src/mesa/tnl/t_vtx_x86.c

   1 /**************************************************************************
   2
   3 Copyright 2004 Tungsten Graphics Inc., Cedar Park, Texas.
   4
   5 All Rights Reserved.
   6
   7 Permission is hereby granted, free of charge, to any person obtaining a
   8 copy of this software and associated documentation files (the "Software"),
   9 to deal in the Software without restriction, including without limitation
  10 on the rights to use, copy, modify, merge, publish, distribute, sub
  11 license, and/or sell copies of the Software, and to permit persons to whom
  12 the Software is furnished to do so, subject to the following conditions:
  13
  14 The above copyright notice and this permission notice (including the next
  15 paragraph) shall be included in all copies or substantial portions of the
  16 Software.
  17
  18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20 FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  21 ATI, TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  22 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  23 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  24 USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26 **************************************************************************/
  27
  28 /*
  29  * Authors:
  30  *   Keith Whitwell <keith@tungstengraphics.com>
  31  *   Daniel Borca <dborca@yahoo.com>
  32  */
  33
  34
  35 #include "glheader.h"
  36 #include "context.h"
  37 #include "macros.h"
  38 #include "vtxfmt.h"
  39 #include "dlist.h"
  40 #include "state.h"
  41 #include "light.h"
  42 #include "api_arrayelt.h"
  43 #include "api_noop.h"
  44 #include "t_vtx_api.h"
  45 #include "simple_list.h"
  46
  47
  48 #if defined(USE_X86_ASM) && !defined(HAVE_NONSTANDARD_GLAPIENTRY)
  49
  50 #define EXTERN( FUNC )          \
  51 extern const char *FUNC;        \
  52 extern const char *FUNC##_end
  53
  54 EXTERN( _tnl_x86_Attribute1fv );
  55 EXTERN( _tnl_x86_Attribute2fv );
  56 EXTERN( _tnl_x86_Attribute3fv );
  57 EXTERN( _tnl_x86_Attribute4fv );
  58 EXTERN( _tnl_x86_Vertex1fv );
  59 EXTERN( _tnl_x86_Vertex2fv );
  60 EXTERN( _tnl_x86_Vertex3fv );
  61 EXTERN( _tnl_x86_Vertex4fv );
  62
  63 EXTERN( _tnl_x86_dispatch_attrf1 );
  64 EXTERN( _tnl_x86_dispatch_attrf2 );
  65 EXTERN( _tnl_x86_dispatch_attrf3 );
  66 EXTERN( _tnl_x86_dispatch_attrf4 );
  67 EXTERN( _tnl_x86_dispatch_attrfv );
  68 EXTERN( _tnl_x86_dispatch_multitexcoordf1 );
  69 EXTERN( _tnl_x86_dispatch_multitexcoordf2 );
  70 EXTERN( _tnl_x86_dispatch_multitexcoordf3 );
  71 EXTERN( _tnl_x86_dispatch_multitexcoordf4 );
  72 EXTERN( _tnl_x86_dispatch_multitexcoordfv );
  73 EXTERN( _tnl_x86_dispatch_vertexattribf1 );
  74 EXTERN( _tnl_x86_dispatch_vertexattribf2 );
  75 EXTERN( _tnl_x86_dispatch_vertexattribf3 );
  76 EXTERN( _tnl_x86_dispatch_vertexattribf4 );
  77 EXTERN( _tnl_x86_dispatch_vertexattribfv );
  78
  79 EXTERN( _tnl_x86_choose_fv );
  80
  81
  82 #define DONT_KNOW_OFFSETS 1
  83
  84
  85 #define DFN( FUNC, CACHE, KEY )                         \
  86    struct _tnl_dynfn *dfn = MALLOC_STRUCT( _tnl_dynfn );\
  87    char *start = (char *)&FUNC;                         \
  88    char *end = (char *)&FUNC##_end;                     \
  89    int offset = 0;                                      \
  90    insert_at_head( &CACHE, dfn );                       \
