1  /* { dg-do compile } */
       2  /* { dg-options "-O2 -ftree-vectorize -ffast-math --param aarch64-sve-compare-costs=0" } */
       3  
       4  #include <stdint.h>
       5  
       6  #define TEST_LOOP(NAME, OUTTYPE, INTYPE, MASKTYPE)		\
       7    void __attribute__ ((noinline, noclone))			\
       8    NAME##_3 (OUTTYPE *__restrict dest, INTYPE *__restrict src,	\
       9  	    MASKTYPE *__restrict cond, intptr_t n)		\
      10    {								\
      11      for (intptr_t i = 0; i < n; ++i)				\
      12        if (cond[i])						\
      13  	dest[i] = (src[i * 3]					\
      14  		   + src[i * 3 + 1]				\
      15  		   + src[i * 3 + 2]);				\
      16    }
      17  
      18  #define TEST2(NAME, OUTTYPE, INTYPE) \
      19    TEST_LOOP (NAME##_i8, OUTTYPE, INTYPE, int8_t) \
      20    TEST_LOOP (NAME##_i16, OUTTYPE, INTYPE, uint16_t) \
      21    TEST_LOOP (NAME##_f32, OUTTYPE, INTYPE, float) \
      22    TEST_LOOP (NAME##_f64, OUTTYPE, INTYPE, double)
      23  
      24  #define TEST1(NAME, OUTTYPE) \
      25    TEST2 (NAME##_i8, OUTTYPE, int8_t) \
      26    TEST2 (NAME##_i16, OUTTYPE, uint16_t) \
      27    TEST2 (NAME##_i32, OUTTYPE, int32_t) \
      28    TEST2 (NAME##_i64, OUTTYPE, uint64_t)
      29  
      30  #define TEST(NAME) \
      31    TEST1 (NAME##_i8, int8_t) \
      32    TEST1 (NAME##_i16, uint16_t) \
      33    TEST1 (NAME##_i32, int32_t) \
      34    TEST1 (NAME##_i64, uint64_t) \
      35    TEST2 (NAME##_f16_f16, _Float16, _Float16) \
      36    TEST2 (NAME##_f32_f32, float, float) \
      37    TEST2 (NAME##_f64_f64, double, double)
      38  
      39  TEST (test)
      40  
      41  /*    Mask |  8 16 32 64
      42      -------+------------
      43      Out  8 |  1  1  1  1
      44          16 |  1  1  1  1
      45          32 |  1  1  1  1
      46          64 |  1  1  1  1.  */
      47  /* { dg-final { scan-assembler-times {\tld3b\t.z[0-9]} 16 } } */
      48  
      49  /*    Mask |  8 16 32 64
      50      -------+------------
      51      Out  8 |  2  2  2  2
      52          16 |  2  1  1  1 x2 (for _Float16)
      53          32 |  2  1  1  1
      54          64 |  2  1  1  1.  */
      55  /* { dg-final { scan-assembler-times {\tld3h\t.z[0-9]} 28 } } */
      56  
      57  /*    Mask |  8 16 32 64
      58      -------+------------
      59      Out  8 |  4  4  4  4
      60          16 |  4  2  2  2
      61          32 |  4  2  1  1 x2 (for float)
      62          64 |  4  2  1  1.  */
      63  /* { dg-final { scan-assembler-times {\tld3w\t.z[0-9]} 50 } } */
      64  
      65  /*    Mask |  8 16 32 64
      66      -------+------------
      67      Out  8 |  8  8  8  8
      68          16 |  8  4  4  4
      69          32 |  8  4  2  2
      70          64 |  8  4  2  1 x2 (for double).  */
      71  /* { dg-final { scan-assembler-times {\tld3d\t.z[0-9]} 98 } } */