clang/test/OpenMP/task_codegen.c

   1 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fopenmp-version=50 -x c -emit-llvm %s -o - | FileCheck %s
   2 // RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -x c -triple x86_64-apple-darwin10 -emit-pch -o %t %s
   3 // RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -x c -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
   4
   5 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fopenmp-version=50 -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
   6 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -x c -triple x86_64-apple-darwin10 -emit-pch -o %t %s
   7 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=50 -x c -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
   8 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
   9 // expected-no-diagnostics
  10 #ifndef HEADER
  11 #define HEADER
  12
  13 typedef void *omp_depend_t;
  14 typedef __UINTPTR_TYPE__ omp_event_handle_t;
  15
  16 void foo(void);
  17
  18 // CHECK-LABEL: @main
  19 int main(void) {
  20   omp_depend_t d, x;
  21   omp_event_handle_t evt;
  22   int a, *b;
  23   // CHECK: [[D_ADDR:%.+]] = alloca ptr,
  24   // CHECK: [[X_ADDR:%.+]] = alloca ptr,
  25   // CHECK: [[EVT_ADDR:%.+]] = alloca i64,
  26   // CHECK: [[A_ADDR:%.+]] = alloca i32,
  27   // CHECK: [[DEPOBJ_SIZE_ADDR:%.+]] = alloca i64,
  28   // CHECK: [[DEPOBJ_SIZE_ADDR1:%.+]] = alloca i64,
  29   // CHECK: = alloca i64,
  30   // CHECK: [[DEP_COUNTER_ADDR:%.+]] = alloca i64,
  31   // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(
  32   // CHECK: [[ALLOC:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 [[GTID]], i32 65, i64 48, i64 0, ptr [[TASK_ENTRY:@.+]])
  33   // CHECK: [[EVT_VAL:%.+]] = call ptr @__kmpc_task_allow_completion_event(ptr @{{.+}}, i32 [[GTID]], ptr [[ALLOC]])
  34   // CHECK: [[CAST_EVT_VAL:%.+]] = ptrtoint ptr [[EVT_VAL]] to i64
  35   // CHECK: store i64 [[CAST_EVT_VAL]], ptr [[EVT_ADDR]], align 8
  36   // CHECK: [[D_DEP:%.+]] = load ptr, ptr [[D_ADDR]], align 8
  37   // CHECK: [[D_DEP_BASE:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[D_DEP]], i{{.+}} -1
  38   // CHECK: [[D_DEP_BASE_SIZE:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[D_DEP_BASE]], i{{.+}} 0, i{{.+}} 0
  39   // CHECK: [[SIZE1:%.+]] = load i64, ptr [[D_DEP_BASE_SIZE]], align 8
  40   // CHECK-DAG: store i64 0, ptr [[DEPOBJ_SIZE_ADDR]], align 8
  41   // CHECK: [[SZ:%.+]] = load i64, ptr [[DEPOBJ_SIZE_ADDR]], align 8
  42   // CHECK: [[SIZE:%.+]] = add nuw i64 [[SZ]], [[SIZE1]]
  43   // CHECK: store i64 [[SIZE]], ptr [[DEPOBJ_SIZE_ADDR]], align 8
  44   // CHECK: [[X_DEP:%.+]] = load ptr, ptr [[X_ADDR]], align 8
  45   // CHECK: [[X_DEP_BASE:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[X_DEP]], i{{.+}} -1
  46   // CHECK: [[X_DEP_BASE_SIZE:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[X_DEP_BASE]], i{{.+}} 0, i{{.+}} 0
  47   // CHECK: [[SIZE2:%.+]] = load i64, ptr [[X_DEP_BASE_SIZE]], align 8
  48   // CHECK-DAG: store i64 0, ptr [[DEPOBJ_SIZE_ADDR1]], align 8
  49   // CHECK: [[SZ:%.+]] = load i64, ptr [[DEPOBJ_SIZE_ADDR1]], align 8
  50   // CHECK: [[SIZE3:%.+]] = add nuw i64 [[SZ]], [[SIZE2]]
  51   // CHECK: store i64 [[SIZE3]], ptr [[DEPOBJ_SIZE_ADDR1]], align 8
  52   // CHECK: [[SZ:%.+]] = load i64, ptr [[DEPOBJ_SIZE_ADDR]], align 8
