clang/test/OpenMP/task_affinity_codegen.cpp

   1 // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s
   2 // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
   3 // RUN: %clang_cc1 -fopenmp -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 -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
   6 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
   7 // RUN: %clang_cc1 -fopenmp-simd -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 // CHECK-LABEL: @main
  14 int main() {
  15   float *p;
  16   int a = 10;
  17   // kmp_task_affinity_info_t affs[1];
  18   // CHECK: [[AFFS_ADDR:%.+]] = alloca [1 x %struct.kmp_task_affinity_info_t],
  19   // CHECK: [[TD:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 [[GTID:%.+]], i32 1, i64 40, i64 1, ptr @{{.+}})
  20   // CHECK: [[AFFINE_LST_ADDR:%.+]] = getelementptr inbounds [1 x %struct.kmp_task_affinity_info_t], ptr [[AFFS_ADDR]], i64 0, i64 0
  21   // CHECK: [[P:%.+]] = load ptr, ptr [[P_ADDR:%.+]],
  22   // CHECK: [[A_VAL:%.+]] = load i32, ptr [[A_ADDR:%.+]],
  23   // CHECK: [[A_SZ:%.+]] = sext i32 [[A_VAL]] to i64
  24   // CHECK: [[BYTES:%.+]] = mul nuw i64 4, [[A_SZ]]
  25   // CHECK: [[SZ:%.+]] = mul nuw i64 [[BYTES]], 10
  26   // CHECK: [[A_VAL:%.+]] = load i32, ptr [[A_ADDR]],
  27   // CHECK: [[A_SZ1:%.+]] = sext i32 [[A_VAL]] to i64
  28   // CHECK: [[SIZE:%.+]] = mul nuw i64 [[SZ]], [[A_SZ]]
  29   // CHECK: [[AFFS_0_ADDR:%.+]] = getelementptr %struct.kmp_task_affinity_info_t, ptr [[AFFINE_LST_ADDR]], i64 0
  30
  31   // affs[0].base = p;
  32   // CHECK: [[AFFS_0_BASE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_task_affinity_info_t, ptr [[AFFS_0_ADDR]], i32 0, i32 0
  33   // CHECK: [[P_INTPTR:%.+]] = ptrtoint ptr [[P]] to i64
  34   // CHECK: store i64 [[P_INTPTR]], ptr [[AFFS_0_BASE_ADDR]],
  35
  36   // affs[0].size = sizeof(*p) * a * 10 * a;
  37   // CHECK: [[AFFS_0_SIZE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_task_affinity_info_t, ptr [[AFFS_0_ADDR]], i32 0, i32 1
  38   // CHECK: store i64 [[SIZE]], ptr [[AFFS_0_SIZE_ADDR]],
  39   // CHECK: call i32 @__kmpc_omp_reg_task_with_affinity(ptr @{{.+}}, i32 [[GTID]], ptr [[TD]], i32 1, ptr [[AFFINE_LST_ADDR]])
  40 #pragma omp task affinity(([a][10][a])p)
  41   ;
  42   // CHECK: [[TD:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 [[GTID]], i32 1, i64 40, i64 1, ptr @{{.+}})
  43   // CHECK: [[A_VAL:%.+]] = load i32, ptr [[A_ADDR]],
  44   // CHECK: [[SUB:%.+]] = sub nsw i32 [[A_VAL]], 0
  45   // CHECK: [[CONV:%.+]] = zext i32 [[SUB]] to i64
  46
  47   // <num_elem> = <num_iters> + 1 constant affinity for affinity(a)
  48   // CHECK: [[NUM_ELEMS:%.+]] = add nuw i64 1, [[CONV]]
  49   // CHECK: [[SV:%.+]] = call ptr @llvm.stacksave.p0()
  50   // CHECK: store ptr [[SV]], ptr [[SV_ADDR:%.+]],
  51
  52   // kmp_task_affinity_info_t affs[<num_elem>];
  53   // CHECK: [[AFFS_ADDR:%.+]] = alloca %struct.kmp_task_affinity_info_t, i64 [[NUM_ELEMS]],
  54   // store i64 %21, ptr %__vla_expr0, align 8
  55   // CHECK: [[NAFFS:%.+]] = trunc i64 [[NUM_ELEMS]] to i32
  56   // CHECK: [[AFFS_0_ADDR:%.+]] = getelementptr %struct.kmp_task_affinity_info_t, ptr [[AFFS_ADDR]], i64 0
  57
  58   // affs[0].base = &a;
