asm.h

   1 /*
   2  * Alpha optimized DSP utils
   3  * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #ifndef AVCODEC_ALPHA_ASM_H
  23 #define AVCODEC_ALPHA_ASM_H
  24
  25 #include <inttypes.h>
  26
  27 #include "libavutil/common.h"
  28
  29 #if AV_GCC_VERSION_AT_LEAST(2,96)
  30 # define likely(x)      __builtin_expect((x) != 0, 1)
  31 # define unlikely(x)    __builtin_expect((x) != 0, 0)
  32 #else
  33 # define likely(x)      (x)
  34 # define unlikely(x)    (x)
  35 #endif
  36
  37 #define AMASK_BWX (1 << 0)
  38 #define AMASK_FIX (1 << 1)
  39 #define AMASK_CIX (1 << 2)
  40 #define AMASK_MVI (1 << 8)
  41
  42 static inline uint64_t BYTE_VEC(uint64_t x)
  43 {
  44     x |= x <<  8;
  45     x |= x << 16;
  46     x |= x << 32;
  47     return x;
  48 }
  49 static inline uint64_t WORD_VEC(uint64_t x)
  50 {
  51     x |= x << 16;
  52     x |= x << 32;
  53     return x;
  54 }
  55
  56 #define sextw(x) ((int16_t) (x))
  57
  58 #ifdef __GNUC__
  59 #define ldq(p)                                                  \
  60     (((const union {                                            \
  61         uint64_t __l;                                           \
  62         __typeof__(*(p)) __s[sizeof (uint64_t) / sizeof *(p)];  \
  63     } *) (p))->__l)
  64 #define ldl(p)                                                  \
  65     (((const union {                                            \
  66         int32_t __l;                                            \
  67         __typeof__(*(p)) __s[sizeof (int32_t) / sizeof *(p)];   \
  68     } *) (p))->__l)
  69 #define stq(l, p)                                                       \
  70     do {                                                                \
  71         (((union {                                                      \
  72             uint64_t __l;                                               \
  73             __typeof__(*(p)) __s[sizeof (uint64_t) / sizeof *(p)];      \
  74         } *) (p))->__l) = l;                                            \
  75     } while (0)
  76 #define stl(l, p)                                                       \
  77     do {                                                                \
  78         (((union {                                                      \
  79             int32_t __l;                                                \
  80             __typeof__(*(p)) __s[sizeof (int32_t) / sizeof *(p)];       \
  81         } *) (p))->__l) = l;                                            \
  82     } while (0)
  83 struct unaligned_long { uint64_t l; } __attribute__((packed));
  84 #define ldq_u(p)        (*(const uint64_t *) (((uint64_t) (p)) & ~7ul))
  85 #define uldq(a)         (((const struct unaligned_long *) (a))->l)
  86
  87 #if AV_GCC_VERSION_AT_LEAST(3,3)
  88 #define prefetch(p)     __builtin_prefetch((p), 0, 1)
  89 #define prefetch_en(p)  __builtin_prefetch((p), 0, 0)
  90 #define prefetch_m(p)   __builtin_prefetch((p), 1, 1)
  91 #define prefetch_men(p) __builtin_prefetch((p), 1, 0)
  92 #define cmpbge          __builtin_alpha_cmpbge
  93 /* Avoid warnings.  */
  94 #define extql(a, b)     __builtin_alpha_extql(a, (uint64_t) (b))
  95 #define extwl(a, b)     __builtin_alpha_extwl(a, (uint64_t) (b))
  96 #define extqh(a, b)     __builtin_alpha_extqh(a, (uint64_t) (b))
  97 #define zap             __builtin_alpha_zap
  98 #define zapnot          __builtin_alpha_zapnot
  99 #define amask           __builtin_alpha_amask
 100 #define implver         __builtin_alpha_implver
 101 #define rpcc            __builtin_alpha_rpcc
 102 #else
 103 #define prefetch(p)     __asm__ volatile("ldl $31,%0"  : : "m"(*(const char *) (p)) : "memory")
 104 #define prefetch_en(p)  __asm__ volatile("ldq $31,%0"  : : "m"(*(const char *) (p)) : "memory")
 105 #define prefetch_m(p)   __asm__ volatile("lds $f31,%0" : : "m"(*(const char *) (p)) : "memory")
 106 #define prefetch_men(p) __asm__ volatile("ldt $f31,%0" : : "m"(*(const char *) (p)) : "memory")
 107 #define cmpbge(a, b) ({ uint64_t __r; __asm__ ("cmpbge  %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 108 #define extql(a, b)  ({ uint64_t __r; __asm__ ("extql   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 109 #define extwl(a, b)  ({ uint64_t __r; __asm__ ("extwl   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 110 #define extqh(a, b)  ({ uint64_t __r; __asm__ ("extqh   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 111 #define zap(a, b)    ({ uint64_t __r; __asm__ ("zap     %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 112 #define zapnot(a, b) ({ uint64_t __r; __asm__ ("zapnot  %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })
 113 #define amask(a)     ({ uint64_t __r; __asm__ ("amask   %1,%0"      : "=r" (__r) : "rI"  (a));           __r; })
 114 #define implver()    ({ uint64_t __r; __asm__ ("implver %0"         : "=r" (__r));                       __r; })
 115 #define rpcc()       ({ uint64_t __r; __asm__ volatile ("rpcc %0"   : "=r" (__r));                       __r; })
 116 #endif
 117 #define wh64(p) __asm__ volatile("wh64 (%0)" : : "r"(p) : "memory")
 118
 119 #if AV_GCC_VERSION_AT_LEAST(3,3) && defined(__alpha_max__)
 120 #define minub8  __builtin_alpha_minub8
 121 #define minsb8  __builtin_alpha_minsb8
 122 #define minuw4  __builtin_alpha_minuw4
 123 #define minsw4  __builtin_alpha_minsw4
 124 #define maxub8  __builtin_alpha_maxub8
 125 #define maxsb8  __builtin_alpha_maxsb8
 126 #define maxuw4  __builtin_alpha_maxuw4
 127 #define maxsw4  __builtin_alpha_maxsw4
 128 #define perr    __builtin_alpha_perr
 129 #define pklb    __builtin_alpha_pklb
 130 #define pkwb    __builtin_alpha_pkwb
 131 #define unpkbl  __builtin_alpha_unpkbl
 132 #define unpkbw  __builtin_alpha_unpkbw
 133 #else
 134 #define minub8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minub8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 135 #define minsb8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minsb8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 136 #define minuw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minuw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 137 #define minsw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minsw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 138 #define maxub8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxub8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 139 #define maxsb8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxsb8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 140 #define maxuw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxuw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 141 #define maxsw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxsw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
 142 #define perr(a, b)   ({ uint64_t __r; __asm__ (".arch ev6; perr    %r1,%r2,%0" : "=r" (__r) : "%rJ" (a), "rJ" (b)); __r; })
 143 #define pklb(a)      ({ uint64_t __r; __asm__ (".arch ev6; pklb    %r1,%0"     : "=r" (__r) : "rJ"  (a));           __r; })
 144 #define pkwb(a)      ({ uint64_t __r; __asm__ (".arch ev6; pkwb    %r1,%0"     : "=r" (__r) : "rJ"  (a));           __r; })
 145 #define unpkbl(a)    ({ uint64_t __r; __asm__ (".arch ev6; unpkbl  %r1,%0"     : "=r" (__r) : "rJ"  (a));           __r; })
 146 #define unpkbw(a)    ({ uint64_t __r; __asm__ (".arch ev6; unpkbw  %r1,%0"     : "=r" (__r) : "rJ"  (a));           __r; })
 147 #endif
 148
 149 #elif defined(__DECC)           /* Digital/Compaq/hp "ccc" compiler */
 150
 151 #include <c_asm.h>
 152 #define ldq(p) (*(const uint64_t *) (p))
 153 #define ldl(p) (*(const int32_t *)  (p))
 154 #define stq(l, p) do { *(uint64_t *) (p) = (l); } while (0)
 155 #define stl(l, p) do { *(int32_t *)  (p) = (l); } while (0)
 156 #define ldq_u(a)     asm ("ldq_u   %v0,0(%a0)", a)
 157 #define uldq(a)      (*(const __unaligned uint64_t *) (a))
 158 #define cmpbge(a, b) asm ("cmpbge  %a0,%a1,%v0", a, b)
 159 #define extql(a, b)  asm ("extql   %a0,%a1,%v0", a, b)
 160 #define extwl(a, b)  asm ("extwl   %a0,%a1,%v0", a, b)
 161 #define extqh(a, b)  asm ("extqh   %a0,%a1,%v0", a, b)
 162 #define zap(a, b)    asm ("zap     %a0,%a1,%v0", a, b)
 163 #define zapnot(a, b) asm ("zapnot  %a0,%a1,%v0", a, b)
 164 #define amask(a)     asm ("amask   %a0,%v0", a)
 165 #define implver()    asm ("implver %v0")
 166 #define rpcc()       asm ("rpcc           %v0")
 167 #define minub8(a, b) asm ("minub8  %a0,%a1,%v0", a, b)
 168 #define minsb8(a, b) asm ("minsb8  %a0,%a1,%v0", a, b)
 169 #define minuw4(a, b) asm ("minuw4  %a0,%a1,%v0", a, b)
 170 #define minsw4(a, b) asm ("minsw4  %a0,%a1,%v0", a, b)
 171 #define maxub8(a, b) asm ("maxub8  %a0,%a1,%v0", a, b)
 172 #define maxsb8(a, b) asm ("maxsb8  %a0,%a1,%v0", a, b)
 173 #define maxuw4(a, b) asm ("maxuw4  %a0,%a1,%v0", a, b)
 174 #define maxsw4(a, b) asm ("maxsw4  %a0,%a1,%v0", a, b)
 175 #define perr(a, b)   asm ("perr    %a0,%a1,%v0", a, b)
 176 #define pklb(a)      asm ("pklb    %a0,%v0", a)
 177 #define pkwb(a)      asm ("pkwb    %a0,%v0", a)
 178 #define unpkbl(a)    asm ("unpkbl  %a0,%v0", a)
 179 #define unpkbw(a)    asm ("unpkbw  %a0,%v0", a)
 180 #define wh64(a)      asm ("wh64    %a0", a)
 181
 182 #else
 183 #error "Unknown compiler!"
 184 #endif
 185
 186 #endif /* AVCODEC_ALPHA_ASM_H */