  53   // CHECK: [[SZ1:%.+]] = load i64, ptr [[DEPOBJ_SIZE_ADDR1]], align 8
  54   // CHECK: [[SIZE1:%.+]] = add nuw i64 0, [[SZ]]
  55   // CHECK: [[SIZE2:%.+]] = add nuw i64 [[SIZE1]], [[SZ1]]
  56   // CHECK: [[SIZE:%.+]] = add nuw i64 [[SIZE2]], 2
  57   // CHECK: [[SV:%.+]] = call ptr @llvm.stacksave.p0()
  58   // CHECK: store ptr [[SV]], ptr [[SV_ADDR:%.+]], align 8
  59   // CHECK: [[VLA:%.+]] = alloca %struct.kmp_depend_info, i64 [[SIZE]],
  60   // CHECK: [[SIZE32:%.+]] = trunc i64 [[SIZE]] to i32
  61   // CHECK: [[A_ADDR_CAST:%.+]] = ptrtoint ptr [[A_ADDR]] to i64
  62   // CHECK: [[VLA0:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[VLA]], i64 0
  63   // CHECK: [[BASE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[VLA0]], i{{.+}} 0, i{{.+}} 0
  64   // CHECK: store i64 [[A_ADDR_CAST]], ptr [[BASE_ADDR]], align 16
  65   // CHECK: [[SIZE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[VLA0]], i{{.+}} 0, i{{.+}} 1
  66   // CHECK: store i64 4, ptr [[SIZE_ADDR]], align 8
  67   // CHECK: [[FLAGS_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[VLA0]], i{{.+}} 0, i{{.+}} 2
  68   // CHECK: store i8 1, ptr [[FLAGS_ADDR]], align 1
  69   // CHECK: [[A:%.+]] = load i32, ptr [[A_ADDR]], align 4
  70   // CHECK: [[A_CAST:%.+]] = sext i32 [[A]] to i64
  71   // CHECK: [[SZ1:%.+]] = mul nuw i64 24, [[A_CAST]]
  72   // CHECK: [[A:%.+]] = load i32, ptr [[A_ADDR]], align 4
  73   // CHECK: [[A_CAST:%.+]] = sext i32 [[A]] to i64
  74   // CHECK: [[SZ:%.+]] = mul nuw i64 [[SZ1]], [[A_CAST]]
  75   // CHECK: [[B_ADDR_CAST:%.+]] = ptrtoint ptr %{{.+}} to i64
  76   // CHECK: [[VLA1:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[VLA]], i64 1
  77   // CHECK: [[BASE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[VLA1]], i{{.+}} 0, i{{.+}} 0
  78   // CHECK: store i64 [[B_ADDR_CAST]], ptr [[BASE_ADDR]], align 8
  79   // CHECK: [[SIZE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[VLA1]], i{{.+}} 0, i{{.+}} 1
  80   // CHECK: store i64 [[SZ]], ptr [[SIZE_ADDR]], align 8
  81   // CHECK: [[FLAGS_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[VLA1]], i{{.+}} 0, i{{.+}} 2
  82   // CHECK: store i8 1, ptr [[FLAGS_ADDR]], align 8
  83   // CHECK: store i64 2, ptr [[DEP_COUNTER_ADDR]], align 8
  84   // CHECK: [[BC:%.+]] = load ptr, ptr [[D_ADDR]], align 8
  85   // CHECK: [[PREV:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[BC]], i64 -1
  86   // CHECK: [[SIZE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[PREV]], i{{.+}} 0, i{{.+}} 0
  87   // CHECK: [[SIZE:%.+]] = load i64, ptr [[SIZE_ADDR]], align 8
  88   // CHECK: [[BYTES:%.+]] = mul nuw i64 24, [[SIZE]]
  89   // CHECK: [[POS:%.+]] = load i64, ptr [[DEP_COUNTER_ADDR]], align 8
  90   // CHECK: [[VLA_D:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[VLA]], i64 [[POS]]
  91   // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{.+}} [[VLA_D]], ptr align {{.+}} [[BC]], i64 [[BYTES]], i1 false)
  92   // CHECK: [[ADD:%.+]] = add nuw i64 [[POS]], [[SIZE]]
  93   // CHECK: store i64 [[ADD]], ptr [[DEP_COUNTER_ADDR]], align 8
  94   // CHECK: [[BC:%.+]] = load ptr, ptr [[X_ADDR]], align 8
  95   // CHECK: [[PREV:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[BC]], i64 -1
  96   // CHECK: [[SIZE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[PREV]], i{{.+}} 0, i{{.+}} 0
  97   // CHECK: [[SIZE:%.+]] = load i64, ptr [[SIZE_ADDR]], align 8
  98   // CHECK: [[BYTES:%.+]] = mul nuw i64 24, [[SIZE]]
  99   // CHECK: [[POS:%.+]] = load i64, ptr [[DEP_COUNTER_ADDR]], align 8
 100   // CHECK: [[VLA_X:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[VLA]], i64 [[POS]]
 101   // CHECK: call void @llvm.memcpy.p0.p0.i64(ptr align {{.+}} [[VLA_X]], ptr align {{.+}} [[BC]], i64 [[BYTES]], i1 false)
 102   // CHECK: [[ADD:%.+]] = add nuw i64 [[POS]], [[SIZE]]
 103   // CHECK: store i64 [[ADD]], ptr [[DEP_COUNTER_ADDR]], align 8
 104   // CHECK: call i32 @__kmpc_omp_task_with_deps(ptr @{{.+}}, i32 [[GTID]], ptr [[ALLOC]], i32 [[SIZE32]], ptr [[VLA]], i32 0, ptr null)
 105   // CHECK: [[SV:%.+]] = load ptr, ptr [[SV_ADDR]], align 8
 106   // CHECK: call void @llvm.stackrestore.p0(ptr [[SV]])
 107 #pragma omp task depend(in: a, ([3][a][a])&b) depend(depobj: d, x) detach(evt)
 108   {
 109 #pragma omp taskgroup
 110     {
 111 #pragma omp task
 112       foo();
 113     }
 114   }
 115   // CHECK: ret i32 0
 116   return 0;
 117 }
 118 // CHECK: call void @__kmpc_taskgroup(
 119 // CHECK: call ptr @__kmpc_omp_task_alloc(
 120 // CHECK: call i32 @__kmpc_omp_task(
 121 // CHECK: call void @__kmpc_end_taskgroup(
 122
 123 // CHECK-LINE: @bar
 124 void bar(void) {
 125   int **a;
 126   // CHECK: call void @__kmpc_for_static_init_4(
 127 #pragma omp for
 128 for (int i = 0; i < 10; ++i)
 129   // CHECK: [[BUF:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 %{{.+}}, i32 1, i64 48,
 130   // CHECK: [[PRIVS:%.+]] = getelementptr inbounds nuw [[TT_WITH_PRIVS:%.+]], ptr [[BUF]], i32 0, i32 1
 131   // CHECK: [[I_PRIV:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr [[PRIVS]], i32 0, i32 0
 132   // CHECK: [[I:%.+]] = load i32, ptr [[I_ADDR:%.+]],
 133   // CHECK: store i32 %{{.+}}, ptr [[I_PRIV]],
 134
 135   // NELEMS = 1 * ((i - 0 + 2 - 1) / 2);
 136   // CHECK: [[END:%.+]] = load i32, ptr [[I_ADDR]],
 137   // CHECK: [[EB_SUB:%.+]] = sub i32 [[END]], 0
 138   // CHECK: [[EB_SUB_2_ADD:%.+]] = add i32 [[EB_SUB]], 2
 139   // CHECK: [[EB_SUB_2_ADD_1_SUB:%.+]] = sub i32 [[EB_SUB_2_ADD]], 1
 140   // CHECK: [[EB_SUB_2_ADD_1_SUB_2_DIV:%.+]] = udiv i32 [[EB_SUB_2_ADD_1_SUB]], 2
 141   // CHECK: [[ELEMS:%.+]] = zext i32 [[EB_SUB_2_ADD_1_SUB_2_DIV]] to i64
 142   // CHECK: [[ELEMS2:%.+]] = mul nuw i64 [[ELEMS]], 1
 143   // CHECK: [[NELEMS:%.+]] = mul nuw i64 [[ELEMS2]], 1
 144
 145   // ITERATOR_TOTAL = NELEMS + 0;
 146   // CHECK: [[ITERATOR_TOTAL:%.+]] = add nuw i64 0, [[NELEMS]]
 147   // NELEMS = ITERATOR_TOTAL + non-iterator-deps (=0)
 148   // CHECK: [[TOTAL:%.+]] = add nuw i64 [[ITERATOR_TOTAL]], 0
 149
 150   // %struct.kmp_depend_info DEPS[TOTAL];
 151   // CHECK: [[DEPS:%.+]] = alloca %struct.kmp_depend_info, i64 [[TOTAL]],
 152   // CHECK: [[NDEPS:%.+]] = trunc i64 [[TOTAL]] to i32
 153
 154   // i64 DEP_COUNTER = 0;
 155   // CHECK: store i64 0, ptr [[DEP_COUNTER_ADDR:%.+]],
 156
 157   // NELEMS = ((i - 0 + 2 - 1) / 2);
 158   // CHECK: [[END:%.+]] = load i32, ptr [[I_ADDR]],
 159   // CHECK: [[EB_SUB:%.+]] = sub i32 [[END]], 0
 160   // CHECK: [[EB_SUB_2_ADD:%.+]] = add i32 [[EB_SUB]], 2
 161   // CHECK: [[EB_SUB_2_ADD_1_SUB:%.+]] = sub i32 [[EB_SUB_2_ADD]], 1
 162   // CHECK: [[ELEMS:%.+]] = udiv i32 [[EB_SUB_2_ADD_1_SUB]], 2
 163
 164   // i32 COUNTER = 0;
 165   // CHECK: store i32 0, ptr [[COUNTER_ADDR:%.+]],
 166   // CHECK: br label %[[CONT:.+]]
 167
 168   // Loop.
