gfxrecon-replay crashes on a trace of vulkaninfo about gfxreconstruct HOT 6 CLOSED

I don't see a segfault, but I do see it failing in vkGetPhysicalDeviceSurfacePreentModesKHR on Fedora Core 28.

With this error:

[gfxrecon] ERROR - API call (vkGetPhysicalDeviceSurfacePresentModesKHR) returned value VK_INCOMPLETE that does not match return value from capture file VK_SUCCESS.

from gfxreconstruct.

I am assuming this is the pNext handle mapping issue, which may produce different results on different drivers.

from gfxreconstruct.

I'm wondering the same.

from gfxreconstruct.

It looks like what I am seeing is entirely related to the VK_KHR_wayland_surface extension. If I remove only that extension from the list, it works for me. Must be something to do with the handle logic.

(Updated with more info I tracked down)

It appears to be failing on the replay on the vkGetPhysicalDeviceSurfacePresentModesKHR call for the Wayland window surface. So, maybe the Wayland surface isn't properly returned by the tools vkCreateWaylandSurfaceKHR or passed into vkGetPhysicalDeviceSurfacePresentModesKHR.

from gfxreconstruct.

Ok, so my problem is that VulkanInfo is recording surface presentation information under Wayland during the recording. [Actually it records the info for Xlib, XCB and Wayland]. When it records the Wayland information, it only gets 2 surface present modes. So the first call to vkGetPhysicalDeviceSurfacePresentModesKHR returns 2, then VulkanInfo creates an array of 2 and fills in the contents.

On playback, the replay tool is running each of those surface queries through it's XCB surface which has more than 2 present modes. So when vkGetPhysicalDeviceSurfacePresentModesKHR is called the first time, it returns 4. The second time vkGetPhysicalDeviceSurfacePresentModesKHR with an array of size 2, vkGetPhysicalDeviceSurfacePresentModesKHR returns VK_INCOMPLETE.

from gfxreconstruct.

The crash is happening because replay does not set the sType and pNext values for the VkQueueFamilyProperties2 structures that are used when calling GetPhysicalDeviceQueueFamilyProperties2.

Should be fixed by #196

from gfxreconstruct.