mdw@git.distorted.org.uk Git - catacomb/blob - base/asm-common.h

   1 /// -*- mode: asm; asm-comment-char: ?/ -*-
   2 ///
   3 /// Fancy SIMD implementation of Salsa20
   4 ///
   5 /// (c) 2015 Straylight/Edgeware
   6 ///
   7
   8 ///----- Licensing notice ---------------------------------------------------
   9 ///
  10 /// This file is part of Catacomb.
  11 ///
  12 /// Catacomb is free software; you can redistribute it and/or modify
  13 /// it under the terms of the GNU Library General Public License as
  14 /// published by the Free Software Foundation; either version 2 of the
  15 /// License, or (at your option) any later version.
  16 ///
  17 /// Catacomb is distributed in the hope that it will be useful,
  18 /// but WITHOUT ANY WARRANTY; without even the implied warranty of
  19 /// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20 /// GNU Library General Public License for more details.
  21 ///
  22 /// You should have received a copy of the GNU Library General Public
  23 /// License along with Catacomb; if not, write to the Free
  24 /// Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
  25 /// MA 02111-1307, USA.
  26
  27 ///--------------------------------------------------------------------------
  28 /// General definitions.
  29
  30 // Some useful variables.
  31         .L$_subsec = 0
  32
  33 // Literal pools done the hard way.
  34 #define _LIT .text .L$_subsec + 1
  35 #define _ENDLIT .text .L$_subsec
  36 #define _LTORG .L$_subsec = .L$_subsec + 2; .text .L$_subsec
  37
  38 // Announcing an external function.
  39 #define FUNC(name)                                                      \
  40         .globl  F(name);                                                \
  41         TYPE_FUNC(name);                                                \
  42         .macro ENDFUNC; _ENDFUNC(name); .endm;                          \
  43         FUNC_PREHOOK(name);                                             \
  44 F(name):                                                                \
  45         FUNC_POSTHOOK(name)
  46
  47 // Marking the end of a function.
  48 #define _ENDFUNC(name)                                                  \
  49         .purgem ENDFUNC;                                                \
  50         SIZE_OBJ(name);                                                 \
  51         ENDFUNC_HOOK(name);                                             \
  52         _LTORG
  53
  54 // Make a helper function, if necessary.
  55 #define AUXFN(name)                                                     \
  56   .ifndef .L$_auxfn_def.name;                                           \
  57         .text 7128;                                                     \
  58         .macro _ENDAUXFN; _ENDAUXFN_TAIL(name); .endm;                  \
  59         FUNC_PREHOOK(name);                                             \
  60 name:
  61 #define _ENDAUXFN_TAIL(name)                                            \
  62         .purgem _ENDAUXFN;                                              \
  63         .text .L$_subsec;                                               \
  64         .L$_auxfn_def.name = 1
  65 #define ENDAUXFN _ENDAUXFN; .endif
  66
  67 ///--------------------------------------------------------------------------
  68 /// ELF-specific hacking.
  69
  70 #if __ELF__
  71
  72 #if __PIC__ || __PIE__
  73 #  define WANT_PIC 1
  74 #endif
  75
  76 #define TYPE_FUNC(name) .type name, STT_FUNC
  77
  78 #define SIZE_OBJ(name) .size name, . - name
  79
  80 #endif
  81
  82 ///--------------------------------------------------------------------------
  83 /// Windows-specific hacking.
  84
  85 #if ABI_WIN
  86
  87 #if CPUFAM_X86
  88 #  define F(name) _##name
  89 #endif
  90
  91 #endif
  92
  93 ///--------------------------------------------------------------------------
  94 /// x86- and amd64-specific hacking.
  95 ///
  96 /// It's (slightly) easier to deal with both of these in one go.
  97
  98 #if CPUFAM_X86 || CPUFAM_AMD64
  99
 100 // Set the function hooks.
 101 #define FUNC_PREHOOK(_) .balign 16
 102
 103 // On Windows, arrange to install stack-unwinding data.
 104 #if CPUFAM_AMD64 && ABI_WIN
 105 #  define FUNC_POSTHOOK(name) .seh_proc name
 106 #  define ENDFUNC_HOOK(_) .seh_endproc
 107 // Procedures are expected to invoke `.seh_setframe' if necessary, and
 108 // `.seh_pushreg' and friends, and `.seh_endprologue'.
 109 #endif
 110
 111 // Don't use the wretched AT&T syntax.  It's festooned with pointless
 112 // punctuation, and all of the data movement is backwards.  Ugh!
 113         .intel_syntax noprefix
 114
 115 // Call external subroutine at ADDR, possibly via PLT.
 116 .macro  callext addr
 117 #if WANT_PIC
 118         call    \addr@PLT
 119 #else
 120         call    \addr
 121 #endif
 122 .endm
 123
 124 // Do I need to arrange a spare GOT register?
 125 #if WANT_PIC && CPUFAM_X86
 126 #  define NEED_GOT 1
 127 #endif
 128 #define GOTREG ebx                      // Not needed in AMD64 so don't care.
 129
 130 // Maybe load GOT address into GOT.
 131 .macro  ldgot got=GOTREG
 132 #if WANT_PIC && CPUFAM_X86
 133   AUXFN(_ldgot.\got)
 134         mov     \got, [esp]
 135         ret
 136   ENDAUXFN
 137         call    _ldgot.\got
 138         add     \got, offset _GLOBAL_OFFSET_TABLE_
 139 #endif
 140 .endm
 141
 142 // Load address of external symbol ADDR into REG, maybe using GOT.
 143 .macro  leaext reg, addr, got=GOTREG
 144 #if WANT_PIC
 145 #  if CPUFAM_X86
 146         mov     \reg, [\got + \addr@GOT]
 147 #  endif
 148 #  if CPUFAM_AMD64
 149         mov     \reg, \addr@GOTPCREL[rip]
 150 #  endif
 151 #else
 152 #  if CPUFAM_X86
 153         mov     \reg, offset \addr
 154 #  endif
 155 #  if CPUFAM_AMD64
 156         lea     \reg, \addr[rip]
 157 #  endif
 158 #endif
 159 .endm
 160
 161 // Address expression (possibly using a base register, and a displacement)
 162 // referring to ADDR, which is within our module, maybe using GOT.
 163 #define INTADDR(...) INTADDR__0(__VA_ARGS__, GOTREG, dummy)
 164 #define INTADDR__0(addr, got, ...) INTADDR__1(addr, got)
 165 #if CPUFAM_AMD64
 166 #  define INTADDR__1(addr, got) addr + rip
 167 #elif WANT_PIC
 168 #  define INTADDR__1(addr, got) got + addr@GOTOFF
 169 #else
 170 #  define INTADDR__1(addr, got) addr
 171 #endif
 172
 173 // Permutations for SIMD instructions.  SHUF(D, C, B, A) is an immediate,
 174 // suitable for use in `pshufd' or `shufpd', which copies element D
 175 // (0 <= D < 4) of the source to element 3 of the destination, element C to
 176 // element 2, element B to element 1, and element A to element 0.
 177 #define SHUF(d, c, b, a) (64*(d) + 16*(c) + 4*(b) + (a))
 178
 179 #endif
 180
 181 ///--------------------------------------------------------------------------
 182 /// ARM-specific hacking.
 183
 184 #if CPUFAM_ARMEL
 185
 186 // ARM/Thumb mode things.  Use ARM by default.
 187 #define ARM .arm; .L$_pcoff = 8
 188 #define THUMB .thumb; .L$_pcoff = 4
 189         ARM
 190
 191 // Set the function hooks.
 192 #define FUNC_PREHOOK(_) .balign 4
 193 #define ENDFUNC_HOOK(name) .ltorg
 194
 195 // Call external subroutine at ADDR, possibly via PLT.
 196 .macro  callext addr, cond=
 197 #if WANT_PIC
 198         bl\cond \addr(PLT)
 199 #else
 200         bl\cond \addr
 201 #endif
 202 .endm
 203
 204 // Do I need to arrange a spare GOT register?
 205 #if WANT_PIC
 206 #  define NEED_GOT 1
 207 #endif
 208 #define GOTREG r9
 209
 210 // Maybe load GOT address into GOT.
 211 .macro  ldgot   cond=, got=GOTREG
 212 #if WANT_PIC
 213         ldr\cond \got, .L$_ldgot$\@
 214 .L$_ldgot_pc$\@:
 215         add\cond \got, pc, \got
 216   _LIT
 217         .balign 4
 218 .L$_ldgot$\@:
 219         .word   _GLOBAL_OFFSET_TABLE_ - .L$_ldgot_pc$\@ - .L$_pcoff
 220   _ENDLIT
 221 #endif
 222 .endm
 223
 224 // Load address of external symbol ADDR into REG, maybe using GOT.
 225 .macro  leaext  reg, addr, cond=, got=GOTREG
 226 #if WANT_PIC
 227         ldr\cond \reg, .L$_leaext$\@
 228         ldr\cond \reg, [\got, \reg]
 229   _LIT
 230         .balign 4
 231 .L$_leaext$\@:
 232         .word   \addr(GOT)
 233   _ENDLIT
 234 #else
 235         ldr\cond \reg, =\addr
 236 #endif
 237 .endm
 238
 239 // Load address of external symbol ADDR into REG directly.
 240 .macro  leaextq reg, addr, cond=
 241 #if WANT_PIC
 242         ldr\cond \reg, .L$_leaextq$\@
 243 .L$_leaextq_pc$\@:
 244         .if     .L$_pcoff == 8
 245         ldr\cond \reg, [pc, \reg]
 246         .else
 247         add\cond \reg, pc
 248         ldr\cond \reg, [\reg]
 249         .endif
 250   _LIT
 251         .balign 4
 252 .L$_leaextq$\@:
 253         .word   \addr(GOT_PREL) + (. - .L$_leaextq_pc$\@ - .L$_pcoff)
 254   _ENDLIT
 255 #else
 256         ldr\cond \reg, =\addr
 257 #endif
 258 .endm
 259
 260 #endif
 261
 262 ///--------------------------------------------------------------------------
 263 /// Final stuff.
 264
 265 // Default values for the various hooks.
 266 #ifndef FUNC_PREHOOK
 267 #  define FUNC_PREHOOK(name)
 268 #endif
 269 #ifndef FUNC_POSTHOOK
 270 #  define FUNC_POSTHOOK(name)
 271 #endif
 272 #ifndef ENDFUNC_HOOK
 273 #  define ENDFUNC_HOOK(name)
 274 #endif
 275
 276 #ifndef F
 277 #  define F(name) name
 278 #endif
 279
 280 #ifndef TYPE_FUNC
 281 #  define TYPE_FUNC(name)
 282 #endif
 283
 284 #ifndef SIZE_OBJ
 285 #  define SIZE_OBJ(name)
 286 #endif
 287
 288 ///----- That's all, folks --------------------------------------------------