src/mesa/drivers/dri/i915tex/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/tnl.h"
  33 #include "tnl/t_context.h"
  34 #include "intel_batchbuffer.h"
  35
  36 #include "i915_reg.h"
  37 #include "i915_context.h"
  38 #include "i915_program.h"
  39
  40 #include "program_instruction.h"
  41 #include "program.h"
  42
  43
  44
  45 /* 1, -1/3!, 1/5!, -1/7! */
  46 static const GLfloat sin_constants[4] = { 1.0,
  47    -1.0 / (3 * 2 * 1),
  48    1.0 / (5 * 4 * 3 * 2 * 1),
  49    -1.0 / (7 * 6 * 5 * 4 * 3 * 2 * 1)
  50 };
  51
  52 /* 1, -1/2!, 1/4!, -1/6! */
  53 static const GLfloat cos_constants[4] = { 1.0,
  54    -1.0 / (2 * 1),
  55    1.0 / (4 * 3 * 2 * 1),
  56    -1.0 / (6 * 5 * 4 * 3 * 2 * 1)
  57 };
  58
  59 /**
  60  * Retrieve a ureg for the given source register.  Will emit
  61  * constants, apply swizzling and negation as needed.
  62  */
  63 static GLuint
  64 src_vector(struct i915_fragment_program *p,
  65            const struct prog_src_register *source,
  66            const struct gl_fragment_program *program)
  67 {
  68    GLuint src;
  69
  70    switch (source->File) {
  71
  72       /* Registers:
  73        */
  74    case PROGRAM_TEMPORARY:
  75       if (source->Index >= I915_MAX_TEMPORARY) {
  76          i915_program_error(p, "Exceeded max temporary reg");
  77          return 0;
  78       }
  79       src = UREG(REG_TYPE_R, source->Index);
  80       break;
  81    case PROGRAM_INPUT:
  82       switch (source->Index) {
  83       case FRAG_ATTRIB_WPOS:
  84          src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
  85          break;
  86       case FRAG_ATTRIB_COL0:
  87          src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
  88          break;
  89       case FRAG_ATTRIB_COL1:
  90          src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
  91          src = swizzle(src, X, Y, Z, ONE);
  92          break;
  93       case FRAG_ATTRIB_FOGC:
  94          src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
  95          src = swizzle(src, W, W, W, W);
  96          break;
  97       case FRAG_ATTRIB_TEX0:
  98       case FRAG_ATTRIB_TEX1:
  99       case FRAG_ATTRIB_TEX2:
 100       case FRAG_ATTRIB_TEX3:
 101       case FRAG_ATTRIB_TEX4:
 102       case FRAG_ATTRIB_TEX5:
 103       case FRAG_ATTRIB_TEX6:
 104       case FRAG_ATTRIB_TEX7:
 105          src = i915_emit_decl(p, REG_TYPE_T,
 106                               T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
 107                               D0_CHANNEL_ALL);
 108          break;
 109
 110       default:
 111          i915_program_error(p, "Bad source->Index");
 112          return 0;
 113       }
 114       break;
 115
 116       /* Various paramters and env values.  All emitted to
 117        * hardware as program constants.
 118        */
 119    case PROGRAM_LOCAL_PARAM:
 120       src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
 121       break;
 122
 123    case PROGRAM_ENV_PARAM:
 124       src =
 125          i915_emit_param4fv(p,
 126                             p->ctx->FragmentProgram.Parameters[source->
 127                                                                Index]);
 128       break;
 129
 130    case PROGRAM_STATE_VAR:
 131    case PROGRAM_NAMED_PARAM:
 132       src =
 133          i915_emit_param4fv(p,
 134                             program->Base.Parameters->ParameterValues[source->
 135                                                                       Index]);
 136       break;
 137
 138    default:
 139       i915_program_error(p, "Bad source->File");
 140       return 0;
 141    }
 142
 143    src = swizzle(src,
 144                  GET_SWZ(source->Swizzle, 0),
 145                  GET_SWZ(source->Swizzle, 1),
 146                  GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
 147
 148    if (source->NegateBase)
 149       src = negate(src,
 150                    GET_BIT(source->NegateBase, 0),
 151                    GET_BIT(source->NegateBase, 1),
 152                    GET_BIT(source->NegateBase, 2),
 153                    GET_BIT(source->NegateBase, 3));
 154
 155    return src;
 156 }
 157
 158
 159 static GLuint
 160 get_result_vector(struct i915_fragment_program *p,
 161                   const struct prog_instruction *inst)
 162 {
 163    switch (inst->DstReg.File) {
 164    case PROGRAM_OUTPUT:
 165       switch (inst->DstReg.Index) {
 166       case FRAG_RESULT_COLR:
 167          return UREG(REG_TYPE_OC, 0);
 168       case FRAG_RESULT_DEPR:
 169          p->depth_written = 1;
 170          return UREG(REG_TYPE_OD, 0);
 171       default:
 172          i915_program_error(p, "Bad inst->DstReg.Index");
 173          return 0;
 174       }
 175    case PROGRAM_TEMPORARY:
 176       return UREG(REG_TYPE_R, inst->DstReg.Index);
 177    default:
 178       i915_program_error(p, "Bad inst->DstReg.File");
 179       return 0;
 180    }
 181 }
 182
 183 static GLuint
 184 get_result_flags(const struct prog_instruction *inst)
 185 {
 186    GLuint flags = 0;
 187
 188    if (inst->SaturateMode == SATURATE_ZERO_ONE)
 189       flags |= A0_DEST_SATURATE;
 190    if (inst->DstReg.WriteMask & WRITEMASK_X)
 191       flags |= A0_DEST_CHANNEL_X;
 192    if (inst->DstReg.WriteMask & WRITEMASK_Y)
 193       flags |= A0_DEST_CHANNEL_Y;
 194    if (inst->DstReg.WriteMask & WRITEMASK_Z)
 195       flags |= A0_DEST_CHANNEL_Z;
 196    if (inst->DstReg.WriteMask & WRITEMASK_W)
 197       flags |= A0_DEST_CHANNEL_W;
 198
 199    return flags;
 200 }
 201
 202 static GLuint
 203 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 204 {
 205    switch (bit) {
 206    case TEXTURE_1D_INDEX:
 207       return D0_SAMPLE_TYPE_2D;
 208    case TEXTURE_2D_INDEX:
 209       return D0_SAMPLE_TYPE_2D;
 210    case TEXTURE_RECT_INDEX:
 211       return D0_SAMPLE_TYPE_2D;
 212    case TEXTURE_3D_INDEX:
 213       return D0_SAMPLE_TYPE_VOLUME;
 214    case TEXTURE_CUBE_INDEX:
 215       return D0_SAMPLE_TYPE_CUBE;
 216    default:
 217       i915_program_error(p, "TexSrcBit");
 218       return 0;
 219    }
 220 }
 221
 222 #define EMIT_TEX( OP )                                          \
 223 do {                                                            \
 224    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 225    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 226                                   inst->TexSrcUnit, dim);       \
 227    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 228    /* Texel lookup */                                           \
 229                                                                 \
 230    i915_emit_texld( p,                                          \
 231                get_result_vector( p, inst ),                    \
 232                get_result_flags( inst ),                        \
 233                sampler,                                         \
 234                coord,                                           \
 235                OP);                                             \
 236 } while (0)
 237
 238 #define EMIT_ARITH( OP, N )                                             \
 239 do {                                                                    \
 240    i915_emit_arith( p,                                                  \
 241                OP,                                                      \
 242                get_result_vector( p, inst ),                            \
 243                get_result_flags( inst ), 0,                     \
 244                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 245                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 246                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 247 } while (0)
 248
 249 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 250 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 251 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 252
 253
 254 /* Possible concerns:
 255  *
 256  * SIN, COS -- could use another taylor step?
 257  * LIT      -- results seem a little different to sw mesa
 258  * LOG      -- different to mesa on negative numbers, but this is conformant.
 259  *
 260  * Parse failures -- Mesa doesn't currently give a good indication
 261  * internally whether a particular program string parsed or not.  This
 262  * can lead to confusion -- hopefully we cope with it ok now.
 263  *
 264  */
 265 static void
 266 upload_program(struct i915_fragment_program *p)
 267 {
 268    const struct gl_fragment_program *program =
 269       p->ctx->FragmentProgram._Current;
 270    const struct prog_instruction *inst = program->Base.Instructions;
 271
 272 /*    _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */
 273
 274    /* Is this a parse-failed program?  Ensure a valid program is
 275     * loaded, as the flagging of an error isn't sufficient to stop
 276     * this being uploaded to hardware.
