src/mesa/drivers/dri/i915/i915_fragprog.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "glheader.h"
  29 #include "macros.h"
  30 #include "enums.h"
  31
  32 #include "tnl/t_context.h"
  33 #include "intel_batchbuffer.h"
  34
  35 #include "i915_reg.h"
  36 #include "i915_context.h"
  37 #include "i915_program.h"
  38
  39 #include "program_instruction.h"
  40 #include "program.h"
  41
  42
  43
  44 /* 1, -1/3!, 1/5!, -1/7! */
  45 static const GLfloat sin_constants[4] = {  1.0,
  46                                            -1.0/(3*2*1),
  47                                            1.0/(5*4*3*2*1),
  48                                            -1.0/(7*6*5*4*3*2*1) };
  49
  50 /* 1, -1/2!, 1/4!, -1/6! */
  51 static const GLfloat cos_constants[4] = {  1.0,
  52                                            -1.0/(2*1),
  53                                            1.0/(4*3*2*1),
  54                                            -1.0/(6*5*4*3*2*1) };
  55
  56 /**
  57  * Retrieve a ureg for the given source register.  Will emit
  58  * constants, apply swizzling and negation as needed.
  59  */
  60 static GLuint src_vector( struct i915_fragment_program *p,
  61                           const struct prog_src_register *source,
  62                           const struct gl_fragment_program *program )
  63 {
  64    GLuint src;
  65
  66    switch (source->File) {
  67
  68       /* Registers:
  69        */
  70       case PROGRAM_TEMPORARY:
  71          if (source->Index >= I915_MAX_TEMPORARY) {
  72             i915_program_error( p, "Exceeded max temporary reg" );
  73             return 0;
  74          }
  75          src = UREG( REG_TYPE_R, source->Index );
  76          break;
  77       case PROGRAM_INPUT:
  78          switch (source->Index) {
  79          case FRAG_ATTRIB_WPOS:
  80             src = i915_emit_decl( p,  REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL );
  81             break;
  82          case FRAG_ATTRIB_COL0:
  83             src = i915_emit_decl( p,  REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL );
  84             break;
  85          case FRAG_ATTRIB_COL1:
  86             src = i915_emit_decl( p,  REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ );
  87             src = swizzle( src, X, Y, Z, ONE );
  88             break;
  89          case FRAG_ATTRIB_FOGC:
  90             src = i915_emit_decl( p,  REG_TYPE_T, T_FOG_W, D0_CHANNEL_W );
  91             src = swizzle( src, W, W, W, W );
  92             break;
  93          case FRAG_ATTRIB_TEX0:
  94          case FRAG_ATTRIB_TEX1:
  95          case FRAG_ATTRIB_TEX2:
  96          case FRAG_ATTRIB_TEX3:
  97          case FRAG_ATTRIB_TEX4:
  98          case FRAG_ATTRIB_TEX5:
  99          case FRAG_ATTRIB_TEX6:
 100          case FRAG_ATTRIB_TEX7:
 101             src = i915_emit_decl( p,  REG_TYPE_T,
 102                                  T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
 103                                  D0_CHANNEL_ALL );
 104             break;
 105
 106          default:
 107             i915_program_error( p, "Bad source->Index" );
 108             return 0;
 109          }
 110          break;
 111
 112          /* Various paramters and env values.  All emitted to
 113           * hardware as program constants.
 114           */
 115       case PROGRAM_LOCAL_PARAM:
 116          src = i915_emit_param4fv(
 117             p, program->Base.LocalParams[source->Index]);
 118          break;
 119
 120       case PROGRAM_ENV_PARAM:
 121          src = i915_emit_param4fv(
 122             p, p->ctx->FragmentProgram.Parameters[source->Index]);
 123          break;
 124
 125       case PROGRAM_STATE_VAR:
 126       case PROGRAM_NAMED_PARAM:
 127          src = i915_emit_param4fv(
 128             p, program->Base.Parameters->ParameterValues[source->Index] );
 129          break;
 130
 131       default:
 132          i915_program_error( p, "Bad source->File" );
 133          return 0;
 134    }
 135
 136    src = swizzle(src,
 137                  GET_SWZ(source->Swizzle, 0),
 138                  GET_SWZ(source->Swizzle, 1),
 139                  GET_SWZ(source->Swizzle, 2),
 140                  GET_SWZ(source->Swizzle, 3));
 141
 142    if (source->NegateBase)
 143       src = negate( src,
 144                     GET_BIT(source->NegateBase, 0),
 145                     GET_BIT(source->NegateBase, 1),
 146                     GET_BIT(source->NegateBase, 2),
 147                     GET_BIT(source->NegateBase, 3));
 148
 149    return src;
 150 }
 151
 152
 153 static GLuint get_result_vector( struct i915_fragment_program *p,
 154                                  const struct prog_instruction *inst )
 155 {
 156    switch (inst->DstReg.File) {
 157    case PROGRAM_OUTPUT:
 158       switch (inst->DstReg.Index) {
 159       case FRAG_RESULT_COLR:
 160          return UREG(REG_TYPE_OC, 0);
 161       case FRAG_RESULT_DEPR:
 162          p->depth_written = 1;
 163          return UREG(REG_TYPE_OD, 0);
 164       default:
 165          i915_program_error( p, "Bad inst->DstReg.Index" );
 166          return 0;
 167       }
 168    case PROGRAM_TEMPORARY:
 169       return UREG(REG_TYPE_R, inst->DstReg.Index);
 170    default:
 171       i915_program_error( p, "Bad inst->DstReg.File" );
 172       return 0;
 173    }
 174 }
 175
 176 static GLuint get_result_flags( const struct prog_instruction *inst )
 177 {
 178    GLuint flags = 0;
 179
 180    if (inst->SaturateMode == SATURATE_ZERO_ONE) flags |= A0_DEST_SATURATE;
 181    if (inst->DstReg.WriteMask & WRITEMASK_X) flags |= A0_DEST_CHANNEL_X;
 182    if (inst->DstReg.WriteMask & WRITEMASK_Y) flags |= A0_DEST_CHANNEL_Y;
 183    if (inst->DstReg.WriteMask & WRITEMASK_Z) flags |= A0_DEST_CHANNEL_Z;
 184    if (inst->DstReg.WriteMask & WRITEMASK_W) flags |= A0_DEST_CHANNEL_W;
 185
 186    return flags;
 187 }
 188
 189 static GLuint translate_tex_src_target( struct i915_fragment_program *p,
 190                                      GLubyte bit )
 191 {
 192    switch (bit) {
 193    case TEXTURE_1D_INDEX:   return D0_SAMPLE_TYPE_2D;
 194    case TEXTURE_2D_INDEX:   return D0_SAMPLE_TYPE_2D;
 195    case TEXTURE_RECT_INDEX: return D0_SAMPLE_TYPE_2D;
 196    case TEXTURE_3D_INDEX:   return D0_SAMPLE_TYPE_VOLUME;
 197    case TEXTURE_CUBE_INDEX: return D0_SAMPLE_TYPE_CUBE;
 198    default: i915_program_error(p, "TexSrcBit"); return 0;
 199    }
 200 }
 201
 202 #define EMIT_TEX( OP )                                          \
 203 do {                                                            \
 204    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 205    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 206                                   inst->TexSrcUnit, dim);       \
 207    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 208    /* Texel lookup */                                           \
 209                                                                 \
 210    i915_emit_texld( p,                                          \
 211                get_result_vector( p, inst ),                    \
 212                get_result_flags( inst ),                        \
 213                sampler,                                         \
 214                coord,                                           \
 215                OP);                                             \
 216 } while (0)
 217
 218 #define EMIT_ARITH( OP, N )                                             \
 219 do {                                                                    \
 220    i915_emit_arith( p,                                                  \
 221                OP,                                                      \
 222                get_result_vector( p, inst ),                            \
 223                get_result_flags( inst ), 0,                     \
 224                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 225                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 226                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 227 } while (0)
 228
 229 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 230 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 231 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 232
 233
 234 /* Possible concerns:
 235  *
 236  * SIN, COS -- could use another taylor step?
 237  * LIT      -- results seem a little different to sw mesa
 238  * LOG      -- different to mesa on negative numbers, but this is conformant.
 239  *
 240  * Parse failures -- Mesa doesn't currently give a good indication
 241  * internally whether a particular program string parsed or not.  This
 242  * can lead to confusion -- hopefully we cope with it ok now.
 243  *
 244  */
 245 static void upload_program( struct i915_fragment_program *p )
 246 {
 247    const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
 248    const struct prog_instruction *inst = program->Base.Instructions;
 249
 250 /*    _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */
 251
 252    /* Is this a parse-failed program?  Ensure a valid program is
 253     * loaded, as the flagging of an error isn't sufficient to stop
 254     * this being uploaded to hardware.
 255     */
 256    if (inst[0].Opcode == OPCODE_END) {
 257       GLuint tmp = i915_get_utemp( p );
 258       i915_emit_arith( p,
 259                       A0_MOV,
 260                       UREG(REG_TYPE_OC, 0),
 261                       A0_DEST_CHANNEL_ALL, 0,
 262                       swizzle(tmp,ONE,ZERO,ONE,ONE), 0, 0);
 263       return;
 264    }
 265
 266    while (1) {
 267       GLuint src0, src1, src2, flags;
 268       GLuint tmp = 0;
 269
 270       switch (inst->Opcode) {
 271       case OPCODE_ABS:
 272          src0 = src_vector( p, &inst->SrcReg[0], program);
 273          i915_emit_arith( p,
 274                          A0_MAX,
 275                          get_result_vector( p, inst ),
 276                          get_result_flags( inst ), 0,
 277                          src0, negate(src0, 1,1,1,1), 0);
 278          break;
 279
 280       case OPCODE_ADD:
 281          EMIT_2ARG_ARITH( A0_ADD );
 282          break;
 283
 284       case OPCODE_CMP:
 285          src0 = src_vector( p, &inst->SrcReg[0], program);
 286          src1 = src_vector( p, &inst->SrcReg[1], program);
 287          src2 = src_vector( p, &inst->SrcReg[2], program);
 288          i915_emit_arith( p,
 289                          A0_CMP,
 290                          get_result_vector( p, inst ),
 291                          get_result_flags( inst ), 0,
 292                          src0, src2, src1);     /* NOTE: order of src2, src1 */
 293          break;
 294
 295       case OPCODE_COS:
 296          src0 = src_vector( p, &inst->SrcReg[0], program);
 297          tmp = i915_get_utemp( p );
 298
 299          i915_emit_arith( p,
 300                          A0_MUL,
 301                          tmp, A0_DEST_CHANNEL_X, 0,
 302                          src0,
 303                          i915_emit_const1f(p, 1.0/(M_PI * 2)),
 304                          0);
 305
 306          i915_emit_arith( p,
 307                          A0_MOD,
 308                          tmp, A0_DEST_CHANNEL_X, 0,
 309                          tmp,
 310                          0, 0 );
 311
 312          /* By choosing different taylor constants, could get rid of this mul:
 313           */
 314          i915_emit_arith( p,
 315                          A0_MUL,
 316                          tmp, A0_DEST_CHANNEL_X, 0,
 317                          tmp,
 318                          i915_emit_const1f(p, (M_PI * 2)),
 319                          0);
 320
 321          /*
 322           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 323           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
 324           * t0 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 325           * result = DP4 t0, cos_constants
 326           */
 327          i915_emit_arith( p,
 328                          A0_MUL,
 329                          tmp, A0_DEST_CHANNEL_XY, 0,
 330                          swizzle(tmp, X,X,ONE,ONE),
 331                          swizzle(tmp, X,ONE,ONE,ONE), 0);
 332
 333          i915_emit_arith( p,
 334                          A0_MUL,
 335                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 336                          swizzle(tmp, X,Y,X,ONE),
 337                          swizzle(tmp, X,X,ONE,ONE), 0);
 338
 339          i915_emit_arith( p,
 340                          A0_MUL,
 341                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 342                          swizzle(tmp, X,X,Z,ONE),
 343                          swizzle(tmp, Z,ONE,ONE,ONE), 0);
 344
 345          i915_emit_arith( p,
 346                          A0_DP4,
 347                          get_result_vector( p, inst ),
 348                          get_result_flags( inst ), 0,
 349                          swizzle(tmp, ONE,Z,Y,X),
 350                          i915_emit_const4fv( p, cos_constants ), 0);
 351
 352          break;
 353
 354       case OPCODE_DP3:
 355          EMIT_2ARG_ARITH( A0_DP3 );
 356          break;
 357
 358       case OPCODE_DP4:
 359          EMIT_2ARG_ARITH( A0_DP4 );
 360          break;
 361
 362       case OPCODE_DPH:
 363          src0 = src_vector( p, &inst->SrcReg[0], program);
 364          src1 = src_vector( p, &inst->SrcReg[1], program);
 365
 366          i915_emit_arith( p,
 367                          A0_DP4,
 368                          get_result_vector( p, inst ),
 369                          get_result_flags( inst ), 0,
 370                          swizzle(src0, X,Y,Z,ONE), src1, 0);
 371          break;
 372
 373       case OPCODE_DST:
 374          src0 = src_vector( p, &inst->SrcReg[0], program);
 375          src1 = src_vector( p, &inst->SrcReg[1], program);
 376
 377          /* result[0] = 1    * 1;
 378           * result[1] = a[1] * b[1];
 379           * result[2] = a[2] * 1;
 380           * result[3] = 1    * b[3];
 381           */
 382          i915_emit_arith( p,
 383                          A0_MUL,
 384                          get_result_vector( p, inst ),
 385                          get_result_flags( inst ), 0,
 386                          swizzle(src0, ONE, Y, Z,   ONE),
 387                          swizzle(src1, ONE, Y, ONE, W  ),
 388                          0);
 389          break;
 390
 391       case OPCODE_EX2:
 392          src0 = src_vector( p, &inst->SrcReg[0], program);
 393
 394          i915_emit_arith( p,
 395                          A0_EXP,
 396                          get_result_vector( p, inst ),
 397                          get_result_flags( inst ), 0,
 398                          swizzle(src0,X,X,X,X), 0, 0);
 399          break;
 400
 401       case OPCODE_FLR:
 402          EMIT_1ARG_ARITH( A0_FLR );
 403          break;
 404
 405       case OPCODE_FRC:
 406          EMIT_1ARG_ARITH( A0_FRC );
 407          break;
 408
 409       case OPCODE_KIL:
 410          src0 = src_vector( p, &inst->SrcReg[0], program);
 411          tmp = i915_get_utemp( p );
 412
 413          i915_emit_texld( p,
 414                          tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
 415                          0,
 416                          src0,
 417                          T0_TEXKILL );
 418          break;
 419
 420       case OPCODE_LG2:
 421          src0 = src_vector( p, &inst->SrcReg[0], program);
 422
 423          i915_emit_arith( p,
 424                          A0_LOG,
 425                          get_result_vector( p, inst ),
 426                          get_result_flags( inst ), 0,
 427                          swizzle(src0,X,X,X,X), 0, 0);
 428          break;
 429
 430       case OPCODE_LIT:
 431          src0 = src_vector( p, &inst->SrcReg[0], program);
 432          tmp = i915_get_utemp( p );
 433
 434          /* tmp = max( a.xyzw, a.00zw )
 435           * XXX: Clamp tmp.w to -128..128
 436           * tmp.y = log(tmp.y)
 437           * tmp.y = tmp.w * tmp.y
 438           * tmp.y = exp(tmp.y)
 439           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 440           */
 441          i915_emit_arith( p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 442                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0 );
 443
 444          i915_emit_arith( p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 445                          swizzle(tmp, Y, Y, Y, Y), 0, 0 );
 446
 447          i915_emit_arith( p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 448                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 449                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0 );
 450
 451          i915_emit_arith( p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 452                          swizzle(tmp, Y, Y, Y, Y), 0, 0 );
 453
 454          i915_emit_arith( p, A0_CMP,
 455                          get_result_vector( p, inst ),
 456                          get_result_flags( inst ), 0,
 457                          negate(swizzle(tmp, ONE, ONE, X, ONE),0,0,1,0),
 458                          swizzle(tmp, ONE, X, ZERO, ONE),
 459                          swizzle(tmp, ONE, X, Y, ONE));
 460
 461          break;
 462
 463       case OPCODE_LRP:
 464          src0 = src_vector( p, &inst->SrcReg[0], program);
 465          src1 = src_vector( p, &inst->SrcReg[1], program);
 466          src2 = src_vector( p, &inst->SrcReg[2], program);
 467          flags = get_result_flags( inst );
 468          tmp = i915_get_utemp( p );
 469
 470          /* b*a + c*(1-a)
 471           *
 472           * b*a + c - ca
 473           *
 474           * tmp = b*a + c,
 475           * result = (-c)*a + tmp
 476           */
 477          i915_emit_arith( p, A0_MAD, tmp,
 478                          flags & A0_DEST_CHANNEL_ALL, 0,
 479                          src1, src0, src2 );
 480
 481          i915_emit_arith( p, A0_MAD,
 482                          get_result_vector( p, inst ),
 483                          flags, 0,
 484                          negate(src2, 1,1,1,1), src0, tmp );
 485          break;
 486
 487       case OPCODE_MAD:
 488          EMIT_3ARG_ARITH( A0_MAD );
 489          break;
 490
 491       case OPCODE_MAX:
 492          EMIT_2ARG_ARITH( A0_MAX );
 493          break;
 494
 495       case OPCODE_MIN:
 496          src0 = src_vector( p, &inst->SrcReg[0], program);
 497          src1 = src_vector( p, &inst->SrcReg[1], program);
 498          tmp = i915_get_utemp( p );
 499          flags = get_result_flags( inst );
 500
 501          i915_emit_arith( p,
 502                          A0_MAX,
 503                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 504                          negate(src0,1,1,1,1),
 505                          negate(src1,1,1,1,1), 0);
 506
 507          i915_emit_arith( p,
 508                          A0_MOV,
 509                          get_result_vector( p, inst ),
 510                          flags, 0,
 511                          negate(tmp, 1,1,1,1), 0, 0);
 512          break;
 513
 514       case OPCODE_MOV:
 515          EMIT_1ARG_ARITH( A0_MOV );
 516          break;
 517
 518       case OPCODE_MUL:
 519          EMIT_2ARG_ARITH( A0_MUL );
 520          break;
 521
 522       case OPCODE_POW:
 523          src0 = src_vector( p, &inst->SrcReg[0], program);
 524          src1 = src_vector( p, &inst->SrcReg[1], program);
 525          tmp = i915_get_utemp( p );
 526          flags = get_result_flags( inst );
 527
 528          /* XXX: masking on intermediate values, here and elsewhere.
