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 "prog_instruction.h"
  41 #include "prog_parameter.h"
  42 #include "program.h"
  43 #include "programopt.h"
  44
  45
  46
  47 /* 1, -1/3!, 1/5!, -1/7! */
  48 static const GLfloat sin_constants[4] = { 1.0,
  49    -1.0 / (3 * 2 * 1),
  50    1.0 / (5 * 4 * 3 * 2 * 1),
  51    -1.0 / (7 * 6 * 5 * 4 * 3 * 2 * 1)
  52 };
  53
  54 /* 1, -1/2!, 1/4!, -1/6! */
  55 static const GLfloat cos_constants[4] = { 1.0,
  56    -1.0 / (2 * 1),
  57    1.0 / (4 * 3 * 2 * 1),
  58    -1.0 / (6 * 5 * 4 * 3 * 2 * 1)
  59 };
  60
  61 /**
  62  * Retrieve a ureg for the given source register.  Will emit
  63  * constants, apply swizzling and negation as needed.
  64  */
  65 static GLuint
  66 src_vector(struct i915_fragment_program *p,
  67            const struct prog_src_register *source,
  68            const struct gl_fragment_program *program)
  69 {
  70    GLuint src;
  71
  72    switch (source->File) {
  73
  74       /* Registers:
  75        */
  76    case PROGRAM_TEMPORARY:
  77       if (source->Index >= I915_MAX_TEMPORARY) {
  78          i915_program_error(p, "Exceeded max temporary reg");
  79          return 0;
  80       }
  81       src = UREG(REG_TYPE_R, source->Index);
  82       break;
  83    case PROGRAM_INPUT:
  84       switch (source->Index) {
  85       case FRAG_ATTRIB_WPOS:
  86          src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
  87          break;
  88       case FRAG_ATTRIB_COL0:
  89          src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
  90          break;
  91       case FRAG_ATTRIB_COL1:
  92          src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
  93          src = swizzle(src, X, Y, Z, ONE);
  94          break;
  95       case FRAG_ATTRIB_FOGC:
  96          src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
  97          src = swizzle(src, W, W, W, W);
  98          break;
  99       case FRAG_ATTRIB_TEX0:
 100       case FRAG_ATTRIB_TEX1:
 101       case FRAG_ATTRIB_TEX2:
 102       case FRAG_ATTRIB_TEX3:
 103       case FRAG_ATTRIB_TEX4:
 104       case FRAG_ATTRIB_TEX5:
 105       case FRAG_ATTRIB_TEX6:
 106       case FRAG_ATTRIB_TEX7:
 107          src = i915_emit_decl(p, REG_TYPE_T,
 108                               T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
 109                               D0_CHANNEL_ALL);
 110          break;
 111
 112       default:
 113          i915_program_error(p, "Bad source->Index");
 114          return 0;
 115       }
 116       break;
 117
 118       /* Various paramters and env values.  All emitted to
 119        * hardware as program constants.
