reinterpret_cast in C++

Question

 uint32_t r,g,b;
 r = (uint32_t)145;
 g = (uint32_t)131;
 b = (uint32_t)139;
 uint32_t rgb = ((uint32_t)r << 16 | (uint32_t)g << 8 | (uint32_t)b);
 float rgbf = *reinterpret_cast<float*>(&rgb);

 uint32_t rgbnew = *(reinterpret_cast<uint32_t *>(&rgbf));
 uint8_t rnew = (rgbnew >> 16) & 0x0000ff;
 uint8_t gnew = (rgbnew >> 8) & 0x0000ff;
 uint8_t bnew = (rgbnew) & 0x0000ff;

When I try to run this code there is segmentation fault at line

uint32_t rgb = ((uint32_t)r << 16 | (uint32_t)g << 8 | (uint32_t)b);

In fact, at one place this is running ok. At another place it is giving seg fault.

Please tell which compiler (and version) are you using, and for which target system. Also give the compilation flags. — Basile Starynkevitch, Jan 05 '13 at 13:42
Did you try to isolate this snippet and see what happens? I put this exact code in a simple main function and there is no segfault. — coelhudo, Jan 05 '13 at 13:47
I guess the explicit conversions to `uint32` in `uint32_t rgb = ((uint32_t)r << 16 | (uint32_t)g << 8 | (uint32_t)b);` are not needed, because you declare `r`, `g`, and `b` to be `uint32`. Is there a place where they are needed, because those variables have a different type? — Andy Prowl, Jan 05 '13 at 13:53

Basile Starynkevitch · Answer 1 · 2013-01-05T13:51:32.113

Please try to compile your code with all warnings and debug info (e.g. with g++ -Wall -g on Linux) and to improve it till no warnings are given. Learn to use the debugger (i.e. gdb on Linux)

I would guess that the fault is probably at

 float rgbf = *reinterpret_cast<float*>(&rgb);

because this may trigger a fault if rgb (i.e. uint32_t) and float don't have the same alignment or size constraints. Some systems (processors, ABIs, compilers) may have different & incompatible constraints for them.

BTW, your code works well with GCC 4.7 on Debian/GNU/Linux/x86-64 invoked as

   g++-4.7 -std=c++11 -Wall -g ramji.cc -o ramji

when I have

 #include <cstdint>
 #include <iostream>

 int main(int argc, char**argv)
 {
   uint32_t r,g,b;
   r = (uint32_t)145;
   g = (uint32_t)131;
   b = (uint32_t)139;
   std::cout << "r=" << r << " g=" << g << " b=" << b << std::endl;
   uint32_t rgb = ((uint32_t)r << 16 | (uint32_t)g << 8 | (uint32_t)b);
   std::cout << "rgb=" << rgb << std::endl;
   float rgbf = *reinterpret_cast<float*>(&rgb);
   std::cout << "rgbf=" << rgbf << std::endl;
   uint32_t rgbnew = *(reinterpret_cast<uint32_t *>(&rgbf));
   std::cout << "rgbnew=" << rgb << std::endl;
   uint8_t rnew = (rgbnew >> 16) & 0x0000ff;
   uint8_t gnew = (rgbnew >> 8) & 0x0000ff;
   uint8_t bnew = (rgbnew) & 0x0000ff;
   std::cout << "rnew=" << rnew << " gnew=" << gnew 
        << " bnew=" << bnew << std::endl;
   return 0;
 }

No warnings, no crashes at execution.

Notice that <cstdint> requires a C++11 conforming compiler.

reinterpret_cast in C++

1 Answers1