 529           */
 530          i915_emit_arith( p,
 531                          A0_LOG,
 532                          tmp, A0_DEST_CHANNEL_X, 0,
 533                          swizzle(src0,X,X,X,X), 0, 0);
 534
 535          i915_emit_arith( p,
 536                          A0_MUL,
 537                          tmp, A0_DEST_CHANNEL_X, 0,
 538                          tmp, src1, 0);
 539
 540
 541          i915_emit_arith( p,
 542                          A0_EXP,
 543                          get_result_vector( p, inst ),
 544                          flags, 0,
 545                          swizzle(tmp,X,X,X,X), 0, 0);
 546
 547          break;
 548
 549       case OPCODE_RCP:
 550          src0 = src_vector( p, &inst->SrcReg[0], program);
 551
 552          i915_emit_arith( p,
 553                          A0_RCP,
 554                          get_result_vector( p, inst ),
 555                          get_result_flags( inst ), 0,
 556                          swizzle(src0,X,X,X,X), 0, 0);
 557          break;
 558
 559       case OPCODE_RSQ:
 560
 561          src0 = src_vector( p, &inst->SrcReg[0], program);
 562
 563          i915_emit_arith( p,
 564                          A0_RSQ,
 565                          get_result_vector( p, inst ),
 566                          get_result_flags( inst ), 0,
 567                          swizzle(src0,X,X,X,X), 0, 0);
 568          break;
 569
 570       case OPCODE_SCS:
 571          src0 = src_vector( p, &inst->SrcReg[0], program);
 572          tmp = i915_get_utemp( p );
 573
 574          /*
 575           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 576           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 577           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 578           * scs.x = DP4 t1, sin_constants
 579           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 580           * scs.y = DP4 t1, cos_constants
 581           */
 582          i915_emit_arith( p,
 583                          A0_MUL,
 584                          tmp, A0_DEST_CHANNEL_XY, 0,
 585                          swizzle(src0, X,X,ONE,ONE),
 586                          swizzle(src0, X,ONE,ONE,ONE), 0);
 587
 588          i915_emit_arith( p,
 589                          A0_MUL,
 590                          tmp, A0_DEST_CHANNEL_ALL, 0,
 591                          swizzle(tmp, X,Y,X,Y),
 592                          swizzle(tmp, X,X,ONE,ONE), 0);
 593
 594          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 595             GLuint tmp1;
 596
 597             if (inst->DstReg.WriteMask & WRITEMASK_X)
 598                tmp1 = i915_get_utemp( p );
 599             else
 600                tmp1 = tmp;
 601
 602             i915_emit_arith( p,
 603                             A0_MUL,
 604                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 605                             swizzle(tmp, X,Y,Y,W),
 606                             swizzle(tmp, X,Z,ONE,ONE), 0);
 607
 608             i915_emit_arith( p,
 609                             A0_DP4,
 610                             get_result_vector( p, inst ),
 611                             A0_DEST_CHANNEL_Y, 0,
 612                             swizzle(tmp1, W,Z,Y,X),
 613                             i915_emit_const4fv( p, sin_constants ), 0);
 614          }
 615
 616          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 617             i915_emit_arith( p,
 618                             A0_MUL,
 619                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 620                             swizzle(tmp, X,X,Z,ONE),
 621                             swizzle(tmp, Z,ONE,ONE,ONE), 0);
 622
 623             i915_emit_arith( p,
 624                             A0_DP4,
 625                             get_result_vector( p, inst ),
 626                             A0_DEST_CHANNEL_X, 0,
 627                             swizzle(tmp, ONE,Z,Y,X),
 628                             i915_emit_const4fv( p, cos_constants ), 0);
 629          }
 630          break;
 631
 632       case OPCODE_SGE:
 633          EMIT_2ARG_ARITH( A0_SGE );
 634          break;
 635
 636       case OPCODE_SIN:
 637          src0 = src_vector( p, &inst->SrcReg[0], program);
 638          tmp = i915_get_utemp( p );
 639
 640          i915_emit_arith( p,
 641                          A0_MUL,
 642                          tmp, A0_DEST_CHANNEL_X, 0,
 643                          src0,
 644                          i915_emit_const1f(p, 1.0/(M_PI * 2)),
 645                          0);
 646
 647          i915_emit_arith( p,
 648                          A0_MOD,
 649                          tmp, A0_DEST_CHANNEL_X, 0,
 650                          tmp,
 651                          0, 0 );
