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 "shader/prog_instruction.h"
  33 #include "shader/prog_parameter.h"
  34 #include "shader/program.h"
  35 #include "shader/programopt.h"
  36
  37 #include "tnl/tnl.h"
  38 #include "tnl/t_context.h"
  39
  40 #include "intel_batchbuffer.h"
  41
  42 #include "i915_reg.h"
  43 #include "i915_context.h"
  44 #include "i915_program.h"
  45
  46
  47
  48 /* 1, -1/3!, 1/5!, -1/7! */
  49 static const GLfloat sin_constants[4] = { 1.0,
  50    -1.0 / (3 * 2 * 1),
  51    1.0 / (5 * 4 * 3 * 2 * 1),
  52    -1.0 / (7 * 6 * 5 * 4 * 3 * 2 * 1)
  53 };
  54
  55 /* 1, -1/2!, 1/4!, -1/6! */
  56 static const GLfloat cos_constants[4] = { 1.0,
  57    -1.0 / (2 * 1),
  58    1.0 / (4 * 3 * 2 * 1),
  59    -1.0 / (6 * 5 * 4 * 3 * 2 * 1)
  60 };
  61
  62 /**
  63  * Retrieve a ureg for the given source register.  Will emit
  64  * constants, apply swizzling and negation as needed.
  65  */
  66 static GLuint
  67 src_vector(struct i915_fragment_program *p,
  68            const struct prog_src_register *source,
  69            const struct gl_fragment_program *program)
  70 {
  71    GLuint src;
  72
  73    switch (source->File) {
  74
  75       /* Registers:
  76        */
  77    case PROGRAM_TEMPORARY:
  78       if (source->Index >= I915_MAX_TEMPORARY) {
  79          i915_program_error(p, "Exceeded max temporary reg");
  80          return 0;
  81       }
  82       src = UREG(REG_TYPE_R, source->Index);
  83       break;
  84    case PROGRAM_INPUT:
  85       switch (source->Index) {
  86       case FRAG_ATTRIB_WPOS:
  87          src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
  88          break;
  89       case FRAG_ATTRIB_COL0:
  90          src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
  91          break;
  92       case FRAG_ATTRIB_COL1:
  93          src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
  94          src = swizzle(src, X, Y, Z, ONE);
  95          break;
  96       case FRAG_ATTRIB_FOGC:
  97          src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
  98          src = swizzle(src, W, W, W, W);
  99          break;
 100       case FRAG_ATTRIB_TEX0:
 101       case FRAG_ATTRIB_TEX1:
 102       case FRAG_ATTRIB_TEX2:
 103       case FRAG_ATTRIB_TEX3:
 104       case FRAG_ATTRIB_TEX4:
 105       case FRAG_ATTRIB_TEX5:
 106       case FRAG_ATTRIB_TEX6:
 107       case FRAG_ATTRIB_TEX7:
 108          src = i915_emit_decl(p, REG_TYPE_T,
 109                               T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
 110                               D0_CHANNEL_ALL);
 111          break;
 112
 113       default:
 114          i915_program_error(p, "Bad source->Index");
 115          return 0;
 116       }
 117       break;
 118
 119       /* Various paramters and env values.  All emitted to
 120        * hardware as program constants.
 121        */
 122    case PROGRAM_LOCAL_PARAM:
 123       src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
 124       break;
 125
 126    case PROGRAM_ENV_PARAM:
 127       src =
 128          i915_emit_param4fv(p,
 129                             p->ctx->FragmentProgram.Parameters[source->
 130                                                                Index]);
 131       break;
 132
 133    case PROGRAM_CONSTANT:
 134    case PROGRAM_STATE_VAR:
 135    case PROGRAM_NAMED_PARAM:
 136       src =
 137          i915_emit_param4fv(p,
 138                             program->Base.Parameters->ParameterValues[source->
 139                                                                       Index]);
 140       break;
 141
 142    default:
 143       i915_program_error(p, "Bad source->File");
 144       return 0;
 145    }
 146
 147    src = swizzle(src,
 148                  GET_SWZ(source->Swizzle, 0),
 149                  GET_SWZ(source->Swizzle, 1),
 150                  GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
 151
 152    if (source->NegateBase)
 153       src = negate(src,
 154                    GET_BIT(source->NegateBase, 0),
 155                    GET_BIT(source->NegateBase, 1),
 156                    GET_BIT(source->NegateBase, 2),
 157                    GET_BIT(source->NegateBase, 3));
 158
 159    return src;
 160 }
 161
 162
 163 static GLuint
 164 get_result_vector(struct i915_fragment_program *p,
 165                   const struct prog_instruction *inst)
 166 {
 167    switch (inst->DstReg.File) {
 168    case PROGRAM_OUTPUT:
 169       switch (inst->DstReg.Index) {
 170       case FRAG_RESULT_COLR:
 171          return UREG(REG_TYPE_OC, 0);
 172       case FRAG_RESULT_DEPR:
 173          p->depth_written = 1;
 174          return UREG(REG_TYPE_OD, 0);
 175       default:
 176          i915_program_error(p, "Bad inst->DstReg.Index");
 177          return 0;
 178       }
 179    case PROGRAM_TEMPORARY:
 180       return UREG(REG_TYPE_R, inst->DstReg.Index);
 181    default:
 182       i915_program_error(p, "Bad inst->DstReg.File");
 183       return 0;
 184    }
 185 }
 186
 187 static GLuint
 188 get_result_flags(const struct prog_instruction *inst)
 189 {
 190    GLuint flags = 0;
 191
 192    if (inst->SaturateMode == SATURATE_ZERO_ONE)
 193       flags |= A0_DEST_SATURATE;
 194    if (inst->DstReg.WriteMask & WRITEMASK_X)
 195       flags |= A0_DEST_CHANNEL_X;
 196    if (inst->DstReg.WriteMask & WRITEMASK_Y)
 197       flags |= A0_DEST_CHANNEL_Y;
 198    if (inst->DstReg.WriteMask & WRITEMASK_Z)
 199       flags |= A0_DEST_CHANNEL_Z;
 200    if (inst->DstReg.WriteMask & WRITEMASK_W)
 201       flags |= A0_DEST_CHANNEL_W;
 202
 203    return flags;
 204 }
 205
 206 static GLuint
 207 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 208 {
 209    switch (bit) {
 210    case TEXTURE_1D_INDEX:
 211       return D0_SAMPLE_TYPE_2D;
 212    case TEXTURE_2D_INDEX:
 213       return D0_SAMPLE_TYPE_2D;
 214    case TEXTURE_RECT_INDEX:
 215       return D0_SAMPLE_TYPE_2D;
 216    case TEXTURE_3D_INDEX:
 217       return D0_SAMPLE_TYPE_VOLUME;
 218    case TEXTURE_CUBE_INDEX:
 219       return D0_SAMPLE_TYPE_CUBE;
 220    default:
 221       i915_program_error(p, "TexSrcBit");
 222       return 0;
 223    }
 224 }
 225
 226 #define EMIT_TEX( OP )                                          \
 227 do {                                                            \
 228    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 229    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 230                                   inst->TexSrcUnit, dim);       \
 231    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 232    /* Texel lookup */                                           \
 233                                                                 \
 234    i915_emit_texld( p,                                          \
 235                get_result_vector( p, inst ),                    \
 236                get_result_flags( inst ),                        \
 237                sampler,                                         \
 238                coord,                                           \
 239                OP);                                             \
 240 } while (0)
 241
 242 #define EMIT_ARITH( OP, N )                                             \
 243 do {                                                                    \
 244    i915_emit_arith( p,                                                  \
 245                OP,                                                      \
 246                get_result_vector( p, inst ),                            \
 247                get_result_flags( inst ), 0,                     \
 248                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 249                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 250                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 251 } while (0)
 252
 253 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 254 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 255 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 256
 257
 258 /* Possible concerns:
 259  *
 260  * SIN, COS -- could use another taylor step?
 261  * LIT      -- results seem a little different to sw mesa
 262  * LOG      -- different to mesa on negative numbers, but this is conformant.
 263  *
 264  * Parse failures -- Mesa doesn't currently give a good indication
 265  * internally whether a particular program string parsed or not.  This
 266  * can lead to confusion -- hopefully we cope with it ok now.
 267  *
 268  */
 269 static void
 270 upload_program(struct i915_fragment_program *p)
 271 {
 272    const struct gl_fragment_program *program =
 273       p->ctx->FragmentProgram._Current;
 274    const struct prog_instruction *inst = program->Base.Instructions;
 275
 276 /*    _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */
 277
 278    /* Is this a parse-failed program?  Ensure a valid program is
 279     * loaded, as the flagging of an error isn't sufficient to stop
 280     * this being uploaded to hardware.
