[PATCH] D58388: [OpenCL] Simplify LLVM IR generated for OpenCL blocks

[Anastasia Stulova via Phabricator via cfe-commits]

Anastasia added inline comments.


================
Comment at: lib/CodeGen/CGBlocks.cpp:1275
+    // We *can* call the block directly unless it is a function argument.
+    if (!isa<ParmVarDecl>(E->getCalleeDecl()))
+      Func = CGM.getOpenCLRuntime().getInvokeFunction(E->getCallee());
----------------
I think it's reasonable enough... if we restrict blocks as parameters in the spec later it should be easy enough to modify this code. 


================
Comment at: lib/CodeGen/CGOpenCLRuntime.cpp:131
+static const BlockExpr *getBlockExpr(const Expr *E) {
+  if (auto Cast = dyn_cast<CastExpr>(E)) {
+    E = Cast->getSubExpr();
----------------
Btw, does this handle the case when we assign a variable multiple time? I was just wondering if we need a loop somewhere?

I.e. does something like this work now:

```
typedef void (^bl_t)(local void *);

bl_t a = ...;
bl_t b = a;
bl_t c = b;

c();
enqueue_kernel(... c, ...);
```




CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D58388/new/

https://reviews.llvm.org/D58388