 652
 653          /* By choosing different taylor constants, could get rid of this mul:
 654           */
 655          i915_emit_arith( p,
 656                          A0_MUL,
 657                          tmp, A0_DEST_CHANNEL_X, 0,
 658                          tmp,
 659                          i915_emit_const1f(p, (M_PI * 2)),
 660                          0);
 661
 662          /*
 663           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 664           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 665           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 666           * result = DP4 t1.wzyx, sin_constants
 667           */
 668          i915_emit_arith( p,
 669                          A0_MUL,
 670                          tmp, A0_DEST_CHANNEL_XY, 0,
 671                          swizzle(tmp, X,X,ONE,ONE),
 672                          swizzle(tmp, X,ONE,ONE,ONE), 0);
 673
 674          i915_emit_arith( p,
 675                          A0_MUL,
 676                          tmp, A0_DEST_CHANNEL_ALL, 0,
 677                          swizzle(tmp, X,Y,X,Y),
 678                          swizzle(tmp, X,X,ONE,ONE), 0);
 679
 680          i915_emit_arith( p,
 681                          A0_MUL,
 682                          tmp, A0_DEST_CHANNEL_ALL, 0,
 683                          swizzle(tmp, X,Y,Y,W),
 684                          swizzle(tmp, X,Z,ONE,ONE), 0);
 685
 686          i915_emit_arith( p,
 687                          A0_DP4,
 688                          get_result_vector( p, inst ),
 689                          get_result_flags( inst ), 0,
 690                          swizzle(tmp, W, Z, Y, X ),
 691                          i915_emit_const4fv( p, sin_constants ), 0);
 692          break;
 693
 694       case OPCODE_SLT:
 695          EMIT_2ARG_ARITH( A0_SLT );
 696          break;
 697
 698       case OPCODE_SUB:
 699          src0 = src_vector( p, &inst->SrcReg[0], program);
 700          src1 = src_vector( p, &inst->SrcReg[1], program);
 701
 702          i915_emit_arith( p,
 703                          A0_ADD,
 704                          get_result_vector( p, inst ),
 705                          get_result_flags( inst ), 0,
 706                          src0, negate(src1, 1,1,1,1), 0);
 707          break;
 708
 709       case OPCODE_SWZ:
 710          EMIT_1ARG_ARITH( A0_MOV ); /* extended swizzle handled natively */
 711          break;
 712
 713       case OPCODE_TEX:
 714          EMIT_TEX( T0_TEXLD );
 715          break;
 716
 717       case OPCODE_TXB:
 718          EMIT_TEX( T0_TEXLDB );
 719          break;
 720
 721       case OPCODE_TXP:
 722          EMIT_TEX( T0_TEXLDP );
 723          break;
 724
 725       case OPCODE_XPD:
 726          /* Cross product:
 727           *      result.x = src0.y * src1.z - src0.z * src1.y;
 728           *      result.y = src0.z * src1.x - src0.x * src1.z;
 729           *      result.z = src0.x * src1.y - src0.y * src1.x;
 730           *      result.w = undef;
 731           */
 732          src0 = src_vector( p, &inst->SrcReg[0], program);
 733          src1 = src_vector( p, &inst->SrcReg[1], program);
 734          tmp = i915_get_utemp( p );
 735
 736          i915_emit_arith( p,
 737                          A0_MUL,
 738                          tmp, A0_DEST_CHANNEL_ALL, 0,
 739                          swizzle(src0,Z,X,Y,ONE),
 740                          swizzle(src1,Y,Z,X,ONE), 0);
 741
 742          i915_emit_arith( p,
 743                          A0_MAD,
 744                          get_result_vector( p, inst ),
 745                          get_result_flags( inst ), 0,
 746                          swizzle(src0,Y,Z,X,ONE),
 747                          swizzle(src1,Z,X,Y,ONE),
 748                          negate(tmp,1,1,1,0));
 749          break;
 750
 751       case OPCODE_END:
 752          return;
 753
 754       default:
 755          i915_program_error( p, "bad opcode" );
 756          return;
 757       }
 758
 759       inst++;
 760       i915_release_utemps( p );
 761    }
 762 }
 763
 764 /* Rather than trying to intercept and jiggle depth writes during
 765  * emit, just move the value into its correct position at the end of
 766  * the program:
 767  */
 768 static void fixup_depth_write( struct i915_fragment_program *p )
 769 {
 770    if (p->depth_written) {
 771       GLuint depth = UREG(REG_TYPE_OD, 0);
 772
 773       i915_emit_arith( p,
 774                       A0_MOV,
 775                       depth, A0_DEST_CHANNEL_W, 0,
 776                       swizzle(depth,X,Y,Z,Z),
 777                       0, 0);
 778    }
 779 }
 780
 781
 782 #define FRAG_BIT_TEX(n)  (FRAG_BIT_TEX0 << (n))
 783
 784
 785 static void check_wpos( struct i915_fragment_program *p )
