[SOLVED] vulkan using llvmpipe instead of amdgpu on slackware64-current

sasha69 · 12-06-2020, 11:54 AM

At some point this was working, but now when I start vkquake it's not using my GPU and the performance is terrible. With R9 390. opengl works fine.

I know it's using amdgpu instead of radeon, this is my kernel parameters:
root=/dev/mapper/myvg-root radeon.cik_support=0 radeon.si_support=0 amdgpu.cik_support=1 amdgpu.si_support=1

and in lspci:

Code:

03:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Hawaii PRO [Radeon R9 290/390] (rev 80)
	Subsystem: Micro-Star International Co., Ltd. [MSI] Hawaii PRO [Radeon R9 290/390]
	Kernel driver in use: amdgpu
	Kernel modules: radeon, amdgpu

This is what glxinfo | grep OpenGL shows:

Code:

blanc% glxinfo | grep OpenGL
OpenGL vendor string: AMD
OpenGL renderer string: AMD Radeon (TM) R9 390 Series (HAWAII, DRM 3.35.0, 5.4.81-rt44, LLVM 11.0.0)
OpenGL core profile version string: 4.6 (Core Profile) Mesa 20.3.0
OpenGL core profile shading language version string: 4.60
OpenGL core profile context flags: (none)
OpenGL core profile profile mask: core profile
OpenGL core profile extensions:
OpenGL version string: 4.6 (Compatibility Profile) Mesa 20.3.0
OpenGL shading language version string: 4.60
OpenGL context flags: (none)
OpenGL profile mask: compatibility profile
OpenGL extensions:
OpenGL ES profile version string: OpenGL ES 3.2 Mesa 20.3.0
OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.20
OpenGL ES profile extensions:

I get these errors:

Code:

ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_radeon.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_intel.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
WARNING: radv is not a conformant vulkan implementation, testing use only.

sasha69 · 12-06-2020, 12:03 PM

I downgraded to mesa 20.2.3 and that fixed it.

RadicalDreamer · 12-06-2020, 01:14 PM

It did the same with my NVIDIA card. Somehow vkquake needs to be told to use our video cards instead of llvmpipe. I suspect there could be a console command for this in vkquake.

I tried rebuilding vkquake and it failed.

Code:

gl_rmisc.c:1429:32: error: ‘VK_DYNAMIC_STATE_RANGE_SIZE’ undeclared (first use in this function); did you mean ‘VK_DYNAMIC_STATE_LINE_WIDTH’?
 1429 |  VkDynamicState dynamic_states[VK_DYNAMIC_STATE_RANGE_SIZE];
      |                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~
      |                                VK_DYNAMIC_STATE_LINE_WIDTH
gl_rmisc.c:1429:32: note: each undeclared identifier is reported only once for each function it appears in
make: *** [Makefile:200: gl_rmisc.o] Error 1
make: Leaving directory '/tmp/SBo/vkQuake-1.04.1/Quake'

I doubt rebuilding would fix the issue though.

Daedra · 12-06-2020, 01:24 PM

Quote:

Originally Posted by RadicalDreamer

It did the same with my NVIDIA card. Somehow vkquake needs to be told to use our video cards instead of llvmpipe. I suspect there could be a console command for this in vkquake.

I tried rebuilding vkquake and it failed.

Code:

gl_rmisc.c:1429:32: error: ‘VK_DYNAMIC_STATE_RANGE_SIZE’ undeclared (first use in this function); did you mean ‘VK_DYNAMIC_STATE_LINE_WIDTH’?
 1429 |  VkDynamicState dynamic_states[VK_DYNAMIC_STATE_RANGE_SIZE];
      |                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~
      |                                VK_DYNAMIC_STATE_LINE_WIDTH
gl_rmisc.c:1429:32: note: each undeclared identifier is reported only once for each function it appears in
make: *** [Makefile:200: gl_rmisc.o] Error 1
make: Leaving directory '/tmp/SBo/vkQuake-1.04.1/Quake'

I doubt rebuilding would fix the issue though.

Try building 1.05.1, I forgot to submit that to SBo. Will do it later today.

sasha69 · 12-06-2020, 01:48 PM

Oh yeah I forgot to say I tried with vkquake 1.05.1 and still had the same issue with mesa 20.3.0. And that builds OK.

bassmadrigal · 12-06-2020, 02:53 PM

Quote:

Originally Posted by sasha69

I get these errors:

Code:

ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_radeon.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_intel.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
WARNING: radv is not a conformant vulkan implementation, testing use only.

