clang/test/OpenMP/ompx_attributes_codegen.cpp

   1 // REQUIRES: amdgpu-registered-target
   2
   3 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host.bc
   4 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=AMD
   5 // RUN: %clang_cc1 -target-cpu gfx900 -fopenmp -x c++ -std=c++11 -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=AMD
   6 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple nvptx64 -fopenmp-targets=nvptx64 -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=NVIDIA
   7 // expected-no-diagnostics
   8
   9
  10 // Check that the target attributes are set on the generated kernel
  11 void func() {
  12   // AMD: amdgpu_kernel void @__omp_offloading[[HASH:.*]]_l16(ptr {{[^,]+}}) #0
  13   // AMD: amdgpu_kernel void @__omp_offloading[[HASH:.*]]_l18(ptr {{[^,]+}})
  14   // AMD: amdgpu_kernel void @__omp_offloading[[HASH:.*]]_l20(ptr {{[^,]+}}) #4
  15
  16   #pragma omp target ompx_attribute([[clang::amdgpu_flat_work_group_size(10, 20)]])
  17   {}
  18   #pragma omp target teams ompx_attribute(__attribute__((launch_bounds(45, 90))))
  19   {}
  20   #pragma omp target teams distribute parallel for simd ompx_attribute([[clang::amdgpu_flat_work_group_size(3, 17)]]) device(3) ompx_attribute(__attribute__((amdgpu_waves_per_eu(3, 7))))
  21   for (int i = 0; i < 1000; ++i)
  22   {}
  23 }
  24
  25 // AMD: attributes #0
  26 // AMD-SAME: "amdgpu-flat-work-group-size"="10,20"
  27 // AMD-SAME: "omp_target_thread_limit"="20"
  28 // AMD: "omp_target_thread_limit"="45"
  29 // AMD: attributes #4
  30 // AMD-SAME: "amdgpu-flat-work-group-size"="3,17"
  31 // AMD-SAME: "amdgpu-waves-per-eu"="3,7"
  32 // AMD-SAME: "omp_target_thread_limit"="17"
  33
  34 // It is unclear if we should use the AMD annotations for other targets, we do for now.
  35 // NVIDIA: "omp_target_thread_limit"="20"
  36 // NVIDIA: "omp_target_thread_limit"="45"
  37 // NVIDIA: "omp_target_thread_limit"="17"
  38 // NVIDIA: !{ptr @__omp_offloading[[HASH1:.*]]_l16, !"maxntidx", i32 20}
  39 // NVIDIA: !{ptr @__omp_offloading[[HASH2:.*]]_l18, !"minctasm", i32 90}
  40 // NVIDIA: !{ptr @__omp_offloading[[HASH2]]_l18, !"maxntidx", i32 45}
  41 // NVIDIA: !{ptr @__omp_offloading[[HASH3:.*]]_l20, !"maxntidx", i32 17}