ROCm: Blender 2.80 Cycles testing, compiling kernel hangs, ROCm 1.8.0
Hi all,
I’m just testing Blender Cycles from the bleeding edge branch as well as 2.79b with ROCm 1.8.0 OpenCL on a Vega10 Frontier Edition, running on a HP z620 Workstation (2x E5-2680v2’s, 64GB RAM, Quadro k2000 for display). Currently this is the debug I get from the command line using --debug-cycles:
Experimental features: Off
Max nodes group: 3
Nodes features: 6
Use Hair: True
Use Object Motion: False
Use Camera Motion: True
Use Baking: False
Use Subsurface: True
Use Volume: True
Use Branched Integrator: True
Use Patch Evaluation: False
Use Transparent Shadows: True
Use Principled BSDF: False
Use Denoising: False
I0505 14:02:50.857064 5927 opencl_base.cpp:216] Loading kernels for platform AMD Accelerated Parallel Processing, device Vega 10 XTX [Radeon Vega Frontier Edition].
I0505 14:02:50.857336 5927 opencl_util.cpp:288] OpenCL program split not found in cache.
I0505 14:02:50.897501 5927 opencl_util.cpp:288] Kernel file /home/alex/.cache/cycles/kernels/cycles_kernel_split_33D08D521DA35BC482B41E5979556C99_DBB1D123AFA735A3DDBE5D0BD2C271B9.clbin either doesn't exist or failed to be loaded by driver.
Compiling OpenCL program split
I0505 14:02:50.926681 5927 opencl_util.cpp:288] Build flags: -D__SPLIT_KERNEL__ -D__NODES_MAX_GROUP__=3 -D__NODES_FEATURES__=6 -D__NO_OBJECT_MOTION__ -D__NO_BAKING__ -D__NO_PATCH_EVAL__ -D__NO_SHADOW_TRICKS__ -D__NO_PRINCIPLED__ -D__NO_DENOISING__ -D__NO_SHADER_RAYTRACE__ -D__COMPUTE_DEVICE_GPU__
I0505 14:02:50.926702 5927 opencl_util.cpp:315] Build options passed to clBuildProgram: '-cl-no-signed-zeros -cl-mad-enable -D__KERNEL_OPENCL_AMD__ -D__SPLIT_KERNEL__ -D__NODES_MAX_GROUP__=3 -D__NODES_FEATURES__=6 -D__NO_OBJECT_MOTION__ -D__NO_BAKING__ -D__NO_PATCH_EVAL__ -D__NO_SHADOW_TRICKS__ -D__NO_PRINCIPLED__ -D__NO_DENOISING__ -D__NO_SHADER_RAYTRACE__ -D__COMPUTE_DEVICE_GPU__'.
Currently it looks like the compiler just hangs here for quite some time, I can let it run for awhile and see if it eventually compiles, but it seems like something to look into. I can try to help test here as best as I can to assist the ROCm developers.
Cheers!
About this issue
- Original URL
- State: closed
- Created 6 years ago
- Comments: 25
Just a monthly update: Performance of the just released ROCm 2.5 is still very bad with Blender
2.80-749d53effd58.@iszotic Thanks for this great tip! I installed only the ROCm
rocm-openclpackage and its immediate dependencies and then manually copiedlibamdocl64.soextracted from the Radeon Pro driver packageopencl-amdgpu-pro-icd_19.10-785425_amd64.debto/opt/rocm/opencl/lib/x86_64as you said. Now Blender runs very fast. Thank you so much!Any update on this?
Hi, Recent testing of the latest Blender 2.80 and ROCm 1.8.3, I’ve got a patch pending with the Blender/Cycles devs here: https://developer.blender.org/D3669
I’ve successfully compiled Blender/Cycles with this patch and the compiling of the kernel doesn’t complain about this previous missing pragma now, however it is displaying the same problem as Blender 2.79 now (hangs and just leaks memory slowly).
Cheers.
Hi @boberfly Thanks for reaching out. The issue mentioned in this ticket is no more now but now Blender is broken working and we have tickets to track this: #1210 and #1106 Request to track these. Thank you.
Thank you! That works perfectly for me with:
rocm-openclapt package (3.0) and ‘libamdocl64.so’ from ‘opencl-amdgpu-pro-icd_19.50-967956_amd64.deb’@boberfly I guess is more elaborate than just copying libamdocl64.so file from pro drivers and replace the one in opt/rocm/opencl/lib/x86_64, btw, rocm 2.5 compiles with low performance at rendering