gcc/testsuite/gcc.target/aarch64/sve/mask_struct_load_5.c

   1 /* { dg-do compile } */
   2 /* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
   3
   4 #define TEST_LOOP(NAME, OUTTYPE, INTYPE, MASKTYPE)              \
   5   void __attribute__ ((noinline, noclone))                      \
   6   NAME##_4 (OUTTYPE *__restrict dest, INTYPE *__restrict src,   \
   7             MASKTYPE *__restrict cond, int n)                   \
   8   {                                                             \
   9     for (int i = 0; i < n; ++i)                                 \
  10       if (cond[i])                                              \
  11         dest[i] = src[i * 4] + src[i * 4 + 3];                  \
  12   }
  13
  14 #define TEST2(NAME, OUTTYPE, INTYPE) \
  15   TEST_LOOP (NAME##_i8, OUTTYPE, INTYPE, signed char) \
  16   TEST_LOOP (NAME##_i16, OUTTYPE, INTYPE, unsigned short) \
  17   TEST_LOOP (NAME##_f32, OUTTYPE, INTYPE, float) \
  18   TEST_LOOP (NAME##_f64, OUTTYPE, INTYPE, double)
  19
  20 #define TEST1(NAME, OUTTYPE) \
  21   TEST2 (NAME##_i8, OUTTYPE, signed char) \
  22   TEST2 (NAME##_i16, OUTTYPE, unsigned short) \
  23   TEST2 (NAME##_i32, OUTTYPE, int) \
  24   TEST2 (NAME##_i64, OUTTYPE, unsigned long)
  25
  26 #define TEST(NAME) \
  27   TEST1 (NAME##_i8, signed char) \
  28   TEST1 (NAME##_i16, unsigned short) \
  29   TEST1 (NAME##_i32, int) \
  30   TEST1 (NAME##_i64, unsigned long) \
  31   TEST2 (NAME##_f16_f16, _Float16, _Float16) \
  32   TEST2 (NAME##_f32_f32, float, float) \
  33   TEST2 (NAME##_f64_f64, double, double)
  34
  35 TEST (test)
  36
  37 /*    Mask |  8 16 32 64
  38     -------+------------
  39     Out  8 |  1  1  1  1
  40         16 |  1  1  1  1
  41         32 |  1  1  1  1
  42         64 |  1  1  1  1.  */
  43 /* { dg-final { scan-assembler-times {\tld4b\t.z[0-9]} 16 } } */
  44
  45 /*    Mask |  8 16 32 64
  46     -------+------------
  47     Out  8 |  2  2  2  2
  48         16 |  2  1  1  1 x2 (for half float)
  49         32 |  2  1  1  1
  50         64 |  2  1  1  1.  */
  51 /* { dg-final { scan-assembler-times {\tld4h\t.z[0-9]} 28 } } */
  52
  53 /*    Mask |  8 16 32 64
  54     -------+------------
  55     Out  8 |  4  4  4  4
  56         16 |  4  2  2  2
  57         32 |  4  2  1  1 x2 (for float)
  58         64 |  4  2  1  1.  */
  59 /* { dg-final { scan-assembler-times {\tld4w\t.z[0-9]} 50 } } */
  60
  61 /*    Mask |  8 16 32 64
  62     -------+------------
  63     Out  8 |  8  8  8  8
  64         16 |  8  4  4  4
  65         32 |  8  4  2  2
  66         64 |  8  4  2  1 x2 (for double).  */
  67 /* { dg-final { scan-assembler-times {\tld4d\t.z[0-9]} 98 } } */