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 "main/glheader.h"
  29 #include "main/macros.h"
  30 #include "main/enums.h"
  31
  32 #include "program/prog_instruction.h"
  33 #include "program/prog_parameter.h"
  34 #include "program/program.h"
  35 #include "program/programopt.h"
  36 #include "program/prog_print.h"
  37
  38 #include "tnl/tnl.h"
  39 #include "tnl/t_context.h"
  40
  41 #include "intel_batchbuffer.h"
  42
  43 #include "i915_reg.h"
  44 #include "i915_context.h"
  45 #include "i915_program.h"
  46
  47 static const GLfloat sin_quad_constants[2][4] = {
  48    {
  49       2.0,
  50       -1.0,
  51       .5,
  52       .75
  53    },
  54    {
  55       4.0,
  56       -4.0,
  57       1.0 / (2.0 * M_PI),
  58       .2225
  59    }
  60 };
  61
  62 static const GLfloat sin_constants[4] = { 1.0,
  63    -1.0 / (3 * 2 * 1),
  64    1.0 / (5 * 4 * 3 * 2 * 1),
  65    -1.0 / (7 * 6 * 5 * 4 * 3 * 2 * 1)
  66 };
  67
  68 /* 1, -1/2!, 1/4!, -1/6! */
  69 static const GLfloat cos_constants[4] = { 1.0,
  70    -1.0 / (2 * 1),
  71    1.0 / (4 * 3 * 2 * 1),
  72    -1.0 / (6 * 5 * 4 * 3 * 2 * 1)
  73 };
  74
  75 /**
  76  * Retrieve a ureg for the given source register.  Will emit
  77  * constants, apply swizzling and negation as needed.
  78  */
  79 static GLuint
  80 src_vector(struct i915_fragment_program *p,
  81            const struct prog_src_register *source,
  82            const struct gl_fragment_program *program)
  83 {
  84    GLuint src;
  85
  86    switch (source->File) {
  87
  88       /* Registers:
  89        */
  90    case PROGRAM_TEMPORARY:
  91       if (source->Index >= I915_MAX_TEMPORARY) {
  92          i915_program_error(p, "Exceeded max temporary reg: %d/%d",
  93                             source->Index, I915_MAX_TEMPORARY);
  94          return 0;
  95       }
  96       src = UREG(REG_TYPE_R, source->Index);
  97       break;
  98    case PROGRAM_INPUT:
  99       switch (source->Index) {
 100       case FRAG_ATTRIB_WPOS:
 101          src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
 102          break;
 103       case FRAG_ATTRIB_COL0:
 104          src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
 105          break;
 106       case FRAG_ATTRIB_COL1:
 107          src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
 108          src = swizzle(src, X, Y, Z, ONE);
 109          break;
 110       case FRAG_ATTRIB_FOGC:
 111          src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
 112          src = swizzle(src, W, ZERO, ZERO, ONE);
 113          break;
 114       case FRAG_ATTRIB_TEX0:
 115       case FRAG_ATTRIB_TEX1:
 116       case FRAG_ATTRIB_TEX2:
 117       case FRAG_ATTRIB_TEX3:
 118       case FRAG_ATTRIB_TEX4:
 119       case FRAG_ATTRIB_TEX5:
 120       case FRAG_ATTRIB_TEX6:
 121       case FRAG_ATTRIB_TEX7:
 122          src = i915_emit_decl(p, REG_TYPE_T,
 123                               T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
 124                               D0_CHANNEL_ALL);
 125          break;
 126
 127       case FRAG_ATTRIB_VAR0:
 128       case FRAG_ATTRIB_VAR0 + 1:
 129       case FRAG_ATTRIB_VAR0 + 2:
 130       case FRAG_ATTRIB_VAR0 + 3:
 131       case FRAG_ATTRIB_VAR0 + 4:
 132       case FRAG_ATTRIB_VAR0 + 5:
 133       case FRAG_ATTRIB_VAR0 + 6:
 134       case FRAG_ATTRIB_VAR0 + 7:
 135          src = i915_emit_decl(p, REG_TYPE_T,
 136                               T_TEX0 + (source->Index - FRAG_ATTRIB_VAR0),
 137                               D0_CHANNEL_ALL);
 138          break;
 139
 140       default:
 141          i915_program_error(p, "Bad source->Index: %d", source->Index);
 142          return 0;
 143       }
 144       break;
 145
 146    case PROGRAM_OUTPUT:
 147       switch (source->Index) {
 148       case FRAG_RESULT_COLOR:
 149          src = UREG(REG_TYPE_OC, 0);
 150          break;
 151       case FRAG_RESULT_DEPTH:
 152          src = UREG(REG_TYPE_OD, 0);
 153          break;
 154       default:
 155          i915_program_error(p, "Bad source->Index: %d", source->Index);
 156          return 0;
 157       }
 158       break;
 159
 160       /* Various paramters and env values.  All emitted to
 161        * hardware as program constants.