 120        */
 121    case PROGRAM_LOCAL_PARAM:
 122       src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
 123       break;
 124
 125    case PROGRAM_ENV_PARAM:
 126       src =
 127          i915_emit_param4fv(p,
 128                             p->ctx->FragmentProgram.Parameters[source->
 129                                                                Index]);
 130       break;
 131
 132    case PROGRAM_CONSTANT:
 133    case PROGRAM_STATE_VAR:
 134    case PROGRAM_NAMED_PARAM:
 135       src =
 136          i915_emit_param4fv(p,
 137                             program->Base.Parameters->ParameterValues[source->
 138                                                                       Index]);
 139       break;
 140
 141    default:
 142       i915_program_error(p, "Bad source->File");
 143       return 0;
 144    }
 145
 146    src = swizzle(src,
 147                  GET_SWZ(source->Swizzle, 0),
 148                  GET_SWZ(source->Swizzle, 1),
 149                  GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
 150
 151    if (source->NegateBase)
 152       src = negate(src,
 153                    GET_BIT(source->NegateBase, 0),
 154                    GET_BIT(source->NegateBase, 1),
 155                    GET_BIT(source->NegateBase, 2),
 156                    GET_BIT(source->NegateBase, 3));
 157
 158    return src;
 159 }
 160
 161
 162 static GLuint
 163 get_result_vector(struct i915_fragment_program *p,
 164                   const struct prog_instruction *inst)
 165 {
 166    switch (inst->DstReg.File) {
 167    case PROGRAM_OUTPUT:
 168       switch (inst->DstReg.Index) {
 169       case FRAG_RESULT_COLR:
 170          return UREG(REG_TYPE_OC, 0);
 171       case FRAG_RESULT_DEPR:
 172          p->depth_written = 1;
 173          return UREG(REG_TYPE_OD, 0);
 174       default:
 175          i915_program_error(p, "Bad inst->DstReg.Index");
 176          return 0;
 177       }
 178    case PROGRAM_TEMPORARY:
 179       return UREG(REG_TYPE_R, inst->DstReg.Index);
 180    default:
 181       i915_program_error(p, "Bad inst->DstReg.File");
 182       return 0;
 183    }
 184 }
 185
 186 static GLuint
 187 get_result_flags(const struct prog_instruction *inst)
 188 {
 189    GLuint flags = 0;
 190
 191    if (inst->SaturateMode == SATURATE_ZERO_ONE)
 192       flags |= A0_DEST_SATURATE;
 193    if (inst->DstReg.WriteMask & WRITEMASK_X)
 194       flags |= A0_DEST_CHANNEL_X;
 195    if (inst->DstReg.WriteMask & WRITEMASK_Y)
 196       flags |= A0_DEST_CHANNEL_Y;
 197    if (inst->DstReg.WriteMask & WRITEMASK_Z)
 198       flags |= A0_DEST_CHANNEL_Z;
 199    if (inst->DstReg.WriteMask & WRITEMASK_W)
 200       flags |= A0_DEST_CHANNEL_W;
 201
 202    return flags;
 203 }
 204
 205 static GLuint
 206 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 207 {
 208    switch (bit) {
 209    case TEXTURE_1D_INDEX:
 210       return D0_SAMPLE_TYPE_2D;
 211    case TEXTURE_2D_INDEX:
 212       return D0_SAMPLE_TYPE_2D;
 213    case TEXTURE_RECT_INDEX:
 214       return D0_SAMPLE_TYPE_2D;
 215    case TEXTURE_3D_INDEX:
 216       return D0_SAMPLE_TYPE_VOLUME;
 217    case TEXTURE_CUBE_INDEX:
 218       return D0_SAMPLE_TYPE_CUBE;
 219    default:
 220       i915_program_error(p, "TexSrcBit");
 221       return 0;
 222    }
 223 }
 224
 225 #define EMIT_TEX( OP )                                          \
 226 do {                                                            \
 227    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 228    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 229                                   inst->TexSrcUnit, dim);       \
 230    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 231    /* Texel lookup */                                           \
 232                                                                 \
 233    i915_emit_texld( p,                                          \
 234                get_result_vector( p, inst ),                    \
 235                get_result_flags( inst ),                        \
 236                sampler,                                         \
 237                coord,                                           \
 238                OP);                                             \
 239 } while (0)
 240
 241 #define EMIT_ARITH( OP, N )                                             \
 242 do {                                                                    \
 243    i915_emit_arith( p,                                                  \
 244                OP,                                                      \
 245                get_result_vector( p, inst ),                            \
 246                get_result_flags( inst ), 0,                     \
 247                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 248                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 249                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 250 } while (0)
 251
 252 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 253 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 254 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 255
 256
 257 /* Possible concerns:
 258  *
 259  * SIN, COS -- could use another taylor step?
 260  * LIT      -- results seem a little different to sw mesa
 261  * LOG      -- different to mesa on negative numbers, but this is conformant.
 262  *
 263  * Parse failures -- Mesa doesn't currently give a good indication
 264  * internally whether a particular program string parsed or not.  This
 265  * can lead to confusion -- hopefully we cope with it ok now.
