[GPU] Gracefully handle zero batch size

[mostafafaheem]

Details:

• Avoid division by zero error when creating memory descriptor for GPU plugin

Tickets:

• closes [Good First Issue]: Trap divide error in GPU plugin for batch size 0 #24243

AI Assistance:

• AI assistance used: yes
• To generate tests

[mostafafaheem]

Hello @p-durandin, @p-wysocki, @praasz!

Is this the right approach, or should it just return an exception? When I tested the following snippet using CPU, it caused a segmentation fault that was fixed when I set batch size to 1, so if the exception approach was to be taken, should it be at the model validation level, or the plugin level? Since CPU plugin also seems to have a problem with 0 batch size.

#include <chrono>
#include <thread>
#include <string>
#include <iostream>

#include <openvino/openvino.hpp>

int main(int argc, char *argv[]){
    ov::Core ieCore;
    auto model = ieCore.read_model(argv[1]);

    // Reshape model
    model->reshape({0, 3, 300, 300});

    ov::CompiledModel compiledModel = ieCore.compile_model(model, argv[2]);

    // Create an inference request
    ov::InferRequest infer_request = compiledModel.create_infer_request();

    // Get input port for model with one input
    auto input_port = compiledModel.input();

    // Create tensor from external memory
    ov::Tensor input_tensor(input_port.get_element_type(), {0, 3, 300, 300});

    // Set input tensor for model with one input
    infer_request.set_input_tensor(input_tensor);
    infer_request.start_async();
    infer_request.wait();

    std::this_thread::sleep_for(std::chrono::seconds(1));
    compiledModel = {};
    std::cout << "Hello\n" << std::flush;
}