cancel
Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for 
Search instead for 
Did you mean: 

OpenCL

acminor
Journeyman III
‎02-16-2021 12:10 PM

OpenCL Nested Parallelism Limits and Error Code

I was doing a simple test to determine the limits of OpenCL nested parallelism on an AMD RX5700XT.

 

Below is my code.

 

#define LOOP_SZ 1024
kernel void _test_loop(global float *gpu_buf, int k) {
  gpu_buf[k % 256] = k;
  printf("k: %d, LOOP_SZ: %d, k < LOOP_SZ, %d\n", k, LOOP_SZ, k < LOOP_SZ);

  if (k < LOOP_SZ) {
    queue_t q = get_default_queue();
    int err = enqueue_kernel(q, CLK_ENQUEUE_FLAGS_WAIT_KERNEL,
                   ndrange_1D(1),
                   ^{
                     _test_loop(gpu_buf, k+1);
                   });
    printf("Error: %d\n", err);

    switch(err) {
    case CLK_INVALID_QUEUE:
      break;
    case CLK_INVALID_NDRANGE:
      break;
    case CLK_INVALID_EVENT_WAIT_LIST:
      break;
    case CLK_DEVICE_QUEUE_FULL:
      printf("device queue\n");
      break;
    case CLK_INVALID_ARG_SIZE:
      break;
    case CLK_EVENT_ALLOCATION_FAILURE:
      break;
    case CLK_OUT_OF_RESOURCES:
      printf("resources\n");
      break;
    default:
      printf("other\n");
    }
  }
}

kernel void test_loop(global float *gpu_buf) {
  printf("k: %d\n", 0);
  queue_t q = get_default_queue();
  enqueue_kernel(q, CLK_ENQUEUE_FLAGS_WAIT_KERNEL,
                 ndrange_1D(1),
                 ^{
                   _test_loop(gpu_buf, 0);
                 });
}

 

I end up with a strange error code -101. I looked this up and could not find an OpenCL error code to match.

Has anyone encountered this OpenCL error code before. Does anyone know if there is a limit to nested parallelism on the RX5700XT.

0 Likes
8 Replies
acminor
Journeyman III
‎02-16-2021 02:31 PM

Also, this code would always fail at k=127 which is why I think it is a nested parallelism depth issue as 0-127 is a binary boundary. In this case, 2^7. This would make sense if nested parallelism depth was handled using an 8-bit number where one bit was a boolean flag.

0 Likes
acminor
In response to acminor
Journeyman III
‎02-16-2021 02:41 PM

I also seem to be running into the same issue. This time it is not about nested parallelism depth but about the amount queued.

 

kernel void _test_loop_2(global float *gpu_buf, int k);

kernel void test_loop_2(global float *gpu_buf) {
  queue_t q = get_default_queue();

  for (int i = 0; i < LOOP_SZ; i++) {
    _test_loop_2(gpu_buf, i);
  }
}

kernel void _test_loop_2_work(global float *gpu_buf, int k) {
  gpu_buf[k % 256] = k % 256;
}

kernel void _test_loop_2(global float *gpu_buf, int k) {
  printf("k: %d\n", k);

  queue_t q = get_default_queue();
  int err = enqueue_kernel(q, CLK_ENQUEUE_FLAGS_NO_WAIT,
                           ndrange_1D(1),
                           ^{
                             _test_loop_2_work(gpu_buf, k);
                           });
}

 

This will go through all k, but it will stop queuing kernels. I checked the queue size. Changing it led to no change in results. It will change the gpu_buf up to 127 and then stop. Adding a printf for err will show that at 128 through the end, it starts to error with the same -101 error code. This seems strange to me. My only thoughts is that this is related to a limit on the number of nested kernels that can be launch (not their depth). I have not found AMD documentation on this. Even so, 128 kernels seems too small a limit for nested parallelism.

0 Likes
dipak
In response to acminor
Big Boss
‎02-17-2021 01:51 AM

Hi @acminor ,

Thank you for reporting it. Please provide your setup details such as OS, driver version etc. and attach the clinfo output. 

Please also share the host-side code that was used to call the kernel.

 

Thanks.

0 Likes
acminor
In response to dipak
Journeyman III
‎02-17-2021 09:18 AM

According to clinfo, I am running the 3180.7 (PAL, LC) OpenCL driver (Platform version: OpenCL 2.1 AMD-APP (3180.7).

I am unsure if I am using a specific AMDGPU driver or the builtin kernel driver. I know that I installed the OpenCL components which is why I list that clinfo version.

I am running Ubuntu 20.04.1 with Linux kernel 5.6.15-050615-generic (which I believe is not the default Linux kernel).

If you need more information, let me know.

Number of platforms                               2
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.1 AMD-APP (3180.7)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback cl_amd_offline_devices 
  Platform Host timer resolution                  1ns
  Platform Extensions function suffix             AMD

  Platform Name                                   Portable Computing Language
  Platform Vendor                                 The pocl project
  Platform Version                                OpenCL 1.2 pocl 1.4, None+Asserts, LLVM 9.0.1, RELOC, SLEEF, DISTRO, POCL_DEBUG
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd
  Platform Extensions function suffix             POCL

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 1
  Device Name                                     gfx1010
  Device Vendor                                   Advanced Micro Devices, Inc.
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 2.0 AMD-APP (3180.7)
  Driver Version                                  3180.7 (PAL,LC)
  Device OpenCL C Version                         OpenCL C 2.0 
  Device Type                                     GPU
  Device Board Name (AMD)                         AMD Radeon RX 5700 XT
  Device Topology (AMD)                           PCI-E, 2f:00.0
  Device Profile                                  FULL_PROFILE
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Max compute units                               20
  SIMD per compute unit (AMD)                     4
  SIMD width (AMD)                                32
  SIMD instruction width (AMD)                    1
  Max clock frequency                             2100MHz
  Graphics IP (AMD)                               10.10
  Device Partition                                (core)
    Max number of sub-devices                     20
    Supported partition types                     None
    Supported affinity domains                    (n/a)
  Max work item dimensions                        3
  Max work item sizes                             1024x1024x1024
  Max work group size                             256
  Preferred work group size (AMD)                 256
  Max work group size (AMD)                       1024
  Preferred work group size multiple              32
  Wavefront width (AMD)                           32
  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
  Single-precision Floating-point support         (core)
    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  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
  Address bits                                    64, Little-Endian
  Global memory size                              8573157376 (7.984GiB)
  Global free memory (AMD)                        8306688 (7.922GiB)
  Global memory channels (AMD)                    8
  Global memory banks per channel (AMD)           4
  Global memory bank width (AMD)                  256 bytes
  Error Correction support                        No
  Max memory allocation                           7059013632 (6.574GiB)
  Unified memory for Host and Device              No
  Shared Virtual Memory (SVM) capabilities        (core)
    Coarse-grained buffer sharing                 Yes
    Fine-grained buffer sharing                   Yes
    Fine-grained system sharing                   No
    Atomics                                       No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       2048 bits (256 bytes)
  Preferred alignment for atomics                 
    SVM                                           0 bytes
    Global                                        0 bytes
    Local                                         0 bytes
  Max size for global variable                    6353112064 (5.917GiB)
  Preferred total size of global vars             8573157376 (7.984GiB)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        16384 (16KiB)
  Global Memory cache line size                   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 pixels
    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                64
    Max number of read/write image args           64
  Max number of pipe args                         16
  Max active pipe reservations                    16
  Max pipe packet size                            2764046336 (2.574GiB)
  Local memory type                               Local
  Local memory size                               65536 (64KiB)
  Local memory syze per CU (AMD)                  65536 (64KiB)
  Local memory banks (AMD)                        32
  Max number of constant args                     8
  Max constant buffer size                        7059013632 (6.574GiB)
  Preferred constant buffer size (AMD)            16384 (16KiB)
  Max size of kernel argument                     1024
  Queue properties (on host)                      
    Out-of-order execution                        No
    Profiling                                     Yes
  Queue properties (on device)                    
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Preferred size                                262144 (256KiB)
    Max size                                      8388608 (8MiB)
  Max queues on device                            1
  Max events on device                            1024
  Prefer user sync for interop                    Yes
  Number of P2P devices (AMD)                     0
  P2P devices (AMD)                               <printDeviceInfo:147: get number of CL_DEVICE_P2P_DEVICES_AMD : error -30>
  Profiling timer resolution                      1ns
  Profiling timer offset since Epoch (AMD)        1613489455036584008ns (Tue Feb 16 09:30:55 2021)
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
    Thread trace supported (AMD)                  Yes
    Number of async queues (AMD)                  4
    Max real-time compute queues (AMD)            1
    Max real-time compute units (AMD)             0
  printf() buffer size                            4194304 (4MiB)
  Built-in kernels                                (n/a)
  Device Extensions                               cl_khr_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_khr_gl_depth_images cl_amd_device_attribute_query cl_amd_media_ops cl_amd_media_ops2 cl_khr_image2d_from_buffer cl_khr_subgroups cl_khr_gl_event cl_khr_depth_images cl_khr_mipmap_image cl_khr_mipmap_image_writes cl_amd_copy_buffer_p2p 