 266  *
 267  */
 268 static void
 269 upload_program(struct i915_fragment_program *p)
 270 {
 271    const struct gl_fragment_program *program =
 272       p->ctx->FragmentProgram._Current;
 273    const struct prog_instruction *inst = program->Base.Instructions;
 274
 275 /*    _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */
 276
 277    /* Is this a parse-failed program?  Ensure a valid program is
 278     * loaded, as the flagging of an error isn't sufficient to stop
 279     * this being uploaded to hardware.
 280     */
 281    if (inst[0].Opcode == OPCODE_END) {
 282       GLuint tmp = i915_get_utemp(p);
 283       i915_emit_arith(p,
 284                       A0_MOV,
 285                       UREG(REG_TYPE_OC, 0),
 286                       A0_DEST_CHANNEL_ALL, 0,
 287                       swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
 288       return;
 289    }
 290
 291    while (1) {
 292       GLuint src0, src1, src2, flags;
 293       GLuint tmp = 0;
 294
 295       switch (inst->Opcode) {
 296       case OPCODE_ABS:
 297          src0 = src_vector(p, &inst->SrcReg[0], program);
 298          i915_emit_arith(p,
 299                          A0_MAX,
 300                          get_result_vector(p, inst),
 301                          get_result_flags(inst), 0,
 302                          src0, negate(src0, 1, 1, 1, 1), 0);
 303          break;
 304
 305       case OPCODE_ADD:
 306          EMIT_2ARG_ARITH(A0_ADD);
 307          break;
 308
 309       case OPCODE_CMP:
 310          src0 = src_vector(p, &inst->SrcReg[0], program);
 311          src1 = src_vector(p, &inst->SrcReg[1], program);
 312          src2 = src_vector(p, &inst->SrcReg[2], program);
 313          i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
 314          break;
 315
 316       case OPCODE_COS:
 317          src0 = src_vector(p, &inst->SrcReg[0], program);
 318          tmp = i915_get_utemp(p);
 319
 320          i915_emit_arith(p,
 321                          A0_MUL,
 322                          tmp, A0_DEST_CHANNEL_X, 0,
 323                          src0, i915_emit_const1f(p, 1.0 / (M_PI * 2)), 0);
 324
 325          i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 326
 327          /* By choosing different taylor constants, could get rid of this mul:
 328           */
 329          i915_emit_arith(p,
 330                          A0_MUL,
 331                          tmp, A0_DEST_CHANNEL_X, 0,
 332                          tmp, i915_emit_const1f(p, (M_PI * 2)), 0);
 333
 334          /*
 335           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 336           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
 337           * t0 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 338           * result = DP4 t0, cos_constants
 339           */
 340          i915_emit_arith(p,
 341                          A0_MUL,
 342                          tmp, A0_DEST_CHANNEL_XY, 0,
 343                          swizzle(tmp, X, X, ONE, ONE),
 344                          swizzle(tmp, X, ONE, ONE, ONE), 0);
 345
 346          i915_emit_arith(p,
 347                          A0_MUL,
 348                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 349                          swizzle(tmp, X, Y, X, ONE),
 350                          swizzle(tmp, X, X, ONE, ONE), 0);
 351
 352          i915_emit_arith(p,
 353                          A0_MUL,
 354                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 355                          swizzle(tmp, X, X, Z, ONE),
 356                          swizzle(tmp, Z, ONE, ONE, ONE), 0);
 357
 358          i915_emit_arith(p,
 359                          A0_DP4,
 360                          get_result_vector(p, inst),
 361                          get_result_flags(inst), 0,
 362                          swizzle(tmp, ONE, Z, Y, X),
 363                          i915_emit_const4fv(p, cos_constants), 0);
 364
 365          break;
 366
 367       case OPCODE_DP3:
 368          EMIT_2ARG_ARITH(A0_DP3);
 369          break;
 370
 371       case OPCODE_DP4:
 372          EMIT_2ARG_ARITH(A0_DP4);
 373          break;
 374
 375       case OPCODE_DPH:
 376          src0 = src_vector(p, &inst->SrcReg[0], program);
 377          src1 = src_vector(p, &inst->SrcReg[1], program);
 378
 379          i915_emit_arith(p,
 380                          A0_DP4,
 381                          get_result_vector(p, inst),
 382                          get_result_flags(inst), 0,
 383                          swizzle(src0, X, Y, Z, ONE), src1, 0);
 384          break;
 385
 386       case OPCODE_DST:
 387          src0 = src_vector(p, &inst->SrcReg[0], program);
 388          src1 = src_vector(p, &inst->SrcReg[1], program);
 389
 390          /* result[0] = 1    * 1;
 391           * result[1] = a[1] * b[1];
 392           * result[2] = a[2] * 1;
 393           * result[3] = 1    * b[3];
 394           */
 395          i915_emit_arith(p,
 396                          A0_MUL,
 397                          get_result_vector(p, inst),
 398                          get_result_flags(inst), 0,
 399                          swizzle(src0, ONE, Y, Z, ONE),
 400                          swizzle(src1, ONE, Y, ONE, W), 0);
 401          break;
 402
 403       case OPCODE_EX2:
 404          src0 = src_vector(p, &inst->SrcReg[0], program);
 405
 406          i915_emit_arith(p,
 407                          A0_EXP,
 408                          get_result_vector(p, inst),
 409                          get_result_flags(inst), 0,
 410                          swizzle(src0, X, X, X, X), 0, 0);
 411          break;
 412
 413       case OPCODE_FLR:
 414          EMIT_1ARG_ARITH(A0_FLR);
 415          break;
 416
 417       case OPCODE_FRC:
 418          EMIT_1ARG_ARITH(A0_FRC);
 419          break;
 420
 421       case OPCODE_KIL:
 422          src0 = src_vector(p, &inst->SrcReg[0], program);
 423          tmp = i915_get_utemp(p);
 424
 425          i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
 426                          0, src0, T0_TEXKILL);
 427          break;
 428
 429       case OPCODE_LG2:
 430          src0 = src_vector(p, &inst->SrcReg[0], program);
 431
 432          i915_emit_arith(p,
 433                          A0_LOG,
 434                          get_result_vector(p, inst),
 435                          get_result_flags(inst), 0,
 436                          swizzle(src0, X, X, X, X), 0, 0);
 437          break;
 438
 439       case OPCODE_LIT:
 440          src0 = src_vector(p, &inst->SrcReg[0], program);
 441          tmp = i915_get_utemp(p);
 442
 443          /* tmp = max( a.xyzw, a.00zw )
 444           * XXX: Clamp tmp.w to -128..128
 445           * tmp.y = log(tmp.y)
 446           * tmp.y = tmp.w * tmp.y
 447           * tmp.y = exp(tmp.y)
 448           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 449           */
 450          i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 451                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
 452
 453          i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 454                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 455
 456          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 457                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 458                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
 459
 460          i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 461                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 462
 463          i915_emit_arith(p, A0_CMP,
 464                          get_result_vector(p, inst),
 465                          get_result_flags(inst), 0,
 466                          negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
 467                          swizzle(tmp, ONE, X, ZERO, ONE),
 468                          swizzle(tmp, ONE, X, Y, ONE));
 469
 470          break;
 471
 472       case OPCODE_LRP:
 473          src0 = src_vector(p, &inst->SrcReg[0], program);
 474          src1 = src_vector(p, &inst->SrcReg[1], program);
 475          src2 = src_vector(p, &inst->SrcReg[2], program);
 476          flags = get_result_flags(inst);
 477          tmp = i915_get_utemp(p);
 478
 479          /* b*a + c*(1-a)
 480           *
 481           * b*a + c - ca
 482           *
 483           * tmp = b*a + c,
 484           * result = (-c)*a + tmp
 485           */
 486          i915_emit_arith(p, A0_MAD, tmp,
 487                          flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
