February 21, 2024 18:05
Show Gist options
  • Star 5 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save cellularmitosis/6c3a17b54881899b645aac1b6d37564a to your computer and use it in GitHub Desktop.
Save cellularmitosis/6c3a17b54881899b645aac1b6d37564a to your computer and use it in GitHub Desktop.
Compilation speed: tcc vs gcc (sudoku.c)

Blog 2020/5/28

<- previous | index | next ->

Compilation speed: tcc vs gcc (sudoku.c)

After writing a trivial sudoku solver in C yesterday, today I'll use it to benchmark the compilation speed of TCC vs GCC across a variety of processor speeds.

Compilation time (seconds)

Screenshot_2020-05-28_14-44-27

Screenshot_2020-05-28_14-46-04

Screenshot_2020-05-28_14-46-16

model name : Geode(TM) Integrated Processor by National Semi
tcc version 0.9.27 (prerelease) (i386 Linux)
gcc (Debian 6.3.0-18+deb9u1) 6.3.0 20170516
time tcc sudoku.c 2>&1 && mv a.out a.out.tcc
0.11user 0.03system 0:00.15elapsed 95%CPU (0avgtext+0avgdata 2340maxresident)k
0inputs+16outputs (0major+236minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0 2>&1
1.78user 0.27system 0:02.09elapsed 98%CPU (0avgtext+0avgdata 18944maxresident)k
0inputs+96outputs (0major+3319minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1 2>&1
3.40user 0.32system 0:03.75elapsed 99%CPU (0avgtext+0avgdata 20044maxresident)k
0inputs+96outputs (0major+3550minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2 2>&1
6.07user 0.33system 0:06.44elapsed 99%CPU (0avgtext+0avgdata 21920maxresident)k
0inputs+104outputs (0major+3804minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs 2>&1
4.18user 0.27system 0:04.54elapsed 97%CPU (0avgtext+0avgdata 21176maxresident)k
0inputs+88outputs (0major+3582minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3 2>&1
23.83user 0.40system 0:24.31elapsed 99%CPU (0avgtext+0avgdata 28560maxresident)k
0inputs+200outputs (0major+5593minor)pagefaults 0swaps
model name : Intel(R) Core(TM) i3-2350M CPU @ 2.30GHz
tcc version 0.9.27 (x86_64 Linux)
gcc (Debian 8.3.0-6) 8.3.0
time tcc sudoku.c 2>&1 && mv a.out a.out.tcc
0.01user 0.00system 0:00.01elapsed 90%CPU (0avgtext+0avgdata 2836maxresident)k
0inputs+24outputs (0major+313minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0 2>&1
0.06user 0.02system 0:00.09elapsed 100%CPU (0avgtext+0avgdata 21464maxresident)k
0inputs+96outputs (0major+4847minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1 2>&1
0.11user 0.02system 0:00.13elapsed 99%CPU (0avgtext+0avgdata 24224maxresident)k
0inputs+88outputs (0major+5086minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2 2>&1
0.19user 0.02system 0:00.21elapsed 100%CPU (0avgtext+0avgdata 26164maxresident)k
0inputs+88outputs (0major+5407minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs 2>&1
0.14user 0.02system 0:00.17elapsed 99%CPU (0avgtext+0avgdata 25632maxresident)k
0inputs+80outputs (0major+5163minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3 2>&1
0.51user 0.04system 0:00.56elapsed 99%CPU (0avgtext+0avgdata 39968maxresident)k
0inputs+240outputs (0major+8583minor)pagefaults 0swaps
a.out: sudoku.c
gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out
ccbench:
@cat /proc/cpuinfo | grep 'model name' | head -n1
@tcc -v || true
@gcc --version | head -n1
@echo
@tcc sudoku.c
time tcc sudoku.c 2>&1 && mv a.out a.out.tcc
@echo
@gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0
time gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0 2>&1
@echo
@gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1
time gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1 2>&1
@echo
@gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2
time gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2 2>&1
@echo
@gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs
time gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs 2>&1
@echo
@gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3
time gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3 2>&1
clean:
rm -f a.out a.out.*
.PHONY: ccbench clean
model name : Intel(R) Atom(TM) CPU N270 @ 1.60GHz
tcc version 0.9.27 (i386 Linux)
gcc (Debian 8.3.0-6) 8.3.0
time tcc sudoku.c 2>&1 && mv a.out a.out.tcc
0.01user 0.01system 0:00.02elapsed 91%CPU (0avgtext+0avgdata 2592maxresident)k
0inputs+16outputs (0major+257minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0 2>&1
0.22user 0.08system 0:00.31elapsed 99%CPU (0avgtext+0avgdata 18864maxresident)k
0inputs+104outputs (0major+3475minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1 2>&1
0.45user 0.06system 0:00.52elapsed 99%CPU (0avgtext+0avgdata 21172maxresident)k
0inputs+104outputs (0major+3698minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2 2>&1
0.81user 0.06system 0:00.88elapsed 99%CPU (0avgtext+0avgdata 23384maxresident)k
0inputs+112outputs (0major+3975minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs 2>&1
0.57user 0.05system 0:00.64elapsed 99%CPU (0avgtext+0avgdata 22876maxresident)k
0inputs+96outputs (0major+3766minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3 2>&1
2.22user 0.07system 0:02.30elapsed 99%CPU (0avgtext+0avgdata 29940maxresident)k
0inputs+192outputs (0major+5540minor)pagefaults 0swaps
model name : Geode(TM) Integrated Processor by AMD PCS
tcc version 0.9.27 (prerelease) (i386 Linux)
gcc (Debian 6.3.0-18+deb9u1) 6.3.0 20170516
time tcc sudoku.c 2>&1 && mv a.out a.out.tcc
0.06user 0.00system 0:00.06elapsed 89%CPU (0avgtext+0avgdata 1388maxresident)k
0inputs+0outputs (0major+403minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0 2>&1
0.92user 0.07system 0:01.02elapsed 97%CPU (0avgtext+0avgdata 10932maxresident)k
0inputs+0outputs (0major+6296minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1 2>&1
1.70user 0.06system 0:01.82elapsed 96%CPU (0avgtext+0avgdata 13036maxresident)k
0inputs+0outputs (0major+6886minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2 2>&1
2.95user 0.04system 0:03.06elapsed 97%CPU (0avgtext+0avgdata 14768maxresident)k
0inputs+0outputs (0major+7333minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs 2>&1
2.00user 0.12system 0:02.17elapsed 97%CPU (0avgtext+0avgdata 14040maxresident)k
0inputs+0outputs (0major+7104minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3 2>&1
11.03user 0.14system 0:11.23elapsed 99%CPU (0avgtext+0avgdata 21780maxresident)k
0inputs+0outputs (0major+9177minor)pagefaults 0swaps
model name : Intel(R) Core(TM) i5-4590 CPU @ 3.30GHz
tcc version 0.9.27 (x86_64 Linux)
gcc (Debian 8.3.0-6) 8.3.0
time tcc sudoku.c 2>&1 && mv a.out a.out.tcc
0.00user 0.00system 0:00.00elapsed 100%CPU (0avgtext+0avgdata 2604maxresident)k
0inputs+24outputs (0major+308minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0 2>&1
0.03user 0.00system 0:00.04elapsed 100%CPU (0avgtext+0avgdata 21408maxresident)k
0inputs+96outputs (0major+4828minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1 2>&1
0.06user 0.00system 0:00.06elapsed 98%CPU (0avgtext+0avgdata 24068maxresident)k
0inputs+88outputs (0major+5078minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2 2>&1
0.09user 0.00system 0:00.10elapsed 100%CPU (0avgtext+0avgdata 26196maxresident)k
0inputs+88outputs (0major+5410minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs 2>&1
0.07user 0.00system 0:00.08elapsed 98%CPU (0avgtext+0avgdata 25612maxresident)k
0inputs+80outputs (0major+5158minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3 2>&1
0.25user 0.02system 0:00.27elapsed 99%CPU (0avgtext+0avgdata 39944maxresident)k
0inputs+240outputs (0major+8574minor)pagefaults 0swaps
model name : Intel(R) Core(TM)2 CPU 6300 @ 1.86GHz
tcc version 0.9.27 (x86_64 Linux)
gcc (Debian 8.3.0-6) 8.3.0
time tcc sudoku.c 2>&1 && mv a.out a.out.tcc
0.00user 0.00system 0:00.00elapsed 88%CPU (0avgtext+0avgdata 2808maxresident)k
0inputs+24outputs (0major+314minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O0 sudoku.c -o a.out.gccO0 2>&1
0.08user 0.02system 0:00.11elapsed 99%CPU (0avgtext+0avgdata 21568maxresident)k
0inputs+96outputs (0major+4851minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O1 sudoku.c -o a.out.gccO1 2>&1
0.13user 0.04system 0:00.18elapsed 99%CPU (0avgtext+0avgdata 24296maxresident)k
0inputs+88outputs (0major+5112minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O2 sudoku.c -o a.out.gccO2 2>&1
0.25user 0.02system 0:00.28elapsed 99%CPU (0avgtext+0avgdata 26128maxresident)k
0inputs+88outputs (0major+5430minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -Os sudoku.c -o a.out.gccOs 2>&1
0.19user 0.01system 0:00.21elapsed 100%CPU (0avgtext+0avgdata 25636maxresident)k
0inputs+80outputs (0major+5150minor)pagefaults 0swaps
time gcc -std=c89 -Wall -Werror -O3 sudoku.c -o a.out.gccO3 2>&1
0.75user 0.05system 0:00.81elapsed 99%CPU (0avgtext+0avgdata 40160maxresident)k
0inputs+240outputs (0major+8605minor)pagefaults 0swaps
#include <stdlib.h>
#include <stdint.h>
#include <stdio.h>
#include <assert.h>
#include <stdbool.h>
/* A sudoku solver. */
/*
Copyright (c) 2020 Jason Pepas
Released under the terms of the MIT license.
See https://opensource.org/licenses/MIT
*/
/*
Board file format: plain text, 9 lines of 9 numbers (zero for empty box).
Example:
530070000
600195000
098000060
800060003
400803001
700020006
060000280
000419005
000080079
*/
/*
Data structures:
- A 'box' is a set of the remaining possible solutions for a spot in the grid.
This set is represented as a uint16_t bitvector, with the sudoke numbers 1-9
represented as the first 9 bits in the bitvector.
- A solved box has only one bit set.
- A 'board' is a 2-dimensional array (9x9) of boxes (bitvectors).
- A solved board is a 9x9 grid of bitvectors which each have only one item.
*/
/*
The general approach is to iterate over the board and use logic to reduce
the number of possible solutions in each box until the board is solved.
Algorithm:
- For each row, build a set of the solved numbers in that row, then
remove those solutions from the remaining possibilities in each of the
unsolved boxes in that row.
- Do the same for each column.
- Do the same for each 3x3 grid.
- Repeat the above process until either the board is solved, or until
deadlock is detected (i.e. the board state remains unchanged after an
iteration).
*/
/* the board. */
uint16_t g_board[9][9];
/* set all 9 bits in this box. */
void board_set_all(int row, int col) {
uint16_t all_nums = /* 0b0000000111111111 */ 0x1FF;
g_board[row][col] = all_nums;
}
/* convert a number to a bitvector. */
uint16_t num_to_bitvec(int num) {
return 1 << (num-1);
}
/* add num to the box's bitvector. */
void board_set_num(int row, int col, int num) {
g_board[row][col] |= num_to_bitvec(num);
}
/* count the number of bits which are set in the bitvector. */
int count_set_bits(uint16_t box) {
int set_bits_count = 0;
int i = 0;
while (i < 16) {
if (box & /* 0b1 */ 0x1) {
set_bits_count++;
}
box = (box >> 1);
i++;
continue;
}
return set_bits_count;
}
/* is the box solved? */
bool box_is_solved(int row, int col) {
uint16_t box = g_board[row][col];
int bitcount = count_set_bits(box);
return bitcount == 1;
}
/* is the board solved? */
bool board_is_solved() {
int row = 0;
int col = 0;
while (row < 9) {
while (col < 9) {
if (!box_is_solved(row, col)) {
return false;
}
col++;
continue;
}
row++;
continue;
}
return true;
}
/* return the character representation of the box. */
char board_get_ch(int row, int col) {
if (!box_is_solved(row, col)) {
return '0';
}
uint16_t box = g_board[row][col];
uint8_t num = 1;
while (num < 10) {
if (box & /* 0b1 */ 0x1) {
return (char)(num + '0');
}
box = (box >> 1);
num++;
continue;
}
assert(false);
exit(2);
}
/* print the board state. prints '0' for unsolved boxes. */
void print_board() {
int row = 0;
while (row < 9) {
int col = 0;
while (col < 9) {
char ch = board_get_ch(row, col);
putchar(ch);