  Platform Name                                   Portable Computing Language
Number of devices                                 1
  Device Name                                     pthread-AMD Ryzen 5 3600 6-Core Processor
  Device Vendor                                   AuthenticAMD
  Device Vendor ID                                0x6c636f70
  Device Version                                  OpenCL 1.2 pocl HSTR: pthread-x86_64-pc-linux-gnu-znver1
  Driver Version                                  1.4
  Device OpenCL C Version                         OpenCL C 1.2 pocl
  Device Type                                     CPU
  Device Profile                                  FULL_PROFILE
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Max compute units                               12
  Max clock frequency                             3600MHz
  Device Partition                                (core)
    Max number of sub-devices                     12
    Supported partition types                     equally, by counts
    Supported affinity domains                    (n/a)
  Max work item dimensions                        3
  Max work item sizes                             4096x4096x4096
  Max work group size                             4096
  Preferred work group size multiple              8
  Preferred / native vector sizes                 
    char                                                16 / 16      
    short                                               16 / 16      
    int                                                  8 / 8       
    long                                                 4 / 4       
    half                                                 0 / 0        (n/a)
    float                                                8 / 8       
    double                                               4 / 4        (cl_khr_fp64)
  Half-precision Floating-point support           (n/a)
  Single-precision Floating-point support         (core)
    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  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
  Address bits                                    64, Little-Endian
  Global memory size                              31542484992 (29.38GiB)
  Error Correction support                        No
  Max memory allocation                           8589934592 (8GiB)
  Unified memory for Host and Device              Yes
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        16777216 (16MiB)
  Global Memory cache line size                   64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             16
    Max size for 1D images from buffer            536870912 pixels
    Max 1D or 2D image array size                 2048 images
    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                128
  Local memory type                               Global
  Local memory size                               8388608 (8MiB)
  Max number of constant args                     8
  Max constant buffer size                        8388608 (8MiB)
  Max size of kernel argument                     1024
  Queue properties                                
    Out-of-order execution                        Yes
    Profiling                                     Yes
  Prefer user sync for interop                    Yes
  Profiling timer resolution                      1ns
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            Yes
  printf() buffer size                            16777216 (16MiB)
  Built-in kernels                                (n/a)
  Device Extensions                               cl_khr_byte_addressable_store 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_3d_image_writes cl_khr_fp64 cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_fp64


NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  No platform
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   No platform
  clCreateContext(NULL, ...) [default]            No platform
  clCreateContext(NULL, ...) [other]              Success [AMD]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_DEFAULT)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1010
  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                                   gfx1010
  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                                   gfx1010

 

0 Likes
dipak
In response to acminor
Big Boss
‎02-17-2021 10:11 AM

Thank you for the above information. Could you please also provide the host-side code?

 

 

0 Likes
acminor
In response to dipak
Journeyman III
‎02-17-2021 10:26 AM

Snippet for the opencl_base.h I used in programming the example.

https://gitlab.com/-/snippets/2078483

Snippet for the actual host code (some stuff commented out because it related to other kernels I was testing). If it does not compile, it will probably be because I commented out a variable.

https://gitlab.com/-/snippets/2078485

The OpenCL kernels are above. See the comment in the code about switching out the two different kernels.

The directory structure of the original code:

- dir

-- opencl_base.h

-- dir

--- test1.c

It was compiled with "clang -lOpenCL test1.c". (Clang version 10.0.0-4ubuntu1).

0 Likes
dipak
In response to acminor
Big Boss
‎02-18-2021 01:33 AM

Thank you for sharing the source files. We'll look into this and get back to you shortly.

 

0 Likes
dipak
In response to dipak
Big Boss
‎02-19-2021 01:28 AM

Could you please try the below and let me know your observations?

1. When checking the error code for enqueue_kernel, check for CLK_ENQUEUE_FAILURE which represents the generic error code.

Please note: enqueue_kernel may not provide any extra error information when "-g" compilation option is used. As per the spec, it's an optional feature, not a requirement. Hence, implementation may ignore it. You may find a similar discussion here: https://community.amd.com/t5/opencl/device-queue-is-not-blocking-if-it-is-full-some-child-kernels/td...

2. When creating the device queue, set the queue size (in "cl_queue_properties")  to its maximum limit (as per the above clinfo, the max. limit is  8388608 byte or 8MB).

(P.S. I also observed an error when queue size was set as the "preferred size". When I increased its value to the max. limit, the code ran successfully.)

 

Thanks.

 

0 Likes