  91    dfn->key = KEY;                                      \
  92    dfn->code = ALIGN_MALLOC( end - start, 16 );         \
  93    memcpy (dfn->code, start, end - start)
  94
  95
  96
  97 #define FIXUP( CODE, KNOWN_OFFSET, CHECKVAL, NEWVAL )   \
  98 do {                                                    \
  99    GLint subst = 0x10101010 + CHECKVAL;                 \
 100                                                         \
 101    if (DONT_KNOW_OFFSETS) {                             \
 102       while (*(int *)(CODE+offset) != subst) offset++;  \
 103       *(int *)(CODE+offset) = (int)(NEWVAL);            \
 104       if (0) fprintf(stderr, "%s/%d: offset %d, new value: 0x%x\n", __FILE__, __LINE__, offset, (int)(NEWVAL)); \
 105       offset += 4;                                      \
 106    }                                                    \
 107    else {                                               \
 108       int *icode = (int *)(CODE+KNOWN_OFFSET);          \
 109       assert (*icode == subst);                         \
 110       *icode = (int)NEWVAL;                             \
 111    }                                                    \
 112 } while (0)
 113
 114
 115
 116 #define FIXUPREL( CODE, KNOWN_OFFSET, CHECKVAL, NEWVAL )\
 117 do {                                                    \
 118    GLint subst = 0x10101010 + CHECKVAL;                 \
 119                                                         \
 120    if (DONT_KNOW_OFFSETS) {                             \
 121       while (*(int *)(CODE+offset) != subst) offset++;  \
 122       *(int *)(CODE+offset) = (int)(NEWVAL) - ((int)(CODE)+offset) - 4; \
 123       if (0) fprintf(stderr, "%s/%d: offset %d, new value: 0x%x\n", __FILE__, __LINE__, offset, (int)(NEWVAL) - ((int)(CODE)+offset) - 4); \
 124       offset += 4;                                      \
 125    }                                                    \
 126    else {                                               \
 127       int *icode = (int *)(CODE+KNOWN_OFFSET);          \
 128       assert (*icode == subst);                         \
 129       *icode = (int)(NEWVAL) - (int)(icode) - 4;        \
 130    }                                                    \
 131 } while (0)
 132
 133
 134
 135
 136 /* Build specialized versions of the immediate calls on the fly for
 137  * the current state.  Generic x86 versions.
 138  */
 139
 140 static struct _tnl_dynfn *makeX86Vertex1fv( GLcontext *ctx, int vertex_size )
 141 {
 142    TNLcontext *tnl = TNL_CONTEXT(ctx);
 143    DFN ( _tnl_x86_Vertex1fv, tnl->vtx.cache.Vertex[1-1], vertex_size );
 144
 145    FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 146    FIXUP(dfn->code, 0, 1, vertex_size - 1);
 147    FIXUP(dfn->code, 0, 2, (int)&tnl->vtx.vertex[1]);
 148    FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 149    FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 150    FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 151    FIXUP(dfn->code, 0, 4, (int)ctx);
 152    FIXUPREL(dfn->code, 0, 5, (int)&_tnl_wrap_filled_vertex);
 153
 154    return dfn;
 155 }
 156
 157 static struct _tnl_dynfn *makeX86Vertex2fv( GLcontext *ctx, int vertex_size )
 158 {
 159    TNLcontext *tnl = TNL_CONTEXT(ctx);
 160    DFN ( _tnl_x86_Vertex2fv, tnl->vtx.cache.Vertex[2-1], vertex_size );
 161
