libmpeg2/alpha_asm.h

   1 /*
   2  * Alpha assembly macros
   3  * Copyright (c) 2002-2003 Falk Hueffner <falk@debian.org>
   4  *
   5  * This file is part of mpeg2dec, a free MPEG-2 video stream decoder.
   6  * See http://libmpeg2.sourceforge.net/ for updates.
   7  *
   8  * mpeg2dec is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * mpeg2dec is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  21  */
  22
  23 #ifndef LIBMPEG2_ALPHA_ASM_H
  24 #define LIBMPEG2_ALPHA_ASM_H
  25
  26 #include <inttypes.h>
  27
  28 #if defined __GNUC__
  29 # define GNUC_PREREQ(maj, min) \
  30         ((__GNUC__ << 16) + __GNUC_MINOR__ >= ((maj) << 16) + (min))
  31 #else
  32 # define GNUC_PREREQ(maj, min) 0
  33 #endif
  34
  35 #define AMASK_BWX (1 << 0)
  36 #define AMASK_FIX (1 << 1)
  37 #define AMASK_CIX (1 << 2)
  38 #define AMASK_MVI (1 << 8)
  39
  40 #ifdef __alpha_bwx__
  41 # define HAVE_BWX() 1
  42 #else
  43 # define HAVE_BWX() (amask(AMASK_BWX) == 0)
  44 #endif
  45 #ifdef __alpha_fix__
  46 # define HAVE_FIX() 1
  47 #else
  48 # define HAVE_FIX() (amask(AMASK_FIX) == 0)
  49 #endif
  50 #ifdef __alpha_max__
  51 # define HAVE_MVI() 1
  52 #else
  53 # define HAVE_MVI() (amask(AMASK_MVI) == 0)
  54 #endif
  55 #ifdef __alpha_cix__
  56 # define HAVE_CIX() 1
  57 #else
  58 # define HAVE_CIX() (amask(AMASK_CIX) == 0)
  59 #endif
  60
  61 inline static uint64_t BYTE_VEC(uint64_t x)
  62 {
  63     x |= x <<  8;
  64     x |= x << 16;
  65     x |= x << 32;
  66     return x;
  67 }
  68 inline static uint64_t WORD_VEC(uint64_t x)
  69 {
  70     x |= x << 16;
  71     x |= x << 32;
  72     return x;
  73 }
  74
  75 #define ldq(p) (*(const uint64_t *) (p))
  76 #define ldl(p) (*(const int32_t *) (p))
  77 #define stl(l, p) do { *(uint32_t *) (p) = (l); } while (0)
  78 #define stq(l, p) do { *(uint64_t *) (p) = (l); } while (0)
  79 #define sextw(x) ((int16_t) (x))
  80
  81 #ifdef __GNUC__
  82 struct unaligned_long { uint64_t l; } __attribute__((packed));
  83 #define ldq_u(p)     (*(const uint64_t *) (((uint64_t) (p)) & ~7ul))
  84 #define uldq(a)      (((const struct unaligned_long *) (a))->l)
  85
  86 #if GNUC_PREREQ(3,3)
  87 #define prefetch(p)     __builtin_prefetch((p), 0, 1)
  88 #define prefetch_en(p)  __builtin_prefetch((p), 0, 0)
  89 #define prefetch_m(p)   __builtin_prefetch((p), 1, 1)
  90 #define prefetch_men(p) __builtin_prefetch((p), 1, 0)
  91 #define cmpbge  __builtin_alpha_cmpbge
  92 /* Avoid warnings.  */
  93 #define extql(a, b)     __builtin_alpha_extql(a, (uint64_t) (b))
  94 #define extwl(a, b)     __builtin_alpha_extwl(a, (uint64_t) (b))
  95 #define extqh(a, b)     __builtin_alpha_extqh(a, (uint64_t) (b))
  96 #define zap     __builtin_alpha_zap
  97 #define zapnot  __builtin_alpha_zapnot
  98 #define amask   __builtin_alpha_amask
  99 #define implver __builtin_alpha_implver
 100 #define rpcc    __builtin_alpha_rpcc
 101 #else
 102 #define prefetch(p)     asm volatile("ldl $31,%0"  : : "m"(*(const char *) (p)) : "memory")
 103 #define prefetch_en(p)  asm volatile("ldq $31,%0"  : : "m"(*(const char *) (p)) : "memory")
 104 #define prefetch_m(p)   asm volatile("lds $f31,%0" : : "m"(*(const char *) (p)) : "memory")
 105 #define prefetch_men(p) asm volatile("ldt $f31,%0" : : "m"(*(const char *) (p)) : "memory")
 106 #define cmpbge(a, b) ({ uint64_t __r; asm ("cmpbge  %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 107 #define extql(a, b)  ({ uint64_t __r; asm ("extql   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 108 #define extwl(a, b)  ({ uint64_t __r; asm ("extwl   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 109 #define extqh(a, b)  ({ uint64_t __r; asm ("extqh   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 110 #define zap(a, b)    ({ uint64_t __r; asm ("zap     %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 111 #define zapnot(a, b) ({ uint64_t __r; asm ("zapnot  %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 112 #define amask(a)     ({ uint64_t __r; asm ("amask   %1,%0"      : "=r" (__r) : "rI"  (a));           __r; })
 113 #define implver()    ({ uint64_t __r; asm ("implver %0"         : "=r" (__r));                       __r; })
 114 #define rpcc()       ({ uint64_t __r; asm volatile ("rpcc %0"   : "=r" (__r));                       __r; })
 115 #endif
 116 #define wh64(p) asm volatile("wh64 (%0)" : : "r"(p) : "memory")
 117
 118 #if GNUC_PREREQ(3,3) && defined(__alpha_max__)
 119 #define minub8  __builtin_alpha_minub8
 120 #define minsb8  __builtin_alpha_minsb8
 121 #define minuw4  __builtin_alpha_minuw4
 122 #define minsw4  __builtin_alpha_minsw4
 123 #define maxub8  __builtin_alpha_maxub8
 124 #define maxsb8  __builtin_alpha_maxsb8
 125 #define maxuw4  __builtin_alpha_maxuw4
 126 #define maxsw4  __builtin_alpha_maxsw4
 127 #define perr    __builtin_alpha_perr
 128 #define pklb    __builtin_alpha_pklb
 129 #define pkwb    __builtin_alpha_pkwb
 130 #define unpkbl  __builtin_alpha_unpkbl
 131 #define unpkbw  __builtin_alpha_unpkbw
 132 #else
 133 #define minub8(a, b) ({ uint64_t __r; asm (".arch ev6; minub8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 134 #define minsb8(a, b) ({ uint64_t __r; asm (".arch ev6; minsb8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 135 #define minuw4(a, b) ({ uint64_t __r; asm (".arch ev6; minuw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 136 #define minsw4(a, b) ({ uint64_t __r; asm (".arch ev6; minsw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 137 #define maxub8(a, b) ({ uint64_t __r; asm (".arch ev6; maxub8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 138 #define maxsb8(a, b) ({ uint64_t __r; asm (".arch ev6; maxsb8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 139 #define maxuw4(a, b) ({ uint64_t __r; asm (".arch ev6; maxuw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 140 #define maxsw4(a, b) ({ uint64_t __r; asm (".arch ev6; maxsw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 141 #define perr(a, b)   ({ uint64_t __r; asm (".arch ev6; perr    %r1,%r2,%0" : "=r" (__r) : "%rJ" (a), "rJ" (b)); __r; })
 142 #define pklb(a)      ({ uint64_t __r; asm (".arch ev6; pklb    %r1,%0"     : "=r" (__r) : "rJ"  (a));        __r; })
 143 #define pkwb(a)      ({ uint64_t __r; asm (".arch ev6; pkwb    %r1,%0"     : "=r" (__r) : "rJ"  (a));        __r; })
 144 #define unpkbl(a)    ({ uint64_t __r; asm (".arch ev6; unpkbl  %r1,%0"     : "=r" (__r) : "rJ"  (a));        __r; })
 145 #define unpkbw(a)    ({ uint64_t __r; asm (".arch ev6; unpkbw  %r1,%0"     : "=r" (__r) : "rJ"  (a));        __r; })
 146 #endif
 147
 148 #elif defined(__DECC)           /* Digital/Compaq/hp "ccc" compiler */
 149
 150 #include <c_asm.h>
 151 #define ldq_u(a)     asm ("ldq_u   %v0,0(%a0)", a)
 152 #define uldq(a)      (*(const __unaligned uint64_t *) (a))
 153 #define cmpbge(a, b) asm ("cmpbge  %a0,%a1,%v0", a, b)
 154 #define extql(a, b)  asm ("extql   %a0,%a1,%v0", a, b)
 155 #define extwl(a, b)  asm ("extwl   %a0,%a1,%v0", a, b)
 156 #define extqh(a, b)  asm ("extqh   %a0,%a1,%v0", a, b)
 157 #define zap(a, b)    asm ("zap     %a0,%a1,%v0", a, b)
 158 #define zapnot(a, b) asm ("zapnot  %a0,%a1,%v0", a, b)
 159 #define amask(a)     asm ("amask   %a0,%v0", a)
 160 #define implver()    asm ("implver %v0")
 161 #define rpcc()       asm ("rpcc    %v0")
 162 #define minub8(a, b) asm ("minub8  %a0,%a1,%v0", a, b)
 163 #define minsb8(a, b) asm ("minsb8  %a0,%a1,%v0", a, b)
 164 #define minuw4(a, b) asm ("minuw4  %a0,%a1,%v0", a, b)
 165 #define minsw4(a, b) asm ("minsw4  %a0,%a1,%v0", a, b)
 166 #define maxub8(a, b) asm ("maxub8  %a0,%a1,%v0", a, b)
 167 #define maxsb8(a, b) asm ("maxsb8  %a0,%a1,%v0", a, b)
 168 #define maxuw4(a, b) asm ("maxuw4  %a0,%a1,%v0", a, b)
 169 #define maxsw4(a, b) asm ("maxsw4  %a0,%a1,%v0", a, b)
 170 #define perr(a, b)   asm ("perr    %a0,%a1,%v0", a, b)
 171 #define pklb(a)      asm ("pklb    %a0,%v0", a)
 172 #define pkwb(a)      asm ("pkwb    %a0,%v0", a)
 173 #define unpkbl(a)    asm ("unpkbl  %a0,%v0", a)
 174 #define unpkbw(a)    asm ("unpkbw  %a0,%v0", a)
 175 #define wh64(a)      asm ("wh64    %a0", a)
 176
 177 #else
 178 #error "Unknown compiler!"
 179 #endif
 180
 181 #endif /* LIBMPEG2_ALPHA_ASM_H */