Closed OursDesCavernes closed 7 years ago
For the first one, it doesnt look very serious, just an extra filename line. I'm not sure where that comes from. I grepped the repository for source_filename
, but didnt find it. I'm wondering if it's something like, you're using a slightly different version of llvm somehow? Do you have some way of confirming which versoin of llvm you are using?
(the second one is trickier. Unfortunatley I dont have access to a Radeon; you will basically need to poke around, to find out how far the 123.0f is getting. It should be being set by the kernel, https://github.com/hughperkins/cuda-on-cl/blob/master/test/gtest/test_hostside_opencl_funcs.cpp#L34 This should be a relatively straightforward test, nothing too weird happening. The test is pretty much written in OpenCL, no CUDA involved, so should be fairly easy to poke around. clenqueuereadbuffer is here: https://www.khronos.org/registry/OpenCL/sdk/1.0/docs/man/xhtml/clEnqueueReadBuffer.html
Oh, it might be that you need to pass the queue into the run_1d
, like run_1d(v->currentContext->default_stream.get()->clqueue->queue, 32, 32)
(maybe wit ha *
or a &
in front of that queue expression possibly. I dont know if that will fix it, but it might.
exact llvm version: sys-devel/llvm 3.9.1-r1 Options:
# equery u sys-devel/llvm
[ Legend : U - final flag setting for installation]
[ : I - package is installed with flag ]
[ Colors : set, unset ]
* Found these USE flags for sys-devel/llvm-3.9.1-r1:
U I
- - abi_x86_32 : 32-bit (x86) libraries
+ + clang : Build the clang C/C++ compiler
- - debug : Enable extra debug codepaths, like asserts and extra output. If you want to get meaningful backtraces see
https://wiki.gentoo.org/wiki/Project:Quality_Assurance/Backtraces
- - default-compiler-rt : Use compiler-rt instead of libgcc as the default rtlib for clang
- - default-libcxx : Use libc++ instead of libstdc++ as the default stdlib for clang
- - doc : Build and install the HTML documentation and regenerate the man pages
- - gold : Build the gold linker plugin
- - libedit : Use the libedit library (replacement for readline)
+ + libffi : Enable support for Foreign Function Interface library
- - llvm_targets_AArch64 : AArch64 CPU target (arm64 in Gentoo)
+ + llvm_targets_AMDGPU : AMDGPU target (supports R600 and GCN GPUs)
- - llvm_targets_ARM : ARM CPU target
- - llvm_targets_BPF : Berkeley Packet Filter target
- - llvm_targets_Hexagon : Qualcomm Hexagon DSP target
- - llvm_targets_MSP430 : MSP430 CPU target (experimental)
- - llvm_targets_Mips : MIPS CPU target (includes MIPS64)
- - llvm_targets_NVPTX : NVIDIA PTX (GPU) target (32-bit and 64-bit)
- - llvm_targets_PowerPC : PowerPC CPU target (PPC32 and PPC64)
- - llvm_targets_Sparc : Sparc CPU target
- - llvm_targets_SystemZ : SystemZ (s390x) CPU target
- - llvm_targets_XCore : XCore CPU target
- - multitarget : Build all host targets (default: host only)
+ + ncurses : Support querying terminal properties using ncurses' terminfo
- - ocaml : Add support/bindings for the Ocaml language
+ + python : Add optional support/bindings for the Python language
+ + python_targets_python2_7 : Build with Python 2.7
+ + sanitize : Build compiler-rt's sanitizers
+ + static-analyzer : Install the Clang static analyzer (requires USE=clang)
- - test : Workaround to pull in packages needed to run with FEATURES=test. Portage-2.1.2 handles this internally, so don't set it in
make.conf/package.use anymore
+ + xml : Add support for XML files
(+ + means option is enabled and curently installed, - - is the oposite) Does it help you in any way ?
I'm trying to get the same tests running on beignet, I can't get opencl working with it, nothing to do with cuda-on-cl.
Please use llvm 3.8
test_hostside_opencl_funcs fixed, in c9d684c. Piccie on radeon and hd:
(and for the other one, its a trivial issue you can ignore; but if you really want to fix it, the current supported fix is: install llvm-3.8 :-) )
Hello, I ran the unit test and 2 of them fail:
test_struct_cloner.test_clone
test_hostside_opencl_funcs.test_create_cl_kernel
System info:
kernel: 4.10.10-gentoo opencl: amdgpu-pro-opencl-16.60.1.379184 on top of amdgpu open source stack llvm: 3.9.1 harware: AMD RX480
clinfo output:
How would you read these results ?
Thanks, Thomas.