 162        */
 163    case PROGRAM_LOCAL_PARAM:
 164       src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
 165       break;
 166
 167    case PROGRAM_ENV_PARAM:
 168       src =
 169          i915_emit_param4fv(p,
 170                             p->ctx->FragmentProgram.Parameters[source->
 171                                                                Index]);
 172       break;
 173
 174    case PROGRAM_CONSTANT:
 175    case PROGRAM_STATE_VAR:
 176    case PROGRAM_NAMED_PARAM:
 177    case PROGRAM_UNIFORM:
 178       src =
 179          i915_emit_param4fv(p,
 180                             program->Base.Parameters->ParameterValues[source->
 181                                                                       Index]);
 182       break;
 183
 184    default:
 185       i915_program_error(p, "Bad source->File: %d", source->File);
 186       return 0;
 187    }
 188
 189    src = swizzle(src,
 190                  GET_SWZ(source->Swizzle, 0),
 191                  GET_SWZ(source->Swizzle, 1),
 192                  GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
 193
 194    if (source->Negate)
 195       src = negate(src,
 196                    GET_BIT(source->Negate, 0),
 197                    GET_BIT(source->Negate, 1),
 198                    GET_BIT(source->Negate, 2),
 199                    GET_BIT(source->Negate, 3));
 200
 201    return src;
 202 }
 203
 204
 205 static GLuint
 206 get_result_vector(struct i915_fragment_program *p,
 207                   const struct prog_instruction *inst)
 208 {
 209    switch (inst->DstReg.File) {
 210    case PROGRAM_OUTPUT:
 211       switch (inst->DstReg.Index) {
 212       case FRAG_RESULT_COLOR:
 213       case FRAG_RESULT_DATA0:
 214          return UREG(REG_TYPE_OC, 0);
 215       case FRAG_RESULT_DEPTH:
 216          p->depth_written = 1;
 217          return UREG(REG_TYPE_OD, 0);
 218       default:
 219          i915_program_error(p, "Bad inst->DstReg.Index: %d",
 220                             inst->DstReg.Index);
 221          return 0;
 222       }
 223    case PROGRAM_TEMPORARY:
 224       return UREG(REG_TYPE_R, inst->DstReg.Index);
 225    default:
 226       i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
 227       return 0;
 228    }
 229 }
 230
 231 static GLuint
 232 get_result_flags(const struct prog_instruction *inst)
 233 {
 234    GLuint flags = 0;
 235
 236    if (inst->SaturateMode == SATURATE_ZERO_ONE)
 237       flags |= A0_DEST_SATURATE;
 238    if (inst->DstReg.WriteMask & WRITEMASK_X)
 239       flags |= A0_DEST_CHANNEL_X;
 240    if (inst->DstReg.WriteMask & WRITEMASK_Y)
 241       flags |= A0_DEST_CHANNEL_Y;
 242    if (inst->DstReg.WriteMask & WRITEMASK_Z)
 243       flags |= A0_DEST_CHANNEL_Z;
 244    if (inst->DstReg.WriteMask & WRITEMASK_W)
 245       flags |= A0_DEST_CHANNEL_W;
 246
 247    return flags;
 248 }
 249
 250 static GLuint
 251 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 252 {
 253    switch (bit) {
 254    case TEXTURE_1D_INDEX:
 255       return D0_SAMPLE_TYPE_2D;
 256    case TEXTURE_2D_INDEX:
 257       return D0_SAMPLE_TYPE_2D;
 258    case TEXTURE_RECT_INDEX:
 259       return D0_SAMPLE_TYPE_2D;
 260    case TEXTURE_3D_INDEX:
 261       return D0_SAMPLE_TYPE_VOLUME;
 262    case TEXTURE_CUBE_INDEX:
 263       return D0_SAMPLE_TYPE_CUBE;
 264    default:
 265       i915_program_error(p, "TexSrcBit: %d", bit);
 266       return 0;
 267    }
 268 }
 269
 270 #define EMIT_TEX( OP )                                          \
 271 do {                                                            \
 272    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 273    const struct gl_fragment_program *program = &p->FragProg;    \
 274    GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit];  \
 275    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 276                                    unit, dim);                  \
 277    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 278    /* Texel lookup */                                           \
 279                                                                 \
 280    i915_emit_texld( p, get_live_regs(p, inst),                                          \
 281                get_result_vector( p, inst ),                    \
 282                get_result_flags( inst ),                        \
 283                sampler,                                         \
 284                coord,                                           \
 285                OP);                                             \
 286 } while (0)
 287
 288 #define EMIT_ARITH( OP, N )                                             \
 289 do {                                                                    \
 290    i915_emit_arith( p,                                                  \
 291                OP,                                                      \
 292                get_result_vector( p, inst ),                            \
 293                get_result_flags( inst ), 0,                     \
 294                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 295                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 296                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 297 } while (0)
 298
 299 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 300 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 301 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 302
 303 /*
 304  * TODO: consider moving this into core
 305  */
 306 static void calc_live_regs( struct i915_fragment_program *p )
 307 {
 308     const struct gl_fragment_program *program = &p->FragProg;
 309     GLuint regsUsed = 0xffff0000;
 310     uint8_t live_components[16] = { 0, };
 311     GLint i;
 312
 313     for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
 314         struct prog_instruction *inst = &program->Base.Instructions[i];
 315         int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
 316         int a;
 317
 318         /* Register is written to: unmark as live for this and preceeding ops */
 319         if (inst->DstReg.File == PROGRAM_TEMPORARY) {
 320             live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
 321             if (live_components[inst->DstReg.Index] == 0)
 322                 regsUsed &= ~(1 << inst->DstReg.Index);
 323         }
 324
 325         for (a = 0; a < opArgs; a++) {
 326             /* Register is read from: mark as live for this and preceeding ops */
 327             if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
 328                 unsigned c;
 329
 330                 regsUsed |= 1 << inst->SrcReg[a].Index;
 331
 332                 for (c = 0; c < 4; c++) {
 333                     const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
 334
 335                     if (field <= SWIZZLE_W)
 336                         live_components[inst->SrcReg[a].Index] |= (1U << field);
 337                 }
 338             }
 339         }
 340
 341         p->usedRegs[i] = regsUsed;
 342     }
 343 }
 344
 345 static GLuint get_live_regs( struct i915_fragment_program *p,
 346                              const struct prog_instruction *inst )
 347 {
 348     const struct gl_fragment_program *program = &p->FragProg;
 349     GLuint nr = inst - program->Base.Instructions;
 350
 351     return p->usedRegs[nr];
 352 }
 353
 354
 355 /* Possible concerns:
 356  *
 357  * SIN, COS -- could use another taylor step?
 358  * LIT      -- results seem a little different to sw mesa
 359  * LOG      -- different to mesa on negative numbers, but this is conformant.
 360  *
 361  * Parse failures -- Mesa doesn't currently give a good indication
 362  * internally whether a particular program string parsed or not.  This
 363  * can lead to confusion -- hopefully we cope with it ok now.
 364  *
 365  */
 366 static void
 367 upload_program(struct i915_fragment_program *p)
 368 {
 369    const struct gl_fragment_program *program = &p->FragProg;
 370    const struct prog_instruction *inst = program->Base.Instructions;
 371
 372    if (INTEL_DEBUG & DEBUG_WM)
 373       _mesa_print_program(&program->Base);
 374
 375    /* Is this a parse-failed program?  Ensure a valid program is
 376     * loaded, as the flagging of an error isn't sufficient to stop
 377     * this being uploaded to hardware.
