Thread: why program crashes?

I have problem with this program:
[C++] bitpacking working version - Pastebin.com
this is the code I need to cope with in this thread it is temporal version.

The original version is here:
[C++] bitpacking.cpp - Pastebin.com
this is just for comparison.

Purpose: I am rewriting the original code in C++ to the new version which will be in C. The code does bitpacking using benchmark.

It crashes on line 2975:

Code:

void __fastpack1(const uint32_t *  __restrict__ in, uint32_t *  __restrict__  out) {
    *out |= (*in)   & 1 ; // here

Mmm - 6000+ lines of the most awful code I've seen in a long while.

> It crashes on line 2975:
Run it in the debugger and figure out which of your pointers is invalid.
- are they both valid (ie memory aligned) to access uint32_t's ?
- are they both pointing to different places in memory, to honor the restrict qualifier?

When the debugger has trapped, then do a stack trace to find out the caller chain back to main.

Is it really any quicker than say 10 lines of code to do the same thing?
For one thing, a short piece of code in a loop would benefit greatly from being cache resident all the time, than several 1000 lines of straight-line code hammering the external bus for both instructions and data.

Not to mention that several 1000 lines are far more prone to errors, much less readable and more time consuming for the programmer.

I have problem to understand this command:

Code:

compressed.resize(N*bit/32, 0);

so when I have rewritten it to C style I did this:

Code:

free(compressed);
free(recovered);
t = N*bit/32;
compressed = (uint32_t *) malloc(t); // in
recovered = (uint32_t *) malloc(t); // out

Hence the size is not uint32_t * N but t. But what does really do vector.resize() ? Does it realloc memory? I am not used to C++ I just try to rewrite the C++ version of the code to C version.

I think what resize() does is copy the old data to the larger buffer and frees the older one. Of course it also copies the new buffer pointer back to the relevant member of compressed.

It doesn't just copy a buffer. It creates copies of each element in the vector, which includes calling the copy constructor. I suspect in C++11 and later, it uses the move constructor, if available.

Originally Posted by the_jackass

I think what resize() does is copy the old data to the larger buffer and frees the older one. Of course it also copies the new buffer pointer back to the relevant member of compressed.

It resizes the size of the vector (remember that a vector is essentially an array and an array has a fixed number of elements; in case of a vector, it can change the number of elements it contains at any time).

Originally Posted by Elkvis

It doesn't just copy a buffer. It creates copies of each element in the vector, which includes calling the copy constructor. I suspect in C++11 and later, it uses the move constructor, if available.

It uses the move constructor/move assignment only if they do not throw. If they do, the copy constructor is used because resize has strong exception safety guarantee.