OpenCL device not recognized inside Conda environment

[yipinghuang1991]

I have a conda environment with Julia from conda-forge.

Julia Version 0.6.1
Platform Info:
  OS: Linux (x86_64-redhat-linux)
  CPU: Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas.so
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, ivybridge)

If I do cl.platforms(), it says
CLError(code=-1001, CL_PLATFORM_NOT_FOUND_KHR)

But the Julia installed from official Arch Repo does work.
OpenCL.Platform('AMD Accelerated Parallel Processing' @0x00007fcc83edd098)

Julia Version 0.6.1
Commit 0d7248e* (2017-10-24 22:15 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas
  LIBM: libm
  LLVM: libLLVM-3.9.1 (ORCJIT, ivybridge)

My clinfo reads:

Number of platforms                               1
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.0 AMD-APP (2482.3)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback cl_amd_offline_devices 
  Platform Extensions function suffix             AMD

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 1
  Device Name                                     Ellesmere
  Device Vendor                                   Advanced Micro Devices, Inc.
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 1.2 AMD-APP (2482.3)
  Driver Version                                  2482.3
  Device OpenCL C Version                         OpenCL C 1.2 
  Device Type                                     GPU
  Device Available                                Yes
  Device Profile                                  FULL_PROFILE
  Device Board Name (AMD)                         AMD Radeon (TM) RX 480 Graphics
  Device Topology (AMD)                           PCI-E, 02:00.0
  Max compute units                               36
  SIMD per compute unit (AMD)                     4
  SIMD width (AMD)                                16
  SIMD instruction width (AMD)                    1
  Max clock frequency                             1303MHz
  Graphics IP (AMD)                               8.0
  Device Partition                                (core)
    Max number of sub-devices                     36
    Supported partition types                     none specified
  Max work item dimensions                        3
  Max work item sizes                             256x256x256
  Max work group size                             256
  Compiler Available                              Yes
  Linker Available                                Yes
  Preferred work group size multiple              64
  Wavefront width (AMD)                           64
  Preferred / native vector sizes                 
    char                                                 4 / 4       
    short                                                2 / 2       
    int                                                  1 / 1       
    long                                                 1 / 1       
    half                                                 1 / 1        (cl_khr_fp16)
    float                                                1 / 1       
    double                                               1 / 1        (cl_khr_fp64)
  Half-precision Floating-point support           (cl_khr_fp16)
    Denormals                                     No
    Infinity and NANs                             No
    Round to nearest                              No
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Single-precision Floating-point support         (core)
    Denormals                                     No
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  Yes
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Address bits                                    64, Little-Endian
  Global memory size                              8479633408 (7.897GiB)
  Global free memory (AMD)                        8259764 (7.877GiB)
  Global memory channels (AMD)                    8
  Global memory banks per channel (AMD)           16
  Global memory bank width (AMD)                  256 bytes
  Error Correction support                        No
  Max memory allocation                           4244635648 (3.953GiB)
  Unified memory for Host and Device              No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       2048 bits (256 bytes)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        16384 (16KiB)
  Global Memory cache line                        64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             16
    Max size for 1D images from buffer            134217728 pixels
    Max 1D or 2D image array size                 2048 images
    Base address alignment for 2D image buffers   256 bytes
    Pitch alignment for 2D image buffers          256 bytes
    Max 2D image size                             16384x16384 pixels
    Max 3D image size                             2048x2048x2048 pixels
    Max number of read image args                 128
    Max number of write image args                8
  Local memory type                               Local
  Local memory size                               32768 (32KiB)
  Local memory syze per CU (AMD)                  65536 (64KiB)
  Local memory banks (AMD)                        32
  Max constant buffer size                        4244635648 (3.953GiB)
  Max number of constant args                     8
  Max size of kernel argument                     1024
  Queue properties                                
    Out-of-order execution                        No
    Profiling                                     Yes
  Prefer user sync for interop                    Yes
  Profiling timer resolution                      1ns
  Profiling timer offset since Epoch (AMD)        1512831448530689299ns (Sat Dec  9 22:57:28 2017)
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
    Thread trace supported (AMD)                  Yes
    SPIR versions                                 1.2
  printf() buffer size                            1048576 (1024KiB)
  Built-in kernels                                
  Device Extensions                               cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_gl_sharing cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_media_ops2 cl_amd_popcnt cl_khr_image2d_from_buffer cl_khr_spir cl_khr_gl_event 

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  AMD Accelerated Parallel Processing
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   Success [AMD]
  clCreateContext(NULL, ...) [default]            Success [AMD]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   Ellesmere
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   Ellesmere

ICD loader properties
  ICD loader Name                                 OpenCL ICD Loader
  ICD loader Vendor                               OCL Icd free software
  ICD loader Version                              2.2.11
  ICD loader Profile                              OpenCL 2.1

[znmeb]

@Yiping-Huang I have scripts that build all of this on Arch - I too am using Julia from Arch repositories. It turns out you need the opencl-amd package and some kernel variables with the latest kernels. See https://wiki.archlinux.org/index.php/AMDGPU for the setup. The open source "Clover" version of OpenCL doesn't work.

This should also work with the Julia binary tarball from the Julia download page and with Julia built from source.

[yipinghuang1991]

@znmeb That's exactly what I did to make OpenCL working.
The problem is that the specific version of Julia inside my miniconda environment doesn't work.

And I managed to solve the problem!

The package ocl-icd (as a dependent of packages clblas and clfft) in my miniconda environment overrides the ocl-icd of my system's.

Creating a symlink of system's libOpenCL.so.1.0.0 works.
ln -sf /usr/lib/libOpenCL.so.1.0.0 /home/yp/miniconda3/envs/py36/lib/libOpenCL.so.1.0.0

And I think editting ocl-icd.pc inside conda environment should work
/home/yp/miniconda3/envs/py36/lib/pkgconfig/ocl-icd.pc

prefix=/home/yp/miniconda3/envs/py36
exec_prefix=${prefix}
libdir=${exec_prefix}/lib    <==Change this to /lib
includedir=${prefix}/include

Name: ocl-icd
Description: Open Computing Language generic Installable Client Driver support
Version: 2.2.9
Libs: -L${libdir}
Cflags: -I${includedir}

[vchuravy]

Yeah I don't think there is anything we can do about conda clobbering the systems ocl-icd

[znmeb]

Is Anaconda on GitHub? Maybe they need an issue for this? Seems like their Julia should work with OpenCL if they support Julia and OpenCL.

[yipinghuang1991]

It seems to be the default behavior.
conda-forge/ocl-icd-feedstock#5
We should symlink *.icds from /etc/OpenCL/vendors/ to $ENV_DIR/etc/OpenCL/vendors/.

[ToucheSir]

For anyone who runs into this issue like I did, conda install ocl-icd-system will expose the system ICD loader without having to manually symlink libraries and such.

[mjdiff]

For me, conda install ocl-icd-system does not work. Still after activating env devices available there are not visible for the OpenCL extension.