cchen marked an inline comment as done.
cchen added inline comments.
================
Comment at: clang/lib/Sema/SemaOpenMP.cpp:15554
+ LocatorChecker Checker;
+ if (Checker.Visit(OrigExpr)) {
+ llvm::copy(Checker.getComponents(),
----------------
ABataev wrote:
> cchen wrote:
> > ABataev wrote:
> > > General question about several cases. How we're going to support them?
> > > 1. (a ? b : c).
> > > 2. __builtin_choose_expr(a, b, c).
> > > 3. a = b.
> > > 4. a?:b
> > > 5. __builtin_convertvector(x, ty)
> > > 6. (int&)a
> > > 7. v.xy, where v is an extended vector
> > > 8. a.b, where b is a bitfield
> > > 9. __builtin_bit_cast(v, ty)
> > > 10. const_cast<ty &>(a)
> > > 11. dynamic_cast<ty &>(a)
> > > 12. reinterpret_cast
> > > 13. static_cast
> > > 14. typeid() (also is an lvalue).
> > > 15. __uuidof(*comPtr)
> > > 16. lambda calls
> > > 17. User defined literals
> > >
> > I think we could first evaluate the lvalue, and then pass the result of
> > `&lvalue` to device?
> What do you mean when you say `evaluate lvalue`? In veneral, it can be
> evaluated at the runtime. Do you propose to implement dyic translation? It
> will definetely slow down the execution on the device.
I mean evaluate lvalue before sending &lvalue to device. For example:
```
int a, b;
b = 5;
#pragma omp target map(a = b) // assign b to a before sending `&(a=b)` to device
{
a++;
}
```
Therefore, the above example have the same semantics as this:
```
int a, b;
b = 5;
int &c = (a=b)
#pragma omp target map(c)
{
a++;
}
```
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D72811/new/
https://reviews.llvm.org/D72811
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
