falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1179 Commits
32 Branches
552MB
Tree: e41e834272
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e41e834272'
${ noResults }
FFmpeg/libavcodec/alpha/asm.h

135 lines
6.1KB
Raw Blame History 

  1. /*

  2.  * Alpha optimized DSP utils

  3.  * Copyright (c) 2002 Falk Hueffner <falk@debian.org>

  4.  *

  5.  * This library is free software; you can redistribute it and/or

  6.  * modify it under the terms of the GNU Lesser General Public

  7.  * License as published by the Free Software Foundation; either

  8.  * version 2 of the License, or (at your option) any later version.

  9.  *

  10.  * This library is distributed in the hope that it will be useful,

  11.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  12.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  13.  * Lesser General Public License for more details.

  14.  *

  15.  * You should have received a copy of the GNU Lesser General Public

  16.  * License along with this library; if not, write to the Free Software

  17.  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  18.  */

  19. 

  20. #ifndef LIBAVCODEC_ALPHA_ASM_H

  21. #define LIBAVCODEC_ALPHA_ASM_H

  22. 

  23. #include <inttypes.h>

  24. 

  25. #define AMASK_BWX (1 << 0)

  26. #define AMASK_FIX (1 << 1)

  27. #define AMASK_CIX (1 << 2)

  28. #define AMASK_MVI (1 << 8)

  29. 

  30. inline static uint64_t BYTE_VEC(uint64_t x)

  31. {

  32.     x |= x <<  8;

  33.     x |= x << 16;

  34.     x |= x << 32;

  35.     return x;

  36. }

  37. inline static uint64_t WORD_VEC(uint64_t x)

  38. {

  39.     x |= x << 16;

  40.     x |= x << 32;

  41.     return x;

  42. }

  43. 

  44. #define ldq(p) (*(const uint64_t *) (p))

  45. #define ldl(p) (*(const int32_t *) (p))

  46. #define stl(l, p) do { *(uint32_t *) (p) = (l); } while (0)

  47. #define stq(l, p) do { *(uint64_t *) (p) = (l); } while (0)

  48. 

  49. #ifdef __GNUC__

  50. #define ASM_ACCEPT_MVI asm (".arch pca56")

  51. struct unaligned_long { uint64_t l; } __attribute__((packed));

  52. #define ldq_u(p)     (*(const uint64_t *) (((uint64_t) (p)) & ~7ul))

  53. #define uldq(a)	     (((const struct unaligned_long *) (a))->l)

  54. 

  55. #if __GNUC__ >= 3 && __GNUC_MINOR__ >= 3

  56. #define cmpbge	__builtin_alpha_cmpbge

  57. /* Avoid warnings.  */

  58. #define extql(a, b)	__builtin_alpha_extql(a, (uint64_t) (b))

  59. #define extqh(a, b)	__builtin_alpha_extqh(a, (uint64_t) (b))

  60. #define zap	__builtin_alpha_zap

  61. #define zapnot	__builtin_alpha_zapnot

  62. #define amask	__builtin_alpha_amask

  63. #define implver	__builtin_alpha_implver

  64. #define rpcc	__builtin_alpha_rpcc

  65. #define minub8	__builtin_alpha_minub8

  66. #define minsb8	__builtin_alpha_minsb8

  67. #define minuw4	__builtin_alpha_minuw4

  68. #define minsw4	__builtin_alpha_minsw4

  69. #define maxub8	__builtin_alpha_maxub8

  70. #define maxsb8	__builtin_alpha_maxsb8

  71. #define maxuw4	__builtin_alpha_maxuw4	

  72. #define maxsw4	__builtin_alpha_maxsw4

  73. #define perr	__builtin_alpha_perr

  74. #define pklb	__builtin_alpha_pklb

  75. #define pkwb	__builtin_alpha_pkwb

  76. #define unpkbl	__builtin_alpha_unpkbl

  77. #define unpkbw	__builtin_alpha_unpkbw

  78. #else

  79. #define cmpbge(a, b) ({ uint64_t __r; asm ("cmpbge  %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })

  80. #define extql(a, b)  ({ uint64_t __r; asm ("extql   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })

  81. #define extqh(a, b)  ({ uint64_t __r; asm ("extqh   %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })

  82. #define zap(a, b)    ({ uint64_t __r; asm ("zap     %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })

  83. #define zapnot(a, b) ({ uint64_t __r; asm ("zapnot  %r1,%2,%0"  : "=r" (__r) : "rJ"  (a), "rI" (b)); __r; })