 488
 489          i915_emit_arith(p, A0_MAD,
 490                          get_result_vector(p, inst),
 491                          flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
 492          break;
 493
 494       case OPCODE_MAD:
 495          EMIT_3ARG_ARITH(A0_MAD);
 496          break;
 497
 498       case OPCODE_MAX:
 499          EMIT_2ARG_ARITH(A0_MAX);
 500          break;
 501
 502       case OPCODE_MIN:
 503          src0 = src_vector(p, &inst->SrcReg[0], program);
 504          src1 = src_vector(p, &inst->SrcReg[1], program);
 505          tmp = i915_get_utemp(p);
 506          flags = get_result_flags(inst);
 507
 508          i915_emit_arith(p,
 509                          A0_MAX,
 510                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 511                          negate(src0, 1, 1, 1, 1),
 512                          negate(src1, 1, 1, 1, 1), 0);
 513
 514          i915_emit_arith(p,
 515                          A0_MOV,
 516                          get_result_vector(p, inst),
 517                          flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
 518          break;
 519
 520       case OPCODE_MOV:
 521          EMIT_1ARG_ARITH(A0_MOV);
 522          break;
 523
 524       case OPCODE_MUL:
 525          EMIT_2ARG_ARITH(A0_MUL);
 526          break;
 527
 528       case OPCODE_POW:
 529          src0 = src_vector(p, &inst->SrcReg[0], program);
 530          src1 = src_vector(p, &inst->SrcReg[1], program);
 531          tmp = i915_get_utemp(p);
 532          flags = get_result_flags(inst);
 533
 534          /* XXX: masking on intermediate values, here and elsewhere.
 535           */
 536          i915_emit_arith(p,
 537                          A0_LOG,
 538                          tmp, A0_DEST_CHANNEL_X, 0,
 539                          swizzle(src0, X, X, X, X), 0, 0);
 540
 541          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
 542
 543
 544          i915_emit_arith(p,
 545                          A0_EXP,
 546                          get_result_vector(p, inst),
 547                          flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
 548
 549          break;
 550
 551       case OPCODE_RCP:
 552          src0 = src_vector(p, &inst->SrcReg[0], program);
 553
 554          i915_emit_arith(p,
 555                          A0_RCP,
 556                          get_result_vector(p, inst),
 557                          get_result_flags(inst), 0,
 558                          swizzle(src0, X, X, X, X), 0, 0);
 559          break;
 560
 561       case OPCODE_RSQ:
 562
 563          src0 = src_vector(p, &inst->SrcReg[0], program);
 564
 565          i915_emit_arith(p,
 566                          A0_RSQ,
 567                          get_result_vector(p, inst),
 568                          get_result_flags(inst), 0,
 569                          swizzle(src0, X, X, X, X), 0, 0);
 570          break;
 571
 572       case OPCODE_SCS:
 573          src0 = src_vector(p, &inst->SrcReg[0], program);
 574          tmp = i915_get_utemp(p);
 575
 576          /*
 577           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 578           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 579           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 580           * scs.x = DP4 t1, sin_constants
 581           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 582           * scs.y = DP4 t1, cos_constants
 583           */
 584          i915_emit_arith(p,
 585                          A0_MUL,
 586                          tmp, A0_DEST_CHANNEL_XY, 0,
 587                          swizzle(src0, X, X, ONE, ONE),
 588                          swizzle(src0, X, ONE, ONE, ONE), 0);
 589
 590          i915_emit_arith(p,
 591                          A0_MUL,
 592                          tmp, A0_DEST_CHANNEL_ALL, 0,
 593                          swizzle(tmp, X, Y, X, Y),
 594                          swizzle(tmp, X, X, ONE, ONE), 0);
 595
 596          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 597             GLuint tmp1;
 598
 599             if (inst->DstReg.WriteMask & WRITEMASK_X)
 600                tmp1 = i915_get_utemp(p);
 601             else
 602                tmp1 = tmp;
 603
 604             i915_emit_arith(p,
