[PATCH] D153924: [OpenMP] Allow exceptions in target regions when offloading to GPUs

Nico Weber via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Mon Sep 11 15:03:05 PDT 2023


thakis added inline comments.


================
Comment at: clang/test/OpenMP/amdgpu_throw_trap.cpp:4
+// RUN: %clang_cc1 -fopenmp -triple amdgcn-amd-amdhsa -fopenmp-is-target-device %s -emit-llvm -S -Wno-openmp-target-exception -o - | FileCheck -check-prefix=DEVICE %s
+// RUN: %clang_cc1 -fopenmp -triple x86_64-pc-linux-gnu -fopenmp-is-target-device -fcxx-exceptions %s -emit-llvm -S -Wno-openmp-target-exception -o - | FileCheck -check-prefix=HOST %s
+// DEVICE: s_trap
----------------
aeubanks wrote:
> thakis wrote:
> > aeubanks wrote:
> > > thakis wrote:
> > > > This test fails if X86 isn't in `LLVM_TARGETS_TO_BUILD` and the host system is some non-x86 system (e.g. arm64).
> > > > 
> > > > (This is the only test in check-clang that fails then.)
> > > > 
> > > > Should this test grow a `REQUIRES: x86-registered-target`? Should it use `%itanium_abi_triple` instead of `x86_64-pc-linux-gnu`? (It seems to pass when replacing `x86_64-pc-linux-gnu` with `%itanium_abi_triple` on my arm mac.)
> > > added x86-registered-target in 238a1ef44f4f2361205e538b3cb7ebc5ec70894d
> > Is that better than `%itanium_abi_triple`?
> I was worried about LLVM failing if the calculated `%itanium_abi_triple` wasn't supported in that build of LLVM, but TIL that clang/LLVM can handle triples it doesn't recognize all the way until the codegen phase. But IIUC optimizations can change depending on whether or not LLVM recognizes the triple so it's still a little inconsistent.
> 
> so yeah `%itanium_abi_triple` would probably work, but it seems susceptible to configuration differences
In return, the test then runs on arm machines if you have the x86 target disabled.

But I'm happy as-is too, and to be honest with x86 target disabled so many tests don't run that it can't _really_ be used for development anyways 😛 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D153924/new/

https://reviews.llvm.org/D153924



More information about the cfe-commits mailing list