[Openmp-commits] [PATCH] D110193: [RFC] Initial documentation for declare target indirect support.

[Vyacheslav Zakharin via Phabricator via Openmp-commits]

vzakhari added inline comments.


================
Comment at: openmp/libomptarget/docs/declare_target_indirect.md:35
+
+FE generates runtime lookup code to match the function address against the key `host_ptr` and produce the new function address `tgt_ptr` that is then used for the indirect function call.
+
----------------
jdoerfert wrote:
> vzakhari wrote:
> > If needed, we can isolate the lookup implementation into a device RTL API, so that FE just generates a lookup call to map the original function address to the device function address.
> API, please: `fp()` -> `(lookupDevPtr(fp))()`
Okay, I will add a definition for the lookup API.


================
Comment at: openmp/libomptarget/docs/declare_target_indirect.md:39
+
+Since all pointers are supposed to be translated/mapped, when program does not use **required unified_shared_memory**, it is possible to avoid generating the runtime dispatch code for indirect function calls. The mapping between host and device address of an indirect function will be established by `libomptarget` during processing of the offload entries table.
+
----------------
jdoerfert wrote:
> How would we translate/map all pointers? I fail to see how we could even identify all (function) pointers (see below). Or did we add wording to the standard to forbid this?
> 
> ```
> char data[sizeof(fp)];
> memcpy(data, &fp, sizeof(fp));
> 
> #pragma omp target map(to:data[:sizeof(fp)])
> {
>   void (*fpc)();
>   memcpy(&fpc, data, sizeof(fp));
>   fpc(); 
> }
> ```
> 
> Modulo my syntax errors ;)
I agree that in this case there will be no translation.  I will remove the words about non-unified_shared_memory "optimization".  Does it sound right?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D110193/new/

https://reviews.llvm.org/D110193