Intended behavior of operator* in id<1>

[Pennycook]

What is the intended behavior of the following code?

#include <sycl/sycl.hpp>

int foo(sycl::id<1> i, int j)
{
    return i * j;
}

It compiles with DPC++, but fails with SimSYCL (celerity/SimSYCL#40) due to an ambiguous overload.

I wrote a toy implementation of id based on what I think the specification says (https://godbolt.org/z/P8MTWW1s7) and the operator overload works as I'd expect.

[fknorr]

When you add the argument-swapped overload (the spec lists both), the call becomes ambiguous: https://godbolt.org/z/74MYofMhY.

Also see the DPC++ workaround for this situation.

[Pennycook]

When you add the argument-swapped overload (the spec lists both), the call becomes ambiguous: https://godbolt.org/z/74MYofMhY.

Oh, you're right! I missed that the second overload was defined, because of the way the specification groups them in the synopsis:

  // OP is: +, -, *, /, %, <<, >>, &, |, ^, &&, ||, <, >, <=, >=
  friend id operatorOP(const id& lhs, const id& rhs) { /* ... */
  }
  friend id operatorOP(const id& lhs, const size_t& rhs) { /* ... */
  }

  // OP is: +=, -=, *=, /=, %=, <<=, >>=, &=, |=, ^=
  friend id& operatorOP(id& lhs, const id& rhs) { /* ... */
  }
  friend id& operatorOP(id& lhs, const size_t& rhs) { /* ... */
  }

  // OP is: +, -, *, /, %, <<, >>, &, |, ^, &&, ||, <, >, <=, >=
  friend id operatorOP(const size_t& lhs, const id& rhs) { /* ... */
  }

Re-ordering those for clarity might be a good idea.

Also see the DPC++ workaround for this situation.

I want to make sure I'm reading this right. The workaround here is that DPC++ defines all of these operators not just for size_t, but for all integral types?

[fknorr]

Also see the DPC++ workaround for this situation.

I want to make sure I'm reading this right. The workaround here is that DPC++ defines all of these operators not just for size_t, but for all integral types?

That's how I interpret the code. It appears these templated conversion operators were first introduced here: intel/llvm#939

[Pennycook]

I've just discovered this is a duplicate of #539.

[keryell]

I've just discovered this is a duplicate of #539.

We have a huge issue backlog unfortunately. 🤒