1  /* { dg-do compile } */
       2  /* { dg-options "-O2 -ftree-vectorize" } */
       3  
       4  #include <stdint.h>
       5  
       6  #define N 2000
       7  
       8  #define TEST_LOOP(NAME, TYPE)					\
       9    void __attribute__ ((noinline, noclone))			\
      10    NAME (TYPE *restrict dest, TYPE *restrict src)		\
      11    {								\
      12      for (int i = 0; i < N; ++i)					\
      13        dest[i] += src[i * 3];					\
      14    }
      15  
      16  #define TEST(NAME) \
      17    TEST_LOOP (NAME##_i8, int8_t) \
      18    TEST_LOOP (NAME##_i16, uint16_t) \
      19    TEST_LOOP (NAME##_f32, float) \
      20    TEST_LOOP (NAME##_f64, double)
      21  
      22  TEST (test)
      23  
      24  /* Check the vectorized loop.  */
      25  /* { dg-final { scan-assembler-times {\tld1b\t} 1 } } */
      26  /* { dg-final { scan-assembler-times {\tld3b\t} 1 } } */
      27  /* { dg-final { scan-assembler-times {\tst1b\t} 1 } } */
      28  /* { dg-final { scan-assembler-times {\tld1h\t} 1 } } */
      29  /* { dg-final { scan-assembler-times {\tld3h\t} 1 } } */
      30  /* { dg-final { scan-assembler-times {\tst1h\t} 1 } } */
      31  /* { dg-final { scan-assembler-times {\tld1w\t} 1 } } */
      32  /* { dg-final { scan-assembler-times {\tld3w\t} 1 } } */
      33  /* { dg-final { scan-assembler-times {\tst1w\t} 1 } } */
      34  /* { dg-final { scan-assembler-times {\tld1d\t} 1 } } */
      35  /* { dg-final { scan-assembler-times {\tld3d\t} 1 } } */
      36  /* { dg-final { scan-assembler-times {\tst1d\t} 1 } } */
      37  
      38  /* Check the scalar tail.  */
      39  /* { dg-final { scan-assembler-times {\tldrb\tw} 2 } } */
      40  /* { dg-final { scan-assembler-times {\tstrb\tw} 1 } } */
      41  /* { dg-final { scan-assembler-times {\tldrh\tw} 2 } } */
      42  /* { dg-final { scan-assembler-times {\tstrh\tw} 1 } } */
      43  /* { dg-final { scan-assembler-times {\tldr\ts} 2 } } */
      44  /* { dg-final { scan-assembler-times {\tstr\ts} 1 } } */
      45  /* { dg-final { scan-assembler-times {\tldr\td} 2 } } */
      46  /* { dg-final { scan-assembler-times {\tstr\td} 1 } } */
      47  
      48  /* The only branches should be in the vectorized loop.  */
      49  /* { dg-final { scan-assembler-times {\tb[.a-z]+\t} 4 } } */