DPCT1120

Message

<variable name> is in device memory and passed into device functions as an extra argument. As <variable name> is a template variable here, then the declaration of an extra parameter in device functions, the variable referenced in command group, and device functions may be incorrect as a result. You may need to adjust the code.

Detailed Help

<variable name> is in device memory and passed into device functions as an extra argument. As <variable name> is a template variable here, then the declaration of an extra parameter in device functions, the variable referenced in command group, and device functions should be template variable also, the migration result may be incorrect. You may need to adjust the code.

Suggestions to Fix

For example, this original CUDA* code:

struct Foo {
  int x;
  __host__ __device__ Foo() {}
  __host__ __device__ ~Foo() {}
};

template <typename T> __constant__ T cmem;

__global__ void kernel() {
  ...
  int x = cmem<Foo>.x;
  ...
}

void foo() {
  ...
  kernel<<<1, 1>>>();
  ...
}

results in the following migrated SYCL* code:

struct Foo {
  int x;
  Foo() {}
  ~Foo() {}
};

/*
DPCT1120:0: cmem is in device memory and passed into device functions as an
extra arguments. As cmem is a template variable here, then the declaration of
an extra parameter in device functions, the variable referenced in command
group, and device functions may be incorrect as a result. You may need to
adjust the code.
*/
template <typename T> static dpct::constant_memory<T, 0> cmem;

void kernel(T cmem) {
  ...
  int x = cmem<Foo>.x;
  ...
}

void foo() {
  ...
  cmem.init();

  dpct::get_in_order_queue().submit([&](sycl::handler &cgh) {
    auto cmem_ptr_ct1 = cmem.get_ptr();

    cgh.parallel_for(
        sycl::nd_range<3>(sycl::range<3>(1, 1, 1), sycl::range<3>(1, 1, 1)),
        [=](sycl::nd_item<3> item_ct1) {
          kernel(*cmem_ptr_ct1);
        });
  });
  ...
}

which need be rewritten to:

struct Foo {
  int x;
  Foo() {}
  ~Foo() {}
};

template <typename T> static dpct::constant_memory<T, 0> cmem;

template <class T>
void kernel(T cmem) {
  ...
  int x = cmem.x;
  ...
}

void foo() {
  ...
  cmem<Foo>.init();

  dpct::get_in_order_queue().submit([&](sycl::handler &cgh) {
    auto cmem_ptr_ct1 = cmem<Foo>.get_ptr();

    cgh.parallel_for(
        sycl::nd_range<3>(sycl::range<3>(1, 1, 1), sycl::range<3>(1, 1, 1)),
        [=](sycl::nd_item<3> item_ct1) {
          kernel(*cmem_ptr_ct1);
        });
  });
  ...
}