clang/test/OpenMP/tile_codegen_for_dependent.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 // The loop trip count used by #pragma omp for depends on code generated
  10 // by #pragma omp file. Check that theses PreInits are emitted before
  11 // the code generated by #pragma omp for.
  12
  13 #ifndef HEADER
  14 #define HEADER
  15
  16 // placeholder for loop body code.
  17 extern "C" void body(...) {}
  18
  19
  20 // IR-LABEL: define {{.*}}@func(
  21 // IR-NEXT:  [[ENTRY:.*]]:
  22 // IR-NEXT:    %[[START_ADDR:.+]] = alloca i32, align 4
  23 // IR-NEXT:    %[[END_ADDR:.+]] = alloca i32, align 4
  24 // IR-NEXT:    %[[STEP_ADDR:.+]] = alloca i32, align 4
  25 // IR-NEXT:    %[[DOTOMP_IV:.+]] = alloca i32, align 4
  26 // IR-NEXT:    %[[TMP:.+]] = alloca i32, align 4
  27 // IR-NEXT:    %[[I:.+]] = alloca i32, align 4
  28 // IR-NEXT:    %[[DOTCAPTURE_EXPR_:.+]] = alloca i32, align 4
  29 // IR-NEXT:    %[[DOTCAPTURE_EXPR_1:.+]] = alloca i32, align 4
  30 // IR-NEXT:    %[[DOTNEW_STEP:.+]] = alloca i32, align 4
  31 // IR-NEXT:    %[[DOTCAPTURE_EXPR_2:.+]] = alloca i32, align 4
  32 // IR-NEXT:    %[[DOTCAPTURE_EXPR_5:.+]] = alloca i32, align 4
  33 // IR-NEXT:    %[[DOTCAPTURE_EXPR_7:.+]] = alloca i32, align 4
  34 // IR-NEXT:    %[[DOTFLOOR_0_IV_I:.+]] = alloca i32, align 4
  35 // IR-NEXT:    %[[DOTOMP_LB:.+]] = alloca i32, align 4
  36 // IR-NEXT:    %[[DOTOMP_UB:.+]] = alloca i32, align 4
  37 // IR-NEXT:    %[[DOTOMP_STRIDE:.+]] = alloca i32, align 4
  38 // IR-NEXT:    %[[DOTOMP_IS_LAST:.+]] = alloca i32, align 4
  39 // IR-NEXT:    %[[DOTFLOOR_0_IV_I11:.+]] = alloca i32, align 4
  40 // IR-NEXT:    %[[DOTTILE_0_IV_I:.+]] = alloca i32, align 4
  41 // IR-NEXT:    %[[TMP0:.+]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB2:.+]])
  42 // IR-NEXT:    store i32 %[[START:.+]], ptr %[[START_ADDR]], align 4
  43 // IR-NEXT:    store i32 %[[END:.+]], ptr %[[END_ADDR]], align 4
  44 // IR-NEXT:    store i32 %[[STEP:.+]], ptr %[[STEP_ADDR]], align 4
  45 // IR-NEXT:    %[[TMP1:.+]] = load i32, ptr %[[START_ADDR]], align 4
  46 // IR-NEXT:    store i32 %[[TMP1]], ptr %[[I]], align 4
  47 // IR-NEXT:    %[[TMP2:.+]] = load i32, ptr %[[START_ADDR]], align 4
  48 // IR-NEXT:    store i32 %[[TMP2]], ptr %[[DOTCAPTURE_EXPR_]], align 4
  49 // IR-NEXT:    %[[TMP3:.+]] = load i32, ptr %[[END_ADDR]], align 4
  50 // IR-NEXT:    store i32 %[[TMP3]], ptr %[[DOTCAPTURE_EXPR_1]], align 4
  51 // IR-NEXT:    %[[TMP4:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
  52 // IR-NEXT:    store i32 %[[TMP4]], ptr %[[DOTNEW_STEP]], align 4
  53 // IR-NEXT:    %[[TMP5:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_1]], align 4
  54 // IR-NEXT:    %[[TMP6:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
  55 // IR-NEXT:    %[[SUB:.+]] = sub i32 %[[TMP5]], %[[TMP6]]
  56 // IR-NEXT:    %[[SUB3:.+]] = sub i32 %[[SUB]], 1
  57 // IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
  58 // IR-NEXT:    %[[ADD:.+]] = add i32 %[[SUB3]], %[[TMP7]]
  59 // IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
  60 // IR-NEXT:    %[[DIV:.+]] = udiv i32 %[[ADD]], %[[TMP8]]
  61 // IR-NEXT:    %[[SUB4:.+]] = sub i32 %[[DIV]], 1
  62 // IR-NEXT:    store i32 %[[SUB4]], ptr %[[DOTCAPTURE_EXPR_2]], align 4
  63 // IR-NEXT:    %[[TMP9:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
  64 // IR-NEXT:    %[[ADD6:.+]] = add i32 %[[TMP9]], 1
  65 // IR-NEXT:    store i32 %[[ADD6]], ptr %[[DOTCAPTURE_EXPR_5]], align 4
  66 // IR-NEXT:    %[[TMP10:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_5]], align 4
  67 // IR-NEXT:    %[[SUB8:.+]] = sub i32 %[[TMP10]], -3
  68 // IR-NEXT:    %[[DIV9:.+]] = udiv i32 %[[SUB8]], 4
  69 // IR-NEXT:    %[[SUB10:.+]] = sub i32 %[[DIV9]], 1
  70 // IR-NEXT:    store i32 %[[SUB10]], ptr %[[DOTCAPTURE_EXPR_7]], align 4
  71 // IR-NEXT:    store i32 0, ptr %[[DOTFLOOR_0_IV_I]], align 4
  72 // IR-NEXT:    %[[TMP11:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_5]], align 4
  73 // IR-NEXT:    %[[CMP:.+]] = icmp ult i32 0, %[[TMP11]]
  74 // IR-NEXT:    br i1 %[[CMP]], label %[[OMP_PRECOND_THEN:.+]], label %[[OMP_PRECOND_END:.+]]
  75 // IR-EMPTY:
  76 // IR-NEXT:  [[OMP_PRECOND_THEN]]:
  77 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_LB]], align 4
  78 // IR-NEXT:    %[[TMP12:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
  79 // IR-NEXT:    store i32 %[[TMP12]], ptr %[[DOTOMP_UB]], align 4
  80 // IR-NEXT:    store i32 1, ptr %[[DOTOMP_STRIDE]], align 4
  81 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_IS_LAST]], align 4
  82 // IR-NEXT:    call void @__kmpc_for_static_init_4u(ptr @[[GLOB1:.+]], i32 %[[TMP0]], i32 34, ptr %[[DOTOMP_IS_LAST]], ptr %[[DOTOMP_LB]], ptr %[[DOTOMP_UB]], ptr %[[DOTOMP_STRIDE]], i32 1, i32 1)
  83 // IR-NEXT:    %[[TMP13:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
  84 // IR-NEXT:    %[[TMP14:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
  85 // IR-NEXT:    %[[CMP12:.+]] = icmp ugt i32 %[[TMP13]], %[[TMP14]]
  86 // IR-NEXT:    br i1 %[[CMP12]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
  87 // IR-EMPTY:
  88 // IR-NEXT:  [[COND_TRUE]]:
  89 // IR-NEXT:    %[[TMP15:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_7]], align 4
  90 // IR-NEXT:    br label %[[COND_END:.+]]
  91 // IR-EMPTY:
  92 // IR-NEXT:  [[COND_FALSE]]:
  93 // IR-NEXT:    %[[TMP16:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
  94 // IR-NEXT:    br label %[[COND_END]]
  95 // IR-EMPTY:
  96 // IR-NEXT:  [[COND_END]]:
  97 // IR-NEXT:    %[[COND:.+]] = phi i32 [ %[[TMP15]], %[[COND_TRUE]] ], [ %[[TMP16]], %[[COND_FALSE]] ]
  98 // IR-NEXT:    store i32 %[[COND]], ptr %[[DOTOMP_UB]], align 4
  99 // IR-NEXT:    %[[TMP17:.+]] = load i32, ptr %[[DOTOMP_LB]], align 4
 100 // IR-NEXT:    store i32 %[[TMP17]], ptr %[[DOTOMP_IV]], align 4
 101 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND:.+]]
 102 // IR-EMPTY:
 103 // IR-NEXT:  [[OMP_INNER_FOR_COND]]:
 104 // IR-NEXT:    %[[TMP18:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 105 // IR-NEXT:    %[[TMP19:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
 106 // IR-NEXT:    %[[ADD13:.+]] = add i32 %[[TMP19]], 1
 107 // IR-NEXT:    %[[CMP14:.+]] = icmp ult i32 %[[TMP18]], %[[ADD13]]
 108 // IR-NEXT:    br i1 %[[CMP14]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
 109 // IR-EMPTY:
 110 // IR-NEXT:  [[OMP_INNER_FOR_BODY]]:
 111 // IR-NEXT:    %[[TMP20:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 112 // IR-NEXT:    %[[MUL:.+]] = mul i32 %[[TMP20]], 4
 113 // IR-NEXT:    %[[ADD15:.+]] = add i32 0, %[[MUL]]
 114 // IR-NEXT:    store i32 %[[ADD15]], ptr %[[DOTFLOOR_0_IV_I11]], align 4
 115 // IR-NEXT:    %[[TMP21:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I11]], align 4
 116 // IR-NEXT:    store i32 %[[TMP21]], ptr %[[DOTTILE_0_IV_I]], align 4
 117 // IR-NEXT:    br label %[[FOR_COND:.+]]
 118 // IR-EMPTY:
 119 // IR-NEXT:  [[FOR_COND]]:
 120 // IR-NEXT:    %[[TMP22:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
 121 // IR-NEXT:    %[[TMP23:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
 122 // IR-NEXT:    %[[ADD16:.+]] = add i32 %[[TMP23]], 1
 123 // IR-NEXT:    %[[TMP24:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I11]], align 4
 124 // IR-NEXT:    %[[ADD17:.+]] = add i32 %[[TMP24]], 4
 125 // IR-NEXT:    %[[CMP18:.+]] = icmp ult i32 %[[ADD16]], %[[ADD17]]
 126 // IR-NEXT:    br i1 %[[CMP18]], label %[[COND_TRUE19:.+]], label %[[COND_FALSE21:.+]]
 127 // IR-EMPTY:
 128 // IR-NEXT:  [[COND_TRUE19]]:
 129 // IR-NEXT:    %[[TMP25:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
 130 // IR-NEXT:    %[[ADD20:.+]] = add i32 %[[TMP25]], 1
 131 // IR-NEXT:    br label %[[COND_END23:.+]]
 132 // IR-EMPTY:
 133 // IR-NEXT:  [[COND_FALSE21]]:
 134 // IR-NEXT:    %[[TMP26:.+]] = load i32, ptr %[[DOTFLOOR_0_IV_I11]], align 4
 135 // IR-NEXT:    %[[ADD22:.+]] = add i32 %[[TMP26]], 4
 136 // IR-NEXT:    br label %[[COND_END23]]
 137 // IR-EMPTY:
 138 // IR-NEXT:  [[COND_END23]]:
 139 // IR-NEXT:    %[[COND24:.+]] = phi i32 [ %[[ADD20]], %[[COND_TRUE19]] ], [ %[[ADD22]], %[[COND_FALSE21]] ]
