src/gallium/auxiliary/rtasm/rtasm_x86sse.h

   1 /**************************************************************************
   2  *
   3  * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be included
  13  * in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  16  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  19  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  20  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  21  *
  22  **************************************************************************/
  23
  24 #ifndef _RTASM_X86SSE_H_
  25 #define _RTASM_X86SSE_H_
  26
  27 #include "pipe/p_config.h"
  28
  29 #if defined(PIPE_ARCH_X86)
  30
  31 /* It is up to the caller to ensure that instructions issued are
  32  * suitable for the host cpu.  There are no checks made in this module
  33  * for mmx/sse/sse2 support on the cpu.
  34  */
  35 struct x86_reg {
  36    unsigned file:3;
  37    unsigned idx:3;
  38    unsigned mod:2;              /* mod_REG if this is just a register */
  39    int      disp:24;            /* only +/- 23bits of offset - should be enough... */
  40 };
  41
  42 struct x86_function {
  43    unsigned size;
  44    unsigned char *store;
  45    unsigned char *csr;
  46
  47    unsigned stack_offset:16;
  48    unsigned need_emms:8;
  49    int x87_stack:8;
  50
  51    unsigned char error_overflow[4];
  52 };
  53
  54 enum x86_reg_file {
  55    file_REG32,
  56    file_MMX,
  57    file_XMM,
  58    file_x87
  59 };
  60
  61 /* Values for mod field of modr/m byte
  62  */
  63 enum x86_reg_mod {
  64    mod_INDIRECT,
  65    mod_DISP8,
  66    mod_DISP32,
  67    mod_REG
  68 };
  69
  70 enum x86_reg_name {
  71    reg_AX,
  72    reg_CX,
  73    reg_DX,
  74    reg_BX,
  75    reg_SP,
  76    reg_BP,
  77    reg_SI,
  78    reg_DI
  79 };
  80
  81
  82 enum x86_cc {
  83    cc_O,                        /* overflow */
  84    cc_NO,                       /* not overflow */
  85    cc_NAE,                      /* not above or equal / carry */
  86    cc_AE,                       /* above or equal / not carry */
  87    cc_E,                        /* equal / zero */
  88    cc_NE                        /* not equal / not zero */
  89 };
  90
  91 enum sse_cc {
  92    cc_Equal,
  93    cc_LessThan,
  94    cc_LessThanEqual,
  95    cc_Unordered,
  96    cc_NotEqual,
  97    cc_NotLessThan,
  98    cc_NotLessThanEqual,
  99    cc_Ordered
 100 };
 101
 102 #define cc_Z  cc_E
 103 #define cc_NZ cc_NE
 104
 105 /* Begin/end/retrieve function creation:
 106  */
 107
 108
 109 void x86_init_func( struct x86_function *p );
 110 void x86_init_func_size( struct x86_function *p, unsigned code_size );
 111 void x86_release_func( struct x86_function *p );
 112 void (*x86_get_func( struct x86_function *p ))( void );
 113
 114 /* Debugging:
 115  */
 116 void x86_print_reg( struct x86_reg reg );
 117
 118
 119 /* Create and manipulate registers and regmem values:
 120  */
 121 struct x86_reg x86_make_reg( enum x86_reg_file file,
 122                              enum x86_reg_name idx );
 123
 124 struct x86_reg x86_make_disp( struct x86_reg reg,
 125                               int disp );
 126
 127 struct x86_reg x86_deref( struct x86_reg reg );
 128
 129 struct x86_reg x86_get_base_reg( struct x86_reg reg );
 130
 131
 132 /* Labels, jumps and fixup:
 133  */
 134 int x86_get_label( struct x86_function *p );
 135
 136 void x86_jcc( struct x86_function *p,
 137               enum x86_cc cc,
 138               int label );
 139
 140 int x86_jcc_forward( struct x86_function *p,
 141                           enum x86_cc cc );
 142
 143 int x86_jmp_forward( struct x86_function *p);
 144
 145 int x86_call_forward( struct x86_function *p);
 146
 147 void x86_fixup_fwd_jump( struct x86_function *p,
 148                          int fixup );
 149
 150 void x86_jmp( struct x86_function *p, int label );
 151
 152 /* void x86_call( struct x86_function *p, void (*label)() ); */
 153 void x86_call( struct x86_function *p, struct x86_reg reg);
 154
 155 void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm );
 156 void x86_add_imm( struct x86_function *p, struct x86_reg dst, int imm );
 157 void x86_or_imm( struct x86_function *p, struct x86_reg dst, int imm );
 158 void x86_and_imm( struct x86_function *p, struct x86_reg dst, int imm );
 159 void x86_sub_imm( struct x86_function *p, struct x86_reg dst, int imm );
 160 void x86_xor_imm( struct x86_function *p, struct x86_reg dst, int imm );
 161 void x86_cmp_imm( struct x86_function *p, struct x86_reg dst, int imm );
 162
 163
 164 /* Macro for sse_shufps() and sse2_pshufd():
 165  */
 166 #define SHUF(_x,_y,_z,_w)       (((_x)<<0) | ((_y)<<2) | ((_z)<<4) | ((_w)<<6))
 167 #define SHUF_NOOP               RSW(0,1,2,3)
 168 #define GET_SHUF(swz, idx)      (((swz) >> ((idx)*2)) & 0x3)
 169
 170 void mmx_emms( struct x86_function *p );
 171 void mmx_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 172 void mmx_movq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 173 void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 174 void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 175
 176 void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 177 void sse2_cvttps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 178 void sse2_cvtdq2ps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 179 void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 180 void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 181 void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 182 void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 183 void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
 184                   unsigned char shuf );
 185 void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 186 void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 187
 188
 189 void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr);
 190 void sse_prefetch0( struct x86_function *p, struct x86_reg ptr);
 191 void sse_prefetch1( struct x86_function *p, struct x86_reg ptr);
 192
 193 void sse_movntps( struct x86_function *p, struct x86_reg dst, struct x86_reg src);
 194
 195 void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 196 void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 197 void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 198 void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 199 void sse_andnps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 200 void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 201 void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src,
 202                 enum sse_cc cc );
 203 void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 204 void sse_maxss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 205 void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 206 void sse_movaps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 207 void sse_movhlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 208 void sse_movhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 209 void sse_movlhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 210 void sse_movlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 211 void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 212 void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 213 void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 214 void sse_mulss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 215 void sse_orps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 216 void sse_xorps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 217 void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 218 void sse_rsqrtps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 219 void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 220 void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
 221                  unsigned char shuf );
 222 void sse_unpckhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 223 void sse_unpcklps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 224 void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src );
 225 void sse2_punpcklbw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 226 void sse_movmskps( struct x86_function *p, struct x86_reg dst, struct x86_reg src);
 227
 228 void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 229 void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 230 void x86_cmp( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 231 void x86_dec( struct x86_function *p, struct x86_reg reg );
 232 void x86_inc( struct x86_function *p, struct x86_reg reg );
 233 void x86_lea( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 234 void x86_mov( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 235 void x86_mul( struct x86_function *p, struct x86_reg src );
 236 void x86_imul( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 237 void x86_or( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 238 void x86_pop( struct x86_function *p, struct x86_reg reg );
 239 void x86_push( struct x86_function *p, struct x86_reg reg );
 240 void x86_push_imm32( struct x86_function *p, int imm );
 241 void x86_ret( struct x86_function *p );
 242 void x86_retw( struct x86_function *p, unsigned short imm );
 243 void x86_sub( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 244 void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 245 void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 246 void x86_sahf( struct x86_function *p );
 247 void x86_div( struct x86_function *p, struct x86_reg src );
 248
 249
 250 void x86_cdecl_caller_push_regs( struct x86_function *p );
 251 void x86_cdecl_caller_pop_regs( struct x86_function *p );
 252
 253 void x87_assert_stack_empty( struct x86_function *p );
 254
 255 void x87_f2xm1( struct x86_function *p );
 256 void x87_fabs( struct x86_function *p );
 257 void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 258 void x87_faddp( struct x86_function *p, struct x86_reg dst );
 259 void x87_fchs( struct x86_function *p );
 260 void x87_fclex( struct x86_function *p );
 261 void x87_fcmovb( struct x86_function *p, struct x86_reg src );
 262 void x87_fcmovbe( struct x86_function *p, struct x86_reg src );
 263 void x87_fcmove( struct x86_function *p, struct x86_reg src );
 264 void x87_fcmovnb( struct x86_function *p, struct x86_reg src );
 265 void x87_fcmovnbe( struct x86_function *p, struct x86_reg src );
 266 void x87_fcmovne( struct x86_function *p, struct x86_reg src );
 267 void x87_fcom( struct x86_function *p, struct x86_reg dst );
 268 void x87_fcomi( struct x86_function *p, struct x86_reg dst );
 269 void x87_fcomip( struct x86_function *p, struct x86_reg dst );
 270 void x87_fcomp( struct x86_function *p, struct x86_reg dst );
 271 void x87_fcos( struct x86_function *p );
 272 void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 273 void x87_fdivp( struct x86_function *p, struct x86_reg dst );
 274 void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 275 void x87_fdivrp( struct x86_function *p, struct x86_reg dst );
 276 void x87_fild( struct x86_function *p, struct x86_reg arg );
 277 void x87_fist( struct x86_function *p, struct x86_reg dst );
 278 void x87_fistp( struct x86_function *p, struct x86_reg dst );
 279 void x87_fld( struct x86_function *p, struct x86_reg arg );
 280 void x87_fld1( struct x86_function *p );
 281 void x87_fldcw( struct x86_function *p, struct x86_reg arg );
 282 void x87_fldl2e( struct x86_function *p );
 283 void x87_fldln2( struct x86_function *p );
 284 void x87_fldz( struct x86_function *p );
 285 void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 286 void x87_fmulp( struct x86_function *p, struct x86_reg dst );
 287 void x87_fnclex( struct x86_function *p );
 288 void x87_fprndint( struct x86_function *p );
 289 void x87_fpop( struct x86_function *p );
 290 void x87_fscale( struct x86_function *p );
 291 void x87_fsin( struct x86_function *p );
 292 void x87_fsincos( struct x86_function *p );
 293 void x87_fsqrt( struct x86_function *p );
 294 void x87_fst( struct x86_function *p, struct x86_reg dst );
 295 void x87_fstp( struct x86_function *p, struct x86_reg dst );
 296 void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 297 void x87_fsubp( struct x86_function *p, struct x86_reg dst );
 298 void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 299 void x87_fsubrp( struct x86_function *p, struct x86_reg dst );
 300 void x87_ftst( struct x86_function *p );
 301 void x87_fxch( struct x86_function *p, struct x86_reg dst );
 302 void x87_fxtract( struct x86_function *p );
 303 void x87_fyl2x( struct x86_function *p );
 304 void x87_fyl2xp1( struct x86_function *p );
 305 void x87_fwait( struct x86_function *p );
 306 void x87_fnstcw( struct x86_function *p, struct x86_reg dst );
 307 void x87_fnstsw( struct x86_function *p, struct x86_reg dst );
 308 void x87_fucompp( struct x86_function *p );
 309 void x87_fucomp( struct x86_function *p, struct x86_reg arg );
 310 void x87_fucom( struct x86_function *p, struct x86_reg arg );
 311
 312
 313
 314 /* Retrieve a reference to one of the function arguments, taking into
 315  * account any push/pop activity.  Note - doesn't track explicit
 316  * manipulation of ESP by other instructions.
 317  */
 318 struct x86_reg x86_fn_arg( struct x86_function *p, unsigned arg );
 319
 320 #endif
 321 #endif