1  /* { dg-do compile } */
       2  /* { dg-require-effective-target arm_v8_1m_mve_ok } */
       3  /* { dg-add-options arm_v8_1m_mve } */
       4  /* { dg-additional-options "-O3" } */
       5  
       6  #include <stdint.h>
       7  
       8  #define FUNC(SIGN, TYPE, BITS, NAME)					\
       9    void test_ ## NAME ##_ ## SIGN ## BITS (TYPE##BITS##_t * __restrict__ dest, \
      10  					  TYPE##BITS##_t *a) {		\
      11      int i;								\
      12      for (i=0; i < (128 / BITS); i++) {					\
      13        dest[i] = (TYPE##BITS##_t)__builtin_clz(a[i]);			\
      14      }									\
      15  }
      16  
      17  FUNC(s, int, 32, clz)
      18  FUNC(u, uint, 32, clz)
      19  FUNC(s, int, 16, clz)
      20  FUNC(u, uint, 16, clz)
      21  FUNC(s, int, 8, clz)
      22  FUNC(u, uint, 8, clz)
      23  
      24  /* 16 and 8-bit versions are not vectorized because they need pack/unpack
      25     patterns since __builtin_clz uses 32-bit parameter and return value.  */
      26  /* { dg-final { scan-assembler-times {vclz\.i32\tq[0-9]+, q[0-9]+} 2 } } */
      27  /* { dg-final { scan-assembler-times {vclz\.i16\tq[0-9]+, q[0-9]+} 2 { xfail *-*-* } } } */
      28  /* { dg-final { scan-assembler-times {vclz\.i8\tq[0-9]+, q[0-9]+} 2 { xfail *-*-* } } } */