Segfault when building the following kernel

[bjourne]

Compiling and building the following OpenCL program causes the latest version of the OpenCL runtime to crash:

typedef struct {
    uint n_mails;
    uint mails[100];
} mailbox;

__kernel void
run() {
    uint me = get_local_id(0);
    __local mailbox boxes[1][4];
    for (uint ix = 0; ix < 1; ix++) {
        __local mailbox *mb = &boxes[ix][me];
        uint i = 0;
        while (true) {
            uint swap = (i * 2) + 1;
            if (swap >= mb->n_mails) {
                break;
            }
            mb->mails[i] = mb->mails[swap];
            i = swap;
        }
    }
}

Compilation crashes on the following device:

  Name           : Intel(R) Xeon(R) W-2245 CPU @ 3.90GHz
  Version        : OpenCL 3.0 (Build 0)
  Driver         : 2024.18.7.0.11_160000
  C Version      : OpenCL C 3.0 
  Compute units  : 16 
  Global memory  : 33.33 GB
  Max allocation : 16.66 GB
  Max wg. size   : 8 k
  Local mem. size: 32.77 kB
  Max work items : 8192:8192:8192
  Pipe support   : yes
  Compiler usable: yes
  Image support  : yes

But works fine on the following device:

  Name           : Intel(R) Xeon(R) W-2245 CPU @ 3.90GHz
  Version        : OpenCL 3.0 (Build 0)
  Driver         : 2024.17.3.0.08_160000
  C Version      : OpenCL C 3.0 
  Compute units  : 16 
  Global memory  : 33.33 GB
  Max allocation : 16.66 GB
  Max wg. size   : 8 k
  Local mem. size: 32.77 kB
  Max work items : 8192:8192:8192
  Pipe support   : yes
  Compiler usable: yes
  Image support  : yes

So I guess a bug was introduced relatively recently.

[JablonskiMateusz]

@bjourne Please duplicate the issue to https://github.com/intel/intel-graphics-compiler repo