 605                             A0_MUL,
 606                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 607                             swizzle(tmp, X, Y, Y, W),
 608                             swizzle(tmp, X, Z, ONE, ONE), 0);
 609
 610             i915_emit_arith(p,
 611                             A0_DP4,
 612                             get_result_vector(p, inst),
 613                             A0_DEST_CHANNEL_Y, 0,
 614                             swizzle(tmp1, W, Z, Y, X),
 615                             i915_emit_const4fv(p, sin_constants), 0);
 616          }
 617
 618          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 619             i915_emit_arith(p,
 620                             A0_MUL,
 621                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 622                             swizzle(tmp, X, X, Z, ONE),
 623                             swizzle(tmp, Z, ONE, ONE, ONE), 0);
 624
 625             i915_emit_arith(p,
 626                             A0_DP4,
 627                             get_result_vector(p, inst),
 628                             A0_DEST_CHANNEL_X, 0,
 629                             swizzle(tmp, ONE, Z, Y, X),
 630                             i915_emit_const4fv(p, cos_constants), 0);
 631          }
 632          break;
 633
 634       case OPCODE_SGE:
 635          EMIT_2ARG_ARITH(A0_SGE);
 636          break;
 637
 638       case OPCODE_SIN:
 639          src0 = src_vector(p, &inst->SrcReg[0], program);
 640          tmp = i915_get_utemp(p);
 641
 642          i915_emit_arith(p,
 643                          A0_MUL,
 644                          tmp, A0_DEST_CHANNEL_X, 0,
 645                          src0, i915_emit_const1f(p, 1.0 / (M_PI * 2)), 0);
 646
 647          i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 648
 649          /* By choosing different taylor constants, could get rid of this mul:
 650           */
 651          i915_emit_arith(p,
 652                          A0_MUL,
 653                          tmp, A0_DEST_CHANNEL_X, 0,
 654                          tmp, i915_emit_const1f(p, (M_PI * 2)), 0);
 655
 656          /*
 657           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 658           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 659           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 660           * result = DP4 t1.wzyx, sin_constants
 661           */
 662          i915_emit_arith(p,
 663                          A0_MUL,
 664                          tmp, A0_DEST_CHANNEL_XY, 0,
 665                          swizzle(tmp, X, X, ONE, ONE),
 666                          swizzle(tmp, X, ONE, ONE, ONE), 0);
 667
 668          i915_emit_arith(p,
 669                          A0_MUL,
 670                          tmp, A0_DEST_CHANNEL_ALL, 0,
 671                          swizzle(tmp, X, Y, X, Y),
 672                          swizzle(tmp, X, X, ONE, ONE), 0);
 673
 674          i915_emit_arith(p,
 675                          A0_MUL,
 676                          tmp, A0_DEST_CHANNEL_ALL, 0,
 677                          swizzle(tmp, X, Y, Y, W),
 678                          swizzle(tmp, X, Z, ONE, ONE), 0);
 679
 680          i915_emit_arith(p,
 681                          A0_DP4,
 682                          get_result_vector(p, inst),
 683                          get_result_flags(inst), 0,
 684                          swizzle(tmp, W, Z, Y, X),
 685                          i915_emit_const4fv(p, sin_constants), 0);
 686          break;
 687
 688       case OPCODE_SLT:
 689          EMIT_2ARG_ARITH(A0_SLT);
 690          break;
 691
 692       case OPCODE_SUB:
 693          src0 = src_vector(p, &inst->SrcReg[0], program);
 694          src1 = src_vector(p, &inst->SrcReg[1], program);
 695
 696          i915_emit_arith(p,
 697                          A0_ADD,
 698                          get_result_vector(p, inst),
 699                          get_result_flags(inst), 0,
 700                          src0, negate(src1, 1, 1, 1, 1), 0);
 701          break;
 702
 703       case OPCODE_SWZ:
 704          EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
 705          break;
 706
 707       case OPCODE_TEX:
 708          EMIT_TEX(T0_TEXLD);
 709          break;
 710
