clang/test/OpenMP/unroll_codegen_heuristic.cpp

   1 // Check code generation
   2 // RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fopenmp -emit-llvm %s -o - | FileCheck %s --check-prefix=IR
   3
   4 // Check same results after serialization round-trip
   5 // RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fopenmp -emit-pch -o %t %s
   6 // RUN: %clang_cc1 -verify -triple x86_64-pc-linux-gnu -fopenmp -include-pch %t -emit-llvm %s -o - | FileCheck %s --check-prefix=IR
   7 // expected-no-diagnostics
   8
   9 #ifndef HEADER
  10 #define HEADER
  11
  12 // placeholder for loop body code.
  13 extern "C" void body(...) {}
  14
  15
  16 // IR-LABEL: @func(
  17 // IR-NEXT:  [[ENTRY:.*]]:
  18 // IR-NEXT:    %[[START_ADDR:.+]] = alloca i32, align 4
  19 // IR-NEXT:    %[[END_ADDR:.+]] = alloca i32, align 4
  20 // IR-NEXT:    %[[STEP_ADDR:.+]] = alloca i32, align 4
  21 // IR-NEXT:    %[[I:.+]] = alloca i32, align 4
  22 // IR-NEXT:    store i32 %[[START:.+]], ptr %[[START_ADDR]], align 4
  23 // IR-NEXT:    store i32 %[[END:.+]], ptr %[[END_ADDR]], align 4
  24 // IR-NEXT:    store i32 %[[STEP:.+]], ptr %[[STEP_ADDR]], align 4
  25 // IR-NEXT:    %[[TMP0:.+]] = load i32, ptr %[[START_ADDR]], align 4
  26 // IR-NEXT:    store i32 %[[TMP0]], ptr %[[I]], align 4
  27 // IR-NEXT:    br label %[[FOR_COND:.+]]
  28 // IR-EMPTY:
  29 // IR-NEXT:  [[FOR_COND]]:
  30 // IR-NEXT:    %[[TMP1:.+]] = load i32, ptr %[[I]], align 4
  31 // IR-NEXT:    %[[TMP2:.+]] = load i32, ptr %[[END_ADDR]], align 4
  32 // IR-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP1]], %[[TMP2]]
  33 // IR-NEXT:    br i1 %[[CMP]], label %[[FOR_BODY:.+]], label %[[FOR_END:.+]]
  34 // IR-EMPTY:
  35 // IR-NEXT:  [[FOR_BODY]]:
  36 // IR-NEXT:    %[[TMP3:.+]] = load i32, ptr %[[START_ADDR]], align 4
  37 // IR-NEXT:    %[[TMP4:.+]] = load i32, ptr %[[END_ADDR]], align 4
  38 // IR-NEXT:    %[[TMP5:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
  39 // IR-NEXT:    %[[TMP6:.+]] = load i32, ptr %[[I]], align 4
  40 // IR-NEXT:    call void (...) @body(i32 noundef %[[TMP3]], i32 noundef %[[TMP4]], i32 noundef %[[TMP5]], i32 noundef %[[TMP6]])
  41 // IR-NEXT:    br label %[[FOR_INC:.+]]
  42 // IR-EMPTY:
  43 // IR-NEXT:  [[FOR_INC]]:
  44 // IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
  45 // IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[I]], align 4
  46 // IR-NEXT:    %[[ADD:.+]] = add nsw i32 %[[TMP8]], %[[TMP7]]
  47 // IR-NEXT:    store i32 %[[ADD]], ptr %[[I]], align 4
  48 // IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP2:[0-9]+]]
  49 // IR-EMPTY:
  50 // IR-NEXT:  [[FOR_END]]:
  51 // IR-NEXT:    ret void
  52 // IR-NEXT:  }
  53 extern "C" void func(int start, int end, int step) {
  54   #pragma omp unroll
  55   for (int i = start; i < end; i+=step)
  56     body(start, end, step, i);
  57 }
  58
  59 #endif /* HEADER */
  60
  61
  62 // IR: ![[LOOP2]] = distinct !{![[LOOP2]], ![[LOOPPROP3:[0-9]+]], ![[LOOPPROP4:[0-9]+]]}
  63 // IR: ![[LOOPPROP3]] = !{!"llvm.loop.mustprogress"}
  64 // IR: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}