Intel GPU driver throws a segmentation fault when executing a Vulkan shader program containing nested loops. The fragment shader snippet is as follows:
float f;
uint g = 1u;
const int t = -40000;
for (int i = t; i > t - 4; i -= 2) {
f = float(input1); // input1 is set to 1 during runtime
for (int j = 0; j < input2; j += 2) { // input2 is set to 2 during runtime
g = g == 0u ? uint(f) : uint(f) / g;
}
}
float l = float(g);
fragColor = vec4(l);
When tweaking the constant variable t to a different value, or changing the outer loop from decrementing to incrementing, the shader executes without any issues. The shader can be compiled without any warning or errors by the `glslangValidator` tool, and does not contain any undefined behavior. However, when it is executed on the Intel GPU, the GPU throws a segmentation fault.
Steps to reproduce:
I made a minimal Vulkan cpp code that simply loads and runs the fragment shader above. The full source code, including both the cpp code and fragment shader, and a detailed README describing system environments and reproduction steps, can be found here. It should take less than a minute to reproduce the issue.