 378     */
 379    if (inst[0].Opcode == OPCODE_END) {
 380       GLuint tmp = i915_get_utemp(p);
 381       i915_emit_arith(p,
 382                       A0_MOV,
 383                       UREG(REG_TYPE_OC, 0),
 384                       A0_DEST_CHANNEL_ALL, 0,
 385                       swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
 386       return;
 387    }
 388
 389    if (program->Base.NumInstructions > I915_MAX_INSN) {
 390       i915_program_error(p, "Exceeded max instructions (%d out of %d)",
 391                          program->Base.NumInstructions, I915_MAX_INSN);
 392       return;
 393    }
 394
 395    /* Not always needed:
 396     */
 397    calc_live_regs(p);
 398
 399    while (1) {
 400       GLuint src0, src1, src2, flags;
 401       GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
 402
 403       switch (inst->Opcode) {
 404       case OPCODE_ABS:
 405          src0 = src_vector(p, &inst->SrcReg[0], program);
 406          i915_emit_arith(p,
 407                          A0_MAX,
 408                          get_result_vector(p, inst),
 409                          get_result_flags(inst), 0,
 410                          src0, negate(src0, 1, 1, 1, 1), 0);
 411          break;
 412
 413       case OPCODE_ADD:
 414          EMIT_2ARG_ARITH(A0_ADD);
 415          break;
 416
 417       case OPCODE_CMP:
 418          src0 = src_vector(p, &inst->SrcReg[0], program);
 419          src1 = src_vector(p, &inst->SrcReg[1], program);
 420          src2 = src_vector(p, &inst->SrcReg[2], program);
 421          i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
 422          break;
 423
 424       case OPCODE_COS:
 425          src0 = src_vector(p, &inst->SrcReg[0], program);
 426          tmp = i915_get_utemp(p);
 427          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 428          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 429
 430          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 431          i915_emit_arith(p,
 432                          A0_MAD,
 433                          tmp, A0_DEST_CHANNEL_X, 0,
 434                          src0,
 435                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 436                          swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
 437
 438          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 439
 440          i915_emit_arith(p,
 441                          A0_MAD,
 442                          tmp, A0_DEST_CHANNEL_X, 0,
 443                          tmp,
 444                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 445                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 446
 447          /* Compute COS with the same calculation used for SIN, but a
 448           * different source range has been mapped to [-1,1] this time.
 449           */
 450
 451          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 452          i915_emit_arith(p,
 453                          A0_MAX,
 454                          tmp, A0_DEST_CHANNEL_Y, 0,
 455                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 456                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 457                          0);
 458
 459          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 460          i915_emit_arith(p,
 461                          A0_MUL,
 462                          tmp, A0_DEST_CHANNEL_Y, 0,
 463                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 464                          tmp,
 465                          0);
 466
 467          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 468          i915_emit_arith(p,
 469                          A0_DP3,
 470                          tmp, A0_DEST_CHANNEL_X, 0,
 471                          tmp,
 472                          swizzle(consts1, X, Y, ZERO, ZERO),
 473                          0);
 474
 475          /* tmp.x now contains a first approximation (y).  Now, weight it
 476           * against tmp.y**2 to get closer.
 477           */
 478          i915_emit_arith(p,
 479                          A0_MAX,
 480                          tmp, A0_DEST_CHANNEL_Y, 0,
 481                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 482                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 483                          0);
 484
 485          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 486          i915_emit_arith(p,
 487                          A0_MAD,
 488                          tmp, A0_DEST_CHANNEL_Y, 0,
 489                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 490                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 491                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 492
 493          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 494          i915_emit_arith(p,
 495                          A0_MAD,
 496                          get_result_vector(p, inst),
 497                          get_result_flags(inst), 0,
 498                          swizzle(consts1, W, W, W, W),
 499                          swizzle(tmp, Y, Y, Y, Y),
 500                          swizzle(tmp, X, X, X, X));
 501          break;
