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 #if defined(__i386__) || defined(__386__) || defined(i386)
  28
  29 /* It is up to the caller to ensure that instructions issued are
  30  * suitable for the host cpu.  There are no checks made in this module
  31  * for mmx/sse/sse2 support on the cpu.
  32  */
  33 struct x86_reg {
  34    unsigned file:3;
  35    unsigned idx:3;
  36    unsigned mod:2;              /* mod_REG if this is just a register */
  37    int      disp:24;            /* only +/- 23bits of offset - should be enough... */
  38 };
  39
  40 struct x86_function {
  41    unsigned size;
  42    unsigned char *store;
  43    unsigned char *csr;
  44    unsigned stack_offset;
  45    int need_emms;
  46    unsigned char error_overflow[4];
  47    const char *fn;
  48 };
  49
  50 enum x86_reg_file {
  51    file_REG32,
  52    file_MMX,
  53    file_XMM,
  54    file_x87
  55 };
  56
  57 /* Values for mod field of modr/m byte
  58  */
  59 enum x86_reg_mod {
  60    mod_INDIRECT,
  61    mod_DISP8,
  62    mod_DISP32,
  63    mod_REG
  64 };
  65
  66 enum x86_reg_name {
  67    reg_AX,
  68    reg_CX,
  69    reg_DX,
  70    reg_BX,
  71    reg_SP,
  72    reg_BP,
  73    reg_SI,
  74    reg_DI
  75 };
  76
  77
  78 enum x86_cc {
  79    cc_O,                        /* overflow */
  80    cc_NO,                       /* not overflow */
  81    cc_NAE,                      /* not above or equal / carry */
  82    cc_AE,                       /* above or equal / not carry */
  83    cc_E,                        /* equal / zero */
  84    cc_NE                        /* not equal / not zero */
  85 };
  86
  87 enum sse_cc {
  88    cc_Equal,
  89    cc_LessThan,
  90    cc_LessThanEqual,
  91    cc_Unordered,
  92    cc_NotEqual,
  93    cc_NotLessThan,
  94    cc_NotLessThanEqual,
  95    cc_Ordered
  96 };
  97
  98 #define cc_Z  cc_E
  99 #define cc_NZ cc_NE
 100
 101 /* Begin/end/retreive function creation:
 102  */
 103
 104
 105 void x86_init_func( struct x86_function *p );
 106 void x86_init_func_size( struct x86_function *p, unsigned code_size );
 107 void x86_release_func( struct x86_function *p );
 108 void (*x86_get_func( struct x86_function *p ))( void );
 109
 110
 111
 112 /* Create and manipulate registers and regmem values:
 113  */
 114 struct x86_reg x86_make_reg( enum x86_reg_file file,
 115                              enum x86_reg_name idx );
 116
 117 struct x86_reg x86_make_disp( struct x86_reg reg,
 118                               int disp );
 119
 120 struct x86_reg x86_deref( struct x86_reg reg );
 121
 122 struct x86_reg x86_get_base_reg( struct x86_reg reg );
 123
 124
 125 /* Labels, jumps and fixup:
 126  */
 127 unsigned char *x86_get_label( struct x86_function *p );
 128
 129 void x86_jcc( struct x86_function *p,
 130               enum x86_cc cc,
 131               unsigned char *label );
 132
 133 unsigned char *x86_jcc_forward( struct x86_function *p,
 134                           enum x86_cc cc );
 135
 136 unsigned char *x86_jmp_forward( struct x86_function *p);
 137
 138 unsigned char *x86_call_forward( struct x86_function *p);
 139
 140 void x86_fixup_fwd_jump( struct x86_function *p,
 141                          unsigned char *fixup );
 142
 143 void x86_jmp( struct x86_function *p, unsigned char *label );
 144
 145 /* void x86_call( struct x86_function *p, void (*label)() ); */
 146 void x86_call( struct x86_function *p, struct x86_reg reg);
 147
 148 /* michal:
 149  * Temporary. As I need immediate operands, and dont want to mess with the codegen,
 150  * I load the immediate into general purpose register and use it.
 151  */
 152 void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm );
 153
 154
 155 /* Macro for sse_shufps() and sse2_pshufd():
 156  */
 157 #define SHUF(_x,_y,_z,_w)       (((_x)<<0) | ((_y)<<2) | ((_z)<<4) | ((_w)<<6))
 158 #define SHUF_NOOP               RSW(0,1,2,3)
 159 #define GET_SHUF(swz, idx)      (((swz) >> ((idx)*2)) & 0x3)
 160
 161 void mmx_emms( struct x86_function *p );
 162 void mmx_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 163 void mmx_movq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 164 void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 165 void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 166
 167 void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 168 void sse2_cvttps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 169 void sse2_cvtdq2ps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 170 void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 171 void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 172 void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 173 void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 174 void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
 175                   unsigned char shuf );
 176 void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 177 void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 178
 179 void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 180 void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 181 void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 182 void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 183 void sse_andnps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 184 void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 185 void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src,
 186                 unsigned char cc );
 187 void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 188 void sse_maxss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 189 void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 190 void sse_movaps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 191 void sse_movhlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 192 void sse_movhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 193 void sse_movlhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 194 void sse_movlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 195 void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 196 void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 197 void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 198 void sse_mulss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 199 void sse_orps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 200 void sse_xorps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 201 void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 202 void sse_rsqrtps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 203 void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 204 void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
 205                  unsigned char shuf );
 206 void sse_unpckhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 207 void sse_unpcklps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 208 void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src );
 209 void sse2_punpcklbw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 210
 211 void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 212 void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 213 void x86_cmp( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 214 void x86_dec( struct x86_function *p, struct x86_reg reg );
 215 void x86_inc( struct x86_function *p, struct x86_reg reg );
 216 void x86_lea( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 217 void x86_mov( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 218 void x86_mul( struct x86_function *p, struct x86_reg src );
 219 void x86_imul( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 220 void x86_or( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 221 void x86_pop( struct x86_function *p, struct x86_reg reg );
 222 void x86_push( struct x86_function *p, struct x86_reg reg );
 223 void x86_ret( struct x86_function *p );
 224 void x86_retw( struct x86_function *p, unsigned short imm );
 225 void x86_sub( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 226 void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 227 void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 228 void x86_sahf( struct x86_function *p );
 229
 230 void x87_f2xm1( struct x86_function *p );
 231 void x87_fabs( struct x86_function *p );
 232 void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 233 void x87_faddp( struct x86_function *p, struct x86_reg dst );
 234 void x87_fchs( struct x86_function *p );
 235 void x87_fclex( struct x86_function *p );
 236 void x87_fcom( struct x86_function *p, struct x86_reg dst );
 237 void x87_fcomp( struct x86_function *p, struct x86_reg dst );
 238 void x87_fcos( struct x86_function *p );
 239 void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 240 void x87_fdivp( struct x86_function *p, struct x86_reg dst );
 241 void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 242 void x87_fdivrp( struct x86_function *p, struct x86_reg dst );
 243 void x87_fild( struct x86_function *p, struct x86_reg arg );
 244 void x87_fist( struct x86_function *p, struct x86_reg dst );
 245 void x87_fistp( struct x86_function *p, struct x86_reg dst );
 246 void x87_fld( struct x86_function *p, struct x86_reg arg );
 247 void x87_fld1( struct x86_function *p );
 248 void x87_fldcw( struct x86_function *p, struct x86_reg arg );
 249 void x87_fldl2e( struct x86_function *p );
 250 void x87_fldln2( struct x86_function *p );
 251 void x87_fldz( struct x86_function *p );
 252 void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 253 void x87_fmulp( struct x86_function *p, struct x86_reg dst );
 254 void x87_fnclex( struct x86_function *p );
 255 void x87_fprndint( struct x86_function *p );
 256 void x87_fscale( struct x86_function *p );
 257 void x87_fsin( struct x86_function *p );
 258 void x87_fsincos( struct x86_function *p );
 259 void x87_fsqrt( struct x86_function *p );
 260 void x87_fst( struct x86_function *p, struct x86_reg dst );
 261 void x87_fstp( struct x86_function *p, struct x86_reg dst );
 262 void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 263 void x87_fsubp( struct x86_function *p, struct x86_reg dst );
 264 void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
 265 void x87_fsubrp( struct x86_function *p, struct x86_reg dst );
 266 void x87_fxch( struct x86_function *p, struct x86_reg dst );
 267 void x87_fxtract( struct x86_function *p );
 268 void x87_fyl2x( struct x86_function *p );
 269 void x87_fyl2xp1( struct x86_function *p );
 270 void x87_fwait( struct x86_function *p );
 271 void x87_fnstsw( struct x86_function *p, struct x86_reg dst );
 272 void x87_fucompp( struct x86_function *p );
 273 void x87_fucomp( struct x86_function *p, struct x86_reg arg );
 274 void x87_fucom( struct x86_function *p, struct x86_reg arg );
 275
 276
 277
 278 /* Retreive a reference to one of the function arguments, taking into
 279  * account any push/pop activity.  Note - doesn't track explict
 280  * manipulation of ESP by other instructions.
 281  */
 282 struct x86_reg x86_fn_arg( struct x86_function *p, unsigned arg );
 283
 284 #endif
 285 #endif