gcc/testsuite/gcc.dg/unroll-and-jam.c

   1 /* { dg-do run } */
   2 /* { dg-options "-O3 -floop-unroll-and-jam --param unroll-jam-min-percent=0 -fdump-tree-unrolljam-details" } */
   3 /* { dg-require-effective-target int32plus } */
   4
   5 #include <stdio.h>
   6 extern unsigned int a[];
   7 extern unsigned int b[];
   8 extern unsigned int aa[][1024];
   9 unsigned int checksum;
  10 void checkaa(void)
  11 {
  12   unsigned sum = 1;
  13   unsigned long i, j;
  14   for (i = 0; i < 1024; i++) {
  15       for (j = 0; j < 16; j++) {
  16           sum += aa[j][i]*31+47;
  17       }
  18   }
  19   checksum = checksum * 27 + sum;
  20   //printf("  %d\n", sum);
  21 }
  22
  23 void checkb(void)
  24 {
  25   unsigned sum = 1;
  26   unsigned long i, j;
  27   for (i = 0; i < 1024; i++) {
  28       sum += b[i]*31+47;
  29   }
  30   checksum = checksum * 27 + sum;
  31   //printf("  %d\n", sum);
  32 }
  33
  34 #define TEST(name, body, test) \
  35 static void __attribute__((noinline,noclone)) name (unsigned long n, unsigned long m) \
  36 { \
  37   unsigned long i, j; \
  38   for (i = 1; i < m; i++) { \
  39       for (j = 1; j < n; j++) { \
  40           body; \
  41       } \
  42   } \
  43   test; \
  44 } \
  45 static void __attribute__((noinline,noclone,optimize("O1"))) name ## noopt (unsigned long n, unsigned long m) \
  46 { \
  47   unsigned long i, j; \
  48   for (i = 1; i < m; i++) { \
  49       for (j = 1; j < n; j++) { \
  50           body; \
  51       } \
  52   } \
  53   test; \
  54 }
  55 TEST(foo1, aa[i+1][j+1]=aa[i][j] * aa[i][j] / 2, checkaa()) //ok, -1,-1
  56 TEST(foo2, aa[i][j+1]=3*aa[i+1][j], checkaa()) //notok, 1,-1
  57 TEST(foo3, aa[i+1][j-1]=aa[i][j] * aa[i][j] / 2, checkaa()) //notok, -1,1
  58 TEST(foo4, aa[i][j] = aa[i-1][j+1] * aa[i-1][j+1] / 2, checkaa()) //notok, -1,1
  59 TEST(foo5, aa[i][j] = aa[i+1][j+1] * aa[i+1][j+1] / 2, checkaa()) //ok, 1,1
  60 TEST(foo6, aa[i][j] = aa[i+1][j] * aa[i+1][j] / 2, checkaa()) //ok, -1,0
  61 TEST(foo7, aa[i+1][j] = aa[i][j] * aa[i][j] / 2, checkaa()) //ok, 1,0
  62 TEST(foo9, b[j] = 3*b[j+1] + 1, checkb()) //notok, 0,-1
  63 TEST(foo10, b[j] = 3*b[j] + 1, checkb()) //ok, 0,0
  64
  65 /* foo8 should work as well, but currently doesn't because the distance
  66    vectors we compute are too pessimistic.  We compute
  67      (0,1), (1,1) and (1,-1)
  68    and the last one causes us to lose.  */
  69 TEST(foo8, b[j+1] = 3*b[j] + 1, checkb()) //ok, 0,1
  70
  71 unsigned int a[1024];
  72 unsigned int b[1024];
  73 unsigned int aa[16][1024];
  74 void init(void)
  75 {
  76   unsigned long i,j;
  77   for (i = 0; i < 1024; i++) {
  78       for (j = 0; j < 16; j++) {
  79           aa[j][i] = ((j+1)*2+i+1) % 17;
  80       }
  81       a[i] = ((i+1)*31) % 19;
  82       b[i] = ((i+1)*47) % 23;
  83   }
  84   checksum = 1;
  85 }
  86
  87 #define RUN(name) \
  88     printf(" %s\n", #name); \
  89     init();for(i=0;i<4;i++)name##noopt(32,8); checka = checksum; \
  90     init();for(i=0;i<4;i++)name(32,8); \
  91     printf("%sok %s\n", checka != checksum ? "NOT " : "", #name);
  92
  93 int main()
  94 {
  95   int i;
  96   unsigned checka;
  97   RUN(foo1);
  98   RUN(foo2);
  99   RUN(foo3);
 100   RUN(foo4);
 101   RUN(foo5);
 102   RUN(foo6);
 103   RUN(foo7);
 104   RUN(foo8);
 105   RUN(foo9);
 106   RUN(foo10);
 107   return 0;
 108 }
 109
 110 /* Five loops should be unroll-jammed (actually six, but see above).  */
 111 /* { dg-final { scan-tree-dump-times "applying unroll and jam" 5 "unrolljam" } } */