Using a builtin instruction (on Cortex M3/M4/M7) saves 268 bytes.
Tested that the result is the same for all values:
```c
#include <stdio.h>
#include <limits.h>
#include <inttypes.h>
int cl2_new(unsigned int x) {
const int bitwidth = CHAR_BIT*sizeof(x);
x--;
const int clz = x ? __builtin_clz(x) : bitwidth; // clz(0) is undefined
return bitwidth-clz;
}
int cl2_old(unsigned int x) {
static const uint8_t log_2[256] = { /* log_2[i] = ceil(log2(i - 1)) */
0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8
};
int l = 0;
x--;
while (x >= 256) { l += 8; x >>= 8; }
return l + log_2[x];
}
int main(int argc, char** argv) {
for (unsigned int i=1; i!=0; i++) { // 0 is not in the domain
int v_new = cl2_new(i);
int v_old = cl2_old(i);
if (v_new != v_old) {
printf("%u: %d != %d\n", i, v_new, v_old);
break;
}
}
}
```
In particular this fixes some exceedingly rare/impossible
use-after-frees.
Add the new docs from the distribution and clarify where we get our code
from. To maintain our alterations, the following patches have been
applied to the source from upstream's repository (1221e987...75ea9ccb)
to bring the source up to date:
* Bug: Long brackets with a huge number of '=' causes overflow
A long bracket with too many equal signs can overflow the 'int' used for
the counting and some arithmetic done on the value. Changing the counter
to 'size_t' avoids that. (Because what is counted goes to a buffer, an
overflow in the counter will first raise a buffer-overflow error.)
* Fixed bug in 'lua_upvaluejoin'
Bug-fix: joining an upvalue with itself could cause a use-after-free
crash.
* Fixed typos in comments
* Fixed missing GC barriers in compiler and undump
While building a new prototype, the GC needs barriers for every object
(strings and nested prototypes) that is attached to the new prototype.
* Updated release number and copyright year
* Fixed bug: invalid mode can crash 'io.popen'
* Fixed bug: Negation overflow in getlocal/setlocal
* 'realloc' can fail when shrinking a block
According to ISO C, 'realloc' can fail when shrinking a block. If that
happens, 'l_alloc' simply ignores the fail and returns the original
block.
* Fixed bug of long strings in binary chunks
When "undumping" a long string, the function 'LoadVector' can call the
reader function, which can run the garbage collector, which can collect
the string being read. So, the string must be anchored during the call
to 'LoadVector'.
Thomas Watson
Michael du Breuil