 277     */
 278    if (inst[0].Opcode == OPCODE_END) {
 279       GLuint tmp = i915_get_utemp(p);
 280       i915_emit_arith(p,
 281                       A0_MOV,
 282                       UREG(REG_TYPE_OC, 0),
 283                       A0_DEST_CHANNEL_ALL, 0,
 284                       swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
 285       return;
 286    }
 287
 288    while (1) {
 289       GLuint src0, src1, src2, flags;
 290       GLuint tmp = 0;
 291
 292       switch (inst->Opcode) {
 293       case OPCODE_ABS:
 294          src0 = src_vector(p, &inst->SrcReg[0], program);
 295          i915_emit_arith(p,
 296                          A0_MAX,
 297                          get_result_vector(p, inst),
 298                          get_result_flags(inst), 0,
 299                          src0, negate(src0, 1, 1, 1, 1), 0);
 300          break;
 301
 302       case OPCODE_ADD:
 303          EMIT_2ARG_ARITH(A0_ADD);
 304          break;
 305
 306       case OPCODE_CMP:
 307          src0 = src_vector(p, &inst->SrcReg[0], program);
 308          src1 = src_vector(p, &inst->SrcReg[1], program);
 309          src2 = src_vector(p, &inst->SrcReg[2], program);
 310          i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
 311          break;
 312
 313       case OPCODE_COS:
 314          src0 = src_vector(p, &inst->SrcReg[0], program);
 315          tmp = i915_get_utemp(p);
 316
 317          i915_emit_arith(p,
 318                          A0_MUL,
 319                          tmp, A0_DEST_CHANNEL_X, 0,
 320                          src0, i915_emit_const1f(p, 1.0 / (M_PI * 2)), 0);
 321
 322          i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 323
 324          /* By choosing different taylor constants, could get rid of this mul:
 325           */
 326          i915_emit_arith(p,
 327                          A0_MUL,
 328                          tmp, A0_DEST_CHANNEL_X, 0,
 329                          tmp, i915_emit_const1f(p, (M_PI * 2)), 0);
 330
 331          /*
 332           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 333           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
 334           * t0 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 335           * result = DP4 t0, cos_constants
 336           */
 337          i915_emit_arith(p,
 338                          A0_MUL,
 339                          tmp, A0_DEST_CHANNEL_XY, 0,
 340                          swizzle(tmp, X, X, ONE, ONE),
 341                          swizzle(tmp, X, ONE, ONE, ONE), 0);
 342
 343          i915_emit_arith(p,
 344                          A0_MUL,
 345                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 346                          swizzle(tmp, X, Y, X, ONE),
 347                          swizzle(tmp, X, X, ONE, ONE), 0);
 348
 349          i915_emit_arith(p,
 350                          A0_MUL,
 351                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 352                          swizzle(tmp, X, X, Z, ONE),
 353                          swizzle(tmp, Z, ONE, ONE, ONE), 0);
 354
 355          i915_emit_arith(p,
 356                          A0_DP4,
 357                          get_result_vector(p, inst),
 358                          get_result_flags(inst), 0,
 359                          swizzle(tmp, ONE, Z, Y, X),
 360                          i915_emit_const4fv(p, cos_constants), 0);
 361
 362          break;
 363
 364       case OPCODE_DP3:
 365          EMIT_2ARG_ARITH(A0_DP3);
 366          break;
 367
 368       case OPCODE_DP4:
 369          EMIT_2ARG_ARITH(A0_DP4);
 370          break;
 371
 372       case OPCODE_DPH:
 373          src0 = src_vector(p, &inst->SrcReg[0], program);
 374          src1 = src_vector(p, &inst->SrcReg[1], program);
 375
 376          i915_emit_arith(p,
 377                          A0_DP4,
 378                          get_result_vector(p, inst),
 379                          get_result_flags(inst), 0,
 380                          swizzle(src0, X, Y, Z, ONE), src1, 0);
 381          break;
 382
 383       case OPCODE_DST:
 384          src0 = src_vector(p, &inst->SrcReg[0], program);
 385          src1 = src_vector(p, &inst->SrcReg[1], program);
 386
 387          /* result[0] = 1    * 1;
 388           * result[1] = a[1] * b[1];
 389           * result[2] = a[2] * 1;
 390           * result[3] = 1    * b[3];
 391           */
 392          i915_emit_arith(p,
 393                          A0_MUL,
 394                          get_result_vector(p, inst),
 395                          get_result_flags(inst), 0,
 396                          swizzle(src0, ONE, Y, Z, ONE),
 397                          swizzle(src1, ONE, Y, ONE, W), 0);
 398          break;
 399
 400       case OPCODE_EX2:
 401          src0 = src_vector(p, &inst->SrcReg[0], program);
 402