 162    FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 163    FIXUP(dfn->code, 0, 1, vertex_size - 2);
 164    FIXUP(dfn->code, 0, 2, (int)&tnl->vtx.vertex[2]);
 165    FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 166    FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 167    FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 168    FIXUP(dfn->code, 0, 4, (int)ctx);
 169    FIXUPREL(dfn->code, 0, 5, (int)&_tnl_wrap_filled_vertex);
 170
 171    return dfn;
 172 }
 173
 174 static struct _tnl_dynfn *makeX86Vertex3fv( GLcontext *ctx, int vertex_size )
 175 {
 176    TNLcontext *tnl = TNL_CONTEXT(ctx);
 177    DFN ( _tnl_x86_Vertex3fv, tnl->vtx.cache.Vertex[3-1], vertex_size );
 178
 179    switch (vertex_size) {
 180       default: {
 181          FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 182          FIXUP(dfn->code, 0, 1, vertex_size - 3);
 183          FIXUP(dfn->code, 0, 2, (int)&tnl->vtx.vertex[3]);
 184          FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 185          FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 186          FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 187          FIXUP(dfn->code, 0, 4, (int)ctx);
 188          FIXUPREL(dfn->code, 0, 5, (int)&_tnl_wrap_filled_vertex);
 189          return dfn;
 190       }
 191    }
 192 }
 193
 194 static struct _tnl_dynfn *makeX86Vertex4fv( GLcontext *ctx, int vertex_size )
 195 {
 196    TNLcontext *tnl = TNL_CONTEXT(ctx);
 197    DFN ( _tnl_x86_Vertex4fv, tnl->vtx.cache.Vertex[4-1], vertex_size );
 198
 199    FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 200    FIXUP(dfn->code, 0, 1, vertex_size - 4);
 201    FIXUP(dfn->code, 0, 2, (int)&tnl->vtx.vertex[4]);
 202    FIXUP(dfn->code, 0, 0, (int)&tnl->vtx.vbptr);
 203    FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 204    FIXUP(dfn->code, 0, 3, (int)&tnl->vtx.counter);
 205    FIXUP(dfn->code, 0, 4, (int)ctx);
 206    FIXUPREL(dfn->code, 0, 5, (int)&_tnl_wrap_filled_vertex);
 207
 208    return dfn;
 209 }
 210
 211
 212 static struct _tnl_dynfn *makeX86Attribute1fv( GLcontext *ctx, int dest )
 213 {
 214    TNLcontext *tnl = TNL_CONTEXT(ctx);
 215    DFN ( _tnl_x86_Attribute1fv, tnl->vtx.cache.Attribute[1-1], dest );
 216
 217    FIXUP(dfn->code, 0, 0, dest);
 218
 219    return dfn;
 220 }
 221
 222 static struct _tnl_dynfn *makeX86Attribute2fv( GLcontext *ctx, int dest )
 223 {
 224    TNLcontext *tnl = TNL_CONTEXT(ctx);
 225    DFN ( _tnl_x86_Attribute2fv, tnl->vtx.cache.Attribute[2-1], dest );
 226
 227    FIXUP(dfn->code, 0, 0, dest);
 228    FIXUP(dfn->code, 0, 1, 4+dest);
 229
 230    return dfn;
 231 }
 232
 233 static struct _tnl_dynfn *makeX86Attribute3fv( GLcontext *ctx, int dest )
 234 {
 235    TNLcontext *tnl = TNL_CONTEXT(ctx);
 236    DFN ( _tnl_x86_Attribute3fv, tnl->vtx.cache.Attribute[3-1], dest );
 237
 238    FIXUP(dfn->code, 0, 0, dest);
 239    FIXUP(dfn->code, 0, 1, 4+dest);
 240    FIXUP(dfn->code, 0, 2, 8+dest);
 241
 242    return dfn;
 243 }
 244
 245 static struct _tnl_dynfn *makeX86Attribute4fv( GLcontext *ctx, int dest )
 246 {
 247    TNLcontext *tnl = TNL_CONTEXT(ctx);
 248    DFN ( _tnl_x86_Attribute4fv, tnl->vtx.cache.Attribute[4-1], dest );
 249
 250    FIXUP(dfn->code, 0, 0, dest);
 251    FIXUP(dfn->code, 0, 1, 4+dest);
 252    FIXUP(dfn->code, 0, 2, 8+dest);
 253    FIXUP(dfn->code, 0, 3, 12+dest);