 502
 503       case OPCODE_DP2:
 504          src0 = src_vector(p, &inst->SrcReg[0], program);
 505          src1 = src_vector(p, &inst->SrcReg[1], program);
 506          i915_emit_arith(p,
 507                          A0_DP3,
 508                          get_result_vector(p, inst),
 509                          get_result_flags(inst), 0,
 510                          swizzle(src0, X, Y, ZERO, ZERO),
 511                          swizzle(src1, X, Y, ZERO, ZERO),
 512                          0);
 513          break;
 514
 515       case OPCODE_DP3:
 516          EMIT_2ARG_ARITH(A0_DP3);
 517          break;
 518
 519       case OPCODE_DP4:
 520          EMIT_2ARG_ARITH(A0_DP4);
 521          break;
 522
 523       case OPCODE_DPH:
 524          src0 = src_vector(p, &inst->SrcReg[0], program);
 525          src1 = src_vector(p, &inst->SrcReg[1], program);
 526
 527          i915_emit_arith(p,
 528                          A0_DP4,
 529                          get_result_vector(p, inst),
 530                          get_result_flags(inst), 0,
 531                          swizzle(src0, X, Y, Z, ONE), src1, 0);
 532          break;
 533
 534       case OPCODE_DST:
 535          src0 = src_vector(p, &inst->SrcReg[0], program);
 536          src1 = src_vector(p, &inst->SrcReg[1], program);
 537
 538          /* result[0] = 1    * 1;
 539           * result[1] = a[1] * b[1];
 540           * result[2] = a[2] * 1;
 541           * result[3] = 1    * b[3];
 542           */
 543          i915_emit_arith(p,
 544                          A0_MUL,
 545                          get_result_vector(p, inst),
 546                          get_result_flags(inst), 0,
 547                          swizzle(src0, ONE, Y, Z, ONE),
 548                          swizzle(src1, ONE, Y, ONE, W), 0);
 549          break;
 550
 551       case OPCODE_EX2:
 552          src0 = src_vector(p, &inst->SrcReg[0], program);
 553
 554          i915_emit_arith(p,
 555                          A0_EXP,
 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_FLR:
 562          EMIT_1ARG_ARITH(A0_FLR);
 563          break;
 564
 565       case OPCODE_TRUNC:
 566          EMIT_1ARG_ARITH(A0_TRC);
 567          break;
 568
 569       case OPCODE_FRC:
 570          EMIT_1ARG_ARITH(A0_FRC);
 571          break;
 572
 573       case OPCODE_KIL:
 574          src0 = src_vector(p, &inst->SrcReg[0], program);
 575          tmp = i915_get_utemp(p);
 576
 577          i915_emit_texld(p, get_live_regs(p, inst),
 578                          tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
 579                          0, src0, T0_TEXKILL);
 580          break;
 581
 582       case OPCODE_KIL_NV:
 583          if (inst->DstReg.CondMask == COND_TR) {
 584             tmp = i915_get_utemp(p);
 585
 586             /* The KIL instruction discards the fragment if any component of
 587              * the source is < 0.  Emit an immediate operand of {-1}.xywz.
 588              */
 589             i915_emit_texld(p, get_live_regs(p, inst),
 590                             tmp, A0_DEST_CHANNEL_ALL,
 591                             0, /* use a dummy dest reg */
 592                             negate(swizzle(tmp, ONE, ONE, ONE, ONE),
 593                                    1, 1, 1, 1),
 594                             T0_TEXKILL);
 595          } else {
 596             p->error = 1;
 597             i915_program_error(p, "Unsupported KIL_NV condition code: %d",
 598                                inst->DstReg.CondMask);
 599          }
 600          break;
 601
 602       case OPCODE_LG2:
 603          src0 = src_vector(p, &inst->SrcReg[0], program);
 604
 605          i915_emit_arith(p,
 606                          A0_LOG,
 607                          get_result_vector(p, inst),
 608                          get_result_flags(inst), 0,
 609                          swizzle(src0, X, X, X, X), 0, 0);
 610          break;
 611
 612       case OPCODE_LIT:
 613          src0 = src_vector(p, &inst->SrcReg[0], program);
 614          tmp = i915_get_utemp(p);
 615
 616          /* tmp = max( a.xyzw, a.00zw )
 617           * XXX: Clamp tmp.w to -128..128
 618           * tmp.y = log(tmp.y)
 619           * tmp.y = tmp.w * tmp.y
 620           * tmp.y = exp(tmp.y)
 621           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 622           */
 623          i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 624                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
 625
 626          i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 627                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 628
 629          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 630                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 631                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
 632
 633          i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 634                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 635
 636          i915_emit_arith(p, A0_CMP,
 637                          get_result_vector(p, inst),
 638                          get_result_flags(inst), 0,
 639                          negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
