src/util/cmimemcpy_qpx.c

   1 #include <string.h>
   2 #include "cmimemcpy_qpx.h"
   3
   4 #define QPX_LOAD(si,sb,fp) \
   5   do {                                                                  \
   6   asm volatile("qvlfdx %0,%1,%2": "=f"(fp) : "b" (si), "r" (sb));       \
   7   } while(0);
   8
   9 #define QPX_STORE(si,sb,fp)                                             \
  10   do {                                                                  \
  11   asm volatile("qvstfdx %2,%0,%1": : "b" (si), "r" (sb), "f"(fp) :"memory"); \
  12   } while(0);
  13
  14 #ifndef __GNUC__
  15 #define FP_REG(i)   asm("f"#i)
  16 #define FP_REG1(i)  "fr"#i
  17 #else
  18 #define FP_REG(i)  asm("fr"#i)
  19 #define FP_REG1(i)  "fr"#i
  20 #endif
  21
  22 //Copy 512 bytes from a 32b aligned pointers
  23 static inline size_t quad_copy_512( char* dest, const char* src ) {
  24     register const double *fpp1_1, *fpp1_2;
  25     register double *fpp2_1, *fpp2_2;
  26
  27     register double f0 FP_REG(0);
  28     register double f1 FP_REG(1);
  29     register double f2 FP_REG(2);
  30     register double f3 FP_REG(3);
  31     register double f4 FP_REG(4);
  32     register double f5 FP_REG(5);
  33     register double f6 FP_REG(6);
  34     register double f7 FP_REG(7);
  35
  36     int r0;
  37     int r1;
  38     int r2;
  39     int r3;
  40     int r4;
  41     int r5;
  42     int r6;
  43     int r7;
  44     r0 = 0;
  45     r1 = 64;
  46     r2 = 128;
  47     r3 = 192;
  48     r4 = 256;
  49     r5 = 320;
  50     r6 = 384;
  51     r7 = 448;
  52
  53     fpp1_1 = (const double *)src;
  54     fpp1_2 = (const double *)src +4;
  55
  56     fpp2_1 = (double *)dest;
  57     fpp2_2 = (double *)dest +4;
  58
  59     QPX_LOAD(fpp1_1,r0,f0);
  60     //asm volatile("qvlfdx 0,%0,%1": : "Ob" (fpp1_1), "r"(r0) :"memory");
  61     QPX_LOAD(fpp1_1,r1,f1);
  62     QPX_LOAD(fpp1_1,r2,f2);
  63     QPX_LOAD(fpp1_1,r3,f3);
  64     QPX_LOAD(fpp1_1,r4,f4);
  65     QPX_LOAD(fpp1_1,r5,f5);
  66     QPX_LOAD(fpp1_1,r6,f6);
  67     QPX_LOAD(fpp1_1,r7,f7);
  68
  69     QPX_STORE(fpp2_1,r0,f0);
  70     QPX_LOAD(fpp1_2,r0,f0);
  71     QPX_STORE(fpp2_1,r1,f1);
  72     QPX_LOAD(fpp1_2,r1,f1);
  73     QPX_STORE(fpp2_1,r2,f2);
  74     QPX_LOAD(fpp1_2,r2,f2);
  75     QPX_STORE(fpp2_1,r3,f3);
  76     QPX_LOAD(fpp1_2,r3,f3);
  77     QPX_STORE(fpp2_1,r4,f4);
  78     QPX_LOAD(fpp1_2,r4,f4);
  79     QPX_STORE(fpp2_1,r5,f5);
  80     QPX_LOAD(fpp1_2,r5,f5);
  81     QPX_STORE(fpp2_1,r6,f6);
  82     QPX_LOAD(fpp1_2,r6,f6);
  83     QPX_STORE(fpp2_1,r7,f7);
  84     QPX_LOAD(fpp1_2,r7,f7);
  85
  86     QPX_STORE(fpp2_2,r0,f0);
  87     QPX_STORE(fpp2_2,r1,f1);
  88     QPX_STORE(fpp2_2,r2,f2);
  89     QPX_STORE(fpp2_2,r3,f3);
  90     QPX_STORE(fpp2_2,r4,f4);
  91     QPX_STORE(fpp2_2,r5,f5);
  92     QPX_STORE(fpp2_2,r6,f6);
  93     QPX_STORE(fpp2_2,r7,f7);
  94
  95     return 0;
  96 }
  97
  98 void CmiMemcpy_qpx (void *dst, const void *src, size_t n)
  99 {
 100   const char *s = src;
 101   char *d = dst;
 102   int n512 = n >> 9;
 103   while (n512 --) {
 104     quad_copy_512(d, s);
 105     d += 512;
 106     s += 512;
 107   }
 108
 109   if ( (n & 511UL) != 0 )
 110     memcpy (d, s, n & 511UL);
 111 }