col++;
continue;
}
putchar('\n');
row++;
continue;
}
}
/* return the total count of all remaining possibliities in every box.
a solved board has a possibilities count of 81. */
int possibilities_count() {
int pcount = 0;
int row = 0;
while (row < 9) {
int col = 0;
while (col < 9) {
uint16_t box = g_board[row][col];
int bitcount = count_set_bits(box);
pcount += bitcount;
col++;
continue;
}
row++;
continue;
}
return pcount;
}
/* read the board in from a file. */
void load_board(char* fname) {
FILE* f = fopen(fname, "r");
assert(f != NULL);
const int bufsize = 110; /* 9+2 per line * 9 lines + 1 null */
char buf[bufsize];
size_t count = fread(buf, 1, bufsize, f);
assert(count >= 90);
fclose(f);
int i = 0;
int row = 0;
while (row < 9) {
int col = 0;
while (col < 9) {
assert (i < bufsize);
char ch = buf[i];
if (ch == '0') {
board_set_all(row, col);
} else {
int num = ch - '0';
board_set_num(row, col, num);
}
i++;
col++;
continue;
}
/* skip over the newline. */
char ch = buf[i];
while (ch == '\n' || ch == '\r') {
i++;
ch = buf[i];
continue;
}
row++;
continue;
}
}
/* return the combined bitmask of all solved boxes in the row. */
uint16_t get_row_solved_bitmask(int row) {
uint16_t bitmask = 0;
int col = 0;
while (col < 9) {
if (box_is_solved(row, col)) {
bitmask |= g_board[row][col];
}
col++;
continue;
}
return bitmask;
}
/* iterate over the rows, reducing possibilities. */
void iterate_rows() {
int row = 0;
while (row < 9) {
uint16_t solved = get_row_solved_bitmask(row);
int col = 0;
while (col < 9) {
if (!box_is_solved(row, col)) {
/* subtract the solved bits from this box. */
uint16_t box = g_board[row][col];
box &= ~(solved);
g_board[row][col] = box;
}
col++;
continue;
}
row++;
continue;
}
}
/* return the combined bitmask of all solved boxes in the column. */
uint16_t get_col_solved_bitmask(int col) {
uint16_t bitmask = 0;
int row = 0;
while (row < 9) {
if (box_is_solved(row, col)) {
bitmask |= g_board[row][col];
}
row++;
continue;
}
return bitmask;
}
/* iterate over the columns, reducing possibilities. */
void iterate_columns() {
int col = 0;
while (col < 9) {
uint16_t solved = get_col_solved_bitmask(col);
int row = 0;
while (row < 9) {
if (!box_is_solved(row, col)) {
/* subtract the solved bits from this box. */
uint16_t box = g_board[row][col];
box &= ~(solved);
g_board[row][col] = box;
}
row++;
continue;
}
col++;
continue;
}
}
/* return the combined bitmask of all solved boxes in the grid. */
uint16_t get_grid_solved_bitmask(int grid) {
uint16_t bitmask = 0;
int first_row = (grid / 3) * 3;
int first_col = (grid % 3) * 3;
int row = first_row;
while (row < (first_row + 3)) {
int col = first_col;
while (col < (first_col + 3)) {
if (box_is_solved(row, col)) {
bitmask |= g_board[row][col];
}
col++;
continue;
}
row++;
continue;
}
return bitmask;
}
/* iterate over the grids, reducing possibilities. */
void iterate_grids() {
int grid = 0;
while (grid < 9) {
uint16_t solved = get_grid_solved_bitmask(grid);
int first_row = (grid / 3) * 3;
int first_col = (grid % 3) * 3;
int row = first_row;
while (row < (first_row + 3)) {
int col = first_col;
while (col < (first_col + 3)) {
if (!box_is_solved(row, col)) {
/* subtract the solved bits from this box. */
uint16_t box = g_board[row][col];
box &= ~(solved);
g_board[row][col] = box;
}
col++;
continue;
}
row++;
continue;
}
grid++;
continue;
}
}
/* iterate over the board, reducing possibilities. */
void iterate_board() {
iterate_rows();
iterate_columns();
iterate_grids();
}
int main(int argc, char* argv[]) {
load_board(argv[1]);
int before_pcount = possibilities_count();
while (!board_is_solved()) {
iterate_board();
int after_pcount = possibilities_count();
if (after_pcount == before_pcount) {
fprintf(stderr, "Error: don't know how to solve this board.\n");
exit(3);
} else {
before_pcount = after_pcount;
continue;
}
}
print_board();
return EXIT_SUCCESS;
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment