opencv: Segmentation fault with DNN_TARGET_OPENCL and DNN_TARGET_OPENCL_FP16 in custom OpenCL(VC4CL)

System information (version)

• OpenCV => Latest master(4.0.1-dev)
• Operating System / Platform => Raspbian Stretch
• Compiler => gcc-6.3.0 native

Detailed description

Hi, I’m experimenting with custom Implementation of OpenCL, VC4CL i.e. an implementation of the OpenCL 1.2 standard for the VideoCore IV GPU found on my Raspberry Pi. I’m experiencing segmentation fault when trying to utilize DNN_TARGET_OPENCL and DNN_TARGET_OPENCL_FP16 i.e. net.setPreferableTarget (to exploit its GPU for performance) with default yolov3(I need extra swap space to run it) prebuilt models to classify objects in an image. Here is the resultant error:(Similar error here)

pi@raspberrypi:~/yolo $ sudo python3 yolo3.py
load model time is  2.7082722187042236
OpenCV(ocl4dnn): consider to specify kernel configuration cache directory 
                 via OPENCV_OCL4DNN_CONFIG_PATH parameter.
OpenCL program build log: dnn/dummy
Status -15: CL_COMPILE_PROGRAM_FAILURE
-cl-no-subgroup-ifp
[E] Mon Feb 18 16:02:55 2019: Errors in precompilation:
[E] Mon Feb 18 16:02:55 2019: error: unknown argument: '-cl-no-subgroup-ifp'


OpenCL program build log: dnn/conv_layer_spatial
Status -15: CL_COMPILE_PROGRAM_FAILURE
 -D TYPE=1 -D Dtype=float -D Dtype2=float2 -D Dtype4=float4 -D Dtype8=float8 -D Dtype16=float16 -D as_Dtype=as_float -D as_Dtype2=as_float2 -D as_Dtype4=as_float4 -D as_Dtype8=as_float8 -D KERNEL_WIDTH=3 -D KERNEL_HEIGHT=3 -D STRIDE_X=1 -D STRIDE_Y=1 -D DILATION_X=1 -D DILATION_Y=1 -D KERNEL_BASIC -cl-fast-relaxed-math -D ConvolveBasic=BASIC_k3x3_cn3_g1_s1x1_d1x1_b1_in256x256_p1x1_num1_M16_activ1_eltwise0_FP32_4_1_1_1 -D CHANNELS=3 -D APPLY_BIAS=1 -D OUTPUT_Z=16 -D ZPAR=1 -D FUSED_CONV_RELU=1
[E] Mon Feb 18 16:02:55 2019: Errors in precompilation:
[E] Mon Feb 18 16:02:55 2019: error: PCH file uses a newer PCH format that cannot be read
1 error generated.


Failed to compile kernel: BASIC_k3x3_cn3_g1_s1x1_d1x1_b1_in256x256_p1x1_num1_M16_activ1_eltwise0_FP32_4_1_1_1, buildflags:  -D TYPE=1 -D Dtype=float -D Dtype2=float2 -D Dtype4=float4 -D Dtype8=float8 -D Dtype16=float16 -D as_Dtype=as_float -D as_Dtype2=as_float2 -D as_Dtype4=as_float4 -D as_Dtype8=as_float8 -D KERNEL_WIDTH=3 -D KERNEL_HEIGHT=3 -D STRIDE_X=1 -D STRIDE_Y=1 -D DILATION_X=1 -D DILATION_Y=1 -D KERNEL_BASIC -cl-fast-relaxed-math -D ConvolveBasic=BASIC_k3x3_cn3_g1_s1x1_d1x1_b1_in256x256_p1x1_num1_M16_activ1_eltwise0_FP32_4_1_1_1 -D CHANNELS=3 -D APPLY_BIAS=1 -D OUTPUT_Z=16 -D ZPAR=1 -D FUSED_CONV_RELU=1, errmsg: [E] Mon Feb 18 16:02:55 2019: Errors in precompilation:
[E] Mon Feb 18 16:02:55 2019: error: PCH file uses a newer PCH format that cannot be read
1 error generated.


Segmentation fault

and

Here is the clinfo output of VC4CL’sOpenCL Implementation on my Raspberry Pi 3B rev 1.2:

pi@raspberrypi:~/yolo $ sudo clinfo
Number of platforms                               1
  Platform Name                                   OpenCL for the Raspberry Pi VideoCore IV GPU
  Platform Vendor                                 doe300
  Platform Version                                OpenCL 1.2 VC4CL 0.4
  Platform Profile                                EMBEDDED_PROFILE
  Platform Extensions                             cl_khr_il_program cl_khr_spir cl_altera_device_temperature cl_altera_live_object_tracking cl_khr_icd cl_vc4cl_performance_counters
  Platform Extensions function suffix             VC4CL

  Platform Name                                   OpenCL for the Raspberry Pi VideoCore IV GPU
Number of devices                                 1
  Device Name                                     VideoCore IV GPU
  Device Vendor                                   Broadcom
  Device Vendor ID                                0xa5c
  Device Version                                  OpenCL 1.2 VC4CL 0.4
  Driver Version                                  0.4
  Device OpenCL C Version                         OpenCL C 1.2 
  Device Type                                     GPU
  Device Profile                                  EMBEDDED_PROFILE
  Max compute units                               1
  Max clock frequency                             350MHz
  Device Partition                                (core)
    Max number of sub-devices                     0
    Supported partition types                     None
  Max work item dimensions                        3
  Max work item sizes                             12x12x12
  Max work group size                             12
  Preferred work group size multiple              <getWGsizes:498: build program : error -15>
  Preferred / native vector sizes                 
    char                                                16 / 16      
    short                                               16 / 16      
    int                                                 16 / 16      
    long                                                 0 / 0       
    half                                                 0 / 0        (n/a)
    float                                               16 / 16      
    double                                               0 / 0        (n/a)
  Half-precision Floating-point support           (n/a)
  Single-precision Floating-point support         (core)
    Denormals                                     No
    Infinity and NANs                             No
    Round to nearest                              No
    Round to zero                                 Yes
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Double-precision Floating-point support         (n/a)
  Address bits                                    32, Little-Endian
  Global memory size                              335544320 (320MiB)
  Error Correction support                        No
  Max memory allocation                           335544320 (320MiB)
  Unified memory for Host and Device              Yes
  Minimum alignment for any data type             64 bytes
  Alignment of base address                       512 bits (64 bytes)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        <printDeviceInfo:89: get CL_DEVICE_GLOBAL_MEM_CACHE_SIZE : error -30>
  Global Memory cache line                        64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             64
    Max size for 1D images from buffer            2048 pixels
    Max 1D or 2D image array size                 2048 images
    Max 2D image size                             2048x2048 pixels
    Max 3D image size                             2048x2048x2048 pixels
    Max number of read image args                 64
    Max number of write image args                64
  Local memory type                               Global
  Local memory size                               335544320 (320MiB)
  Max constant buffer size                        335544320 (320MiB)
  Max number of constant args                     64
  Max size of kernel argument                     256
  Queue properties                                
    Out-of-order execution                        No
    Profiling                                     Yes
  Prefer user sync for interop                    Yes
  Profiling timer resolution                      1ns
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
  printf() buffer size                            0
  Built-in kernels                                
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                No
  Device Extensions                               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_byte_addressable_store cl_khr_3d_image_writes cl_intel_packed_yuv cl_nv_pragma_unroll cl_arm_core_id cl_ext_atomic_counters_32 cl_khr_initialize_memory cl_arm_integer_dot_product_int8 cl_arm_integer_dot_product_accumulate_int8 cl_arm_integer_dot_product_accumulate_int16

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  OpenCL for the Raspberry Pi VideoCore IV GPU
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   Success [VC4CL]
  clCreateContext(NULL, ...) [default]            Success [VC4CL]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  Success (1)
    Platform Name                                 OpenCL for the Raspberry Pi VideoCore IV GPU
    Device Name                                   VideoCore IV GPU
  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                                 OpenCL for the Raspberry Pi VideoCore IV GPU
    Device Name                                   VideoCore IV GPU

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

Steps to reproduce

I’m using following example: yolo.txt

Btw It runs fine with CPU backend. Any help regarding is highly appreciated.