 140 // IR-NEXT:    %[[CMP25:.+]] = icmp ult i32 %[[TMP22]], %[[COND24]]
 141 // IR-NEXT:    br i1 %[[CMP25]], label %[[FOR_BODY:.+]], label %[[FOR_END:.+]]
 142 // IR-EMPTY:
 143 // IR-NEXT:  [[FOR_BODY]]:
 144 // IR-NEXT:    %[[TMP27:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
 145 // IR-NEXT:    %[[TMP28:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
 146 // IR-NEXT:    %[[TMP29:.+]] = load i32, ptr %[[DOTNEW_STEP]], align 4
 147 // IR-NEXT:    %[[MUL26:.+]] = mul i32 %[[TMP28]], %[[TMP29]]
 148 // IR-NEXT:    %[[ADD27:.+]] = add i32 %[[TMP27]], %[[MUL26]]
 149 // IR-NEXT:    store i32 %[[ADD27]], ptr %[[I]], align 4
 150 // IR-NEXT:    %[[TMP30:.+]] = load i32, ptr %[[START_ADDR]], align 4
 151 // IR-NEXT:    %[[TMP31:.+]] = load i32, ptr %[[END_ADDR]], align 4
 152 // IR-NEXT:    %[[TMP32:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
 153 // IR-NEXT:    %[[TMP33:.+]] = load i32, ptr %[[I]], align 4
 154 // IR-NEXT:    call void (...) @body(i32 noundef %[[TMP30]], i32 noundef %[[TMP31]], i32 noundef %[[TMP32]], i32 noundef %[[TMP33]])
 155 // IR-NEXT:    br label %[[FOR_INC:.+]]
 156 // IR-EMPTY:
 157 // IR-NEXT:  [[FOR_INC]]:
 158 // IR-NEXT:    %[[TMP34:.+]] = load i32, ptr %[[DOTTILE_0_IV_I]], align 4
 159 // IR-NEXT:    %[[INC:.+]] = add i32 %[[TMP34]], 1
 160 // IR-NEXT:    store i32 %[[INC]], ptr %[[DOTTILE_0_IV_I]], align 4
 161 // IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP3:[0-9]+]]
 162 // IR-EMPTY:
 163 // IR-NEXT:  [[FOR_END]]:
 164 // IR-NEXT:    br label %[[OMP_BODY_CONTINUE:.+]]
 165 // IR-EMPTY:
 166 // IR-NEXT:  [[OMP_BODY_CONTINUE]]:
 167 // IR-NEXT:    br label %[[OMP_INNER_FOR_INC:.+]]
 168 // IR-EMPTY:
 169 // IR-NEXT:  [[OMP_INNER_FOR_INC]]:
 170 // IR-NEXT:    %[[TMP35:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 171 // IR-NEXT:    %[[ADD28:.+]] = add i32 %[[TMP35]], 1
 172 // IR-NEXT:    store i32 %[[ADD28]], ptr %[[DOTOMP_IV]], align 4
 173 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND]]
 174 // IR-EMPTY:
 175 // IR-NEXT:  [[OMP_INNER_FOR_END]]:
 176 // IR-NEXT:    br label %[[OMP_LOOP_EXIT:.+]]
 177 // IR-EMPTY:
 178 // IR-NEXT:  [[OMP_LOOP_EXIT]]:
 179 // IR-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 %[[TMP0]])
 180 // IR-NEXT:    br label %[[OMP_PRECOND_END]]
 181 // IR-EMPTY:
 182 // IR-NEXT:  [[OMP_PRECOND_END]]:
 183 // IR-NEXT:    call void @__kmpc_barrier(ptr @[[GLOB3:.+]], i32 %[[TMP0]])
 184 // IR-NEXT:    ret void
 185 // IR-NEXT:  }
 186 extern "C" void func(int start, int end, int step) {
 187 #pragma omp for
 188 #pragma omp tile sizes(4)
 189   for (int i = start; i < end; i += step)
 190     body(start, end, step, i);
 191 }
 192
 193 #endif /* HEADER */