  59   // CHECK: [[AFFS_0_BASE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_task_affinity_info_t, ptr [[AFFS_0_ADDR]], i32 0, i32 0
  60   // CHECK: [[A_INTPTR:%.+]] = ptrtoint ptr [[A_ADDR]] to i64
  61   // CHECK: store i64 [[A_INTPTR]], ptr [[AFFS_0_BASE_ADDR]],
  62
  63   // affs[0].size = sizeof(a);
  64   // CHECK: [[AFFS_0_SIZE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_task_affinity_info_t, ptr [[AFFS_0_ADDR]], i32 0, i32 1
  65   // CHECK: store i64 4, ptr [[AFFS_0_SIZE_ADDR]],
  66
  67   // affs_cnt = 1;
  68   // CHECK: store i64 1, ptr [[AFFS_CNT_ADDR:%.+]],
  69   // CHECK: [[A_VAL:%.+]] = load i32, ptr [[A_ADDR]],
  70   // CHECK: [[NITERS:%.+]] = sub nsw i32 [[A_VAL]], 0
  71   // CHECK: store i32 0, ptr [[CNT_ADDR:%.+]],
  72   // CHECK: br label %[[CONT:[^,]+]]
  73
  74   //for (int cnt = 0; cnt < (a-0); ++cnt) {
  75   // int i = cnt + 0;
  76   // affs[affs_cnt].base = &p[i];
  77   // affs[affs_cnt].size = sizeof(p[i]);
  78   // ++affs_cnt;
  79   // }
  80
  81   // CHECK: [[CONT]]:
  82   // CHECK: [[CNT:%.+]] = load i32, ptr [[CNT_ADDR]],
  83   // CHECK: [[CMP:%.+]] = icmp slt i32 [[CNT]], [[NITERS]]
  84   // CHECK: br i1 [[CMP]], label %[[BODY:[^,]+]], label %[[DONE:[^,]+]]
  85
  86   // CHECK: [[BODY]]:
  87   // i = cnt + 0;
  88   // CHECK: [[CNT:%.+]] = load i32, ptr [[CNT_ADDR]],
  89   // CHECK: [[VAL:%.+]] = add nsw i32 0, [[CNT]]
  90   // CHECK: store i32 [[VAL]], ptr [[I_ADDR:%.+]],
  91
  92   // &p[i]
  93   // CHECK: [[P:%.+]] = load ptr, ptr [[P_ADDR]],
  94   // CHECK: [[I:%.+]] = load i32, ptr [[I_ADDR]],
  95   // CHECK: [[IDX:%.+]] = sext i32 [[I]] to i64
  96   // CHECK: [[P_I_ADDR:%.+]] = getelementptr inbounds float, ptr [[P]], i64 [[IDX]]
  97
  98   // affs[affs_cnt]
  99   // CHECK: [[AFFS_CNT:%.+]] = load i64, ptr [[AFFS_CNT_ADDR]],
 100   // CHECK: [[AFFS_ELEM_ADDR:%.+]] = getelementptr %struct.kmp_task_affinity_info_t, ptr [[AFFS_ADDR]], i64 [[AFFS_CNT]]
 101
 102   // affs[affs_cnt].base = &p[i];
 103   // CHECK: [[AFFS_ELEM_BASE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_task_affinity_info_t, ptr [[AFFS_ELEM_ADDR]], i32 0, i32 0
 104   // CHECK: [[CAST:%.+]] = ptrtoint ptr [[P_I_ADDR]] to i64
 105   // CHECK: store i64 [[CAST]], ptr [[AFFS_ELEM_BASE_ADDR]],
 106
 107   // affs[affs_cnt].size = sizeof(p[i]);
 108   // CHECK: [[AFFS_ELEM_SIZE_ADDR:%.+]] = getelementptr inbounds nuw %struct.kmp_task_affinity_info_t, ptr [[AFFS_ELEM_ADDR]], i32 0, i32 1
 109   // CHECK: store i64 4, ptr [[AFFS_ELEM_SIZE_ADDR]],
 110
 111   // ++affs_cnt;
 112   // CHECK: [[AFFS_CNT_NEXT:%.+]] = add nuw i64 [[AFFS_CNT]], 1
 113   // CHECK: store i64 [[AFFS_CNT_NEXT]], ptr [[AFFS_CNT_ADDR]],
 114
 115   // ++cnt;
 116   // CHECK: [[CNT:%.+]] = load i32, ptr [[CNT_ADDR]],
 117   // CHECK: [[CNT_NEXT:%.+]] = add nsw i32 [[CNT]], 1
 118   // CHECK: store i32 [[CNT_NEXT]], ptr [[CNT_ADDR]],
 119   // CHECK: br label %[[CONT]]
 120
 121   // CHECK: [[DONE]]:
 122   // CHECK: call i32 @__kmpc_omp_reg_task_with_affinity(ptr @{{.+}} i32 [[GTID]], ptr [[TD]], i32 [[NAFFS]], ptr [[AFFS_ADDR]])
 123   // CHECK: [[SV:%.+]] = load ptr, ptr [[SV_ADDR]],
 124   // CHECK: call void @llvm.stackrestore.p0(ptr [[SV]])
 125 #pragma omp task affinity(iterator(i=0:a): p[i]) affinity(a)
 126   ;
 127   return 0;
 128 }
 129
 130 #endif