 169   // CHECK: [[CONT]]:
 170   // CHECK: [[COUNTER:%.+]] = load i32, ptr [[COUNTER_ADDR]],
 171   // CHECK: [[CMP:%.+]] = icmp ult i32 [[COUNTER]], [[ELEMS]]
 172   // CHECK: br i1 [[CMP]], label %[[BODY:.+]], label %[[EXIT:.+]]
 173
 174   // CHECK: [[BODY]]:
 175
 176   // k = 0 + 2*COUNTER;
 177   // CHECK: [[COUNTER:%.+]] = load i32, ptr [[COUNTER_ADDR]],
 178   // CHECK: [[C2_MUL:%.+]] = mul i32 [[COUNTER]], 2
 179   // CHECK: [[C2_MUL_0_ADD:%.+]] = add i32 0, [[C2_MUL]]
 180   // CHECK: store i32 [[C2_MUL_0_ADD]], ptr [[K_ADDR:%.+]],
 181
 182   // &a[k][i]
 183   // CHECK: [[A:%.+]] = load ptr, ptr [[A_ADDR:%.+]],
 184   // CHECK: [[K:%.+]] = load i32, ptr [[K_ADDR]],
 185   // CHECK: [[IDX:%.+]] = zext i32 [[K]] to i64
 186   // CHECK: [[AK_ADDR:%.+]] = getelementptr inbounds nuw ptr, ptr [[A]], i64 [[IDX]]
 187   // CHECK: [[AK:%.+]] = load ptr, ptr [[AK_ADDR]],
 188   // CHECK: [[I:%.+]] = load i32, ptr [[I_ADDR]],
 189   // CHECK: [[IDX:%.+]] = sext i32 [[I]] to i64
 190   // CHECK: [[AKI_ADDR:%.+]] = getelementptr inbounds i32, ptr [[AK]], i64 [[IDX]]
 191   // CHECK: [[AKI_INT:%.+]] = ptrtoint ptr [[AKI_ADDR]] to i64
 192
 193   // DEPS[DEP_COUNTER].base_addr = &a[k][i];
 194   // CHECK: [[DEP_COUNTER:%.+]] = load i64, ptr [[DEP_COUNTER_ADDR]],
 195   // CHECK: [[DEPS_DC:%.+]] = getelementptr %struct.kmp_depend_info, ptr [[DEPS]], i64 [[DEP_COUNTER]]
 196   // CHECK: [[DEPS_DC_BASE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[DEPS_DC]], i{{.+}} 0, i{{.+}} 0
 197   // CHECK: store i64 [[AKI_INT]], ptr [[DEPS_DC_BASE_ADDR]],
 198
 199   // DEPS[DEP_COUNTER].size = sizeof(a[k][i]);
 200   // CHECK: [[DEPS_DC_SIZE:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[DEPS_DC]], i{{.+}} 0, i{{.+}} 1
 201   // CHECK: store i64 4, ptr [[DEPS_DC_SIZE]],
 202
 203   // DEPS[DEP_COUNTER].flags = in;
 204   // CHECK: [[DEPS_DC_FLAGS:%.+]] = getelementptr inbounds nuw %struct.kmp_depend_info, ptr [[DEPS_DC]], i{{.+}} 0, i{{.+}} 2
 205   // CHECK: store i8 1, ptr [[DEPS_DC_FLAGS]],
 206
 207   // DEP_COUNTER = DEP_COUNTER + 1;
 208   // CHECK: [[DEP_COUNTER:%.+]] = load i64, ptr [[DEP_COUNTER_ADDR]],
 209   // CHECK: [[INC:%.+]] = add nuw i64 [[DEP_COUNTER]], 1
 210   // CHECK: store i64 [[INC]], ptr [[DEP_COUNTER_ADDR]],
 211
 212   // COUNTER = COUNTER + 1;
 213   // CHECK: [[COUNTER:%.+]] = load i32, ptr [[COUNTER_ADDR]],
 214   // CHECK: [[INC:%.+]] = add i32 [[COUNTER]], 1
 215   // CHECK: store i32 [[INC]], ptr [[COUNTER_ADDR]],
 216   // CHECK: br label %[[CONT]]
 217
 218   // CHECK: [[EXIT]]:
 219   // CHECK: = call i32 @__kmpc_omp_task_with_deps(ptr @{{.+}}, i32 %{{.+}}, ptr [[BUF]], i32 [[NDEPS]], ptr [[DEPS]], i32 0, ptr null)
 220 #pragma omp task depend(iterator(unsigned k=0:i:2), in: a[k][i])
 221 ++i;
 222 }
 223 #endif