Enhance Vulkan device discovery

[ylsdamxssjxxdd]

At present, the Vulkan version automatically uses integrated graphics cards. The following is an enhanced implementation for device discovery~~

stable-diffusion.cpp

#ifdef SD_USE_VULKAN
        LOG_DEBUG("Using Vulkan backend");
        int dev_count = ggml_backend_vk_get_device_count();
        int dev = 0;
        if (const char* s = getenv("GGML_VK_DEVICE")) {
            int v = atoi(s);
            if (v >= 0 && v < dev_count) dev = v;
        }
        // prefer a discrete NVIDIA device if available; fallback to first
        int preferred = -1;
        for (int i = 0; i < dev_count; ++i) {
            char desc[256] = {0};
            ggml_backend_vk_get_device_description(i, desc, sizeof(desc));
            // avoid SwiftShader/Software devices and prefer NVIDIA/GeForce/RTX naming
            std::string d(desc);
            if (d.find("NVIDIA") != std::string::npos || d.find("GeForce") != std::string::npos || d.find("RTX") != std::string::npos) {
                preferred = i;
                break;
            }
        }
        if (preferred >= 0) dev = preferred;
        // Log available devices
        for (int i = 0; i < dev_count; ++i) {
            char desc[256] = {0};
            ggml_backend_vk_get_device_description(i, desc, sizeof(desc));
            LOG_INFO("ggml_vulkan: %d = %s", i, desc);
        }
        backend = ggml_backend_vk_init(dev);
        if (!backend) {
            LOG_WARN("Failed to initialize Vulkan backend (device %d)", dev);
        } else {
            char desc[256] = {0};
            ggml_backend_vk_get_device_description(dev, desc, sizeof(desc));
            LOG_INFO("Vulkan selected device: %d - %s", dev, desc);
        }
#endif

[wbruna]

Without considering the heuristic itself, and the badly named GGML_VK_DEVICE: why would you check an environment variable first (which supposedly could be changed by the user), then promptly override its value with a fixed criteria? How would the user fix it if the code keeps choosing wrong?

[jeffbolznv]

Is the existing GGML_VK_VISIBLE_DEVICES env var sufficient? Or does it not work with how sd uses vulkan?

[wbruna]

Is the existing GGML_VK_VISIBLE_DEVICES env var sufficient? Or does it not work with how sd uses vulkan?

It works fine for me.