 254
 255    return dfn;
 256 }
 257
 258
 259 void _tnl_InitX86Codegen( struct _tnl_dynfn_generators *gen )
 260 {
 261    gen->Vertex[0] = makeX86Vertex1fv;
 262    gen->Vertex[1] = makeX86Vertex2fv;
 263    gen->Vertex[2] = makeX86Vertex3fv;
 264    gen->Vertex[3] = makeX86Vertex4fv;
 265    gen->Attribute[0] = makeX86Attribute1fv;
 266    gen->Attribute[1] = makeX86Attribute2fv;
 267    gen->Attribute[2] = makeX86Attribute3fv;
 268    gen->Attribute[3] = makeX86Attribute4fv;
 269 }
 270
 271
 272 #define MKDISP(FUNC, SIZE, ATTR, WARP)                                  \
 273 do {                                                                    \
 274    char *code;                                                          \
 275    char *start = (char *)&WARP;                                         \
 276    char *end = (char *)&WARP##_end;                                     \
 277    int offset = 0;                                                      \
 278    code = ALIGN_MALLOC( end - start, 16 );                              \
 279    memcpy (code, start, end - start);                                   \
 280    FIXUP(code, 0, 0, (int)&(TNL_CONTEXT(ctx)->vtx.tabfv[ATTR][SIZE-1]));\
 281    *(void **)&vfmt->FUNC = code;                                        \
 282 } while (0)
 283
 284
 285 /* Install the codegen'ed versions of the 2nd level dispatch
 286  * functions.  We should keep a list and free them in the end...
 287  */
 288 void _tnl_x86_exec_vtxfmt_init( GLcontext *ctx )
 289 {
 290    GLvertexformat *vfmt = &(TNL_CONTEXT(ctx)->exec_vtxfmt);
 291
 292    MKDISP(Color3f,             3, _TNL_ATTRIB_COLOR0, _tnl_x86_dispatch_attrf3);
 293    MKDISP(Color3fv,            3, _TNL_ATTRIB_COLOR0, _tnl_x86_dispatch_attrfv);
 294    MKDISP(Color4f,             4, _TNL_ATTRIB_COLOR0, _tnl_x86_dispatch_attrf4);
 295    MKDISP(Color4fv,            4, _TNL_ATTRIB_COLOR0, _tnl_x86_dispatch_attrfv);
 296    MKDISP(FogCoordfEXT,        1, _TNL_ATTRIB_FOG,    _tnl_x86_dispatch_attrf1);
 297    MKDISP(FogCoordfvEXT,       1, _TNL_ATTRIB_FOG,    _tnl_x86_dispatch_attrfv);
 298    MKDISP(Normal3f,            3, _TNL_ATTRIB_NORMAL, _tnl_x86_dispatch_attrf3);
 299    MKDISP(Normal3fv,           3, _TNL_ATTRIB_NORMAL, _tnl_x86_dispatch_attrfv);
 300    MKDISP(SecondaryColor3fEXT, 3, _TNL_ATTRIB_COLOR1, _tnl_x86_dispatch_attrf3);
 301    MKDISP(SecondaryColor3fvEXT,3, _TNL_ATTRIB_COLOR1, _tnl_x86_dispatch_attrfv);
 302    MKDISP(TexCoord1f,          1, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrf1);
 303    MKDISP(TexCoord1fv,         1, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrfv);
 304    MKDISP(TexCoord2f,          2, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrf2);
 305    MKDISP(TexCoord2fv,         2, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrfv);
 306    MKDISP(TexCoord3f,          3, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrf3);
 307    MKDISP(TexCoord3fv,         3, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrfv);
 308    MKDISP(TexCoord4f,          4, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrf4);
 309    MKDISP(TexCoord4fv,         4, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_attrfv);
 310    MKDISP(Vertex2f,            2, _TNL_ATTRIB_POS,    _tnl_x86_dispatch_attrf2);
 311    MKDISP(Vertex2fv,           2, _TNL_ATTRIB_POS,    _tnl_x86_dispatch_attrfv);
