[PATCH] D113154: [OpenMP] Add parsing/sema/serialization for 'bind' clause

[Mike Rice via Phabricator via llvm-commits]

mikerice added inline comments.


================
Comment at: clang/lib/Sema/SemaOpenMP.cpp:4706-4713
+    // OpenMP 5.1 [2.11.7, loop Construct, Restrictions]
+    // If a loop construct is not nested inside another OpenMP construct and it
+    // appears in a procedure, the bind clause must be present.
+    if (CurrentRegion == OMPD_loop && ParentRegion == OMPD_unknown &&
+        BindKind == OMPC_BIND_unknown) {
+      SemaRef.Diag(StartLoc, diag::err_omp_loop_directive_without_bind);
+      return true;
----------------
ABataev wrote:
> Do we allow something like this:
> ```
> void foo() {
>   #pragma omp loop // no bind
>   ...
> }
> volid bar() {
>   #pragma omp parallel
>   foo();
> }
> ```
> ?
I thought this is not allowed and a bind clause should be used. But there appears to be some room for allowing it. The spec also says: 


> If the bind clause is not present on the construct and the loop construct is closely nested inside a teams or parallel construct, the binding region is the corresponding teams or parallel region. **If none of those conditions hold, the binding region is not defined.**


> If the binding region is not defined, then the binding thread set is the encountering thread.


Should I remove this check or leave it?  It's not clear to me what the intentions are here.



CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113154/new/

https://reviews.llvm.org/D113154