1  /* { dg-do assemble { target aarch64_asm_sve_ok } } */
       2  /* { dg-options "-O2 -msve-vector-bits=256 --save-temps" } */
       3  
       4  #include <stdint.h>
       5  
       6  typedef int8_t vnx16qi __attribute__((vector_size (32)));
       7  typedef int16_t vnx8hi __attribute__((vector_size (32)));
       8  typedef int32_t vnx4si __attribute__((vector_size (32)));
       9  typedef int64_t vnx2di __attribute__((vector_size (32)));
      10  
      11  typedef _Float16 vnx8hf __attribute__((vector_size (32)));
      12  typedef float vnx4sf __attribute__((vector_size (32)));
      13  typedef double vnx2df __attribute__((vector_size (32)));
      14  
      15  /* Predicate vector: 1 1 0 0 ... */
      16  
      17  #define MASK_32		{ 0, 1, 34, 35, 4, 5, 38, 39, 8, 9, 42, 43, 12, 13,	\
      18  			  46, 47, 16, 17, 50, 51, 20, 21, 54, 55, 24, 25,	\
      19  			  58, 59, 28, 29, 62, 63 } 
      20  
      21  #define MASK_16		{0, 1, 18, 19, 4, 5, 22, 23, 8, 9, 26, 27, 12, 13, 30, 31} 
      22  #define MASK_8		{0, 1, 10, 11, 4, 5, 14, 15} 
      23  #define MASK_4		{0, 1, 6, 7}
      24  
      25  #define INDEX_32 vnx16qi
      26  #define INDEX_16 vnx8hi
      27  #define INDEX_8 vnx4si
      28  #define INDEX_4 vnx2di
      29  
      30  #define PERMUTE(type, nunits)						\
      31  type permute_##type (type x, type y)					\
      32  {									\
      33    return __builtin_shuffle (x, y, (INDEX_##nunits) MASK_##nunits);	\
      34  }
      35  
      36  PERMUTE(vnx16qi, 32)
      37  PERMUTE(vnx8hi, 16)
      38  PERMUTE(vnx4si, 8)
      39  PERMUTE(vnx2di, 4)
      40  
      41  PERMUTE(vnx8hf, 16)
      42  PERMUTE(vnx4sf, 8)
      43  PERMUTE(vnx2df, 4)
      44  
      45  /* { dg-final { scan-assembler-not {\ttbl\t} } } */
      46  
      47  /* { dg-final { scan-assembler-times {\tsel\tz[0-9]+\.b, p[0-9]+, z[0-9]+\.b, z[0-9]+\.b\n} 1 } } */
      48  /* { dg-final { scan-assembler-times {\tsel\tz[0-9]+\.h, p[0-9]+, z[0-9]+\.h, z[0-9]+\.h\n} 2 } } */
      49  /* { dg-final { scan-assembler-times {\tsel\tz[0-9]+\.s, p[0-9]+, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */
      50  /* { dg-final { scan-assembler-times {\tsel\tz[0-9]+\.d, p[0-9]+, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */