polly/test/ScheduleOptimizer/ensure-correct-tile-sizes.ll

   1 ; RUN: opt %loadPolly -polly-process-unprofitable -polly-remarks-minimal \
   2 ; RUN:     -polly-opt-isl -polly-pattern-matching-based-opts=true \
   3 ; RUN:     -polly-target-throughput-vector-fma=1 \
   4 ; RUN:     -polly-target-latency-vector-fma=1 \
   5 ; RUN:     -polly-target-vector-register-bitwidth=4096 \
   6 ; RUN:     -polly-target-1st-cache-level-associativity=3 -polly-print-ast -disable-output < %s | FileCheck %s
   7 ;
   8 ;     /* Test that Polly does not crash due to configurations that can lead to
   9 ;    incorrect tile size computations.
  10 ;    The parameters are setup such that Car in `getMacroKernelParams`
  11 ;    is evaluated to 0. */
  12 ;
  13 ;    static const int N = 3000;
  14 ;
  15 ;    void f(int A[N][N], int B[N][N], int C[N][N]) {
  16 ;      for (int i = 0; i < N; i++) {
  17 ;        for (int j = 0; j < N; j++) {
  18 ;          A[i][j] = 0;
  19 ;          for (int k = 0; k < N; k++) {
  20 ;            A[i][j] += B[i][k] * C[k][j];
  21 ;          }
  22 ;        }
  23 ;      }
  24 ;    }
  25 ;
  26 ; CHECK:           // 1st level tiling - Tiles
  27 ; CHECK-NEXT:      for (int c0 = 0; c0 <= 93; c0 += 1)
  28 ; CHECK-NEXT:        for (int c1 = 0; c1 <= 93; c1 += 1) {
  29 ; CHECK-NEXT:          // 1st level tiling - Points
  30 ; CHECK-NEXT:          for (int c2 = 0; c2 <= min(31, -32 * c0 + 2999); c2 += 1)
  31 ; CHECK-NEXT:            for (int c3 = 0; c3 <= min(31, -32 * c1 + 2999); c3 += 1)
  32 ; CHECK-NEXT:              Stmt_for_body3(32 * c0 + c2, 32 * c1 + c3);
  33 ; CHECK-NEXT:        }
  34 ; CHECK-NEXT:      // Register tiling - Tiles
  35 ; CHECK-NEXT:      for (int c0 = 0; c0 <= 23; c0 += 1)
  36 ; CHECK-NEXT:        for (int c1 = 0; c1 <= 2999; c1 += 1)
  37 ; CHECK-NEXT:          for (int c2 = 0; c2 <= 2999; c2 += 1) {
  38 ; CHECK-NEXT:            // Register tiling - Points
  39 ; CHECK-NEXT:            {
  40 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0, c2);
  41 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 1, c2);
  42 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 2, c2);
  43 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 3, c2);
  44 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 4, c2);
  45 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 5, c2);
  46 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 6, c2);
  47 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 7, c2);
  48 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 8, c2);
  49 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 9, c2);
  50 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 10, c2);
  51 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 11, c2);
  52 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 12, c2);
  53 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 13, c2);
  54 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 14, c2);
  55 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 15, c2);
  56 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 16, c2);
  57 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 17, c2);
  58 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 18, c2);
  59 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 19, c2);
  60 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 20, c2);
  61 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 21, c2);
  62 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 22, c2);
  63 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 23, c2);
  64 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 24, c2);
  65 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 25, c2);
  66 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 26, c2);
  67 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 27, c2);
  68 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 28, c2);
  69 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 29, c2);
  70 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 30, c2);
  71 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 31, c2);
  72 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 32, c2);
  73 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 33, c2);
  74 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 34, c2);
  75 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 35, c2);
  76 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 36, c2);
  77 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 37, c2);
  78 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 38, c2);
  79 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 39, c2);
  80 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 40, c2);
  81 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 41, c2);
  82 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 42, c2);
  83 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 43, c2);
  84 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 44, c2);
  85 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 45, c2);
  86 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 46, c2);
  87 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 47, c2);
  88 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 48, c2);
  89 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 49, c2);
  90 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 50, c2);
  91 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 51, c2);
  92 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 52, c2);
  93 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 53, c2);
  94 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 54, c2);
  95 ; CHECK-NEXT:              Stmt_for_body8(c1, 128 * c0 + 55, c2);
  96 ; CHECK-NEXT:              if (c0 <= 22) {
  97 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 56, c2);
  98 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 57, c2);
  99 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 58, c2);
 100 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 59, c2);
 101 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 60, c2);
 102 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 61, c2);
 103 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 62, c2);
 104 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 63, c2);
 105 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 64, c2);
 106 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 65, c2);
 107 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 66, c2);
 108 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 67, c2);
 109 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 68, c2);
 110 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 69, c2);
 111 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 70, c2);