 312    MKDISP(Vertex3f,            3, _TNL_ATTRIB_POS,    _tnl_x86_dispatch_attrf3);
 313    MKDISP(Vertex3fv,           3, _TNL_ATTRIB_POS,    _tnl_x86_dispatch_attrfv);
 314    MKDISP(Vertex4f,            4, _TNL_ATTRIB_POS,    _tnl_x86_dispatch_attrf4);
 315    MKDISP(Vertex4fv,           4, _TNL_ATTRIB_POS,    _tnl_x86_dispatch_attrfv);
 316
 317    MKDISP(MultiTexCoord1fARB,  1, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordf1);
 318    MKDISP(MultiTexCoord1fvARB, 1, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordfv);
 319    MKDISP(MultiTexCoord2fARB,  2, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordf2);
 320    MKDISP(MultiTexCoord2fvARB, 2, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordfv);
 321    MKDISP(MultiTexCoord3fARB,  3, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordf3);
 322    MKDISP(MultiTexCoord3fvARB, 3, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordfv);
 323    MKDISP(MultiTexCoord4fARB,  4, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordf4);
 324    MKDISP(MultiTexCoord4fvARB, 4, _TNL_ATTRIB_TEX0,   _tnl_x86_dispatch_multitexcoordfv);
 325
 326    MKDISP(VertexAttrib1fNV,    1, 0,                  _tnl_x86_dispatch_vertexattribf1);
 327    MKDISP(VertexAttrib1fvNV,   1, 0,                  _tnl_x86_dispatch_vertexattribfv);
 328    MKDISP(VertexAttrib2fNV,    2, 0,                  _tnl_x86_dispatch_vertexattribf2);
 329    MKDISP(VertexAttrib2fvNV,   2, 0,                  _tnl_x86_dispatch_vertexattribfv);
 330    MKDISP(VertexAttrib3fNV,    3, 0,                  _tnl_x86_dispatch_vertexattribf3);
 331    MKDISP(VertexAttrib3fvNV,   3, 0,                  _tnl_x86_dispatch_vertexattribfv);
 332    MKDISP(VertexAttrib4fNV,    4, 0,                  _tnl_x86_dispatch_vertexattribf4);
 333    MKDISP(VertexAttrib4fvNV,   4, 0,                  _tnl_x86_dispatch_vertexattribfv);
 334 }
 335
 336
 337 /* Install the codegen'ed choosers.
 338  * We should keep a list and free them in the end...
 339  */
 340 void _tnl_x86choosers( tnl_attrfv_func (*choose)[4],
 341                        tnl_attrfv_func (*do_choose)( GLuint attr,
 342                                                  GLuint sz ))
 343 {
 344    int attr, size;
 345
 346    for (attr = 0; attr < _TNL_MAX_ATTR_CODEGEN; attr++) {
 347       for (size = 0; size < 4; size++) {
 348          char *code;
 349          char *start = (char *)&_tnl_x86_choose_fv;
 350          char *end = (char *)&_tnl_x86_choose_fv_end;
 351          int offset = 0;
 352          code = ALIGN_MALLOC( end - start, 16 );
 353          memcpy (code, start, end - start);
 354          FIXUP(code, 0, 0, attr);
 355          FIXUP(code, 0, 1, size + 1);
 356          FIXUPREL(code, 0, 2, do_choose);
 357          choose[attr][size] = (tnl_attrfv_func)code;
 358       }
 359    }
 360 }
 361
 362 #else
 363
 364 void _tnl_InitX86Codegen( struct _tnl_dynfn_generators *gen )
 365 {
 366    (void) gen;
 367 }
 368
 369
 370 void _tnl_x86_exec_vtxfmt_init( GLcontext *ctx )
 371 {
 372    (void) ctx;
 373 }
 374
 375
 376 void _tnl_x86choosers( tnl_attrfv_func (*choose)[4],
 377                        tnl_attrfv_func (*do_choose)( GLuint attr,
 378                                                  GLuint sz ))
 379 {
 380    (void) choose;
 381    (void) do_choose;
 382 }
 383
 384 #endif