 403          i915_emit_arith(p,
 404                          A0_EXP,
 405                          get_result_vector(p, inst),
 406                          get_result_flags(inst), 0,
 407                          swizzle(src0, X, X, X, X), 0, 0);
 408          break;
 409
 410       case OPCODE_FLR:
 411          EMIT_1ARG_ARITH(A0_FLR);
 412          break;
 413
 414       case OPCODE_FRC:
 415          EMIT_1ARG_ARITH(A0_FRC);
 416          break;
 417
 418       case OPCODE_KIL:
 419          src0 = src_vector(p, &inst->SrcReg[0], program);
 420          tmp = i915_get_utemp(p);
 421
 422          i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
 423                          0, src0, T0_TEXKILL);
 424          break;
 425
 426       case OPCODE_LG2:
 427          src0 = src_vector(p, &inst->SrcReg[0], program);
 428
 429          i915_emit_arith(p,
 430                          A0_LOG,
 431                          get_result_vector(p, inst),
 432                          get_result_flags(inst), 0,
 433                          swizzle(src0, X, X, X, X), 0, 0);
 434          break;
 435
 436       case OPCODE_LIT:
 437          src0 = src_vector(p, &inst->SrcReg[0], program);
 438          tmp = i915_get_utemp(p);
 439
 440          /* tmp = max( a.xyzw, a.00zw )
 441           * XXX: Clamp tmp.w to -128..128
 442           * tmp.y = log(tmp.y)
 443           * tmp.y = tmp.w * tmp.y
 444           * tmp.y = exp(tmp.y)
 445           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 446           */
 447          i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 448                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
 449
 450          i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 451                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 452
 453          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 454                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 455                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
 456
 457          i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 458                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 459
 460          i915_emit_arith(p, A0_CMP,
 461                          get_result_vector(p, inst),
 462                          get_result_flags(inst), 0,
 463                          negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
 464                          swizzle(tmp, ONE, X, ZERO, ONE),
 465                          swizzle(tmp, ONE, X, Y, ONE));
 466
 467          break;
 468
 469       case OPCODE_LRP:
 470          src0 = src_vector(p, &inst->SrcReg[0], program);
 471          src1 = src_vector(p, &inst->SrcReg[1], program);
 472          src2 = src_vector(p, &inst->SrcReg[2], program);
 473          flags = get_result_flags(inst);
 474          tmp = i915_get_utemp(p);
 475
 476          /* b*a + c*(1-a)
 477           *
 478           * b*a + c - ca
 479           *
 480           * tmp = b*a + c,
 481           * result = (-c)*a + tmp
 482           */
 483          i915_emit_arith(p, A0_MAD, tmp,
 484                          flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
 485
 486          i915_emit_arith(p, A0_MAD,
 487                          get_result_vector(p, inst),
 488                          flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
 489          break;
 490
 491       case OPCODE_MAD:
 492          EMIT_3ARG_ARITH(A0_MAD);
 493          break;
 494
 495       case OPCODE_MAX:
 496          EMIT_2ARG_ARITH(A0_MAX);
 497          break;
 498
 499       case OPCODE_MIN:
 500          src0 = src_vector(p, &inst->SrcReg[0], program);
 501          src1 = src_vector(p, &inst->SrcReg[1], program);
 502          tmp = i915_get_utemp(p);
 503          flags = get_result_flags(inst);
 504
 505          i915_emit_arith(p,
 506                          A0_MAX,
 507                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 508                          negate(src0, 1, 1, 1, 1),
 509                          negate(src1, 1, 1, 1, 1), 0);
 510
 511          i915_emit_arith(p,
 512                          A0_MOV,
 513                          get_result_vector(p, inst),
 514                          flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
 515          break;
 516
 517       case OPCODE_MOV:
 518          EMIT_1ARG_ARITH(A0_MOV);
 519          break;
 520
 521       case OPCODE_MUL:
 522          EMIT_2ARG_ARITH(A0_MUL);
 523          break;
 524
 525       case OPCODE_POW:
 526          src0 = src_vector(p, &inst->SrcReg[0], program);
 527          src1 = src_vector(p, &inst->SrcReg[1], program);
 528          tmp = i915_get_utemp(p);
 529          flags = get_result_flags(inst);
 530
 531          /* XXX: masking on intermediate values, here and elsewhere.