 112 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 71, c2);
 113 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 72, c2);
 114 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 73, c2);
 115 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 74, c2);
 116 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 75, c2);
 117 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 76, c2);
 118 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 77, c2);
 119 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 78, c2);
 120 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 79, c2);
 121 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 80, c2);
 122 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 81, c2);
 123 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 82, c2);
 124 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 83, c2);
 125 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 84, c2);
 126 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 85, c2);
 127 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 86, c2);
 128 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 87, c2);
 129 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 88, c2);
 130 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 89, c2);
 131 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 90, c2);
 132 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 91, c2);
 133 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 92, c2);
 134 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 93, c2);
 135 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 94, c2);
 136 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 95, c2);
 137 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 96, c2);
 138 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 97, c2);
 139 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 98, c2);
 140 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 99, c2);
 141 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 100, c2);
 142 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 101, c2);
 143 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 102, c2);
 144 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 103, c2);
 145 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 104, c2);
 146 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 105, c2);
 147 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 106, c2);
 148 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 107, c2);
 149 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 108, c2);
 150 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 109, c2);
 151 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 110, c2);
 152 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 111, c2);
 153 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 112, c2);
 154 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 113, c2);
 155 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 114, c2);
 156 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 115, c2);
 157 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 116, c2);
 158 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 117, c2);
 159 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 118, c2);
 160 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 119, c2);
 161 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 120, c2);
 162 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 121, c2);
 163 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 122, c2);
 164 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 123, c2);
 165 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 124, c2);
 166 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 125, c2);
 167 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 126, c2);
 168 ; CHECK-NEXT:                Stmt_for_body8(c1, 128 * c0 + 127, c2);
 169 ; CHECK-NEXT:              }
 170 ; CHECK-NEXT:            }
 171 ; CHECK-NEXT:          }
 172 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
 173
 174 define void @f(ptr %A, ptr %B, ptr %C) {
 175 entry:
 176   br label %for.cond
 177
 178 for.cond:                                         ; preds = %for.inc24, %entry
 179   %indvars.iv4 = phi i64 [ %indvars.iv.next5, %for.inc24 ], [ 0, %entry ]
 180   %exitcond6 = icmp ne i64 %indvars.iv4, 3000
 181   br i1 %exitcond6, label %for.body, label %for.end26
 182
 183 for.body:                                         ; preds = %for.cond
 184   br label %for.cond1
 185
 186 for.cond1:                                        ; preds = %for.inc21, %for.body
 187   %indvars.iv1 = phi i64 [ %indvars.iv.next2, %for.inc21 ], [ 0, %for.body ]
 188   %exitcond3 = icmp ne i64 %indvars.iv1, 3000
 189   br i1 %exitcond3, label %for.body3, label %for.end23
 190
 191 for.body3:                                        ; preds = %for.cond1
 192   %arrayidx5 = getelementptr inbounds [3000 x i32], ptr %A, i64 %indvars.iv4, i64 %indvars.iv1
 193   store i32 0, ptr %arrayidx5, align 4
 194   br label %for.cond6
 195
 196 for.cond6:                                        ; preds = %for.inc, %for.body3
 197   %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %for.body3 ]
 198   %exitcond = icmp ne i64 %indvars.iv, 3000
 199   br i1 %exitcond, label %for.body8, label %for.end
 200
 201 for.body8:                                        ; preds = %for.cond6
 202   %arrayidx12 = getelementptr inbounds [3000 x i32], ptr %B, i64 %indvars.iv4, i64 %indvars.iv
 203   %tmp = load i32, ptr %arrayidx12, align 4
 204   %arrayidx16 = getelementptr inbounds [3000 x i32], ptr %C, i64 %indvars.iv, i64 %indvars.iv1
 205   %tmp7 = load i32, ptr %arrayidx16, align 4
 206   %mul = mul nsw i32 %tmp, %tmp7
 207   %arrayidx20 = getelementptr inbounds [3000 x i32], ptr %A, i64 %indvars.iv4, i64 %indvars.iv1
 208   %tmp8 = load i32, ptr %arrayidx20, align 4
 209   %add = add nsw i32 %tmp8, %mul
 210   store i32 %add, ptr %arrayidx20, align 4
 211   br label %for.inc
 212
 213 for.inc:                                          ; preds = %for.body8
 214   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
 215   br label %for.cond6
 216
 217 for.end:                                          ; preds = %for.cond6
 218   br label %for.inc21
 219
 220 for.inc21:                                        ; preds = %for.end
 221   %indvars.iv.next2 = add nuw nsw i64 %indvars.iv1, 1
 222   br label %for.cond1
 223
 224 for.end23:                                        ; preds = %for.cond1
 225   br label %for.inc24
 226
 227 for.inc24:                                        ; preds = %for.end23
 228   %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 1
 229   br label %for.cond
 230
 231 for.end26:                                        ; preds = %for.cond
 232   ret void
 233 }