When calling vkEnumeratePhysicalDevices it ends up failing at the ioctl call within anv_physical_device_init. It correctly opens the device /dev/dri/renderD128, but then fails with invalid argument when it tries to get the chipset id.
This is run on a i7-3630QM with a HD4000. There is also a GTX 675MX present, but it's not used.
Im running an updated Arch-Linux system (23 May 2016)
I build mesa with debug enabled and used the intel_vulkan library, but when trying to run any vulkan application (like vulkaninfo) it fails with
anv_device.c:77: failed to get chipset id: Invalid argument (VK_ERROR_INITIALIZATION_FAILED)
Can you try it with the wip/anv-multi-device branch on my freedesktop.org:
It has a couple of patches that *should* let it work even if there are multiple GPUs in the system.
Running it with J. Ekstrand git branch:
I just get this warning
WARNING: Ivy Bridge Vulkan support is incomplete
The cube and triangle demo works, but when running vulkaninfo
vulkaninfo: anv_formats.c:268: anv_get_format: Assertion `vk_format_aspects(vk_format) == VK_IMAGE_ASPECT_COLOR_BIT' failed.
Although I think this is now a seperate issue.
(In reply to Paul le Roux from comment #2)
> Running it with J. Ekstrand git branch:
> I just get this warning
> WARNING: Ivy Bridge Vulkan support is incomplete
> The cube and triangle demo works, but when running vulkaninfo
Cool. I just pushed the multi-gpu patches.
> vulkaninfo: anv_formats.c:268: anv_get_format: Assertion
> `vk_format_aspects(vk_format) == VK_IMAGE_ASPECT_COLOR_BIT' failed.
> Although I think this is now a seperate issue.
Yes, and I just fixed it! I'm closing this bug now. If you continue to have issues, please file them under a new bug.