Do you get these errors normally? Because those imply that it's trying to load the wrong arch (or maybe it's the right arch but wrong version). Have you verified you have the right arch installed for mesa (32bit vs 64bit)? If you are running 64bit, are you running multilib? If you are running multilib, did you upgrade the compat32 packages?

RadicalDreamer · 12-06-2020, 02:55 PM

Thanks Daedra!
Here is what I get with the NVIDIA binary driver:

Code:

bash-5.0$ vkquake 
Command line: ./vkquake
Found SDL version 2.0.12
Detected 8 CPUs.
Quake 1.09 (c) id Software
GLQuake 1.00 (c) id Software
FitzQuake 0.85 (c) John Fitzgibbons
FitzQuake SDL port (c) SleepwalkR, Baker
QuakeSpasm 0.93.2 (c) Ozkan Sezer, Eric Wasylishen & others
vkQuake 1.05.1 (c) Axel Gneiting & others
Host_Init
Playing registered version.
Console initialized.
UDP Initialized
Server using protocol 666 (FitzQuake)
Exe: 15:47:20 Dec  6 2020
256.0 megabyte heap

Vulkan Initialization
Using VK_KHR_get_surface_capabilities2
Using VK_KHR_get_physical_device_properties2
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
Vendor: Unknown (0x10005)
Device: llvmpipe (LLVM 11.0.0, 256 bits)

ERROR-OUT BEGIN


QUAKE ERROR: Couldn't find graphics queue

We need to tell it to use our graphics card and not the llvmpipe. I ran into the same situation with QMPlay2 but I could tell that to use my video card.

sasha69 · 12-06-2020, 03:21 PM

Quote:

Originally Posted by bassmadrigal

Do you get these errors normally? Because those imply that it's trying to load the wrong arch (or maybe it's the right arch but wrong version). Have you verified you have the right arch installed for mesa (32bit vs 64bit)? If you are running 64bit, are you running multilib? If you are running multilib, did you upgrade the compat32 packages?

I should clarify it's just "vulkaninfo" that gives those ELF class errors, not vkQuake. Yeah I'm running multilib, I have the compat32 package for mesa and the 64-bit one. I have everything upgraded, well except the mesa that I now downgraded.

I don't know what changed in the new mesa that would make it use llvmpipe but since I still get those errors with 20.2.3 and it works, it's maybe not related.

Oh and I only get the radv warning with mesa 20.2.3, not the lavapipe one.

I also don't get this one with 20.2.3 and 20.2.2 compat32:
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32

RadicalDreamer · 12-06-2020, 03:28 PM

Here is my vulkaninfo. It shows llvmpipe as GPU id = 0.

Code:

bash-5.0$ vulkaninfo | grep GPU
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_radeon.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_intel.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
GPU id : 0 (llvmpipe (LLVM 11.0.0, 256 bits)):
GPU id : 1 (GeForce GTX 1070):
GPU0:
GPU1:
        deviceType     = PHYSICAL_DEVICE_TYPE_DISCRETE_GPU

sasha69 · 12-06-2020, 03:33 PM

Quote:

Originally Posted by RadicalDreamer

Here is my vulkaninfo. It shows llvmpipe as GPU id = 0.

Code:

bash-5.0$ vulkaninfo | grep GPU
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_radeon.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_intel.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
                GPU id = 0 (llvmpipe (LLVM 11.0.0, 256 bits))
                GPU id = 1 (GeForce GTX 1070)
GPU id : 0 (llvmpipe (LLVM 11.0.0, 256 bits)):
GPU id : 1 (GeForce GTX 1070):
GPU0:
GPU1:
        deviceType     = PHYSICAL_DEVICE_TYPE_DISCRETE_GPU

Yeah, I get the same with the newest mesa.

RadicalDreamer · 12-06-2020, 03:52 PM

Quote:

Originally Posted by sasha69

Yeah, I get the same with the newest mesa.

