mirror of
https://github.com/llvm/llvm-project.git
synced 2025-05-03 03:56:06 +00:00
c8c5830715
Currently the diag is emitted even when there is no target feature specified on command line for OpenMP. This is because the function to initialize feature map is also used with cached feature string. The fix is to only diag when the feature map is initialized with feature strings from command line options. Reviewed by: Joseph Huber, Matt Arsenault, Johannes Doerfert Differential Revision: https://reviews.llvm.org/D153123
17 lines
659 B
C
17 lines
659 B
C
// REQUIRES: x86-registered-target
|
|
// REQUIRES: amdgpu-registered-target
|
|
// REQUIRES: clang-target-64-bits
|
|
|
|
// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx1030 \
|
|
// RUN: -fopenmp -nogpulib -fopenmp-is-device -verify %s
|
|
// expected-no-diagnostics
|
|
|
|
// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -target-cpu gfx1030 \
|
|
// RUN: -fopenmp -nogpulib -target-feature -image-insts \
|
|
// RUN: -fopenmp-is-device -emit-llvm -S -o - %s 2>&1 | FileCheck %s
|
|
// CHECK: warning: feature flag '-image-insts' is ignored since the feature is read only
|
|
|
|
#pragma omp begin declare variant match(device = {arch(amdgcn)})
|
|
void foo();
|
|
#pragma omp end declare variant
|