 711       case OPCODE_TXB:
 712          EMIT_TEX(T0_TEXLDB);
 713          break;
 714
 715       case OPCODE_TXP:
 716          EMIT_TEX(T0_TEXLDP);
 717          break;
 718
 719       case OPCODE_XPD:
 720          /* Cross product:
 721           *      result.x = src0.y * src1.z - src0.z * src1.y;
 722           *      result.y = src0.z * src1.x - src0.x * src1.z;
 723           *      result.z = src0.x * src1.y - src0.y * src1.x;
 724           *      result.w = undef;
 725           */
 726          src0 = src_vector(p, &inst->SrcReg[0], program);
 727          src1 = src_vector(p, &inst->SrcReg[1], program);
 728          tmp = i915_get_utemp(p);
 729
 730          i915_emit_arith(p,
 731                          A0_MUL,
 732                          tmp, A0_DEST_CHANNEL_ALL, 0,
 733                          swizzle(src0, Z, X, Y, ONE),
 734                          swizzle(src1, Y, Z, X, ONE), 0);
 735
 736          i915_emit_arith(p,
 737                          A0_MAD,
 738                          get_result_vector(p, inst),
 739                          get_result_flags(inst), 0,
 740                          swizzle(src0, Y, Z, X, ONE),
 741                          swizzle(src1, Z, X, Y, ONE),
 742                          negate(tmp, 1, 1, 1, 0));
 743          break;
 744
 745       case OPCODE_END:
 746          return;
 747
 748       default:
 749          i915_program_error(p, "bad opcode");
 750          return;
 751       }
 752
 753       inst++;
 754       i915_release_utemps(p);
 755    }
 756 }
 757
 758 /* Rather than trying to intercept and jiggle depth writes during
 759  * emit, just move the value into its correct position at the end of
 760  * the program:
 761  */
 762 static void
 763 fixup_depth_write(struct i915_fragment_program *p)
 764 {
 765    if (p->depth_written) {
 766       GLuint depth = UREG(REG_TYPE_OD, 0);
 767
 768       i915_emit_arith(p,
 769                       A0_MOV,
 770                       depth, A0_DEST_CHANNEL_W, 0,
 771                       swizzle(depth, X, Y, Z, Z), 0, 0);
 772    }
 773 }
 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    }
 853 }
 854
 855 static struct gl_program *
 856 i915NewProgram(GLcontext * ctx, GLenum target, GLuint id)
 857 {
 858    switch (target) {
 859    case GL_VERTEX_PROGRAM_ARB:
 860       return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
 861                                        target, id);
 862
 863    case GL_FRAGMENT_PROGRAM_ARB:{
 864          struct i915_fragment_program *prog =
 865             CALLOC_STRUCT(i915_fragment_program);
 866          if (prog) {
 867             i915_init_program(I915_CONTEXT(ctx), prog);
 868
 869             return _mesa_init_fragment_program(ctx, &prog->FragProg,
 870                                                target, id);
 871          }
 872          else
 873             return NULL;
 874       }
 875
 876    default:
 877       /* Just fallback:
 878        */
 879       return _mesa_new_program(ctx, target, id);
 880    }
 881 }
 882
 883 static void
 884 i915DeleteProgram(GLcontext * ctx, struct gl_program *prog)
 885 {
 886    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
 887       struct i915_context *i915 = I915_CONTEXT(ctx);
 888       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 889
 890       if (i915->current_program == p)
 891          i915->current_program = 0;
 892    }
 893
 894    _mesa_delete_program(ctx, prog);
 895 }
 896
 897
 898 static GLboolean
 899 i915IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
 900 {
 901    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 902       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 903
 904       if (!p->translated)
 905          translate_program(p);
 906
 907       return !p->error;
 908    }
 909    else
 910       return GL_TRUE;
 911 }
 912
 913 static void
 914 i915ProgramStringNotify(GLcontext * ctx,
 915                         GLenum target, struct gl_program *prog)
 916 {
 917    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 918       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 919       p->translated = 0;
 920
 921       /* Hack: make sure fog is correctly enabled according to this
 922        * fragment program's fog options.