 532           */
 533          i915_emit_arith(p,
 534                          A0_LOG,
 535                          tmp, A0_DEST_CHANNEL_X, 0,
 536                          swizzle(src0, X, X, X, X), 0, 0);
 537
 538          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
 539
 540
 541          i915_emit_arith(p,
 542                          A0_EXP,
 543                          get_result_vector(p, inst),
 544                          flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
 545
 546          break;
 547
 548       case OPCODE_RCP:
 549          src0 = src_vector(p, &inst->SrcReg[0], program);
 550
 551          i915_emit_arith(p,
 552                          A0_RCP,
 553                          get_result_vector(p, inst),
 554                          get_result_flags(inst), 0,
 555                          swizzle(src0, X, X, X, X), 0, 0);
 556          break;
 557
 558       case OPCODE_RSQ:
 559
 560          src0 = src_vector(p, &inst->SrcReg[0], program);
 561
 562          i915_emit_arith(p,
 563                          A0_RSQ,
 564                          get_result_vector(p, inst),
 565                          get_result_flags(inst), 0,
 566                          swizzle(src0, X, X, X, X), 0, 0);
 567          break;
 568
 569       case OPCODE_SCS:
 570          src0 = src_vector(p, &inst->SrcReg[0], program);
 571          tmp = i915_get_utemp(p);
 572
 573          /*
 574           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 575           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 576           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 577           * scs.x = DP4 t1, sin_constants
 578           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 579           * scs.y = DP4 t1, cos_constants
 580           */
 581          i915_emit_arith(p,
 582                          A0_MUL,
 583                          tmp, A0_DEST_CHANNEL_XY, 0,
 584                          swizzle(src0, X, X, ONE, ONE),
 585                          swizzle(src0, X, ONE, ONE, ONE), 0);
 586
 587          i915_emit_arith(p,
 588                          A0_MUL,
 589                          tmp, A0_DEST_CHANNEL_ALL, 0,
 590                          swizzle(tmp, X, Y, X, Y),
 591                          swizzle(tmp, X, X, ONE, ONE), 0);
 592
 593          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 594             GLuint tmp1;
 595
 596             if (inst->DstReg.WriteMask & WRITEMASK_X)
 597                tmp1 = i915_get_utemp(p);
 598             else
 599                tmp1 = tmp;
 600
 601             i915_emit_arith(p,
 602                             A0_MUL,
 603                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 604                             swizzle(tmp, X, Y, Y, W),
 605                             swizzle(tmp, X, Z, ONE, ONE), 0);
 606
 607             i915_emit_arith(p,
 608                             A0_DP4,
 609                             get_result_vector(p, inst),
 610                             A0_DEST_CHANNEL_Y, 0,
 611                             swizzle(tmp1, W, Z, Y, X),
 612                             i915_emit_const4fv(p, sin_constants), 0);
 613          }
 614
 615          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 616             i915_emit_arith(p,
 617                             A0_MUL,
 618                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 619                             swizzle(tmp, X, X, Z, ONE),
 620                             swizzle(tmp, Z, ONE, ONE, ONE), 0);
 621
 622             i915_emit_arith(p,
 623                             A0_DP4,
 624                             get_result_vector(p, inst),
 625                             A0_DEST_CHANNEL_X, 0,
 626                             swizzle(tmp, ONE, Z, Y, X),
 627                             i915_emit_const4fv(p, cos_constants), 0);
 628          }
 629          break;
 630
 631       case OPCODE_SGE:
 632          EMIT_2ARG_ARITH(A0_SGE);
 633          break;
 634
 635       case OPCODE_SIN:
 636          src0 = src_vector(p, &inst->SrcReg[0], program);
 637          tmp = i915_get_utemp(p);
 638
 639          i915_emit_arith(p,
 640                          A0_MUL,
 641                          tmp, A0_DEST_CHANNEL_X, 0,
 642                          src0, i915_emit_const1f(p, 1.0 / (M_PI * 2)), 0);
 643
 644          i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 645