 786 {
 787    GLuint inputs = p->FragProg.Base.InputsRead;
 788    GLint i;
 789
 790    p->wpos_tex = -1;
 791
 792    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
 793       if (inputs & FRAG_BIT_TEX(i))
 794          continue;
 795       else if (inputs & FRAG_BIT_WPOS) {
 796          p->wpos_tex = i;
 797          inputs &= ~FRAG_BIT_WPOS;
 798       }
 799    }
 800
 801    if (inputs & FRAG_BIT_WPOS) {
 802       i915_program_error(p, "No free texcoord for wpos value");
 803    }
 804 }
 805
 806
 807 static void translate_program( struct i915_fragment_program *p )
 808 {
 809    i915ContextPtr i915 = I915_CONTEXT(p->ctx);
 810
 811    i915_init_program( i915, p );
 812    check_wpos( p );
 813    upload_program( p );
 814    fixup_depth_write( p );
 815    i915_fini_program( p );
 816
 817    p->translated = 1;
 818 }
 819
 820
 821 static void track_params( struct i915_fragment_program *p )
 822 {
 823    GLint i;
 824
 825    if (p->nr_params)
 826       _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
 827
 828    for (i = 0; i < p->nr_params; i++) {
 829       GLint reg = p->param[i].reg;
 830       COPY_4V( p->constant[reg], p->param[i].values );
 831    }
 832
 833    p->params_uptodate = 1;
 834    p->on_hardware = 0;          /* overkill */
 835 }
 836
 837
 838 static void i915BindProgram( GLcontext *ctx,
 839                             GLenum target,
 840                             struct gl_program *prog )
 841 {
 842    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 843       i915ContextPtr i915 = I915_CONTEXT(ctx);
 844       struct i915_fragment_program *p = (struct i915_fragment_program *)prog;
 845
 846       if (i915->current_program == p)
 847          return;
 848
 849       if (i915->current_program) {
 850          i915->current_program->on_hardware = 0;
 851          i915->current_program->params_uptodate = 0;
 852       }
 853
 854       i915->current_program = p;
 855
 856       assert(p->on_hardware == 0);
 857       assert(p->params_uptodate == 0);
 858
 859       /* Hack: make sure fog is correctly enabled according to this
 860        * fragment program's fog options.
 861        */
 862       ctx->Driver.Enable( ctx, GL_FRAGMENT_PROGRAM_ARB,
 863                           ctx->FragmentProgram.Enabled );
 864    }
 865 }
 866
 867 static struct gl_program *i915NewProgram( GLcontext *ctx,
 868                                       GLenum target,
 869                                       GLuint id )
 870 {
 871    switch (target) {
 872    case GL_VERTEX_PROGRAM_ARB:
 873       return _mesa_init_vertex_program( ctx, CALLOC_STRUCT(gl_vertex_program),
 874                                         target, id );
 875
 876    case GL_FRAGMENT_PROGRAM_ARB: {
 877       struct i915_fragment_program *prog = CALLOC_STRUCT(i915_fragment_program);
 878       if (prog) {
 879          i915_init_program( I915_CONTEXT(ctx), prog );
 880
 881          return _mesa_init_fragment_program( ctx, &prog->FragProg,
 882                                              target, id );
 883       }
 884       else
 885          return NULL;
 886    }
 887
 888    default:
 889       /* Just fallback:
 890        */
 891       return _mesa_new_program( ctx, target, id );
 892    }
 893 }
 894
 895 static void i915DeleteProgram( GLcontext *ctx,
 896                               struct gl_program *prog )
 897 {
 898    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
 899       i915ContextPtr i915 = I915_CONTEXT(ctx);
 900       struct i915_fragment_program *p = (struct i915_fragment_program *)prog;
 901
 902       if (i915->current_program == p)
 903          i915->current_program = 0;
 904    }
 905
 906    _mesa_delete_program( ctx, prog );
 907 }
 908
 909
 910 static GLboolean i915IsProgramNative( GLcontext *ctx,
 911                                      GLenum target,
 912                                      struct gl_program *prog )
 913 {
 914    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 915       struct i915_fragment_program *p = (struct i915_fragment_program *)prog;
 916
 917       if (!p->translated)
 918          translate_program( p );
 919
 920       return !p->error;
 921    }
 922    else
 923       return GL_TRUE;
 924 }
 925
 926 static void i915ProgramStringNotify( GLcontext *ctx,
 927                                     GLenum target,
 928                                     struct gl_program *prog )
 929 {
 930    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 931       struct i915_fragment_program *p = (struct i915_fragment_program *)prog;
 932       p->translated = 0;
 933
 934       /* Hack: make sure fog is correctly enabled according to this
 935        * fragment program's fog options.
