CPU ID does not like rocm-opencl-runtime

[RushingAlien]

Describe the bug/Expected behavior

cpu-x: /home/main-builder/pkgwork/src/ROCclr-rocm-5.2.3/device/rocm/rocdevice.cpp:1132: bool roc::Device::populateOCLDeviceConstants(): Assertion `cachesize[0] > 0' failed.

For CPU ID to grab OpenCL device

Additional information

• Operating system name and version: Arch Linux
• CPU-X installation type: AUR

Bug information

========================= Backtrace =========================
CPU-X 4.4.0 (Sep 26 2022 02:48:24, Linux x86_64, GNU 12.2.0)
# 1 /usr/lib/libc.so.6(+0x38a00) [0x7f85d1ff4a00]
# 2 /usr/lib/libc.so.6(+0x8849c) [0x7f85d204449c]
# 3 /usr/lib/libc.so.6(gsignal+0x18) [0x7f85d1ff4958]
# 4 /usr/lib/libc.so.6(abort+0xd7) [0x7f85d1fde53d]
# 5 /usr/lib/libc.so.6(+0x2245c) [0x7f85d1fde45c]
# 6 /usr/lib/libc.so.6(+0x31486) [0x7f85d1fed486]
# 7 /opt/rocm/lib/libamdocl64.so(+0xf5d24) [0x7f85b051ed24]
# 8 /opt/rocm/lib/libamdocl64.so(+0xfa217) [0x7f85b0523217]
# 9 /opt/rocm/lib/libamdocl64.so(+0xfab73) [0x7f85b0523b73]
#10 /opt/rocm/lib/libamdocl64.so(+0xb4176) [0x7f85b04dd176]
#11 /opt/rocm/lib/libamdocl64.so(+0xea3d0) [0x7f85b05133d0]
#12 /opt/rocm/lib/libamdocl64.so(+0x8d736) [0x7f85b04b6736]
#13 /usr/lib/libc.so.6(+0x8b6d7) [0x7f85d20476d7]
#14 /opt/rocm/lib/libamdocl64.so(clIcdGetPlatformIDsKHR+0xa0) [0x7f85b04b7760]
#15 /opt/rocm/lib/libOpenCL.so.1(+0x31cc) [0x7f85d21e11cc]
======================== End Backtrace =======================

cpu-x.log
cpu-x-daemon.log

[Umio-Yasuno]

It may be a problem with the OpenCL runtime.
Do clinfo and rocminfo work correctly?

[RushingAlien]

clinfo reported an error that core is dumped, same error that appears when launching cpu-x, however, rocminfo does work correctly

[Umio-Yasuno]

If the error occurs in clinfo, I think it is a problem on the runtime side.

ROCm does not officially support Raven APU.
Does the same problem occur with AMDGPU-PRO driver?

Note: AMD_LOG_LEVEL=2 clinfo may give a more detailed log.

ref: https://github.com/ROCm-Developer-Tools/ROCclr/blob/rocm-5.2.3/device/rocm/rocdevice.cpp#L1132
ref: https://github.com/ROCm-Developer-Tools/ROCclr/blob/develop/utils/flags.hpp
Related: rocm-arch/rocm-arch#670

[TheTumultuousUnicornOfDarkness]

CPU-X crashes due to a SIGABRT caused somewhere after clGetPlatformIDs call. In that case, it seems to be a bug with ROCm.
I was not able to reproduce this bug with ROCm 5.2.3 on my AMD Radeon RX 580.
However, it seems reasonable to me to handle this signal in that case and try to recover. C language does not have a proper way to handle exceptions unfortunately, so let's say 3ff4a60 is a workaround.

It would produce something like this without crashing the app:

...
find_gpu_device_path: device_path=/sys/bus/pci/devices/0000:26:00.0
Vulkan instance extensions count: 20
Found instance extension: VK_KHR_get_physical_device_properties2
Found instance extension: VK_KHR_portability_enumeration
Vulkan devices count: 1

Oops, something was wrong! CPU-X has received signal 6 (Aborted) and is trying to recover.
========================= Backtrace =========================
CPU-X 4.4.0+git-r20-g5da0020eab12ac33178c682147fd58dfdfbebee1-dirty (Oct  1 2022 12:22:11, Linux x86_64, GNU 12.2.0)
# 1 main.c:720 cpu-x() [0x40d117]
# 2 /usr/lib/libc.so.6(+0x38a00) [0x7f430014da00]
# 3 /usr/lib/libc.so.6(kill+0xb) [0x7f430014dc3b]
# 4 core.c:1202 cpu-x(test_crash+0x18) [0x41528a]
# 5 core.c:1220 cpu-x() [0x41553b]
# 6 core.c:1461 cpu-x() [0x416f64]
# 7 core.c:111 cpu-x(fill_labels+0x76) [0x410499]
# 8 main.c:856 cpu-x(main+0x277) [0x40d4c6]
# 9 /usr/lib/libc.so.6(+0x23290) [0x7f4300138290]
#10 /usr/lib/libc.so.6(__libc_start_main+0x8a) [0x7f430013834a]
#11 start.S:117 cpu-x(_start+0x25) [0x409325]
======================== End Backtrace =======================

There is no platform with OpenCL support (SIGABRT)
request_sensor_path(base_dir=/sys/class/hwmon, cached_path=/sys/class/hwmon/hwmon0/temp1_input, which=0) ==> 0
request_sensor_path(base_dir=/sys/bus/pci/devices/0000:26:00.0/hwmon, cached_path=/sys/bus/pci/devices/0000:26:00.0/hwmon/hwmon2, which=5) ==> 0
request_sensor_path(base_dir=/sys/bus/pci/devices/0000:26:00.0/drm, cached_path=/sys/bus/pci/devices/0000:26:00.0/drm/card0, which=4) ==> 0
...