You Can Help Bring Vulkan Support To Older AMD GCN GPUs

[Nille_kungen]

Someone needs a night sandwich so this craziness can stop.

[ronsaldo]

I started to port the SI driver from the radeon drm module into the AMDGPU module. So far I have already ported the code for some modules such as the MC, DCE, DMA and IH. Currently I am using a laptop with an integrated Intel graphics card and an Oland GPU for testing, and another machine for compiling. Right now I am trying to get the interruptions from the IH working, which are required to signal a fence that is used for testing IB on the DMA. Since this fence is not being signaled, I am having a deadlock on module load. I think that I should be getting this problem solved soon.

My biggest issue with porting the driver is that I am missing documentation. Specially some high level documentations on the IP blocks present in the graphics card, their versions and proper register information, and glossary with the many acronyms that are present. Currently, I am working by using the radeon driver as the documentation.

I will publish soon what I got so if there is more people interested on playing with this.

[haagch]

Very cool. Make sure that you contact AMD so that you don't duplicate work that they may or may not be internally already doing.

[ronsaldo]

Ok, I managed to get the interruptions working and the DMA. I was able to use memory related tests for amdgpu in libdrm, and they are passing. I had to modify the DMA tests in libdrm because the DMA present in SI uses a different packet format to the one present in CIK and newer graphics cards.

I haven't tested modesetting yet. I am still missing support for the graphics, uvd, and vcd ring buffer. Next I will add support for the gfx ring buffer, and I will try to test modesetting.

As for the code, I am pushing the driver into: https://github.com/ronsaldo/linux-amdgpu-si and the modified libdrm into https://github.com/ronsaldo/libdrm .

[haagch]

Hey, I saw that you made some more commits.
Is it a lot of work left? Can you realistically finish this soon by working all alone to the point where power management and 3d rendering work?

[bridgman]

He won't be working all alone. We've done a fair amount of work on this as well (as I said a few times already), but the developer's office was shut down for Chinese New Year holidays so we're just catching up with him now.

Our main focus was on the graphics core, so AFAIK we have OpenGL and OpenCL working-ish. One of the remaining open questions is what to do about display, looking into that now.

[ronsaldo]

I do not think that I can keep working alone. I just got stuck debugging mesa.
Currently I am on vacation with my family faraway, so I won't be working on this until the next week.

I have already ported the DMA, MC, GFX module and the DCE. I tested modeseting in my desktop computer with a HD 7770 graphics cards and the framebuffer console is already working. What is missing is the SMC module(I belive this is related with power management), the UVD and the VCE. What it is also missing is making mesa work. I was not able to run X11 because it uses glamor on my distro. Mesa is not working, it seems to hang the GPU when dispatching the draw array command. I am leaving my modifed version of mesa in this repo: https://github.com/ronsaldo/mesa and the modified ddx driver https://github.com/ronsaldo/xf86-video-amdgpu with the SI PCI IDs.

[bridgman]

Nice. We'll take a look while you're on vacation and sync up when you get back.