 281     */
 282    if (inst[0].Opcode == OPCODE_END) {
 283       GLuint tmp = i915_get_utemp(p);
 284       i915_emit_arith(p,
 285                       A0_MOV,
 286                       UREG(REG_TYPE_OC, 0),
 287                       A0_DEST_CHANNEL_ALL, 0,
 288                       swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
 289       return;
 290    }
 291
 292    while (1) {
 293       GLuint src0, src1, src2, flags;
 294       GLuint tmp = 0;
 295
 296       switch (inst->Opcode) {
 297       case OPCODE_ABS:
 298          src0 = src_vector(p, &inst->SrcReg[0], program);
 299          i915_emit_arith(p,
 300                          A0_MAX,
 301                          get_result_vector(p, inst),
 302                          get_result_flags(inst), 0,
 303                          src0, negate(src0, 1, 1, 1, 1), 0);
 304          break;
 305
 306       case OPCODE_ADD:
 307          EMIT_2ARG_ARITH(A0_ADD);
 308          break;
 309
 310       case OPCODE_CMP:
 311          src0 = src_vector(p, &inst->SrcReg[0], program);
 312          src1 = src_vector(p, &inst->SrcReg[1], program);
 313          src2 = src_vector(p, &inst->SrcReg[2], program);
 314          i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
 315          break;
 316
 317       case OPCODE_COS:
 318          src0 = src_vector(p, &inst->SrcReg[0], program);
 319          tmp = i915_get_utemp(p);
 320
 321          i915_emit_arith(p,
 322                          A0_MUL,
 323                          tmp, A0_DEST_CHANNEL_X, 0,
 324                          src0, i915_emit_const1f(p, 1.0 / (M_PI)), 0);
 325
 326          i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 327
 328          /* By choosing different taylor constants, could get rid of this mul:
 329           */
 330          i915_emit_arith(p,
 331                          A0_MUL,
 332                          tmp, A0_DEST_CHANNEL_X, 0,
 333                          tmp, i915_emit_const1f(p, (M_PI)), 0);
 334
 335          /*
 336           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 337           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
 338           * t0 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 339           * result = DP4 t0, cos_constants
 340           */
 341          i915_emit_arith(p,
 342                          A0_MUL,
 343                          tmp, A0_DEST_CHANNEL_XY, 0,
 344                          swizzle(tmp, X, X, ONE, ONE),
 345                          swizzle(tmp, X, ONE, ONE, ONE), 0);
 346
 347          i915_emit_arith(p,
 348                          A0_MUL,
 349                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 350                          swizzle(tmp, X, Y, X, ONE),
 351                          swizzle(tmp, X, X, ONE, ONE), 0);
 352
 353          i915_emit_arith(p,
 354                          A0_MUL,
 355                          tmp, A0_DEST_CHANNEL_XYZ, 0,
 356                          swizzle(tmp, X, X, Z, ONE),
 357                          swizzle(tmp, Z, ONE, ONE, ONE), 0);
 358
 359          i915_emit_arith(p,
 360                          A0_DP4,
 361                          get_result_vector(p, inst),
 362                          get_result_flags(inst), 0,
 363                          swizzle(tmp, ONE, Z, Y, X),
 364                          i915_emit_const4fv(p, cos_constants), 0);
 365
 366          break;
 367
 368       case OPCODE_DP3:
 369          EMIT_2ARG_ARITH(A0_DP3);
 370          break;
 371
 372       case OPCODE_DP4:
 373          EMIT_2ARG_ARITH(A0_DP4);
 374          break;
 375
 376       case OPCODE_DPH:
 377          src0 = src_vector(p, &inst->SrcReg[0], program);
 378          src1 = src_vector(p, &inst->SrcReg[1], program);
 379
 380          i915_emit_arith(p,
 381                          A0_DP4,
 382                          get_result_vector(p, inst),
 383                          get_result_flags(inst), 0,
 384                          swizzle(src0, X, Y, Z, ONE), src1, 0);
 385          break;
 386
 387       case OPCODE_DST:
 388          src0 = src_vector(p, &inst->SrcReg[0], program);
 389          src1 = src_vector(p, &inst->SrcReg[1], program);
 390
 391          /* result[0] = 1    * 1;
 392           * result[1] = a[1] * b[1];
 393           * result[2] = a[2] * 1;
 394           * result[3] = 1    * b[3];
 395           */
 396          i915_emit_arith(p,
 397                          A0_MUL,
 398                          get_result_vector(p, inst),
 399                          get_result_flags(inst), 0,
 400                          swizzle(src0, ONE, Y, Z, ONE),
 401                          swizzle(src1, ONE, Y, ONE, W), 0);
 402          break;
 403
 404       case OPCODE_EX2:
 405          src0 = src_vector(p, &inst->SrcReg[0], program);
 406
 407          i915_emit_arith(p,