 646          /* By choosing different taylor constants, could get rid of this mul:
 647           */
 648          i915_emit_arith(p,
 649                          A0_MUL,
 650                          tmp, A0_DEST_CHANNEL_X, 0,
 651                          tmp, i915_emit_const1f(p, (M_PI * 2)), 0);
 652
 653          /*
 654           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 655           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 656           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 657           * result = DP4 t1.wzyx, sin_constants
 658           */
 659          i915_emit_arith(p,
 660                          A0_MUL,
 661                          tmp, A0_DEST_CHANNEL_XY, 0,
 662                          swizzle(tmp, X, X, ONE, ONE),
 663                          swizzle(tmp, X, ONE, ONE, ONE), 0);
 664
 665          i915_emit_arith(p,
 666                          A0_MUL,
 667                          tmp, A0_DEST_CHANNEL_ALL, 0,
 668                          swizzle(tmp, X, Y, X, Y),
 669                          swizzle(tmp, X, X, ONE, ONE), 0);
 670
 671          i915_emit_arith(p,
 672                          A0_MUL,
 673                          tmp, A0_DEST_CHANNEL_ALL, 0,
 674                          swizzle(tmp, X, Y, Y, W),
 675                          swizzle(tmp, X, Z, ONE, ONE), 0);
 676
 677          i915_emit_arith(p,
 678                          A0_DP4,
 679                          get_result_vector(p, inst),
 680                          get_result_flags(inst), 0,
 681                          swizzle(tmp, W, Z, Y, X),
 682                          i915_emit_const4fv(p, sin_constants), 0);
 683          break;
 684
 685       case OPCODE_SLT:
 686          EMIT_2ARG_ARITH(A0_SLT);
 687          break;
 688
 689       case OPCODE_SUB:
 690          src0 = src_vector(p, &inst->SrcReg[0], program);
 691          src1 = src_vector(p, &inst->SrcReg[1], program);
 692
 693          i915_emit_arith(p,
 694                          A0_ADD,
 695                          get_result_vector(p, inst),
 696                          get_result_flags(inst), 0,
 697                          src0, negate(src1, 1, 1, 1, 1), 0);
 698          break;
 699
 700       case OPCODE_SWZ:
 701          EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
 702          break;
 703
 704       case OPCODE_TEX:
 705          EMIT_TEX(T0_TEXLD);
 706          break;
 707
 708       case OPCODE_TXB:
 709          EMIT_TEX(T0_TEXLDB);
 710          break;
 711
 712       case OPCODE_TXP:
 713          EMIT_TEX(T0_TEXLDP);
 714          break;
 715
 716       case OPCODE_XPD:
 717          /* Cross product:
 718           *      result.x = src0.y * src1.z - src0.z * src1.y;
 719           *      result.y = src0.z * src1.x - src0.x * src1.z;
 720           *      result.z = src0.x * src1.y - src0.y * src1.x;
 721           *      result.w = undef;
 722           */
 723          src0 = src_vector(p, &inst->SrcReg[0], program);
 724          src1 = src_vector(p, &inst->SrcReg[1], program);
 725          tmp = i915_get_utemp(p);
 726
 727          i915_emit_arith(p,
 728                          A0_MUL,
 729                          tmp, A0_DEST_CHANNEL_ALL, 0,
 730                          swizzle(src0, Z, X, Y, ONE),
 731                          swizzle(src1, Y, Z, X, ONE), 0);
 732
 733          i915_emit_arith(p,
 734                          A0_MAD,
 735                          get_result_vector(p, inst),
 736                          get_result_flags(inst), 0,
 737                          swizzle(src0, Y, Z, X, ONE),
 738                          swizzle(src1, Z, X, Y, ONE),
 739                          negate(tmp, 1, 1, 1, 0));
 740          break;
 741
 742       case OPCODE_END:
 743          return;
 744
 745       default:
 746          i915_program_error(p, "bad opcode");
 747          return;
 748       }
 749
 750       inst++;
 751       i915_release_utemps(p);
 752    }
 753 }
 754
 755 /* Rather than trying to intercept and jiggle depth writes during
 756  * emit, just move the value into its correct position at the end of
 757  * the program:
 758  */
 759 static void
 760 fixup_depth_write(struct i915_fragment_program *p)
 761 {
 762    if (p->depth_written) {
 763       GLuint depth = UREG(REG_TYPE_OD, 0);
 764
 765       i915_emit_arith(p,
 766                       A0_MOV,
 767                       depth, A0_DEST_CHANNEL_W, 0,
 768                       swizzle(depth, X, Y, Z, Z), 0, 0);
 769    }
 770 }
 771