 640                          swizzle(tmp, ONE, X, ZERO, ONE),
 641                          swizzle(tmp, ONE, X, Y, ONE));
 642
 643          break;
 644
 645       case OPCODE_LRP:
 646          src0 = src_vector(p, &inst->SrcReg[0], program);
 647          src1 = src_vector(p, &inst->SrcReg[1], program);
 648          src2 = src_vector(p, &inst->SrcReg[2], program);
 649          flags = get_result_flags(inst);
 650          tmp = i915_get_utemp(p);
 651
 652          /* b*a + c*(1-a)
 653           *
 654           * b*a + c - ca
 655           *
 656           * tmp = b*a + c,
 657           * result = (-c)*a + tmp
 658           */
 659          i915_emit_arith(p, A0_MAD, tmp,
 660                          flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
 661
 662          i915_emit_arith(p, A0_MAD,
 663                          get_result_vector(p, inst),
 664                          flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
 665          break;
 666
 667       case OPCODE_MAD:
 668          EMIT_3ARG_ARITH(A0_MAD);
 669          break;
 670
 671       case OPCODE_MAX:
 672          EMIT_2ARG_ARITH(A0_MAX);
 673          break;
 674
 675       case OPCODE_MIN:
 676          src0 = src_vector(p, &inst->SrcReg[0], program);
 677          src1 = src_vector(p, &inst->SrcReg[1], program);
 678          tmp = i915_get_utemp(p);
 679          flags = get_result_flags(inst);
 680
 681          i915_emit_arith(p,
 682                          A0_MAX,
 683                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 684                          negate(src0, 1, 1, 1, 1),
 685                          negate(src1, 1, 1, 1, 1), 0);
 686
 687          i915_emit_arith(p,
 688                          A0_MOV,
 689                          get_result_vector(p, inst),
 690                          flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
 691          break;
 692
 693       case OPCODE_MOV:
 694          EMIT_1ARG_ARITH(A0_MOV);
 695          break;
 696
 697       case OPCODE_MUL:
 698          EMIT_2ARG_ARITH(A0_MUL);
 699          break;
 700
 701       case OPCODE_POW:
 702          src0 = src_vector(p, &inst->SrcReg[0], program);
 703          src1 = src_vector(p, &inst->SrcReg[1], program);
 704          tmp = i915_get_utemp(p);
 705          flags = get_result_flags(inst);
 706
 707          /* XXX: masking on intermediate values, here and elsewhere.
 708           */
 709          i915_emit_arith(p,
 710                          A0_LOG,
 711                          tmp, A0_DEST_CHANNEL_X, 0,
 712                          swizzle(src0, X, X, X, X), 0, 0);
 713
 714          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
 715
 716
 717          i915_emit_arith(p,
 718                          A0_EXP,
 719                          get_result_vector(p, inst),
 720                          flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
 721
 722          break;
 723
 724       case OPCODE_RCP:
 725          src0 = src_vector(p, &inst->SrcReg[0], program);
 726
 727          i915_emit_arith(p,
 728                          A0_RCP,
 729                          get_result_vector(p, inst),
 730                          get_result_flags(inst), 0,
 731                          swizzle(src0, X, X, X, X), 0, 0);
 732          break;
 733
 734       case OPCODE_RSQ:
 735
 736          src0 = src_vector(p, &inst->SrcReg[0], program);
 737
 738          i915_emit_arith(p,
 739                          A0_RSQ,
 740                          get_result_vector(p, inst),
 741                          get_result_flags(inst), 0,
 742                          swizzle(src0, X, X, X, X), 0, 0);
 743          break;
 744
 745       case OPCODE_SCS:
 746          src0 = src_vector(p, &inst->SrcReg[0], program);
 747          tmp = i915_get_utemp(p);
 748
 749          /*
 750           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 751           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 752           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 753           * scs.x = DP4 t1, sin_constants
 754           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 755           * scs.y = DP4 t1, cos_constants
 756           */
 757          i915_emit_arith(p,
 758                          A0_MUL,
 759                          tmp, A0_DEST_CHANNEL_XY, 0,
 760                          swizzle(src0, X, X, ONE, ONE),
 761                          swizzle(src0, X, ONE, ONE, ONE), 0);
 762
 763          i915_emit_arith(p,
 764                          A0_MUL,
 765                          tmp, A0_DEST_CHANNEL_ALL, 0,
 766                          swizzle(tmp, X, Y, X, Y),
 767                          swizzle(tmp, X, X, ONE, ONE), 0);
 768
 769          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 770             GLuint tmp1;
 771
 772             if (inst->DstReg.WriteMask & WRITEMASK_X)
 773                tmp1 = i915_get_utemp(p);
 774             else
 775                tmp1 = tmp;
 776
 777             i915_emit_arith(p,
 778                             A0_MUL,
 779                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 780                             swizzle(tmp, X, Y, