clang/test/OpenMP/unroll_codegen_tile_for.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:    %[[DOTOMP_IV:.+]] = alloca i32, align 4
  22 // IR-NEXT:    %[[TMP:.+]] = alloca i32, align 4
  23 // IR-NEXT:    %[[I:.+]] = alloca i32, align 4
  24 // IR-NEXT:    %[[DOTCAPTURE_EXPR_:.+]] = alloca i32, align 4
  25 // IR-NEXT:    %[[DOTCAPTURE_EXPR_1:.+]] = alloca i32, align 4
  26 // IR-NEXT:    %[[DOTCAPTURE_EXPR_2:.+]] = alloca i32, align 4
  27 // IR-NEXT:    %[[DOTCAPTURE_EXPR_3:.+]] = alloca i32, align 4
  28 // IR-NEXT:    %[[DOTUNROLLED_IV_I:.+]] = alloca i32, align 4
  29 // IR-NEXT:    %[[DOTCAPTURE_EXPR_6:.+]] = alloca i32, align 4
  30 // IR-NEXT:    %[[DOTCAPTURE_EXPR_8:.+]] = alloca i32, align 4
  31 // IR-NEXT:    %[[DOTCAPTURE_EXPR_12:.+]] = alloca i32, align 4
  32 // IR-NEXT:    %[[DOTCAPTURE_EXPR_14:.+]] = alloca i32, align 4
  33 // IR-NEXT:    %[[DOTFLOOR_0_IV__UNROLLED_IV_I:.+]] = alloca i32, align 4
  34 // IR-NEXT:    %[[DOTOMP_LB:.+]] = alloca i32, align 4
  35 // IR-NEXT:    %[[DOTOMP_UB:.+]] = alloca i32, align 4
  36 // IR-NEXT:    %[[DOTOMP_STRIDE:.+]] = alloca i32, align 4
  37 // IR-NEXT:    %[[DOTOMP_IS_LAST:.+]] = alloca i32, align 4
  38 // IR-NEXT:    %[[DOTFLOOR_0_IV__UNROLLED_IV_I18:.+]] = alloca i32, align 4
  39 // IR-NEXT:    %[[DOTTILE_0_IV__UNROLLED_IV_I:.+]] = alloca i32, align 4
  40 // IR-NEXT:    %[[DOTUNROLL_INNER_IV_I:.+]] = alloca i32, align 4
  41 // IR-NEXT:    %[[TMP0:.+]] = call i32 @__kmpc_global_thread_num(ptr @2)
  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 %[[DOTCAPTURE_EXPR_2]], 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:    %[[SUB4:.+]] = sub i32 %[[SUB]], 1
  57 // IR-NEXT:    %[[TMP7:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
  58 // IR-NEXT:    %[[ADD:.+]] = add i32 %[[SUB4]], %[[TMP7]]
  59 // IR-NEXT:    %[[TMP8:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
  60 // IR-NEXT:    %[[DIV:.+]] = udiv i32 %[[ADD]], %[[TMP8]]
  61 // IR-NEXT:    %[[SUB5:.+]] = sub i32 %[[DIV]], 1
  62 // IR-NEXT:    store i32 %[[SUB5]], ptr %[[DOTCAPTURE_EXPR_3]], align 4
  63 // IR-NEXT:    store i32 0, ptr %[[DOTUNROLLED_IV_I]], align 4
  64 // IR-NEXT:    %[[TMP9:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
  65 // IR-NEXT:    %[[ADD7:.+]] = add i32 %[[TMP9]], 1
  66 // IR-NEXT:    store i32 %[[ADD7]], ptr %[[DOTCAPTURE_EXPR_6]], align 4
  67 // IR-NEXT:    %[[TMP10:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_6]], align 4
  68 // IR-NEXT:    %[[SUB9:.+]] = sub i32 %[[TMP10]], -1
  69 // IR-NEXT:    %[[DIV10:.+]] = udiv i32 %[[SUB9]], 2
  70 // IR-NEXT:    %[[SUB11:.+]] = sub i32 %[[DIV10]], 1
  71 // IR-NEXT:    store i32 %[[SUB11]], ptr %[[DOTCAPTURE_EXPR_8]], align 4
  72 // IR-NEXT:    %[[TMP11:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
  73 // IR-NEXT:    %[[ADD13:.+]] = add i32 %[[TMP11]], 1
  74 // IR-NEXT:    store i32 %[[ADD13]], ptr %[[DOTCAPTURE_EXPR_12]], align 4
  75 // IR-NEXT:    %[[TMP12:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_12]], align 4
  76 // IR-NEXT:    %[[SUB15:.+]] = sub i32 %[[TMP12]], -3
  77 // IR-NEXT:    %[[DIV16:.+]] = udiv i32 %[[SUB15]], 4
  78 // IR-NEXT:    %[[SUB17:.+]] = sub i32 %[[DIV16]], 1
  79 // IR-NEXT:    store i32 %[[SUB17]], ptr %[[DOTCAPTURE_EXPR_14]], align 4
  80 // IR-NEXT:    store i32 0, ptr %[[DOTFLOOR_0_IV__UNROLLED_IV_I]], align 4
  81 // IR-NEXT:    %[[TMP13:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_12]], align 4
  82 // IR-NEXT:    %[[CMP:.+]] = icmp ult i32 0, %[[TMP13]]
  83 // IR-NEXT:    br i1 %[[CMP]], label %[[OMP_PRECOND_THEN:.+]], label %[[OMP_PRECOND_END:.+]]
  84 // IR-EMPTY:
  85 // IR-NEXT:  [[OMP_PRECOND_THEN]]:
  86 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_LB]], align 4
  87 // IR-NEXT:    %[[TMP14:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_14]], align 4
  88 // IR-NEXT:    store i32 %[[TMP14]], ptr %[[DOTOMP_UB]], align 4
  89 // IR-NEXT:    store i32 1, ptr %[[DOTOMP_STRIDE]], align 4
  90 // IR-NEXT:    store i32 0, ptr %[[DOTOMP_IS_LAST]], align 4
  91 // IR-NEXT:    call void @__kmpc_for_static_init_4u(ptr @1, i32 %[[TMP0]], i32 34, ptr %[[DOTOMP_IS_LAST]], ptr %[[DOTOMP_LB]], ptr %[[DOTOMP_UB]], ptr %[[DOTOMP_STRIDE]], i32 1, i32 1)
  92 // IR-NEXT:    %[[TMP15:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
  93 // IR-NEXT:    %[[TMP16:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_14]], align 4
  94 // IR-NEXT:    %[[CMP19:.+]] = icmp ugt i32 %[[TMP15]], %[[TMP16]]
  95 // IR-NEXT:    br i1 %[[CMP19]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
  96 // IR-EMPTY:
  97 // IR-NEXT:  [[COND_TRUE]]:
  98 // IR-NEXT:    %[[TMP17:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_14]], align 4
  99 // IR-NEXT:    br label %[[COND_END:.+]]
 100 // IR-EMPTY:
 101 // IR-NEXT:  [[COND_FALSE]]:
 102 // IR-NEXT:    %[[TMP18:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
 103 // IR-NEXT:    br label %[[COND_END]]
 104 // IR-EMPTY:
 105 // IR-NEXT:  [[COND_END]]:
 106 // IR-NEXT:    %[[COND:.+]] = phi i32 [ %[[TMP17]], %[[COND_TRUE]] ], [ %[[TMP18]], %[[COND_FALSE]] ]
 107 // IR-NEXT:    store i32 %[[COND]], ptr %[[DOTOMP_UB]], align 4
 108 // IR-NEXT:    %[[TMP19:.+]] = load i32, ptr %[[DOTOMP_LB]], align 4
 109 // IR-NEXT:    store i32 %[[TMP19]], ptr %[[DOTOMP_IV]], align 4
 110 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND:.+]]
 111 // IR-EMPTY:
 112 // IR-NEXT:  [[OMP_INNER_FOR_COND]]:
 113 // IR-NEXT:    %[[TMP20:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 114 // IR-NEXT:    %[[TMP21:.+]] = load i32, ptr %[[DOTOMP_UB]], align 4
 115 // IR-NEXT:    %[[ADD20:.+]] = add i32 %[[TMP21]], 1
 116 // IR-NEXT:    %[[CMP21:.+]] = icmp ult i32 %[[TMP20]], %[[ADD20]]
 117 // IR-NEXT:    br i1 %[[CMP21]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
 118 // IR-EMPTY:
 119 // IR-NEXT:  [[OMP_INNER_FOR_BODY]]:
 120 // IR-NEXT:    %[[TMP22:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 121 // IR-NEXT:    %[[MUL:.+]] = mul i32 %[[TMP22]], 4
 122 // IR-NEXT:    %[[ADD22:.+]] = add i32 0, %[[MUL]]
 123 // IR-NEXT:    store i32 %[[ADD22]], ptr %[[DOTFLOOR_0_IV__UNROLLED_IV_I18]], align 4
 124 // IR-NEXT:    %[[TMP23:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__UNROLLED_IV_I18]], align 4
 125 // IR-NEXT:    store i32 %[[TMP23]], ptr %[[DOTTILE_0_IV__UNROLLED_IV_I]], align 4
 126 // IR-NEXT:    br label %[[FOR_COND:.+]]
 127 // IR-EMPTY:
 128 // IR-NEXT:  [[FOR_COND]]:
 129 // IR-NEXT:    %[[TMP24:.+]] = load i32, ptr %[[DOTTILE_0_IV__UNROLLED_IV_I]], align 4
 130 // IR-NEXT:    %[[TMP25:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
 131 // IR-NEXT:    %[[ADD23:.+]] = add i32 %[[TMP25]], 1
 132 // IR-NEXT:    %[[TMP26:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__UNROLLED_IV_I18]], align 4
 133 // IR-NEXT:    %[[ADD24:.+]] = add i32 %[[TMP26]], 4
 134 // IR-NEXT:    %[[CMP25:.+]] = icmp ult i32 %[[ADD23]], %[[ADD24]]
 135 // IR-NEXT:    br i1 %[[CMP25]], label %[[COND_TRUE26:.+]], label %[[COND_FALSE28:.+]]
 136 // IR-EMPTY:
 137 // IR-NEXT:  [[COND_TRUE26]]:
 138 // IR-NEXT:    %[[TMP27:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_8]], align 4
 139 // IR-NEXT:    %[[ADD27:.+]] = add i32 %[[TMP27]], 1
 140 // IR-NEXT:    br label %[[COND_END30:.+]]
 141 // IR-EMPTY:
 142 // IR-NEXT:  [[COND_FALSE28]]:
 143 // IR-NEXT:    %[[TMP28:.+]] = load i32, ptr %[[DOTFLOOR_0_IV__UNROLLED_IV_I18]], align 4
 144 // IR-NEXT:    %[[ADD29:.+]] = add i32 %[[TMP28]], 4
 145 // IR-NEXT:    br label %[[COND_END30]]
 146 // IR-EMPTY:
 147 // IR-NEXT:  [[COND_END30]]:
 148 // IR-NEXT:    %[[COND31:.+]] = phi i32 [ %[[ADD27]], %[[COND_TRUE26]] ], [ %[[ADD29]], %[[COND_FALSE28]] ]
 149 // IR-NEXT:    %[[CMP32:.+]] = icmp ult i32 %[[TMP24]], %[[COND31]]
 150 // IR-NEXT:    br i1 %[[CMP32]], label %[[FOR_BODY:.+]], label %[[FOR_END45:.+]]
 151 // IR-EMPTY:
 152 // IR-NEXT:  [[FOR_BODY]]:
 153 // IR-NEXT:    %[[TMP29:.+]] = load i32, ptr %[[DOTTILE_0_IV__UNROLLED_IV_I]], align 4
 154 // IR-NEXT:    %[[MUL33:.+]] = mul i32 %[[TMP29]], 2
 155 // IR-NEXT:    %[[ADD34:.+]] = add i32 0, %[[MUL33]]
 156 // IR-NEXT:    store i32 %[[ADD34]], ptr %[[DOTUNROLLED_IV_I]], align 4
 157 // IR-NEXT:    %[[TMP30:.+]] = load i32, ptr %[[DOTUNROLLED_IV_I]], align 4
 158 // IR-NEXT:    store i32 %[[TMP30]], ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 159 // IR-NEXT:    br label %[[FOR_COND35:.+]]
 160 // IR-EMPTY:
 161 // IR-NEXT:  [[FOR_COND35]]:
 162 // IR-NEXT:    %[[TMP31:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 163 // IR-NEXT:    %[[TMP32:.+]] = load i32, ptr %[[DOTUNROLLED_IV_I]], align 4
 164 // IR-NEXT:    %[[ADD36:.+]] = add i32 %[[TMP32]], 2
 165 // IR-NEXT:    %[[CMP37:.+]] = icmp ult i32 %[[TMP31]], %[[ADD36]]
 166 // IR-NEXT:    br i1 %[[CMP37]], label %[[LAND_RHS:.+]], label %[[LAND_END:.+]]
 167 // IR-EMPTY:
 168 // IR-NEXT:  [[LAND_RHS]]:
 169 // IR-NEXT:    %[[TMP33:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 170 // IR-NEXT:    %[[TMP34:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_3]], align 4
 171 // IR-NEXT:    %[[ADD38:.+]] = add i32 %[[TMP34]], 1
 172 // IR-NEXT:    %[[CMP39:.+]] = icmp ult i32 %[[TMP33]], %[[ADD38]]
 173 // IR-NEXT:    br label %[[LAND_END]]
 174 // IR-EMPTY:
 175 // IR-NEXT:  [[LAND_END]]:
 176 // IR-NEXT:    %[[TMP35:.+]] = phi i1 [ false, %[[FOR_COND35]] ], [ %[[CMP39]], %[[LAND_RHS]] ]
 177 // IR-NEXT:    br i1 %[[TMP35]], label %[[FOR_BODY40:.+]], label %[[FOR_END:.+]]
 178 // IR-EMPTY:
 179 // IR-NEXT:  [[FOR_BODY40]]:
 180 // IR-NEXT:    %[[TMP36:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_]], align 4
 181 // IR-NEXT:    %[[TMP37:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 182 // IR-NEXT:    %[[TMP38:.+]] = load i32, ptr %[[DOTCAPTURE_EXPR_2]], align 4
 183 // IR-NEXT:    %[[MUL41:.+]] = mul i32 %[[TMP37]], %[[TMP38]]
 184 // IR-NEXT:    %[[ADD42:.+]] = add i32 %[[TMP36]], %[[MUL41]]
 185 // IR-NEXT:    store i32 %[[ADD42]], ptr %[[I]], align 4
 186 // IR-NEXT:    %[[TMP39:.+]] = load i32, ptr %[[START_ADDR]], align 4
 187 // IR-NEXT:    %[[TMP40:.+]] = load i32, ptr %[[END_ADDR]], align 4
 188 // IR-NEXT:    %[[TMP41:.+]] = load i32, ptr %[[STEP_ADDR]], align 4
 189 // IR-NEXT:    %[[TMP42:.+]] = load i32, ptr %[[I]], align 4
 190 // IR-NEXT:    call void (...) @body(i32 noundef %[[TMP39]], i32 noundef %[[TMP40]], i32 noundef %[[TMP41]], i32 noundef %[[TMP42]])
 191 // IR-NEXT:    br label %[[FOR_INC:.+]]
 192 // IR-EMPTY:
 193 // IR-NEXT:  [[FOR_INC]]:
 194 // IR-NEXT:    %[[TMP43:.+]] = load i32, ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 195 // IR-NEXT:    %[[INC:.+]] = add i32 %[[TMP43]], 1
 196 // IR-NEXT:    store i32 %[[INC]], ptr %[[DOTUNROLL_INNER_IV_I]], align 4
 197 // IR-NEXT:    br label %[[FOR_COND35]], !llvm.loop ![[LOOP2:[0-9]+]]
 198 // IR-EMPTY:
 199 // IR-NEXT:  [[FOR_END]]:
 200 // IR-NEXT:    br label %[[FOR_INC43:.+]]
 201 // IR-EMPTY:
 202 // IR-NEXT:  [[FOR_INC43]]:
 203 // IR-NEXT:    %[[TMP44:.+]] = load i32, ptr %[[DOTTILE_0_IV__UNROLLED_IV_I]], align 4
 204 // IR-NEXT:    %[[INC44:.+]] = add i32 %[[TMP44]], 1
 205 // IR-NEXT:    store i32 %[[INC44]], ptr %[[DOTTILE_0_IV__UNROLLED_IV_I]], align 4
 206 // IR-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP5:[0-9]+]]
 207 // IR-EMPTY:
 208 // IR-NEXT:  [[FOR_END45]]:
 209 // IR-NEXT:    br label %[[OMP_BODY_CONTINUE:.+]]
 210 // IR-EMPTY:
 211 // IR-NEXT:  [[OMP_BODY_CONTINUE]]:
 212 // IR-NEXT:    br label %[[OMP_INNER_FOR_INC:.+]]
 213 // IR-EMPTY:
 214 // IR-NEXT:  [[OMP_INNER_FOR_INC]]:
 215 // IR-NEXT:    %[[TMP45:.+]] = load i32, ptr %[[DOTOMP_IV]], align 4
 216 // IR-NEXT:    %[[ADD46:.+]] = add i32 %[[TMP45]], 1
 217 // IR-NEXT:    store i32 %[[ADD46]], ptr %[[DOTOMP_IV]], align 4
 218 // IR-NEXT:    br label %[[OMP_INNER_FOR_COND]]
 219 // IR-EMPTY:
 220 // IR-NEXT:  [[OMP_INNER_FOR_END]]:
 221 // IR-NEXT:    br label %[[OMP_LOOP_EXIT:.+]]
 222 // IR-EMPTY:
 223 // IR-NEXT:  [[OMP_LOOP_EXIT]]:
 224 // IR-NEXT:    call void @__kmpc_for_static_fini(ptr @1, i32 %[[TMP0]])
 225 // IR-NEXT:    br label %[[OMP_PRECOND_END]]
 226 // IR-EMPTY:
 227 // IR-NEXT:  [[OMP_PRECOND_END]]:
 228 // IR-NEXT:    call void @__kmpc_barrier(ptr @3, i32 %[[TMP0]])
 229 // IR-NEXT:    ret void
 230 // IR-NEXT:  }
 231 extern "C" void func(int start, int end, int step) {
 232   #pragma omp for
 233   #pragma omp tile sizes(4)
 234   #pragma omp unroll partial
 235   for (int i = start; i < end; i+=step)
 236     body(start, end, step, i);
 237 }
 238
 239 #endif /* HEADER */
 240
 241
 242 // IR: ![[LOOP2]] = distinct !{![[LOOP2]], ![[LOOPPROP3:[0-9]+]], ![[LOOPPROP4:[0-9]+]]}
 243 // IR: ![[LOOPPROP3]] = !{!"llvm.loop.mustprogress"}
 244 // IR: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.count", i32 2}
 245 // IR: ![[LOOP5]] = distinct !{![[LOOP5]], ![[LOOPPROP3]]}