1  /* { dg-do compile } */
       2  /* Disable loop vectorization to avoid that loop vectorizer
       3     optimizes those two loops that operate tmp array so that
       4     subsequent dse3 won't eliminate expected tmp stores.  */
       5  /* { dg-options "-O2 -funroll-loops -ftree-slp-vectorize -fno-tree-loop-vectorize -fdump-tree-dse-details" } */
       6  
       7  /* Test if scalar cleanup pass takes effects, mainly check
       8     its secondary pass DSE can remove dead stores on array
       9     tmp.  */
      10  
      11  #include "stdint.h"
      12  
      13  static inline void
      14  foo (int16_t *diff, int i_size, uint8_t *val1, int i_val1, uint8_t *val2,
      15       int i_val2)
      16  {
      17    for (int y = 0; y < i_size; y++)
      18      {
      19        for (int x = 0; x < i_size; x++)
      20  	diff[x + y * i_size] = val1[x] - val2[x];
      21        val1 += i_val1;
      22        val2 += i_val2;
      23      }
      24  }
      25  
      26  void
      27  bar (int16_t res[16], uint8_t *val1, uint8_t *val2)
      28  {
      29    int16_t d[16];
      30    int16_t tmp[16];
      31  
      32    foo (d, 4, val1, 16, val2, 32);
      33  
      34    for (int i = 0; i < 4; i++)
      35      {
      36        int s03 = d[i * 4 + 0] + d[i * 4 + 3];
      37        int s12 = d[i * 4 + 1] + d[i * 4 + 2];
      38        int d03 = d[i * 4 + 0] - d[i * 4 + 3];
      39        int d12 = d[i * 4 + 1] - d[i * 4 + 2];
      40  
      41        tmp[0 * 4 + i] = s03 + s12;
      42        tmp[1 * 4 + i] = 2 * d03 + d12;
      43        tmp[2 * 4 + i] = s03 - s12;
      44        tmp[3 * 4 + i] = d03 - 2 * d12;
      45      }
      46  
      47    for (int i = 0; i < 4; i++)
      48      {
      49        int s03 = tmp[i * 4 + 0] + tmp[i * 4 + 3];
      50        int s12 = tmp[i * 4 + 1] + tmp[i * 4 + 2];
      51        int d03 = tmp[i * 4 + 0] - tmp[i * 4 + 3];
      52        int d12 = tmp[i * 4 + 1] - tmp[i * 4 + 2];
      53  
      54        res[i * 4 + 0] = s03 + s12;
      55        res[i * 4 + 1] = 2 * d03 + d12;
      56        res[i * 4 + 2] = s03 - s12;
      57        res[i * 4 + 3] = d03 - 2 * d12;
      58      }
      59  }
      60  
      61  /* { dg-final { scan-tree-dump {Deleted dead store:.*tmp} "dse4" } } */