 408                          A0_EXP,
 409                          get_result_vector(p, inst),
 410                          get_result_flags(inst), 0,
 411                          swizzle(src0, X, X, X, X), 0, 0);
 412          break;
 413
 414       case OPCODE_FLR:
 415          EMIT_1ARG_ARITH(A0_FLR);
 416          break;
 417
 418       case OPCODE_FRC:
 419          EMIT_1ARG_ARITH(A0_FRC);
 420          break;
 421
 422       case OPCODE_KIL:
 423          src0 = src_vector(p, &inst->SrcReg[0], program);
 424          tmp = i915_get_utemp(p);
 425
 426          i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
 427                          0, src0, T0_TEXKILL);
 428          break;
 429
 430       case OPCODE_LG2:
 431          src0 = src_vector(p, &inst->SrcReg[0], program);
 432
 433          i915_emit_arith(p,
 434                          A0_LOG,
 435                          get_result_vector(p, inst),
 436                          get_result_flags(inst), 0,
 437                          swizzle(src0, X, X, X, X), 0, 0);
 438          break;
 439
 440       case OPCODE_LIT:
 441          src0 = src_vector(p, &inst->SrcReg[0], program);
 442          tmp = i915_get_utemp(p);
 443
 444          /* tmp = max( a.xyzw, a.00zw )
 445           * XXX: Clamp tmp.w to -128..128
 446           * tmp.y = log(tmp.y)
 447           * tmp.y = tmp.w * tmp.y
 448           * tmp.y = exp(tmp.y)
 449           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 450           */
 451          i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 452                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
 453
 454          i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 455                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 456
 457          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 458                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 459                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
 460
 461          i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 462                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 463
 464          i915_emit_arith(p, A0_CMP,
 465                          get_result_vector(p, inst),
 466                          get_result_flags(inst), 0,
 467                          negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
 468                          swizzle(tmp, ONE, X, ZERO, ONE),
 469                          swizzle(tmp, ONE, X, Y, ONE));
 470
 471          break;
 472
 473       case OPCODE_LRP:
 474          src0 = src_vector(p, &inst->SrcReg[0], program);
 475          src1 = src_vector(p, &inst->SrcReg[1], program);
 476          src2 = src_vector(p, &inst->SrcReg[2], program);
 477          flags = get_result_flags(inst);
 478          tmp = i915_get_utemp(p);
 479
 480          /* b*a + c*(1-a)
 481           *
 482           * b*a + c - ca
 483           *
 484           * tmp = b*a + c,
 485           * result = (-c)*a + tmp
 486           */
 487          i915_emit_arith(p, A0_MAD, tmp,
 488                          flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
 489
 490          i915_emit_arith(p, A0_MAD,
 491                          get_result_vector(p, inst),
 492                          flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
 493          break;
 494
 495       case OPCODE_MAD:
 496          EMIT_3ARG_ARITH(A0_MAD);
 497          break;
 498
 499       case OPCODE_MAX:
 500          EMIT_2ARG_ARITH(A0_MAX);
 501          break;
 502
 503       case OPCODE_MIN:
 504          src0 = src_vector(p, &inst->SrcReg[0], program);
 505          src1 = src_vector(p, &inst->SrcReg[1], program);
 506          tmp = i915_get_utemp(p);
 507          flags = get_result_flags(inst);
 508
 509          i915_emit_arith(p,
 510                          A0_MAX,
 511                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 512                          negate(src0, 1, 1, 1, 1),
 513                          negate(src1, 1, 1, 1, 1), 0);
 514
 515          i915_emit_arith(p,
 516                          A0_MOV,
 517                          get_result_vector(p, inst),
 518                          flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
 519          break;
 520
 521       case OPCODE_MOV:
 522          EMIT_1ARG_ARITH(A0_MOV);
 523          break;
 524
 525       case OPCODE_MUL:
 526          EMIT_2ARG_ARITH(A0_MUL);
 527          break;
 528
 529       case OPCODE_POW:
 530          src0 = src_vector(p, &inst->SrcReg[0], program);
 531          src1 = src_vector(p, &inst->SrcReg[1], program);
 532          tmp = i915_get_utemp(p);
 533          flags = get_result_flags(inst);
 534
 535          /* XXX: masking on intermediate values, here and elsewhere.
 536           */
 537          i915_emit_arith(p,
 538                          A0_LOG,
 539                          tmp, A0_DEST_CHANNEL_X, 0,
 540                          swizzle(src0, X, X, X, X), 0, 0);
 541
 542          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
 543
 544
 545          i915_emit_arith(p,
 546                          A0_EXP,
 547                          get_result_vector(p, inst),
 548                          flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
 549
 550          break;
 551
 552       case OPCODE_RCP:
 553          src0 = src_vector(p, &inst->SrcReg[0], program);
 554
 555          i915_emit_arith(p,
 556                          A0_RCP,
 557                          get_result_vector(p, inst),
 558                          get_result_flags(inst), 0,
 559                          swizzle(src0, X, X, X, X), 0, 0);
 560          break;
 561
 562       case OPCODE_RSQ:
 563
 564          src0 = src_vector(p, &inst->SrcReg[0], program);
 565
 566          i915_emit_arith(p,
 567                          A0_RSQ,
 568                          get_result_vector(p, inst),
 569                          get_result_flags(inst), 0,
 570                          swizzle(src0, X, X, X, X), 0, 0);
 571          break;
 572
 573       case OPCODE_SCS:
 574          src0 = src_vector(p, &inst->SrcReg[0], program);
 575          tmp = i915_get_utemp(p);
 576
 577          /*
 578           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 579           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 580           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 581           * scs.x = DP4 t1, sin_constants
 582           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 583           * scs.y = DP4 t1, cos_constants
 584           */
 585          i915_emit_arith(p,
 586                          A0_MUL,
 587                          tmp, A0_DEST_CHANNEL_XY, 0,
 588                          swizzle(src0, X, X, ONE, ONE),
 589                          swizzle(src0, X, ONE, ONE, ONE), 0);
 590
 591          i915_emit_arith(p,
 592                          A0_MUL,
 593                          tmp, A0_DEST_CHANNEL_ALL, 0,
 594                          swizzle(tmp, X, Y, X, Y),
 595                          swizzle(tmp, X, X, ONE, ONE), 0);
 596
 597          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 598             GLuint tmp1;
 599
 600             if (inst->DstReg.WriteMask & WRITEMASK_X)
 601                tmp1 = i915_get_utemp(p);
 602             else
 603                tmp1 = tmp;
 604
 605             i915_emit_arith(p,
 606                             A0_MUL,
 607                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 608                             swizzle(tmp, X, Y, Y, W),
 609                             swizzle(tmp, X, Z, ONE, ONE), 0);
 610
 611             i915_emit_arith(p,
 612                             A0_DP4,
 613                             get_result_vector(p, inst),
 614                             A0_DEST_CHANNEL_Y, 0,
 615                             swizzle(tmp1, W, Z, Y, X),
 616                             i915_emit_const4fv(p, sin_constants), 0);
 617          }
 618
 619          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 620             i915_emit_arith(p,
 621                             A0_MUL,
 622                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 623                             swizzle(tmp, X, X, Z, ONE),
 624                             swizzle(tmp, Z, ONE, ONE, ONE), 0);
 625
 626             i915_emit_arith(p,
 627                             A0_DP4,
 628                             get_result_vector(p, inst),
 629                             A0_DEST_CHANNEL_X, 0,
 630                             swizzle(tmp, ONE, Z, Y, X),
 631                             i915_emit_const4fv(p, cos_constants), 0);
 632          }
 633          break;
 634
 635       case OPCODE_SGE:
 636          EMIT_2ARG_ARITH(A0_SGE);
 637          break;
 638
 639       case OPCODE_SIN:
 640          src0 = src_vector(p, &inst->SrcReg[0], program);
 641          tmp = i915_get_utemp(p);
 642
 643          i915_emit_arith(p,
 644                          A0_MUL,
 645                          tmp, A0_DEST_CHANNEL_X, 0,
 646                          src0, i915_emit_const1f(p, 1.0 / (M_PI)), 0);
 647
 648          i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 649
 650          /* By choosing different taylor constants, could get rid of this mul:
 651           */
 652          i915_emit_arith(p,
 653                          A0_MUL,
 654                          tmp, A0_DEST_CHANNEL_X, 0,
 655                          tmp, i915_emit_const1f(p, (M_PI)), 0);
 656
 657          /*
 658           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 659           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 660           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 661           * result = DP4 t1.wzyx, sin_constants
 662           */
 663          i915_emit_arith(p,
 664                          A0_MUL,
 665                          tmp, A0_DEST_CHANNEL_XY, 0,
 666                          swizzle(tmp, X, X, ONE, ONE),
 667                          swizzle(tmp, X, ONE, ONE, ONE), 0);
 668
 669          i915_emit_arith(p,
 670                          A0_MUL,
 671                          tmp, A0_DEST_CHANNEL_ALL, 0,
 672                          swizzle(tmp, X, Y, X, Y),
 673                          swizzle(tmp, X, X, ONE, ONE), 0);
 674
 675          i915_emit_arith(p,
 676                          A0_MUL,
 677                          tmp, A0_DEST_CHANNEL_ALL, 0,
 678                          swizzle(tmp, X, Y, Y, W),
 679                          swizzle(tmp, X, Z, ONE, ONE), 0);
 680
 681          i915_emit_arith(p,
 682                          A0_DP4,
 683                          get_result_vector(p, inst),
 684                          get_result_flags(inst), 0,
 685                          swizzle(tmp, W, Z, Y, X),
 686                          i915_emit_const4fv(p, sin_constants), 0);
 687          break;
 688
 689       case OPCODE_SLT:
 690          EMIT_2ARG_ARITH(A0_SLT);
 691          break;
 692
 693       case OPCODE_SUB:
 694          src0 = src_vector(p, &inst->SrcReg[0], program);
 695          src1 = src_vector(p, &inst->SrcReg[1], program);
 696
 697          i915_emit_arith(p,
 698                          A0_ADD,
 699                          get_result_vector(p, inst),
 700                          get_result_flags(inst), 0,
 701                          src0, negate(src1, 1, 1, 1, 1), 0);
 702          break;
 703
 704       case OPCODE_SWZ:
 705          EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
 706          break;
 707
 708       case OPCODE_TEX:
 709          EMIT_TEX(T0_TEXLD);
 710          break;
 711
 712       case OPCODE_TXB:
 713          EMIT_TEX(T0_TEXLDB);
 714          break;
 715
 716       case OPCODE_TXP:
 717          EMIT_TEX(T0_TEXLDP);
 718          break;
 719
 720       case OPCODE_XPD:
 721          /* Cross product:
 722           *      result.x = src0.y * src1.z - src0.z * src1.y;
 723           *      result.y = src0.z * src1.x - src0.x * src1.z;
 724           *      result.z = src0.x * src1.y - src0.y * src1.x;
 725           *      result.w = undef;
 726           */
 727          src0 = src_vector(p, &inst->SrcReg[0], program);
 728          src1 = src_vector(p, &inst->SrcReg[1], program);
 729          tmp = i915_get_utemp(p);
 730
 731          i915_emit_arith(p,
 732                          A0_MUL,
 733                          tmp, A0_DEST_CHANNEL_ALL, 0,
 734                          swizzle(src0, Z, X, Y, ONE),
 735                          swizzle(src1, Y, Z, X, ONE), 0);
 736
 737          i915_emit_arith(p,
 738                          A0_MAD,
 739                          get_result_vector(p, inst),
 740                          get_result_flags(inst), 0,
 741                          swizzle(src0, Y, Z, X, ONE),
 742                          swizzle(src1, Z, X, Y, ONE),
 743                          negate(tmp, 1, 1, 1, 0));
 744          break;
 745
 746       case OPCODE_END:
 747          return;
 748
 749       default:
 750          i915_program_error(p, "bad opcode");
 751          return;
 752       }
 753
 754       inst++;
 755       i915_release_utemps(p);
 756    }
 757 }
 758
 759 /* Rather than trying to intercept and jiggle depth writes during
 760  * emit, just move the value into its correct position at the end of
 761  * the program:
 762  */
 763 static void
 764 fixup_depth_write(struct i915_fragment_program *p)
 765 {
 766    if (p->depth_written) {
 767       GLuint depth = UREG(REG_TYPE_OD, 0);
 768
 769       i915_emit_arith(p,
 770                       A0_MOV,
 771                       depth, A0_DEST_CHANNEL_W, 0,
 772                       swizzle(depth, X, Y, Z, Z), 0, 0);
 773    }
 774 }
 775
 776
 777 static void
 778 check_wpos(struct i915_fragment_program *p)
 779 {
 780    GLuint inputs = p->FragProg.Base.InputsRead;
 781    GLint i;
 782
 783    p->wpos_tex = -1;
 784
 785    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
 786       if (inputs & FRAG_BIT_TEX(i))
 787          continue;
 788       else if (inputs & FRAG_BIT_WPOS) {
 789          p->wpos_tex = i;
 790          inputs &= ~FRAG_BIT_WPOS;
 791       }
 792    }
 793
 794    if (inputs & FRAG_BIT_WPOS) {
 795       i915_program_error(p, "No free texcoord for wpos value");
 796    }
 797 }
 798
 799
 800 static void
 801 translate_program(struct i915_fragment_program *p)
 802 {
 803    struct i915_context *i915 = I915_CONTEXT(p->ctx);
 804
 805    i915_init_program(i915, p);
 806    check_wpos(p);
 807    upload_program(p);
 808    fixup_depth_write(p);
 809    i915_fini_program(p);
 810
 811    p->translated = 1;
 812 }
 813
 814
 815 static void
 816 track_params(struct i915_fragment_program *p)
 817 {
 818    GLint i;
 819
 820    if (p->nr_params)
 821       _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
 822
 823    for (i = 0; i < p->nr_params; i++) {
 824       GLint reg = p->param[i].reg;
 825       COPY_4V(p->constant[reg], p->param[i].values);
 826    }
 827
 828    p->params_uptodate = 1;
 829    p->on_hardware = 0;          /* overkill */
 830 }
 831
 832
 833 static void
 834 i915BindProgram(GLcontext * ctx, GLenum target, struct gl_program *prog)
 835 {
 836    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 837       struct i915_context *i915 = I915_CONTEXT(ctx);
 838       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 839
 840       if (i915->current_program == p)
 841          return;
 842
 843       if (i915->current_program) {
 844          i915->current_program->on_hardware = 0;
 845          i915->current_program->params_uptodate = 0;
 846       }
 847
 848       i915->current_program = p;
 849
 850       assert(p->on_hardware == 0);
 851       assert(p->params_uptodate == 0);
 852
 853    }
 854 }
 855
 856 static struct gl_program *
 857 i915NewProgram(GLcontext * ctx, GLenum target, GLuint id)
 858 {
 859    switch (target) {
 860    case GL_VERTEX_PROGRAM_ARB:
 861       return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
 862                                        target, id);
 863
 864    case GL_FRAGMENT_PROGRAM_ARB:{
 865          struct i915_fragment_program *prog =
 866             CALLOC_STRUCT(i915_fragment_program);
 867          if (prog) {
 868             i915_init_program(I915_CONTEXT(ctx), prog);
 869
 870             return _mesa_init_fragment_program(ctx, &prog->FragProg,
 871                                                target, id);
 872          }
 873          else
 874             return NULL;
 875       }
 876
 877    default:
 878       /* Just fallback:
 879        */
 880       return _mesa_new_program(ctx, target, id);
 881    }
 882 }
 883
 884 static void
 885 i915DeleteProgram(GLcontext * ctx, struct gl_program *prog)
 886 {
 887    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
 888       struct i915_context *i915 = I915_CONTEXT(ctx);
 889       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 890
 891       if (i915->current_program == p)
 892          i915->current_program = 0;
 893    }
 894
 895    _mesa_delete_program(ctx, prog);
 896 }
 897
 898
 899 static GLboolean
 900 i915IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
 901 {
 902    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 903       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 904
 905       if (!p->translated)
 906          translate_program(p);
 907
 908       return !p->error;
 909    }
 910    else
 911       return GL_TRUE;
 912 }
 913
 914 static void
 915 i915ProgramStringNotify(GLcontext * ctx,
 916                         GLenum target, struct gl_program *prog)
 917 {
 918    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 919       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 920       p->translated = 0;
 921
 922       /* Hack: make sure fog is correctly enabled according to this
 923        * fragment program's fog options.
 924        */
 925       if (p->FragProg.FogOption) {
 926          /* add extra instructions to do fog, then turn off FogOption field */
 927          _mesa_append_fog_code(ctx, &p->FragProg);
 928          p->FragProg.FogOption = GL_NONE;
 929       }
 930    }
 931
 932    _tnl_program_string(ctx, target, prog);
 933 }
 934
 935
 936 void
 937 i915ValidateFragmentProgram(struct i915_context *i915)
 938 {
 939    GLcontext *ctx = &i915->intel.ctx;
 940    struct intel_context *intel = intel_context(ctx);
 941    TNLcontext *tnl = TNL_CONTEXT(ctx);
 942    struct vertex_buffer *VB = &tnl->vb;
 943
 944    struct i915_fragment_program *p =
 945       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
 946
 947    const GLuint inputsRead = p->FragProg.Base.InputsRead;
 948    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
 949    GLuint s2 = S2_TEXCOORD_NONE;
 950    int i, offset = 0;
 951
 952    if (i915->current_program != p) {
 953       if (i915->current_program) {
 954          i915->current_program->on_hardware = 0;
 955          i915->current_program->params_uptodate = 0;
 956       }
 957
 958       i915->current_program = p;
 959    }
 960
 961
 962    /* Important:
 963     */
 964    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
 965
 966    if (!p->translated)
 967       translate_program(p);
 968
 969    intel->vertex_attr_count = 0;
 970    intel->wpos_offset = 0;
 971    intel->wpos_size = 0;
 972    intel->coloroffset = 0;
 973    intel->specoffset = 0;
 974
 975    if (inputsRead & FRAG_BITS_TEX_ANY) {
 976       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
 977    }
 978    else {
 979       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
 980    }
 981
 982    if (inputsRead & FRAG_BIT_COL0) {
 983       intel->coloroffset = offset / 4;
 984       EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
 985    }
 986
 987    if ((inputsRead & (FRAG_BIT_COL1 | FRAG_BIT_FOGC)) ||
 988        i915->vertex_fog != I915_FOG_NONE) {
 989
 990       if (inputsRead & FRAG_BIT_COL1) {
 991          intel->specoffset = offset / 4;
 992          EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, S4_VFMT_SPEC_FOG, 3);
 993       }
 994       else
 995          EMIT_PAD(3);
 996
 997       if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE)
 998          EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1UB_1F, S4_VFMT_SPEC_FOG, 1);
 999       else
1000          EMIT_PAD(1);
1001    }
1002
1003    /* XXX this was disabled, but enabling this code helped fix the Glean
1004     * tfragprog1 fog tests.
1005     */
1006 #if 1
1007    if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1008       EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1009    }
1010 #endif
1011
1012    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1013       if (inputsRead & FRAG_BIT_TEX(i)) {
1014          int sz = VB->TexCoordPtr[i]->size;
1015
1016          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1017          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1018
1019          EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1020       }
1021       else if (i == p->wpos_tex) {
1022
1023          /* If WPOS is required, duplicate the XYZ position data in an
1024           * unused texture coordinate:
1025           */
1026          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1027          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1028
1029          intel->wpos_offset = offset;
1030          intel->wpos_size = 3 * sizeof(GLuint);
1031
1032          EMIT_PAD(intel->wpos_size);
1033       }
1034    }
1035
1036    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1037        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1038       int k;
1039
1040       I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1041
1042       /* Must do this *after* statechange, so as not to affect
1043        * buffered vertices reliant on the old state:
1044        */
1045       intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1046                                               intel->vertex_attrs,
1047                                               intel->vertex_attr_count,
1048                                               intel->ViewportMatrix.m, 0);
1049
1050       intel->vertex_size >>= 2;
1051
1052       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1053       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1054
1055       k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1056       assert(k);
1057    }
1058
1059    if (!p->params_uptodate)
1060       track_params(p);
1061
1062    if (!p->on_hardware)
1063       i915_upload_program(i915, p);
1064 }
1065
1066 void
1067 i915InitFragProgFuncs(struct dd_function_table *functions)
1068 {
1069    functions->BindProgram = i915BindProgram;
1070    functions->NewProgram = i915NewProgram;
1071    functions->DeleteProgram = i915DeleteProgram;
1072    functions->IsProgramNative = i915IsProgramNative;
1073    functions->ProgramStringNotify = i915ProgramStringNotify;
1074 }