 936        */
 937       ctx->Driver.Enable( ctx, GL_FRAGMENT_PROGRAM_ARB,
 938                           ctx->FragmentProgram.Enabled );
 939    }
 940
 941    _tnl_program_string(ctx, target, prog);
 942 }
 943
 944
 945 void i915ValidateFragmentProgram( i915ContextPtr i915 )
 946 {
 947    GLcontext *ctx = &i915->intel.ctx;
 948    intelContextPtr intel = INTEL_CONTEXT(ctx);
 949    TNLcontext *tnl = TNL_CONTEXT(ctx);
 950    struct vertex_buffer *VB = &tnl->vb;
 951
 952    struct i915_fragment_program *p =
 953       (struct i915_fragment_program *)ctx->FragmentProgram._Current;
 954
 955    const GLuint inputsRead = p->FragProg.Base.InputsRead;
 956    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
 957    GLuint s2 = S2_TEXCOORD_NONE;
 958    int i, offset = 0;
 959
 960    if (i915->current_program != p)
 961    {
 962       if (i915->current_program) {
 963          i915->current_program->on_hardware = 0;
 964          i915->current_program->params_uptodate = 0;
 965       }
 966
 967       i915->current_program = p;
 968    }
 969
 970
 971    /* Important:
 972     */
 973    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
 974
 975    if (!p->translated)
 976       translate_program( p );
 977
 978    intel->vertex_attr_count = 0;
 979    intel->wpos_offset = 0;
 980    intel->wpos_size = 0;
 981    intel->coloroffset = 0;
 982    intel->specoffset = 0;
 983
 984    if (inputsRead & FRAG_BITS_TEX_ANY) {
 985       EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16 );
 986    }
 987    else {
 988       EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12 );
 989    }
 990
 991    if (inputsRead & FRAG_BIT_COL0) {
 992       intel->coloroffset = offset / 4;
 993       EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4 );
 994    }
 995
 996    if ((inputsRead & (FRAG_BIT_COL1|FRAG_BIT_FOGC)) ||
 997        i915->vertex_fog != I915_FOG_NONE) {
 998
 999       if (inputsRead & FRAG_BIT_COL1) {
1000          intel->specoffset = offset / 4;
1001          EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, S4_VFMT_SPEC_FOG, 3 );
1002       }
1003       else
1004          EMIT_PAD(3);
1005
1006       if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE)
1007          EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, S4_VFMT_SPEC_FOG, 1 );
1008       else
1009          EMIT_PAD( 1 );
1010    }
1011
1012 #if 0
1013    if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1014       EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4 );
1015    }
1016 #endif
1017
1018    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1019       if (inputsRead & FRAG_BIT_TEX(i)) {
1020          int sz = VB->TexCoordPtr[i]->size;
1021
1022          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1023          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1024
1025          EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_SZ(sz), 0, sz * 4 );
1026       }
1027       else if (i == p->wpos_tex) {
1028
1029          /* If WPOS is required, duplicate the XYZ position data in an
1030           * unused texture coordinate:
1031           */
1032          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1033          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1034
1035          intel->wpos_offset = offset;
1036          intel->wpos_size = 3 * sizeof(GLuint);
1037
1038          EMIT_PAD( intel->wpos_size );
1039       }
1040    }
1041
1042    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1043        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1044
1045       I915_STATECHANGE( i915, I915_UPLOAD_CTX );
1046
1047       /* Must do this *after* statechange, so as not to affect
1048        * buffered vertices reliant on the old state:
1049        */
1050       intel->vertex_size = _tnl_install_attrs( &intel->ctx,
1051                                                intel->vertex_attrs,
1052                                                intel->vertex_attr_count,
1053                                                intel->ViewportMatrix.m, 0 );
1054
1055       intel->vertex_size >>= 2;
1056
1057       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1058       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1059
1060       assert(intel->vtbl.check_vertex_size( intel, intel->vertex_size ));
1061    }
1062
1063    if (!p->params_uptodate)
1064       track_params( p );
1065
1066    if (!p->on_hardware)
1067       i915_upload_program( i915, p );
1068 }
1069
1070 void i915InitFragProgFuncs( struct dd_function_table *functions )
1071 {
1072    functions->BindProgram = i915BindProgram;
1073    functions->NewProgram = i915NewProgram;
1074    functions->DeleteProgram = i915DeleteProgram;
1075    functions->IsProgramNative = i915IsProgramNative;
1076    functions->ProgramStringNotify = i915ProgramStringNotify;
1077 }