DPCT1096

Message

The right-most dimension of the work-group used in the SYCL kernel that calls this function may be less than <value of kernel sub-group size attribute>. The function <help function name> may return an unexpected result on the CPU device. Modify the size of the work-group to ensure that the value of the right-most dimension is a multiple of <value of kernel sub-group size attribute>.

Detailed Help

The dpct::select_from_sub_group, dpct::shift_sub_group_left, dpct::shift_sub_group_right, and dpct::permute_sub_group_by_xor functions may return unexpected results when run on a CPU device with an OpenCL™ backend, if the right-most dimension value of the work-group used in the SYCL* kernel that calls these functions is less than the value of the kernel sub-group size attribute. The real sub-group size may not be the value specified by the kernel sub-group size attribute and could cause the helper function to return unexpected results on a CPU device.

Adjust the code by modifying the size of the work-group to ensure that the value of the right-most dimension is a multiple of the kernel sub-group size attribute.

For example, this original CUDA* code:

__global__ void kernel() {
  ...
  value = __shfl_down(x, delta);
  ...
}

void foo() {
  ...
  auto GridSize = dim3(2);
  auto BlockSize = dim3(8, 8, 1);
  kernel<<<GridSize, BlockSize>>>();
  ...
}

results in the following migrated SYCL code:

void kernel(const sycl::nd_item<3> &item_ct1) {
  ...
  /*
  DPCT1096:0: The right-most dimension of the work-group used in the SYCL kernel
  that calls this function may be less than "32". The function
  "dpct::shift_sub_group_left" may return an unexpected result on the CPU
  device. Modify the size of the work-group to ensure that the value of the
  right-most dimension is a multiple of "32".
  */
  value = dpct::shift_sub_group_left(item_ct1.get_sub_group(), x, delta); // May return unexpected result on CPU
  ...
}

void foo() {
  ...
  auto GridSize = sycl::range<3>(1, 1, 2);
  auto BlockSize = sycl::range<3>(1, 8, 8); // Problem: value of the right-most dimension 8 is less than the kernel sub group size attribute 32.
  dpct::get_in_order_queue().parallel_for(
      sycl::nd_range<3>(GridSize * BlockSize, BlockSize),
      [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
        kernel(item_ct1);
      });
  ...
}

which is rewritten to:

void kernel(const sycl::nd_item<3> &item_ct1) {
  ...
  value = dpct::shift_sub_group_left(item_ct1.get_sub_group(), x, delta);
  ...
}

void foo() {
  ...
  auto GridSize = sycl::range<3>(1, 1, 2);
  auto BlockSize = sycl::range<3>(1, 2, 32); // Fix: modified work group size to make the right-most dimension to be multiple of the kernel sub group size attribute value, which is 32.
  dpct::get_in_order_queue().parallel_for(
      sycl::nd_range<3>(GridSize * BlockSize, BlockSize),
      [=](sycl::nd_item<3> item_ct1) [[intel::reqd_sub_group_size(32)]] {
        kernel(item_ct1);
      });
  ...
}

Suggestions to Fix

If the program needs to execute on a CPU device, you may need to adjust the code.