clang/test/OpenMP/target_parallel_for_uses_allocators_codegen.cpp

   1 // Test host codegen.
   2 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
   3 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
   4 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
   5
   6 // expected-no-diagnostics
   7 #ifndef HEADER
   8 #define HEADER
   9
  10 enum omp_allocator_handle_t {
  11   omp_null_allocator = 0,
  12   omp_default_mem_alloc = 1,
  13   omp_large_cap_mem_alloc = 2,
  14   omp_const_mem_alloc = 3,
  15   omp_high_bw_mem_alloc = 4,
  16   omp_low_lat_mem_alloc = 5,
  17   omp_cgroup_mem_alloc = 6,
  18   omp_pteam_mem_alloc = 7,
  19   omp_thread_mem_alloc = 8,
  20   KMP_ALLOCATOR_MAX_HANDLE = __UINTPTR_MAX__
  21 };
  22
  23 typedef enum omp_alloctrait_key_t { omp_atk_sync_hint = 1,
  24                                     omp_atk_alignment = 2,
  25                                     omp_atk_access = 3,
  26                                     omp_atk_pool_size = 4,
  27                                     omp_atk_fallback = 5,
  28                                     omp_atk_fb_data = 6,
  29                                     omp_atk_pinned = 7,
  30                                     omp_atk_partition = 8
  31 } omp_alloctrait_key_t;
  32 typedef enum omp_alloctrait_value_t {
  33   omp_atv_false = 0,
  34   omp_atv_true = 1,
  35   omp_atv_default = 2,
  36   omp_atv_contended = 3,
  37   omp_atv_uncontended = 4,
  38   omp_atv_sequential = 5,
  39   omp_atv_private = 6,
  40   omp_atv_all = 7,
  41   omp_atv_thread = 8,
  42   omp_atv_pteam = 9,
  43   omp_atv_cgroup = 10,
  44   omp_atv_default_mem_fb = 11,
  45   omp_atv_null_fb = 12,
  46   omp_atv_abort_fb = 13,
  47   omp_atv_allocator_fb = 14,
  48   omp_atv_environment = 15,
  49   omp_atv_nearest = 16,
  50   omp_atv_blocked = 17,
  51   omp_atv_interleaved = 18
  52 } omp_alloctrait_value_t;
  53
  54 typedef struct omp_alloctrait_t {
  55   omp_alloctrait_key_t key;
  56   __UINTPTR_TYPE__ value;
  57 } omp_alloctrait_t;
  58
  59 // Just map the traits variable as a firstprivate variable.
  60 // CHECK-DAG: [[SIZES:@.+]] = private unnamed_addr constant [1 x i64] [i64 160]
  61 // CHECK-DAG: [[MAPTYPES:@.+]] = private unnamed_addr constant [1 x i64] [i64 673]
  62
  63 // CHECK: define {{.*}}[[FOO:@.+]]()
  64 void foo() {
  65   omp_alloctrait_t traits[10];
  66   omp_allocator_handle_t my_allocator;
  67
  68 // CHECK: [[RES:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 -1, i32 1, i32 0, ptr @.[[TGT_REGION:.+]].region_id, ptr %[[KERNEL_ARGS:.+]])
  69 // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0
  70 // CHECK: br i1 [[CMP]], label %[[FAILED:.+]], label %[[DONE:.+]]
  71 // CHECK: [[FAILED]]:
  72 // CHECK: call void @[[TGT_REGION]](ptr %{{[^,]+}})
  73 #pragma omp target parallel for uses_allocators(omp_null_allocator, omp_thread_mem_alloc, my_allocator(traits))
  74   for (int i = 0; i < 10; ++i)
  75     ;
  76 }
  77
  78 // CHECK: define internal void @[[TGT_REGION]](ptr {{.+}})
  79 // CHECK: [[TRAITS_ADDR_REF:%.+]] = alloca ptr,
  80 // CHECK: [[MY_ALLOCATOR_ADDR:%.+]] = alloca i64,
  81 // CHECK: [[TRAITS_ADDR:%.+]] = load ptr, ptr [[TRAITS_ADDR_REF]],
  82 // CHECK: [[ALLOCATOR:%.+]] = call ptr @__kmpc_init_allocator(i32 %{{.+}}, ptr null, i32 10, ptr [[TRAITS_ADDR]])
  83 // CHECK: [[CONV:%.+]] = ptrtoint ptr [[ALLOCATOR]] to i64
  84 // CHECK: store i64 [[CONV]], ptr [[MY_ALLOCATOR_ADDR]],
  85
  86 // Destroy allocator upon exit from the region.
  87 // CHECK: [[ALLOCATOR:%.+]] = load i64, ptr [[MY_ALLOCATOR_ADDR]],
  88 // CHECK: [[CONV:%.+]] = inttoptr i64 [[ALLOCATOR]] to ptr
  89 // CHECK: call void @__kmpc_destroy_allocator(i32 %{{.+}}, ptr [[CONV]])
  90 // CHECK: ret void
  91
  92 #endif