 772
 773 #define FRAG_BIT_TEX(n)  (FRAG_BIT_TEX0 << (n))
 774
 775
 776 static void
 777 check_wpos(struct i915_fragment_program *p)
 778 {
 779    GLuint inputs = p->FragProg.Base.InputsRead;
 780    GLint i;
 781
 782    p->wpos_tex = -1;
 783
 784    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
 785       if (inputs & FRAG_BIT_TEX(i))
 786          continue;
 787       else if (inputs & FRAG_BIT_WPOS) {
 788          p->wpos_tex = i;
 789          inputs &= ~FRAG_BIT_WPOS;
 790       }
 791    }
 792
 793    if (inputs & FRAG_BIT_WPOS) {
 794       i915_program_error(p, "No free texcoord for wpos value");
 795    }
 796 }
 797
 798
 799 static void
 800 translate_program(struct i915_fragment_program *p)
 801 {
 802    struct i915_context *i915 = I915_CONTEXT(p->ctx);
 803
 804    i915_init_program(i915, p);
 805    check_wpos(p);
 806    upload_program(p);
 807    fixup_depth_write(p);
 808    i915_fini_program(p);
 809
 810    p->translated = 1;
 811 }
 812
 813
 814 static void
 815 track_params(struct i915_fragment_program *p)
 816 {
 817    GLint i;
 818
 819    if (p->nr_params)
 820       _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
 821
 822    for (i = 0; i < p->nr_params; i++) {
 823       GLint reg = p->param[i].reg;
 824       COPY_4V(p->constant[reg], p->param[i].values);
 825    }
 826
 827    p->params_uptodate = 1;
 828    p->on_hardware = 0;          /* overkill */
 829 }
 830
 831
 832 static void
 833 i915BindProgram(GLcontext * ctx, GLenum target, struct gl_program *prog)
 834 {
 835    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 836       struct i915_context *i915 = I915_CONTEXT(ctx);
 837       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 838
 839       if (i915->current_program == p)
 840          return;
 841
 842       if (i915->current_program) {
 843          i915->current_program->on_hardware = 0;
 844          i915->current_program->params_uptodate = 0;
 845       }
 846
 847       i915->current_program = p;
 848
 849       assert(p->on_hardware == 0);
 850       assert(p->params_uptodate == 0);
 851
 852       /* Hack: make sure fog is correctly enabled according to this
 853        * fragment program's fog options.
 854        */
 855       ctx->Driver.Enable(ctx, GL_FRAGMENT_PROGRAM_ARB,
 856                          ctx->FragmentProgram.Enabled);
 857    }
 858 }
 859
 860 static struct gl_program *
 861 i915NewProgram(GLcontext * ctx, GLenum target, GLuint id)
 862 {
 863    switch (target) {
 864    case GL_VERTEX_PROGRAM_ARB:
 865       return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
 866                                        target, id);
 867
 868    case GL_FRAGMENT_PROGRAM_ARB:{
 869          struct i915_fragment_program *prog =
 870             CALLOC_STRUCT(i915_fragment_program);
 871          if (prog) {
 872             i915_init_program(I915_CONTEXT(ctx), prog);
 873
 874             return _mesa_init_fragment_program(ctx, &prog->FragProg,
 875                                                target, id);
 876          }
 877          else
 878             return NULL;
 879       }
 880
 881    default:
 882       /* Just fallback:
 883        */
 884       return _mesa_new_program(ctx, target, id);
 885    }
 886 }
 887
 888 static void
 889 i915DeleteProgram(GLcontext * ctx, struct gl_program *prog)
 890 {
 891    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
 892       struct i915_context *i915 = I915_CONTEXT(ctx);
 893       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 894
 895       if (i915->current_program == p)
 896          i915->current_program = 0;
 897    }
 898
 899    _mesa_delete_program(ctx, prog);
 900 }
 901
 902
 903 static GLboolean
 904 i915IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
 905 {
 906    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 907       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 908
 909       if (!p->translated)
 910          translate_program(p);
 911
 912       return !p->error;
 913    }
 914    else
 915       return GL_TRUE;
 916 }
 917
 918 static void
 919 i915ProgramStringNotify(GLcontext * ctx,
 920                         GLenum target, struct gl_program *prog)
 921 {
 922    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 923       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 924       p->translated = 0;
 925
 926       /* Hack: make sure fog is correctly enabled according to this
 927        * fragment program's fog options.
 928        */
 929       ctx->Driver.Enable(ctx, GL_FRAGMENT_PROGRAM_ARB,
 930                          ctx->FragmentProgram.Enabled);
 931    }
 932
 933    _tnl_program_string(ctx, target, prog);
 934 }
 935
 936
 937 void
 938 i915ValidateFragmentProgram(struct i915_context *i915)
 939 {
 940    GLcontext *ctx = &i915->intel.ctx;
 941    struct intel_context *intel = intel_context(ctx);
 942    TNLcontext *tnl = TNL_CONTEXT(ctx);
 943    struct vertex_buffer *VB = &tnl->vb;
 944
 945    struct i915_fragment_program *p =
 946       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
 947
 948    const GLuint inputsRead = p->FragProg.Base.InputsRead;
 949    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
 950    GLuint s2 = S2_TEXCOORD_NONE;
 951    int i, offset = 0;
 952
 953    if (i915->current_program != p) {
 954       if (i915->current_program) {
 955          i915->current_program->on_hardware = 0;
 956          i915->current_program->params_uptodate = 0;
 957       }
 958
 959       i915->current_program = p;
 960    }
 961
 962
 963    /* Important:
 964     */
 965    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
 966
 967    if (!p->translated)
 968       translate_program(p);
 969
 970    intel->vertex_attr_count = 0;
 971    intel->wpos_offset = 0;
 972    intel->wpos_size = 0;
 973    intel->coloroffset = 0;
 974    intel->specoffset = 0;
 975
 976    if (inputsRead & FRAG_BITS_TEX_ANY) {
 977       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
 978    }
 979    else {
 980       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
 981    }
 982
 983    if (inputsRead & FRAG_BIT_COL0) {
 984       intel->coloroffset = offset / 4;
 985       EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
 986    }
 987
 988    if ((inputsRead & (FRAG_BIT_COL1 | FRAG_BIT_FOGC)) ||
 989        i915->vertex_fog != I915_FOG_NONE) {
 990
 991       if (inputsRead & FRAG_BIT_COL1) {
 992          intel->specoffset = offset / 4;
 993          EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, S4_VFMT_SPEC_FOG, 3);
 994       }
 995       else
 996          EMIT_PAD(3);
 997
 998       if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE)
 999          EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1UB_1F, S4_VFMT_SPEC_FOG, 1);
1000       else
1001          EMIT_PAD(1);
1002    }
1003
1004 #if 0
1005    if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1006       EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1007    }
1008 #endif
1009
1010    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1011       if (inputsRead & FRAG_BIT_TEX(i)) {
1012          int sz = VB->TexCoordPtr[i]->size;
1013
1014          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1015          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1016
1017          EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1018       }
1019       else if (i == p->wpos_tex) {
1020
1021          /* If WPOS is required, duplicate the XYZ position data in an
1022           * unused texture coordinate:
1023           */
1024          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1025          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1026
1027          intel->wpos_offset = offset;
1028          intel->wpos_size = 3 * sizeof(GLuint);
1029
1030          EMIT_PAD(intel->wpos_size);
1031       }
1032    }
1033
1034    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1035        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1036       int k;
1037
1038       I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1039
1040       /* Must do this *after* statechange, so as not to affect
1041        * buffered vertices reliant on the old state:
1042        */
1043       intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1044                                               intel->vertex_attrs,
1045                                               intel->vertex_attr_count,
1046                                               intel->ViewportMatrix.m, 0);
1047
1048       intel->vertex_size >>= 2;
1049
1050       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1051       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1052
1053       k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1054       assert(k);
1055    }
1056
1057    if (!p->params_uptodate)
1058       track_params(p);
1059
1060    if (!p->on_hardware)
1061       i915_upload_program(i915, p);
1062 }
1063
1064 void
1065 i915InitFragProgFuncs(struct dd_function_table *functions)
1066 {
1067    functions->BindProgram = i915BindProgram;
1068    functions->NewProgram = i915NewProgram;
1069    functions->DeleteProgram = i915DeleteProgram;
1070    functions->IsProgramNative = i915IsProgramNative;
1071    functions->ProgramStringNotify = i915ProgramStringNotify;
1072 }