QueueFamilyIndices VulkanEngine::find_queue_families(VkPhysicalDevice device)

[EssenOH]

it should be updated as following.

int i = 0;
for (const auto& queueFamily : queue_families) {

	if (queueFamily.queueFlags & VK_QUEUE_GRAPHICS_BIT) {
		indices.graphics_family = i;
	}
	
	if (queueFamily.queueFlags & VK_QUEUE_COMPUTE_BIT) {
		indices.compute_family = i;
	}
	
	if (queueFamily.queueFlags & VK_QUEUE_TRANSFER_BIT) {
		indices.transfer_family = i;
	}

	if (indices.is_complete()) {
		break;
	}

	i++;
}

[hooyuser]

Thanks for your suggestions. Sorry I haven't spent enough time on vendor-specific problems. Here is the queue family information of my device. https://vulkan.gpuinfo.org/displayreport.php?id=11642#queuefamilies

For now, given the objective of asynchronous computing, I use the GRAPHICS|COMPUTE|TRANSFER queue family for graphics and the COMPUTE|TRANSFER queue family for computing. I guess the following treatment works for my purpose as well as produces a fallback.

int i = 0;
for (const auto& queueFamily : queue_families) {
        const VkQueueFlags queue_flags = queueFamily.queueFlags;

        if (queue_flags & VK_QUEUE_TRANSFER_BIT) {
	        if (!indices.transfer_family.has_value() || !(queue_flags & (VK_QUEUE_GRAPHICS_BIT | VK_QUEUE_COMPUTE_BIT))) {
		        indices.transfer_family = i;
	        }
        }

        if (queue_flags & VK_QUEUE_COMPUTE_BIT) {
	        if (!indices.compute_family.has_value() || !(queue_flags & VK_QUEUE_GRAPHICS_BIT)) {
		        indices.compute_family = i;
	        }
        }

        if (queue_flags & VK_QUEUE_GRAPHICS_BIT) {
	        indices.graphics_family = i;
        }

        if (indices.is_complete()) {
	        break;
        }
        
        i++;
}

I don't know your queue family information. Most of them are available on https://vulkan.gpuinfo.org. With that information, there may be better ways to handle your case.

[EssenOH]

You use a high performance desktop GPU, but I am working with a notebook and my computer has only one single queue family. then, checking the available queue family # and deciding the assignment could be a good way to solve the compatibility issue.

…

On Mon, Aug 1, 2022 at 8:16 AM Derived Cat ***@***.***> wrote: Thanks for your suggestions. Sorry I haven't spent enough time on vendor-specific problems. Here is the queue family information of my device. https://vulkan.gpuinfo.org/displayreport.php?id=11642#queuefamilies [image: queue family] <https://user-images.githubusercontent.com/10691820/182175879-900f5e43-82be-4a58-98c9-c7d8039f43fe.png> For now, given the objective of asynchronous computing, I use the GRAPHICS|COMPUTE|TRANSFER queue family for graphics and the COMPUTE|TRANSFER queue family for computing. I guess the following treatment works for my purpose as well as produces a fallback. int i = 0; for (const auto& queueFamily : queue_families) { if (queueFamily.queueFlags & VK_QUEUE_TRANSFER_BIT) { indices.transfer_family = i; } if (queueFamily.queueFlags & VK_QUEUE_COMPUTE_BIT) { indices.compute_family = i; } if (queueFamily.queueFlags & VK_QUEUE_GRAPHICS_BIT) { indices.graphics_family = i; } if (indices.is_complete()) { break; } i++; } I don't know your queue family information. Most of them are available on https://vulkan.gpuinfo.org. With that information, there may be better ways to handle your case. — Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA4KKJYY3EBSFA53TVPIUE3VW7S6BANCNFSM55G5LHHA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[hooyuser]

I have updated the logic for queue family selection. You can pull the latest changes and check whether the fallback works as I expected.