 923        */
 924       if (p->FragProg.FogOption) {
 925          /* add extra instructions to do fog, then turn off FogOption field */
 926          _mesa_append_fog_code(ctx, &p->FragProg);
 927          p->FragProg.FogOption = GL_NONE;
 928       }
 929    }
 930
 931    _tnl_program_string(ctx, target, prog);
 932 }
 933
 934
 935 void
 936 i915ValidateFragmentProgram(struct i915_context *i915)
 937 {
 938    GLcontext *ctx = &i915->intel.ctx;
 939    struct intel_context *intel = intel_context(ctx);
 940    TNLcontext *tnl = TNL_CONTEXT(ctx);
 941    struct vertex_buffer *VB = &tnl->vb;
 942
 943    struct i915_fragment_program *p =
 944       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
 945
 946    const GLuint inputsRead = p->FragProg.Base.InputsRead;
 947    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
 948    GLuint s2 = S2_TEXCOORD_NONE;
 949    int i, offset = 0;
 950
 951    if (i915->current_program != p) {
 952       if (i915->current_program) {
 953          i915->current_program->on_hardware = 0;
 954          i915->current_program->params_uptodate = 0;
 955       }
 956
 957       i915->current_program = p;
 958    }
 959
 960
 961    /* Important:
 962     */
 963    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
 964
 965    if (!p->translated)
 966       translate_program(p);
 967
 968    intel->vertex_attr_count = 0;
 969    intel->wpos_offset = 0;
 970    intel->wpos_size = 0;
 971    intel->coloroffset = 0;
 972    intel->specoffset = 0;
 973
 974    if (inputsRead & FRAG_BITS_TEX_ANY) {
 975       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
 976    }
 977    else {
 978       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
 979    }
 980
 981    if (inputsRead & FRAG_BIT_COL0) {
 982       intel->coloroffset = offset / 4;
 983       EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
 984    }
 985
 986    if ((inputsRead & (FRAG_BIT_COL1 | FRAG_BIT_FOGC)) ||
 987        i915->vertex_fog != I915_FOG_NONE) {
 988
 989       if (inputsRead & FRAG_BIT_COL1) {
 990          intel->specoffset = offset / 4;
 991          EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, S4_VFMT_SPEC_FOG, 3);
 992       }
 993       else
 994          EMIT_PAD(3);
 995
 996       if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE)
 997          EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1UB_1F, S4_VFMT_SPEC_FOG, 1);
 998       else
 999          EMIT_PAD(1);
1000    }
1001
1002    /* XXX this was disabled, but enabling this code helped fix the Glean
1003     * tfragprog1 fog tests.
1004     */
1005 #if 1
1006    if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1007       EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1008    }
1009 #endif
1010
1011    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1012       if (inputsRead & FRAG_BIT_TEX(i)) {
1013          int sz = VB->TexCoordPtr[i]->size;
1014
1015          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1016          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1017
1018          EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1019       }
1020       else if (i == p->wpos_tex) {
1021
1022          /* If WPOS is required, duplicate the XYZ position data in an
1023           * unused texture coordinate:
1024           */
1025          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1026          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1027
1028          intel->wpos_offset = offset;
1029          intel->wpos_size = 3 * sizeof(GLuint);
1030
1031          EMIT_PAD(intel->wpos_size);
1032       }
1033    }
1034
1035    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1036        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1037       int k;
1038
1039       I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1040
1041       /* Must do this *after* statechange, so as not to affect
1042        * buffered vertices reliant on the old state:
1043        */
1044       intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1045                                               intel->vertex_attrs,
1046                                               intel->vertex_attr_count,
1047                                               intel->ViewportMatrix.m, 0);
1048
1049       intel->vertex_size >>= 2;
1050
1051       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1052       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1053
1054       k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1055       assert(k);
1056    }
1057
1058    if (!p->params_uptodate)
1059       track_params(p);
1060
1061    if (!p->on_hardware)
1062       i915_upload_program(i915, p);
1063 }
1064
1065 void
1066 i915InitFragProgFuncs(struct dd_function_table *functions)
1067 {
1068    functions->BindProgram = i915BindProgram;
1069    functions->NewProgram = i915NewProgram;
1070    functions->DeleteProgram = i915DeleteProgram;
1071    functions->IsProgramNative = i915IsProgramNative;
1072    functions->ProgramStringNotify = i915ProgramStringNotify;
1073 }