arch/powerpc/lib/xor_vmx.c

   1 /*
   2  * This program is free software; you can redistribute it and/or modify
   3  * it under the terms of the GNU General Public License as published by
   4  * the Free Software Foundation; either version 2 of the License, or
   5  * (at your option) any later version.
   6  *
   7  * This program is distributed in the hope that it will be useful,
   8  * but WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  10  * GNU General Public License for more details.
  11  *
  12  * You should have received a copy of the GNU General Public License
  13  * along with this program; if not, write to the Free Software
  14  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  15  *
  16  * Copyright (C) IBM Corporation, 2012
  17  *
  18  * Author: Anton Blanchard <anton@au.ibm.com>
  19  */
  20
  21 /*
  22  * Sparse (as at v0.5.0) gets very, very confused by this file.
  23  * Make it a bit simpler for it.
  24  */
  25 #if !defined(__CHECKER__)
  26 #include <altivec.h>
  27 #else
  28 #define vec_xor(a, b) a ^ b
  29 #define vector __attribute__((vector_size(16)))
  30 #endif
  31
  32 #include "xor_vmx.h"
  33
  34 typedef vector signed char unative_t;
  35
  36 #define DEFINE(V)                               \
  37         unative_t *V = (unative_t *)V##_in;     \
  38         unative_t V##_0, V##_1, V##_2, V##_3
  39
  40 #define LOAD(V)                 \
  41         do {                    \
  42                 V##_0 = V[0];   \
  43                 V##_1 = V[1];   \
  44                 V##_2 = V[2];   \
  45                 V##_3 = V[3];   \
  46         } while (0)
  47
  48 #define STORE(V)                \
  49         do {                    \
  50                 V[0] = V##_0;   \
  51                 V[1] = V##_1;   \
  52                 V[2] = V##_2;   \
  53                 V[3] = V##_3;   \
  54         } while (0)
  55
  56 #define XOR(V1, V2)                                     \
  57         do {                                            \
  58                 V1##_0 = vec_xor(V1##_0, V2##_0);       \
  59                 V1##_1 = vec_xor(V1##_1, V2##_1);       \
  60                 V1##_2 = vec_xor(V1##_2, V2##_2);       \
  61                 V1##_3 = vec_xor(V1##_3, V2##_3);       \
  62         } while (0)
  63
  64 void __xor_altivec_2(unsigned long bytes, unsigned long *v1_in,
  65                      unsigned long *v2_in)
  66 {
  67         DEFINE(v1);
  68         DEFINE(v2);
  69         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
  70
  71         do {
  72                 LOAD(v1);
  73                 LOAD(v2);
  74                 XOR(v1, v2);
  75                 STORE(v1);
  76
  77                 v1 += 4;
  78                 v2 += 4;
  79         } while (--lines > 0);
  80 }
  81
  82 void __xor_altivec_3(unsigned long bytes, unsigned long *v1_in,
  83                      unsigned long *v2_in, unsigned long *v3_in)
  84 {
  85         DEFINE(v1);
  86         DEFINE(v2);
  87         DEFINE(v3);
  88         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
  89
  90         do {
  91                 LOAD(v1);
  92                 LOAD(v2);
  93                 LOAD(v3);
  94                 XOR(v1, v2);
  95                 XOR(v1, v3);
  96                 STORE(v1);
  97
  98                 v1 += 4;
  99                 v2 += 4;
 100                 v3 += 4;
 101         } while (--lines > 0);
 102 }
 103
 104 void __xor_altivec_4(unsigned long bytes, unsigned long *v1_in,
 105                      unsigned long *v2_in, unsigned long *v3_in,
 106                      unsigned long *v4_in)
 107 {
 108         DEFINE(v1);
 109         DEFINE(v2);
 110         DEFINE(v3);
 111         DEFINE(v4);
 112         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
 113
 114         do {
 115                 LOAD(v1);
 116                 LOAD(v2);
 117                 LOAD(v3);
 118                 LOAD(v4);
 119                 XOR(v1, v2);
 120                 XOR(v3, v4);
 121                 XOR(v1, v3);
 122                 STORE(v1);
 123
 124                 v1 += 4;
 125                 v2 += 4;
 126                 v3 += 4;
 127                 v4 += 4;
 128         } while (--lines > 0);
 129 }
 130
 131 void __xor_altivec_5(unsigned long bytes, unsigned long *v1_in,
 132                      unsigned long *v2_in, unsigned long *v3_in,
 133                      unsigned long *v4_in, unsigned long *v5_in)
 134 {
 135         DEFINE(v1);
 136         DEFINE(v2);
 137         DEFINE(v3);
 138         DEFINE(v4);
 139         DEFINE(v5);
 140         unsigned long lines = bytes / (sizeof(unative_t)) / 4;
 141
 142         do {
 143                 LOAD(v1);
 144                 LOAD(v2);
 145                 LOAD(v3);
 146                 LOAD(v4);
 147                 LOAD(v5);
 148                 XOR(v1, v2);
 149                 XOR(v3, v4);
 150                 XOR(v1, v5);
 151                 XOR(v1, v3);
 152                 STORE(v1);
 153
 154                 v1 += 4;
 155                 v2 += 4;
 156                 v3 += 4;
 157                 v4 += 4;
 158                 v5 += 4;
 159         } while (--lines > 0);
 160 }