Forums - How to use GPU runtime on QCS605

3 posts / 0 new
or Register
Last post
How to use GPU runtime on QCS605
caoducquan
Join Date: 2 Oct 18
Posts: 6
Posted: Wed, 2019-03-20 18:15
Top

Hello SNPE team,

I am trying snpe-1.23.1.245 on Linux Embedded OS base on QCS605 SoC.

I tried the native example:
snpe-1.23.1.245/examples/NativeCpp/SampleCode
 
with aarch64-oe-linux-gcc6.4 toolchain
make -f Makefile.aarch64-oe-linux-gcc6.4
 
However, when I run that sample binary on target, it could not use GPU
# snpe-sample -b ITENSOR -d ../dlc/bvlc_alexnet.dlc -i target_raw_list.txt -o output_sample -r gpusnpe-sample -b ITENSOR -d ../dlc/bvlc_alexnet.dlc -i target_raw_list.txt -o output_sample -r gpu
SNPE Version: 1.23.1.245
Selected runtime not present. Falling back to CPU.
Batch size for the container is 1
Processing DNN Input: cropped/plastic_cup.raw
Processing DNN Input: cropped/trash_bin.raw
Processing DNN Input: cropped/notice_sign.raw
Processing DNN Input: cropped/handicap_sign.raw
Processing DNN Input: cropped/chairs.raw
 
It looks like SNPE GPU runtime based on openCL and I checked that openCL is available on this OS by porting the clinfo
# /data/bin/clinfo -a
Platform #0
  Name:                                  Snapdragon(TM)
  Vendor:                                QUALCOMM
  Version:                               OpenCL 2.0 QUALCOMM build: commit #c3dd282 changeid # Date: 01/24/19 Thu Local Branch:  Remote Branch: 
  Profile:                               FULL_PROFILE
  Extensions:                            (null)
 
  Device #0
    Name:                                QUALCOMM Adreno(TM)
    Type:                                GPU
    Vendor:                              QUALCOMM
    Vendor ID:                           3209509963
    Profile:                             FULL_PROFILE
    Available:                           Yes
    Version:                             OpenCL 2.0 Adreno(TM) 615
    Driver version:                      OpenCL 2.0 QUALCOMM build: commit #c3dd282 changeid # Date: 01/24/19 Thu Local Branch:  Remote Branch:  Compiler E031.36.04.01
    Compiler available:                  Yes
    Address space size:                  64
    Little endian:                       Yes
    Error correction support:            No
    Address alignment (bits):            1024
    Smallest alignment (bytes):          128
    Resolution of timer (ns):            1000
    Max clock frequency (MHz):           1
    Max compute units:                   1
    Max constant args:                   8
    Max constant buffer size:            64 kB 
    Max mem alloc size:                  256 MB 
    Max parameter size:                  1024
    Command-queue supported props:       Out of order execution
                                         Profiling
    Execution capabilities:              OpenCL kernels
    Global memory size:                  1 GB 
    Global memory cache size:            64 kB 
    Global memory line cache size:       64
    Local memory size:                   32 kB 
    Local memory type:                   Local
    Global memory cache type:            Read write
    Max work group size:                 1024
    Max work item dimensions:            3
    Max work item sizes:                 (1024, 1024, 1024)
    Image support:                       Yes
    Max 2D image height:                 16384
    Max 2D image width:                  16384
    Max 3D image depth:                  2048
    Max 3D image height:                 16384
    Max 3D image width:                  16384
    Max read image args:                 128
    Max write image args:                64
    Max samplers:                        16
    Preferred vector width char:         1
    Preferred vector width short:        1
    Preferred vector width int:          1
    Preferred vector width long:         1
    Preferred vector width float:        1
    Preferred vector width double:       0
    Half precision float capability:     Inf and NaNs
                                         Round to nearest even rounding mode
                                         Round to +ve and -ve infinity rounding modes
    Single precision float capability:   Inf and NaNs
                                         Round to nearest even rounding mode
                                         Round to +ve and -ve infinity rounding modes
    Double precision float capability:   Not supported
    Extensions:                          cl_khr_3d_image_writes
                                         cl_img_egl_image
                                         cl_khr_byte_addressable_store
                                         cl_khr_depth_images
                                         cl_khr_egl_event
                                         cl_khr_egl_image
                                         cl_khr_fp16
                                         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_image2d_from_buffer
                                         cl_khr_mipmap_image
                                         cl_khr_srgb_image_writes
                                         cl_khr_subgroups
                                         cl_qcom_create_buffer_from_image
                                         cl_qcom_ext_host_ptr
                                         cl_qcom_ion_host_ptr
                                         cl_qcom_perf_hint
                                         cl_qcom_other_image
                                         cl_qcom_subgroup_shuffle
                                         cl_qcom_vector_image_ops
                                         cl_qcom_extract_image_plane
                                         cl_qcom_protected_context
                                         cl_qcom_priority_hint
                                         cl_qcom_compressed_yuv_image_read
                                         cl_qcom_compressed_image
                                         cl_qcom_ext_host_ptr_iocoherent
                                         cl_qcom_accelerated_image_ops
 
Could you please guide me how to enable GPU runtime?
 
Thanks and best regards,
CDQ

 

  • Up0
  • Down0
Re: How to use GPU runtime on QCS605
caoducquan
Join Date: 2 Oct 18
Posts: 6
Posted: Thu, 2019-03-21 06:43
Top

From logcat, I got this warning

03-08 07:43:01.565  3266  3266 E Adreno-CB: <clGetPlatformIDs:2399>: Fatal: Failed to open libCB from libOpenCL
03-08 07:43:03.469  3266  3266 E Adreno-CB: <clGetPlatformIDs:2399>: Fatal: Failed to open libCB from libOpenCL
 

But I have libCB.so in /usr/lib64/libCB.so

 

  • Up0
  • Down0
Re: How to use GPU runtime on QCS605
caoducquan
Join Date: 2 Oct 18
Posts: 6
Posted: Thu, 2019-03-21 20:41
Top

Looks like there is a problem with dl loader. I found a workaround solution by init the cl runtime first before using SNPE



int clinit(void)
{
    // Get platform and device information
    cl_platform_id platform_id = NULL;
    cl_device_id device_id = NULL;
    cl_uint ret_num_devices;
    cl_uint ret_num_platforms;
    cl_int ret = clGetPlatformIDs(1, &platform_id, &ret_num_platforms);
    //FORCE CPU running
    //ret = clGetDeviceIDs( platform_id, CL_DEVICE_TYPE_ALL, 1,
    ret = clGetDeviceIDs( platform_id, CL_DEVICE_TYPE_GPU, 1,
            &device_id, &ret_num_devices);
    if (ret != CL_SUCCESS) {
      printf("ERROR: fail to obtain GPU runtine\n");
      return -1;
    } else {
      printf("GPU runtime inited\n");
    }
    return 0;
}
just call clinit before start using SNPE API then it works.
  • Up0
  • Down0
 or Register
Opinions expressed in the content posted here are the personal opinions of the original authors, and do not necessarily reflect those of Qualcomm Incorporated or its subsidiaries (“Qualcomm”). The content is provided for informational purposes only and is not meant to be an endorsement or representation by Qualcomm or any other party. This site may also provide links or references to non-Qualcomm sites and resources. Qualcomm makes no representations, warranties, or other commitments whatsoever about any non-Qualcomm sites or third-party resources that may be referenced, accessible from, or linked to this site.