clang/test/OpenMP/unroll_codegen_parallel_for_factor.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:    store i32 %[[START:.+]], ptr %[[START_ADDR]], align 4
  22 // IR-NEXT:    store i32 %[[END:.+]], ptr %[[END_ADDR]], align 4
  23 // IR-NEXT:    store i32 %[[STEP:.+]], ptr %[[STEP_ADDR]], align 4
  24 // IR-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @2, i32 3, ptr @func.omp_outlined, ptr %[[START_ADDR]], ptr %[[END_ADDR]], ptr %[[STEP_ADDR]])
  25 // IR-NEXT:    ret void
  26 // IR-NEXT:  }
  27 extern "C" void func(int start, int end, int step) {
  28   #pragma omp parallel for
  29   #pragma omp unroll partial(7)
  30   for (int i = start; i < end; i+=step)
  31     body(start, end, step, i);
  32 }
  33
  34
  35 // IR-LABEL: @func.omp_outlined(
  36 // IR-NEXT:  [[ENTRY:.*]]:
  37 // IR-NEXT:    %[[DOTGLOBAL_TID__ADDR:.+]] = alloca ptr, align 8
  38 // IR-NEXT:    %[[DOTBOUND_TID__ADDR:.+]] = alloca ptr, align 8
  39 // IR-NEXT:    %[[START_ADDR:.+]] = alloca ptr, align 8
  40 // IR-NEXT:    %[[END_ADDR:.+]] = alloca ptr, align 8
  41 // IR-NEXT:    %[[STEP_ADDR:.+]] = alloca ptr, align 8
  42 // IR-NEXT:    %[[DOTOMP_IV:.+]] = alloca i32, align 4
  43 // IR-NEXT:    %[[TMP:.+]] = alloca i32, align 4
  44 // IR-NEXT:    %[[I:.+]] = alloca i32, align 4
  45 // IR-NEXT:    %[[DOTCAPTURE_EXPR_:.+]] = alloca i32, align 4
  46 // IR-NEXT:    %[[DOTCAPTURE_EXPR_1:.+]] = alloca i32, align 4
  47 // IR-NEXT:    %[[DOTCAPTURE_EXPR_2:.+]] = alloca i32, align 4
  48 // IR-NEXT:    %[[DOTCAPTURE_EXPR_3:.+]] = alloca i32, align 4
  49 // IR-NEXT:    %[[DOTCAPTURE_EXPR_6:.+]] = alloca i32, align 4
  50 // IR-NEXT:    %[[DOTCAPTURE_EXPR_8:.+]] = alloca i32, align 4
  51 // IR-NEXT:    %[[DOTUNROLLED_IV_I:.+]] = alloca i32, align 4
  52 // IR-NEXT:    %[[DOTOMP_LB:.+]] = alloca i32, align 4
  53 // IR-NEXT:    %[[DOTOMP_UB:.+]] = alloca i32, align 4
  54 // IR-NEXT:    %[[DOTOMP_STRIDE:.+]] = alloca i32, align 4
  55 // IR-NEXT:    %[[DOTOMP_IS_LAST:.+]] = alloca i32, align 4
  56 // IR-NEXT:    %[[DOTUNROLLED_IV_I12:.+]] = alloca i32, align 4
  57 // IR-NEXT:    %[[DOTUNROLL_INNER_IV_I:.+]] = alloca i32, align 4
  58 // IR-NEXT:    store ptr %[[DOTGLOBAL_TID_:.+]], ptr %[[DOTGLOBAL_TID__ADDR]], align 8
  59 // IR-NEXT:    store ptr %[[DOTBOUND_TID_:.+]], ptr %[[DOTBOUND_TID__ADDR]], align 8
  60 // IR-NEXT:    store ptr %[[START:.+]], ptr %[[START_ADDR]], align 8
  61 // IR-NEXT:    store ptr %[[END:.+]], ptr %[[END_ADDR]], align 8
  62 // IR-NEXT:    store ptr %[[STEP:.+]], ptr %[[STEP_ADDR]], align 8
  63 // IR-NEXT:    %[[TMP2:.+]] = load ptr, ptr %[[START_ADDR]], align 8
  64 // IR-NEXT:    %[[TMP0:.+]] = load ptr, ptr %[[END_ADDR]], align 8
  65 // IR-NEXT:    %[[TMP1:.+]] = load ptr, ptr %[[STEP_ADDR]], align 8
  66 // IR-NEXT:    %[[TMP3:.+]] = load i32, ptr %[[TMP2]], align 4
  67 // IR-NEXT:    store i32 %[[TMP3]], ptr %[[I]], align 4
  68 // IR-NEXT:    %[[TMP4:.+]] = load i32, ptr %[[TMP2]], align 4
  69 // IR-NEXT:    store i32 %[[TMP4]], ptr %[[DOTCAPTURE_EXPR_]], align 4
  70 // IR-NEXT:    %[[TMP5:.+]] = load i32, ptr %[[TMP0]], align 4
  71 // IR-NEXT:    store i32 %[[TMP5]], ptr %[[DOTCAPTURE_EXPR_1]], align 4
  72 // IR-NEXT:    %[[TMP6:.+]] = load i32, ptr %[[TMP1]], align 4
  73 // IR-NEXT:    store i32 %[[TMP6]], ptr %[[DOTCAPTURE_EXPR_2]], align 4
  74 // IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_1]], align 4
  75 // IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
  76 // IR-NEXT:    %[[SUB:.+]] = sub i32 %[[TMP7]], %[[TMP8]]
  77 // IR-NEXT:    %[[SUB4:.+]] = sub i32 %[[SUB]], 1
  78 // IR-NEXT:    %[[TMP9:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
  79 // IR-NEXT:    %[[ADD:.+]] = add i32 %[[SUB4]], %[[TMP9]]
  80 // IR-NEXT:    %[[TMP10:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
  81 // IR-NEXT:    %[[DIV:.+]] = udiv i32 %[[ADD]], %[[TMP10]]
  82 // IR-NEXT:    %[[SUB5:.+]] = sub i32 %[[DIV]], 1
  83 // IR-NEXT:    store i32 %[[SUB5]], ptr %[[DOTCAPTURE_EXPR_3]], align 4
  84 // IR-NEXT:    %[[TMP11:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
  85 // IR-NEXT:    %[[ADD7:.+]] = add i32 %[[TMP11]], 1
  86 // IR-NEXT:    store i32 %[[ADD7]], ptr %[[DOTCAPTURE_EXPR_6]], align 4
  87 // IR-NEXT:    %[[TMP12:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_6]], align 4
  88 // IR-NEXT:    %[[SUB9:.+]] = sub i32 %[[TMP12]], -6
  89 // IR-NEXT:    %[[DIV10:.+]] = udiv i32 %[[SUB9]], 7
  90 // IR-NEXT:    %[[SUB11:.+]] = sub i32 %[[DIV10]], 1
  91 // IR-NEXT:    store i32 %[[SUB11]], ptr %[[DOTCAPTURE_EXPR_8]], align 4
  92 // IR-NEXT:    store i32 0, ptr %[[DOTUNROLLED_IV_I]], align 4
  93 // IR-NEXT:    %[[TMP13:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_6]], align 4
  94 // IR-NEXT:    %[[CMP:.+]] = icmp ult i32 0, %[[TMP13]]
  95 // IR-NEXT:    br i1 %[[CMP]], label %[[OMP_PRECOND_THEN:.+]], label %[[OMP_PRECOND_END:.+]]
  96 // IR-EMPTY:
  97 // IR-NEXT:  [[OMP_PRECOND_THEN]]:
  98 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_LB]], align 4
  99 // IR-NEXT:    %[[TMP14:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
 100 // IR-NEXT:    store i32 %[[TMP14]], ptr %[[DOTOMP_UB]], align 4
 101 // IR-NEXT:    store i32 1, ptr %[[DOTOMP_STRIDE]], align 4
 102 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_IS_LAST]], align 4
 103 // IR-NEXT:    %[[TMP15:.+]] = load ptr, ptr %[[DOTGLOBAL_TID__ADDR]], align 8
 104 // IR-NEXT:    %[[TMP16:.+]] = load i32, ptr %[[TMP15]], align 4
 105 // IR-NEXT:    call void @__kmpc_for_static_init_4u(ptr @1, i32 %[[TMP16]], i32 34, ptr %[[DOTOMP_IS_LAST]], ptr %[[DOTOMP_LB]], ptr %[[DOTOMP_UB]], ptr %[[DOTOMP_STRIDE]], i32 1, i32 1)
 106 // IR-NEXT:    %[[TMP17:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
 107 // IR-NEXT:    %[[TMP18:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
 108 // IR-NEXT:    %[[CMP13:.+]] = icmp ugt i32 %[[TMP17]], %[[TMP18]]
 109 // IR-NEXT:    br i1 %[[CMP13]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
 110 // IR-EMPTY:
 111 // IR-NEXT:  [[COND_TRUE]]:
 112 // IR-NEXT:    %[[TMP19:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
 113 // IR-NEXT:    br label %[[COND_END:.+]]
 114 // IR-EMPTY:
 115 // IR-NEXT:  [[COND_FALSE]]:
 116 // IR-NEXT:    %[[TMP20:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
 117 // IR-NEXT:    br label %[[COND_END]]
 118 // IR-EMPTY:
 119 // IR-NEXT:  [[COND_END]]:
 120 // IR-NEXT:    %[[COND:.+]] = phi i32 [ %[[TMP19]], %[[COND_TRUE]] ], [ %[[TMP20]], %[[COND_FALSE]] ]
 121 // IR-NEXT:    store i32 %[[COND]], ptr %[[DOTOMP_UB]], align 4
 122 // IR-NEXT:    %[[TMP21:.+]] = load i32, ptr %[[DOTOMP_LB]], align 4
 123 // IR-NEXT:    store i32 %[[TMP21]], ptr %[[DOTOMP_IV]], align 4
 124 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND:.+]]
 125 // IR-EMPTY:
 126 // IR-NEXT:  [[OMP_INNER_FOR_COND]]:
 127 // IR-NEXT:    %[[TMP22:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 128 // IR-NEXT:    %[[TMP23:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
 129 // IR-NEXT:    %[[ADD14:.+]] = add i32 %[[TMP23]], 1
 130 // IR-NEXT:    %[[CMP15:.+]] = icmp ult i32 %[[TMP22]], %[[ADD14]]
 131 // IR-NEXT:    br i1 %[[CMP15]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
 132 // IR-EMPTY:
 133 // IR-NEXT:  [[OMP_INNER_FOR_BODY]]:
 134 // IR-NEXT:    %[[TMP24:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 135 // IR-NEXT:    %[[MUL:.+]] = mul i32 %[[TMP24]], 7
 136 // IR-NEXT:    %[[ADD16:.+]] = add i32 0, %[[MUL]]
 137 // IR-NEXT:    store i32 %[[ADD16]], ptr %[[DOTUNROLLED_IV_I12]], align 4
 138 // IR-NEXT:    %[[TMP25:.+]] = load i32, ptr %[[DOTUNROLLED_IV_I12]], align 4
 139 // IR-NEXT:    store i32 %[[TMP25]], ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 140 // IR-NEXT:    br label %[[FOR_COND:.+]]
 141 // IR-EMPTY:
 142 // IR-NEXT:  [[FOR_COND]]:
 143 // IR-NEXT:    %[[TMP26:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 144 // IR-NEXT:    %[[TMP27:.+]] = load i32, ptr %[[DOTUNROLLED_IV_I12]], align 4
 145 // IR-NEXT:    %[[ADD17:.+]] = add i32 %[[TMP27]], 7
 146 // IR-NEXT:    %[[CMP18:.+]] = icmp ult i32 %[[TMP26]], %[[ADD17]]
 147 // IR-NEXT:    br i1 %[[CMP18]], label %[[LAND_RHS:.+]], label %[[LAND_END:.+]]
 148 // IR-EMPTY:
 149 // IR-NEXT:  [[LAND_RHS]]:
 150 // IR-NEXT:    %[[TMP28:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 151 // IR-NEXT:    %[[TMP29:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
 152 // IR-NEXT:    %[[ADD19:.+]] = add i32 %[[TMP29]], 1
 153 // IR-NEXT:    %[[CMP20:.+]] = icmp ult i32 %[[TMP28]], %[[ADD19]]
 154 // IR-NEXT:    br label %[[LAND_END]]
 155 // IR-EMPTY:
 156 // IR-NEXT:  [[LAND_END]]:
 157 // IR-NEXT:    %[[TMP30:.+]] = phi i1 [ false, %[[FOR_COND]] ], [ %[[CMP20]], %[[LAND_RHS]] ]
 158 // IR-NEXT:    br i1 %[[TMP30]], label %[[FOR_BODY:.+]], label %[[FOR_END:.+]]
 159 // IR-EMPTY:
 160 // IR-NEXT:  [[FOR_BODY]]:
 161 // IR-NEXT:    %[[TMP31:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
 162 // IR-NEXT:    %[[TMP32:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 163 // IR-NEXT:    %[[TMP33:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
 164 // IR-NEXT:    %[[MUL21:.+]] = mul i32 %[[TMP32]], %[[TMP33]]
 165 // IR-NEXT:    %[[ADD22:.+]] = add i32 %[[TMP31]], %[[MUL21]]
 166 // IR-NEXT:    store i32 %[[ADD22]], ptr %[[I]], align 4
 167 // IR-NEXT:    %[[TMP34:.+]] = load i32, ptr %[[TMP2]], align 4
 168 // IR-NEXT:    %[[TMP35:.+]] = load i32, ptr %[[TMP0]], align 4
 169 // IR-NEXT:    %[[TMP36:.+]] = load i32, ptr %[[TMP1]], align 4
 170 // IR-NEXT:    %[[TMP37:.+]] = load i32, ptr %[[I]], align 4
 171 // IR-NEXT:    call void (...) @body(i32 noundef %[[TMP34]], i32 noundef %[[TMP35]], i32 noundef %[[TMP36]], i32 noundef %[[TMP37]])
 172 // IR-NEXT:    br label %[[FOR_INC:.+]]
 173 // IR-EMPTY:
 174 // IR-NEXT:  [[FOR_INC]]:
 175 // IR-NEXT:    %[[TMP38:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 176 // IR-NEXT:    %[[INC:.+]] = add i32 %[[TMP38]], 1
 177 // IR-NEXT:    store i32 %[[INC]], ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 178 // IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP2:[0-9]+]]
 179 // IR-EMPTY:
 180 // IR-NEXT:  [[FOR_END]]:
 181 // IR-NEXT:    br label %[[OMP_BODY_CONTINUE:.+]]
 182 // IR-EMPTY:
 183 // IR-NEXT:  [[OMP_BODY_CONTINUE]]:
 184 // IR-NEXT:    br label %[[OMP_INNER_FOR_INC:.+]]
 185 // IR-EMPTY:
 186 // IR-NEXT:  [[OMP_INNER_FOR_INC]]:
 187 // IR-NEXT:    %[[TMP39:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 188 // IR-NEXT:    %[[ADD23:.+]] = add i32 %[[TMP39]], 1
 189 // IR-NEXT:    store i32 %[[ADD23]], ptr %[[DOTOMP_IV]], align 4
 190 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND]]
 191 // IR-EMPTY:
 192 // IR-NEXT:  [[OMP_INNER_FOR_END]]:
 193 // IR-NEXT:    br label %[[OMP_LOOP_EXIT:.+]]
 194 // IR-EMPTY:
 195 // IR-NEXT:  [[OMP_LOOP_EXIT]]:
 196 // IR-NEXT:    %[[TMP40:.+]] = load ptr, ptr %[[DOTGLOBAL_TID__ADDR]], align 8
 197 // IR-NEXT:    %[[TMP41:.+]] = load i32, ptr %[[TMP40]], align 4
 198 // IR-NEXT:    call void @__kmpc_for_static_fini(ptr @1, i32 %[[TMP41]])
 199 // IR-NEXT:    br label %[[OMP_PRECOND_END]]
 200 // IR-EMPTY:
 201 // IR-NEXT:  [[OMP_PRECOND_END]]:
 202 // IR-NEXT:    ret void
 203 // IR-NEXT:  }
 204
 205 #endif /* HEADER */
 206
 207
 208 // IR: ![[LOOP2]] = distinct !{![[LOOP2]], ![[LOOPPROP3:[0-9]+]], ![[LOOPPROP4:[0-9]+]]}
 209 // IR: ![[LOOPPROP3]] = !{!"llvm.loop.mustprogress"}
 210 // IR: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.count", i32 7}