clang/test/OpenMP/irbuilder_simd.cpp

   1 // RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
   2 // expected-no-diagnostics
   3
   4 struct S {
   5   int a, b;
   6 };
   7
   8 struct P {
   9   int a, b;
  10 };
  11
  12 void simple(float *a, float *b, int *c) {
  13   S s, *p;
  14   P pp;
  15 #pragma omp simd
  16   for (int i = 3; i < 32; i += 5) {
  17     // llvm.access.group test
  18     // CHECK: %[[A_ADDR:.+]] = alloca ptr, align 8
  19     // CHECK: %[[B_ADDR:.+]] = alloca ptr, align 8
  20     // CHECK: %[[S:.+]] = alloca %struct.S, align 4
  21     // CHECK: %[[P:.+]] = alloca ptr, align 8
  22     // CHECK: %[[I:.+]] = alloca i32, align 4
  23     // CHECK: %[[TMP3:.+]] = load ptr, ptr %[[B_ADDR:.+]], align 8, !llvm.access.group ![[META3:[0-9]+]]
  24     // CHECK-NEXT: %[[TMP4:.+]] = load i32, ptr %[[I:.+]], align 4, !llvm.access.group ![[META3:[0-9]+]]
  25     // CHECK-NEXT: %[[IDXPROM:.+]] = sext i32 %[[TMP4:.+]] to i64
  26     // CHECK-NEXT: %[[ARRAYIDX:.+]] = getelementptr inbounds float, ptr %[[TMP3:.+]], i64 %[[IDXPROM:.+]]
  27     // CHECK-NEXT: %[[TMP5:.+]] = load float, ptr %[[ARRAYIDX:.+]], align 4, !llvm.access.group ![[META3:[0-9]+]]
  28     // CHECK-NEXT: %[[A2:.+]] = getelementptr inbounds nuw %struct.S, ptr %[[S:.+]], i32 0, i32 0
  29     // CHECK-NEXT: %[[TMP6:.+]] = load i32, ptr %[[A2:.+]], align 4, !llvm.access.group ![[META3:[0-9]+]]
  30     // CHECK-NEXT: %[[CONV:.+]] = sitofp i32 %[[TMP6:.+]] to float
  31     // CHECK-NEXT: %[[ADD:.+]] = fadd float %[[TMP5:.+]], %[[CONV:.+]]
  32     // CHECK-NEXT: %[[TMP7:.+]] = load ptr, ptr %[[P:.+]], align 8, !llvm.access.group ![[META3:[0-9]+]]
  33     // CHECK-NEXT: %[[A3:.+]] = getelementptr inbounds nuw %struct.S, ptr %[[TMP7:.+]], i32 0, i32 0
  34     // CHECK-NEXT: %[[TMP8:.+]] = load i32, ptr %[[A3:.+]], align 4, !llvm.access.group ![[META3:[0-9]+]]
  35     // CHECK-NEXT: %[[CONV4:.+]] = sitofp i32 %[[TMP8:.+]] to float
  36     // CHECK-NEXT: %[[ADD5:.+]] = fadd float %[[ADD:.+]], %[[CONV4:.+]]
  37     // CHECK-NEXT: %[[TMP9:.+]] = load ptr, ptr %[[A_ADDR:.+]], align 8, !llvm.access.group ![[META3:[0-9]+]]
  38     // CHECK-NEXT: %[[TMP10:.+]] = load i32, ptr %[[I:.+]], align 4, !llvm.access.group ![[META3:[0-9]+]]
  39     // CHECK-NEXT: %[[IDXPROM6:.+]] = sext i32 %[[TMP10:.+]] to i64
  40     // CHECK-NEXT: %[[ARRAYIDX7:.+]] = getelementptr inbounds float, ptr %[[TMP9:.+]], i64 %[[IDXPROM6:.+]]
  41     // CHECK-NEXT: store float %[[ADD5:.+]], ptr %[[ARRAYIDX7:.+]], align 4, !llvm.access.group ![[META3:[0-9]+]]
  42     // llvm.loop test
  43     // CHECK: %[[OMP_LOOPDOTNEXT:.+]] = add nuw i32 %[[OMP_LOOPDOTIV:.+]], 1
  44     // CHECK-NEXT: br label %omp_loop.header, !llvm.loop ![[META4:[0-9]+]]
  45     a[i] = b[i] + s.a + p->a;
  46   }
  47
  48 #pragma omp simd
  49   for (int j = 3; j < 32; j += 5) {
  50     // test if unique access groups were used for a second loop
  51     // CHECK: %[[A22:.+]] = getelementptr inbounds nuw %struct.P, ptr %[[PP:.+]], i32 0, i32 0
  52     // CHECK-NEXT: %[[TMP14:.+]] = load i32, ptr %[[A22:.+]], align 4, !llvm.access.group ![[META7:[0-9]+]]
  53     // CHECK-NEXT: %[[TMP15:.+]] = load ptr, ptr %[[C_ADDR:.+]], align 8, !llvm.access.group ![[META7:[0-9]+]]
  54     // CHECK-NEXT: %[[TMP16:.+]] = load i32, ptr %[[J:.+]], align 4, !llvm.access.group ![[META7:[0-9]+]]
  55     // CHECK-NEXT: %[[IDXPROM23:.+]] = sext i32 %[[TMP16:.+]] to i64
  56     // CHECK-NEXT: %[[ARRAYIDX24:.+]] = getelementptr inbounds i32, ptr %[[TMP15:.+]], i64 %[[IDXPROM23:.+]]
  57     // CHECK-NEXT: store i32 %[[TMP14:.+]], ptr %[[ARRAYIDX24:.+]], align 4, !llvm.access.group ![[META7:[0-9]+]]
  58     // check llvm.loop metadata
  59     // CHECK: %[[OMP_LOOPDOTNEXT:.+]] = add nuw i32 %[[OMP_LOOPDOTIV:.+]], 1
  60     // CHECK-NEXT: br label %[[OMP_LLOP_BODY:.*]], !llvm.loop ![[META8:[0-9]+]]
  61     c[j] = pp.a;
  62   }
  63 }
  64
  65 // CHECK: ![[META3:[0-9]+]] = distinct !{}
  66 // CHECK-NEXT: ![[META4]]  = distinct !{![[META4]], ![[META5:[0-9]+]], ![[META6:[0-9]+]]}
  67 // CHECK-NEXT: ![[META5]]  = !{!"llvm.loop.parallel_accesses", ![[META3]]}
  68 // CHECK-NEXT: ![[META6]]  = !{!"llvm.loop.vectorize.enable", i1 true}
  69 // CHECK-NEXT: ![[META7:[0-9]+]] = distinct !{}
  70 // CHECK-NEXT: ![[META8]]  = distinct !{![[META8]], ![[META9:[0-9]+]], ![[META6]]}
  71 // CHECK-NEXT: ![[META9]]  = !{!"llvm.loop.parallel_accesses", ![[META7]]}