My best guess is that it is choosing the first device that it sees so our video cards need to be "GPU id = 0" not "GPU id = 1."

sasha69 · 12-06-2020, 04:31 PM

this upstream bug looks like it might have something to do with it. https://gitlab.freedesktop.org/mesa/mesa/-/issues/3935

sasha69 · 12-24-2020, 09:21 AM

With 20.2.3 I actually got a different problem which was glx not working at all.

I had to compile 20.2.6 myself to get both glx and vulkan working.

teerl · 01-08-2021, 10:12 PM

Quote:

Originally Posted by sasha69

With 20.2.3 I actually got a different problem which was glx not working at all.

I had to compile 20.2.6 myself to get both glx and vulkan working.

Would you mind sharing your SlackBuild? I'm having a heck of a time trying to get mesa to build.

Edit:

I'm running nvidia-drivers and was finally able to get vkcube to run with

Code:

VK_ICD_FILENAMES=/etc/vulkan/icd.d/nvidia_icd.json vkcube

Various settings of MESA_VK_DEVICE_SELECT and DRI_PRIME produced no effect for me.

walecha · 01-09-2021, 04:00 AM

Quote:

Originally Posted by teerl

Various settings of MESA_VK_DEVICE_SELECT and DRI_PRIME produced no effect for me.

I finally got vulkan env ENABLE_DEVICE_CHOOSER_LAYER=1 VULKAN_DEVICE_INDEX=$GPU_ID and vkdevicechooser working by rebuilding vulkan-sdk with additional build option for Vulkan-ValidationLayer-sdk cmake:

https://www.linuxquestions.org/quest...ml#post6203697

Here is the output of vulkaninfo without VULKAN_DEVICE_INDEX:

Code:

[16:54:35] ~
 $ ENABLE_DEVICE_CHOOSER_LAYER=1 vulkaninfo | grep -e 'Devices: count' -e 'GPU id' | sort -u
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_intel.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_radeon.so: wrong ELF class: ELFCLASS32
WARNING: radv is not a conformant vulkan implementation, testing use only.
ATTENTION: default value of option mesa_glthread overridden by environment.
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
MESA-INTEL: warning: Haswell Vulkan support is incomplete
        Devices: count = 3
                GPU id = 0 (AMD RADV BONAIRE (ACO))
GPU id : 0 (AMD RADV BONAIRE (ACO)):
                GPU id = 1 (llvmpipe (LLVM 11.0.0, 256 bits))
GPU id : 1 (llvmpipe (LLVM 11.0.0, 256 bits)):
                GPU id = 2 (Intel(R) HD Graphics 4600 (HSW GT2))
GPU id : 2 (Intel(R) HD Graphics 4600 (HSW GT2)):
                physicalDevices: count = 1

vulkaninfo + env VULKAN_DEVICE_INDEX=0

Code:

[16:55:59] ~
 $ ENABLE_DEVICE_CHOOSER_LAYER=1 VULKAN_DEVICE_INDEX=0 vulkaninfo | grep -e 'Devices: count' -e 'GPU id' | sort -u
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_intel.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_radeon.so: wrong ELF class: ELFCLASS32
WARNING: radv is not a conformant vulkan implementation, testing use only.
ATTENTION: default value of option mesa_glthread overridden by environment.
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
MESA-INTEL: warning: Haswell Vulkan support is incomplete
        Devices: count = 1
                GPU id = 0 (AMD RADV BONAIRE (ACO))
GPU id : 0 (AMD RADV BONAIRE (ACO)):
                physicalDevices: count = 1

vulkaninfo + env VULKAN_DEVICE_INDEX=2

Code:

[16:58:53] ~
 $ ENABLE_DEVICE_CHOOSER_LAYER=1 VULKAN_DEVICE_INDEX=2 vulkaninfo | grep -e 'Devices: count' -e 'GPU id' | sort -u
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_lvp.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_intel.so: wrong ELF class: ELFCLASS32
ERROR: [Loader Message] Code 0 : /usr/lib/libvulkan_radeon.so: wrong ELF class: ELFCLASS32
WARNING: radv is not a conformant vulkan implementation, testing use only.
ATTENTION: default value of option mesa_glthread overridden by environment.
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
MESA-INTEL: warning: Haswell Vulkan support is incomplete
        Devices: count = 1
                GPU id = 0 (Intel(R) HD Graphics 4600 (HSW GT2))
GPU id : 0 (Intel(R) HD Graphics 4600 (HSW GT2)):
                physicalDevices: count = 1