1
0
mirror of https://github.com/RIOT-OS/RIOT.git synced 2024-12-29 04:50:03 +01:00
RIOT/sys/fmt/fmt.c
bors[bot] 8c2a4b43b3
Merge #19027
19027: sys/fmt: optimize scn_u32_dec scn_u32_hex r=benpicco a=kfessel

### Contribution description

Improves the compilation result for `scn_u32_dec` `scn_u32_hex` especially on `cortex-m` reducing either stack usage and or code size.

This makes use of unsigned int overflow (slightly less better without doing that `hexn`).

See godbolt (original versions got an `o` attached, modified versions got `k`s) all functions are  marked `_S_` defined to `static`

by assigning the global at end the compiled function can be changed (`deco deck  hexo hexk hexkk hexn`)

this PR is `hexkk` and `deck` 

### Testing procedure

run unit-test/test-fmt

```
<RIOT>/tests/unittests$ make tests-fmt
<RIOT>/tests/unittests$ make term
```

### Issues/PRs references

[godbolt](https://godbolt.org/z/MzT1zh4q1)

Co-authored-by: Karl Fessel <karl.fessel@ovgu.de>
2023-02-17 20:09:53 +00:00

605 lines
11 KiB
C

/*
* Copyright (C) 2015 Kaspar Schleiser <kaspar@schleiser.de>
*
* This file is subject to the terms and conditions of the GNU Lesser
* General Public License v2.1. See the file LICENSE in the top level
* directory for more details.
*/
/**
* @ingroup sys_fmt
* @{
*
* @file
* @brief String formatting library implementation
*
* @author Kaspar Schleiser <kaspar@schleiser.de>
*
* @}
*/
#include <assert.h>
#include <stdarg.h>
#include <stdint.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include "kernel_defines.h"
#include "fmt.h"
#include "stdio_base.h"
static const char _hex_chars[16] = "0123456789ABCDEF";
static const uint32_t _tenmap[] = {
0,
10LU,
100LU,
1000LU,
10000LU,
100000LU,
1000000LU,
10000000LU,
};
#define TENMAP_SIZE ARRAY_SIZE(_tenmap)
static inline char _to_lower(char c)
{
return 'a' + (c - 'A');
}
int fmt_is_number(const char *str)
{
if (!str || !*str) {
return 0;
}
for (; *str; str++) {
if (!fmt_is_digit(*str)) {
return 0;
}
}
return 1;
}
size_t fmt_byte_hex(char *out, uint8_t byte)
{
if (out) {
*out++ = _hex_chars[byte >> 4];
*out = _hex_chars[byte & 0x0F];
}
return 2;
}
size_t fmt_bytes_hex(char *out, const uint8_t *ptr, size_t n)
{
size_t len = n * 2;
if (out) {
while (n--) {
out += fmt_byte_hex(out, *ptr++);
}
}
return len;
}
size_t fmt_strlen(const char *str)
{
const char *tmp = str;
while (*tmp) {
tmp++;
}
return (tmp - str);
}
size_t fmt_strnlen(const char *str, size_t maxlen)
{
const char *tmp = str;
while (*tmp && maxlen--) {
tmp++;
}
return (tmp - str);
}
size_t fmt_str(char *out, const char *str)
{
int len = 0;
if (!out) {
len = fmt_strlen(str);
}
else {
char c;
while ((c = *str++)) {
*out++ = c;
len++;
}
}
return len;
}
size_t fmt_bytes_hex_reverse(char *out, const uint8_t *ptr, size_t n)
{
size_t i = n;
while (i--) {
out += fmt_byte_hex(out, ptr[i]);
}
return (n << 1);
}
static uint8_t _byte_mod25(uint8_t x)
{
for (unsigned divisor = 200; divisor >= 25; divisor >>= 1) {
if (x >= divisor) {
x -= divisor;
}
}
return x;
}
static uint8_t _hex_nib(uint8_t nib)
{
return _byte_mod25((nib & 0x1f) + 9);
}
uint8_t fmt_hex_byte(const char *hex)
{
return (_hex_nib(hex[0]) << 4) | _hex_nib(hex[1]);
}
size_t fmt_hex_bytes(uint8_t *out, const char *hex)
{
size_t len = fmt_strlen(hex);
if (len & 1) {
return 0;
}
size_t final_len = len >> 1;
if (out == NULL) {
return final_len;
}
for (size_t i = 0, j = 0; j < final_len; i += 2, j++) {
out[j] = fmt_hex_byte(hex + i);
}
return final_len;
}
size_t fmt_u16_hex(char *out, uint16_t val)
{
return fmt_bytes_hex_reverse(out, (uint8_t *)&val, 2);
}
size_t fmt_u32_hex(char *out, uint32_t val)
{
return fmt_bytes_hex_reverse(out, (uint8_t *)&val, 4);
}
size_t fmt_u64_hex(char *out, uint64_t val)
{
return fmt_bytes_hex_reverse(out, (uint8_t *)&val, 8);
}
size_t fmt_u64_dec(char *out, uint64_t val)
{
uint32_t d[5];
uint32_t q;
size_t len = 0;
d[0] = val & 0xFFFF;
d[1] = (val >> 16) & 0xFFFF;
d[2] = (val >> 32) & 0xFFFF;
d[3] = (val >> 48) & 0xFFFF;
d[0] = 656 * d[3] + 7296 * d[2] + 5536 * d[1] + d[0];
q = d[0] / 10000;
d[0] = d[0] % 10000;
d[1] = q + 7671 * d[3] + 9496 * d[2] + 6 * d[1];
q = d[1] / 10000;
d[1] = d[1] % 10000;
d[2] = q + 4749 * d[3] + 42 * d[2];
q = d[2] / 10000;
d[2] = d[2] % 10000;
d[3] = q + 281 * d[3];
q = d[3] / 10000;
d[3] = d[3] % 10000;
d[4] = q;
int first = 4;
while (!d[first] && first) {
first--;
}
len = fmt_u32_dec(out, d[first]);
int total_len = len + (first * 4);
if (out) {
out += len;
memset(out, '0', total_len - len);
while (first) {
first--;
if (d[first]) {
size_t tmp = fmt_u32_dec(NULL, d[first]);
fmt_u32_dec(out + (4 - tmp), d[first]);
}
out += 4;
}
}
return total_len;
}
size_t fmt_u32_dec(char *out, uint32_t val)
{
size_t len = 1;
/* count needed characters */
/* avoid multiply overflow: uint32_t max len = 10 digits */
if (val >= 1000000000ul) {
len = 10;
}
else {
for (uint32_t tmp = 10; tmp <= val; len++) {
tmp *= 10;
}
}
if (out) {
char *ptr = out + len;
do {
*--ptr = (val % 10) + '0';
} while ((val /= 10));
}
return len;
}
size_t fmt_u16_dec(char *out, uint16_t val)
{
return fmt_u32_dec(out, val);
}
size_t fmt_s64_dec(char *out, int64_t val)
{
unsigned negative = (val < 0);
uint64_t sval;
if (negative) {
if (out) {
*out++ = '-';
}
sval = -(uint64_t)(val);
}
else {
sval = +(uint64_t)(val);
}
return fmt_u64_dec(out, sval) + negative;
}
size_t fmt_s32_dec(char *out, int32_t val)
{
unsigned negative = (val < 0);
uint32_t sval;
if (negative) {
if (out) {
*out++ = '-';
}
sval = -((uint32_t)(val));
}
else {
sval = +((uint32_t)(val));
}
return fmt_u32_dec(out, sval) + negative;
}
size_t fmt_s16_dec(char *out, int16_t val)
{
return fmt_s32_dec(out, val);
}
size_t fmt_s16_dfp(char *out, int16_t val, int scale)
{
return fmt_s32_dfp(out, val, scale);
}
size_t fmt_s32_dfp(char *out, int32_t val, int scale)
{
unsigned pos = 0;
if (scale == 0) {
pos = fmt_s32_dec(out, val);
}
else if (scale > 0) {
pos = fmt_s32_dec(out, val);
if (out) {
memset(&out[pos], '0', scale);
}
pos += scale;
}
else {
scale = -scale;
char buf[10]; /* "2147483648" */
int negative = val < 0;
uint32_t uval = negative ? -val : val;
int len = fmt_u32_dec(buf, uval);
if (negative) {
if (out) {
out[pos] = '-';
}
pos++;
}
if (len <= scale) {
int zeroes = scale - len + 1;
if (out) {
memset(&out[pos], '0', zeroes);
}
pos += zeroes;
}
if (out) {
memcpy(&out[pos], buf, len);
}
pos += len;
if (out) {
unsigned dot_pos = pos - scale;
for (unsigned i = pos; i >= dot_pos; i--) {
out[i] = out[i - 1];
}
out[dot_pos] = '.';
}
pos += 1;
}
return pos;
}
/* this is very probably not the most efficient implementation, as it at least
* pulls in floating point math. But it works, and it's always nice to have
* low hanging fruits when optimizing. (Kaspar)
*/
size_t fmt_float(char *out, float f, unsigned precision)
{
assert(precision < TENMAP_SIZE);
unsigned negative = (f < 0);
uint32_t integer;
if (negative) {
f = -f;
}
integer = (uint32_t)f;
f -= integer;
uint32_t fraction = f * _tenmap[precision];
if (negative && out) {
*out++ = '-';
}
size_t res = fmt_u32_dec(out, integer);
if (precision && fraction) {
if (out) {
out += res;
*out++ = '.';
size_t tmp = fmt_u32_dec(out, fraction);
fmt_lpad(out, tmp, precision, '0');
}
res += (1 + precision);
}
res += negative;
return res;
}
size_t fmt_lpad(char *out, size_t in_len, size_t pad_len, char pad_char)
{
if (in_len >= pad_len) {
return in_len;
}
if (out) {
size_t n = pad_len - in_len;
if (FMT_USE_MEMMOVE) {
memmove(out + n, out, in_len);
memset(out, pad_char, n);
}
else {
char *pos = out + pad_len - 1;
out += in_len - 1;
while (in_len--) {
*pos-- = *out--;
}
while (n--) {
*pos-- = pad_char;
}
}
}
return pad_len;
}
size_t fmt_char(char *out, char c)
{
if (out) {
*out = c;
}
return 1;
}
size_t fmt_to_lower(char *out, const char *str)
{
size_t len = 0;
while (str && *str) {
if (fmt_is_upper(*str)) {
if (out) {
*out++ = _to_lower(*str);
}
}
else if (out) {
*out++ = *str;
}
str++;
len++;
}
return len;
}
uint32_t scn_u32_dec(const char *str, size_t n)
{
uint32_t res = 0;
while (n--) {
unsigned c = *str++;
unsigned d = c - (unsigned)'0';
if ( !( '0'<= c && c <= '9') ) {
break;
}
res *= 10U;
res += d;
}
return res;
}
uint32_t scn_u32_hex(const char *str, size_t n)
{
uint32_t res = 0;
while (n--) {
unsigned c = *str++;
unsigned d;
if (('0'<= c) && (c <= '9')){
d = c - (unsigned)'0';
}
else if (('A' <= c) && (c <= 'F')) {
d = c - (unsigned)'A' + 0xaU;
}
else if (('a' <= c) && (c <= 'f')) {
d = c - (unsigned)'a' + 0xaU;
}
else {
break;
}
res <<= 4U;
res |= d;
}
return res;
}
/* native gets special treatment as native's stdio code is ... special.
* And when not building for RIOT, there's no `stdio_write()`.
* In those cases, just defer to `printf()`.
*/
#if IS_USED(MODULE_STDIO_NATIVE) || !defined(RIOT_VERSION)
void print(const char *s, size_t n)
{
printf("%.*s", (int)n, s);
}
#else
void print(const char *s, size_t n)
{
/* flush the libc's output buffer so output is not intermingled. */
fflush(stdout);
stdio_write(s, n);
}
#endif
void print_u32_dec(uint32_t val)
{
char buf[10]; /* "4294967295" */
size_t len = fmt_u32_dec(buf, val);
print(buf, len);
}
void print_s32_dec(int32_t val)
{
char buf[11]; /* "-2147483648" */
size_t len = fmt_s32_dec(buf, val);
print(buf, len);
}
void print_byte_hex(uint8_t byte)
{
char buf[2];
fmt_byte_hex(buf, byte);
print(buf, sizeof(buf));
}
void print_bytes_hex(const void *bytes, size_t num)
{
const uint8_t *b = bytes;
while (num--) {
print_byte_hex(*b++);
}
}
void print_u32_hex(uint32_t val)
{
char buf[8];
fmt_u32_hex(buf, val);
print(buf, sizeof(buf));
}
void print_u64_hex(uint64_t val)
{
print_u32_hex(val >> 32);
print_u32_hex(val);
}
void print_u64_dec(uint64_t val)
{
char buf[20]; /* "18446744073709551615" */
size_t len = fmt_u64_dec(buf, val);
print(buf, len);
}
void print_s64_dec(uint64_t val)
{
char buf[20]; /* "-9223372036854775808" */
size_t len = fmt_s64_dec(buf, val);
print(buf, len);
}
void print_float(float f, unsigned precision)
{
char buf[19];
size_t len = fmt_float(buf, f, precision);
print(buf, len);
}
void print_str(const char *str)
{
print(str, fmt_strlen(str));
}