gcc/testsuite/gcc.target/s390/memset-1.c

   1 /* Make sure that short memset's with constant length are emitted
   2    without loop statements.  */
   3
   4 /* { dg-do compile } */
   5 /* { dg-options "-O3 -mzarch -march=z13" } */
   6
   7 /* 1 stc */
   8 void
   9 *memset0(void *s, int c)
  10 {
  11   return __builtin_memset (s, c, 1);
  12 }
  13
  14 /* 3 vst */
  15 void
  16 *memset1(void *s, int c)
  17 {
  18   return __builtin_memset (s, c, 42);
  19 }
  20
  21 /* 3 stc 3 mvc */
  22 void
  23 *memset2(void *s, int c)
  24 {
  25   return __builtin_memset (s, c, 700);
  26 }
  27
  28 /* nop */
  29 void
  30 *memset3(void *s, int c)
  31 {
  32   return __builtin_memset (s, c, 0);
  33 }
  34
  35 /* 1 stc 1 mvc */
  36 void
  37 *memset4(void *s, int c)
  38 {
  39   return __builtin_memset (s, c, 256);
  40 }
  41
  42 /* 2 stc 2 mvc */
  43 void
  44 *memset5(void *s, int c)
  45 {
  46   return __builtin_memset (s, c, 512);
  47 }
  48
  49 /* 2 stc 2 mvc - still due to the stc bytes */
  50 void
  51 *memset6(void *s, int c)
  52 {
  53   return __builtin_memset (s, c, 514);
  54 }
  55
  56 /* 3 stc 2 mvc */
  57 void
  58 *memset7(void *s, int c)
  59 {
  60   return __builtin_memset (s, c, 515);
  61 }
  62
  63 /* 4 stc 4 mvc - 4 * 256 + 4 stc bytes */
  64 void
  65 *memset8(void *s, int c)
  66 {
  67   return __builtin_memset (s, c, 1028);
  68 }
  69
  70 /* 2 stc 1 pfd 2 mvc - start using mvc loop */
  71 void
  72 *memset9(void *s, int c)
  73 {
  74   return __builtin_memset (s, c, 1029);
  75 }
  76
  77 /* 3 stc 3 mvc - displacement overflow after the first */
  78 void
  79 *memset10(void *s, int c)
  80 {
  81   return __builtin_memset ((char*)s + 4000, c, 700);
  82 }
  83
  84 /* 1 mvi */
  85 void
  86 *clrmem0(void *s)
  87 {
  88   return __builtin_memset (s, 0, 1);
  89 }
  90
  91 /* 1 xc */
  92 void
  93 *clrmem1(void *s)
  94 {
  95   return __builtin_memset (s, 0, 42);
  96 }
  97
  98 /* 3 xc */
  99 void
 100 *clrmem2(void *s)
 101 {
 102   return __builtin_memset (s, 0, 700);
 103 }
 104
 105 /* nop */
 106 void
 107 *clrmem3(void *s)
 108 {
 109   return __builtin_memset (s, 0, 0);
 110 }
 111
 112 /* 1 xc */
 113 void
 114 *clrmem4(void *s)
 115 {
 116   return __builtin_memset (s, 0, 256);
 117 }
 118
 119 /* 2 xc */
 120 void
 121 *clrmem5(void *s)
 122 {
 123   return __builtin_memset (s, 0, 512);
 124 }
 125
 126 /* 4 xc */
 127 void
 128 *clrmem6(void *s)
 129 {
 130   return __builtin_memset (s, 0, 1024);
 131 }
 132
 133 /* 2 xc - start using xc loop*/
 134 void
 135 *clrmem7(void *s)
 136 {
 137   return __builtin_memset (s, 0, 1025);
 138 }
 139
 140 /* 5 xc - on z10 PFD would be used in the loop body so the unrolled
 141    variant would still be shorter.  */
 142 __attribute__ ((target("tune=z10")))
 143 void
 144 *clrmem7_z10(void *s)
 145 {
 146   return __builtin_memset (s, 0, 1025);
 147 }
 148
 149 /* 5 xc */
 150 __attribute__ ((target("tune=z10")))
 151 void
 152 *clrmem8_z10(void *s)
 153 {
 154   return __builtin_memset (s, 0, 1280);
 155 }
 156
 157 /* 1 pfd 2 xc - start using xc loop also on z10 */
 158 __attribute__ ((target("tune=z10")))
 159 void
 160 *clrmem9_z10(void *s)
 161 {
 162   return __builtin_memset (s, 0, 1281);
 163 }
 164
 165 /* 3 xc - displacement overflow after the first */
 166 void
 167 *clrmem10(void *s)
 168 {
 169   return __builtin_memset (s + 4000, 0, 700);
 170 }
 171
 172 /* { dg-final { scan-assembler-times "mvi\\s" 1 } } */
 173 /* { dg-final { scan-assembler-times "mvc\\s" 19 } } */
 174 /* { dg-final { scan-assembler-times "xc\\s" 28 } } */
 175 /* { dg-final { scan-assembler-times "stc\\s" 21 } } */
 176 /* { dg-final { scan-assembler-times "stcy\\s" 0 } } */
 177 /* { dg-final { scan-assembler-times "pfd\\s" 2 } } */
 178 /* { dg-final { scan-assembler-times "vst\\s" 3 } } */