  84. #define amask(a)     ({ uint64_t __r; asm ("amask   %1,%0"      : "=r" (__r) : "rI"  (a));	     __r; })

  85. #define implver()    ({ uint64_t __r; asm ("implver %0"         : "=r" (__r));			     __r; })

  86. #define rpcc()	     ({ uint64_t __r; asm volatile ("rpcc %0"   : "=r" (__r));			     __r; })

  87. #define minub8(a, b) ({ uint64_t __r; asm ("minub8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  88. #define minsb8(a, b) ({ uint64_t __r; asm ("minsb8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  89. #define minuw4(a, b) ({ uint64_t __r; asm ("minuw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  90. #define minsw4(a, b) ({ uint64_t __r; asm ("minsw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  91. #define maxub8(a, b) ({ uint64_t __r; asm ("maxub8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  92. #define maxsb8(a, b) ({ uint64_t __r; asm ("maxsb8  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  93. #define maxuw4(a, b) ({ uint64_t __r; asm ("maxuw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  94. #define maxsw4(a, b) ({ uint64_t __r; asm ("maxsw4  %r1,%2,%0"  : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })

  95. #define perr(a, b)   ({ uint64_t __r; asm ("perr    %r1,%r2,%0" : "=r" (__r) : "%rJ" (a), "rJ" (b)); __r; })

  96. #define pklb(a)      ({ uint64_t __r; asm ("pklb    %r1,%0"     : "=r" (__r) : "rJ"  (a));	     __r; })

  97. #define pkwb(a)      ({ uint64_t __r; asm ("pkwb    %r1,%0"     : "=r" (__r) : "rJ"  (a));	     __r; })

  98. #define unpkbl(a)    ({ uint64_t __r; asm ("unpkbl  %r1,%0"     : "=r" (__r) : "rJ"  (a));	     __r; })

  99. #define unpkbw(a)    ({ uint64_t __r; asm ("unpkbw  %r1,%0"     : "=r" (__r) : "rJ"  (a));	     __r; })

  100. #endif

  101. 

  102. #elif defined(__DECC)		/* Digital/Compaq "ccc" compiler */

  103. 

  104. #include <c_asm.h>

  105. #define ASM_ACCEPT_MVI

  106. #define ldq_u(a)     asm ("ldq_u   %v0,0(%a0)", a)

  107. #define uldq(a)	     (*(const __unaligned uint64_t *) (a))

  108. #define cmpbge(a, b) asm ("cmpbge  %a0,%a1,%v0", a, b)

  109. #define extql(a, b)  asm ("extql   %a0,%a1,%v0", a, b)

  110. #define extqh(a, b)  asm ("extqh   %a0,%a1,%v0", a, b)

  111. #define zap(a, b)    asm ("zap     %a0,%a1,%v0", a, b)

  112. #define zapnot(a, b) asm ("zapnot  %a0,%a1,%v0", a, b)

  113. #define amask(a)     asm ("amask   %a0,%v0", a)

  114. #define implver()    asm ("implver %v0")

  115. #define rpcc()	     asm ("rpcc	   %v0")

  116. #define minub8(a, b) asm ("minub8  %a0,%a1,%v0", a, b)

  117. #define minsb8(a, b) asm ("minsb8  %a0,%a1,%v0", a, b)

  118. #define minuw4(a, b) asm ("minuw4  %a0,%a1,%v0", a, b)

  119. #define minsw4(a, b) asm ("minsw4  %a0,%a1,%v0", a, b)

  120. #define maxub8(a, b) asm ("maxub8  %a0,%a1,%v0", a, b)

  121. #define maxsb8(a, b) asm ("maxsb8  %a0,%a1,%v0", a, b)

  122. #define maxuw4(a, b) asm ("maxuw4  %a0,%a1,%v0", a, b)

  123. #define maxsw4(a, b) asm ("maxsw4  %a0,%a1,%v0", a, b)

  124. #define perr(a, b)   asm ("perr    %a0,%a1,%v0", a, b)

  125. #define pklb(a)      asm ("pklb    %a0,%v0", a)

  126. #define pkwb(a)      asm ("pkwb    %a0,%v0", a)

  127. #define unpkbl(a)    asm ("unpkbl  %a0,%v0", a)

  128. #define unpkbw(a)    asm ("unpkbw  %a0,%v0", a)

  129. 

  130. #else

  131. #error "Unknown compiler!"

  132. #endif

  133. 

  134. #endif /* LIBAVCODEC_ALPHA_ASM_H */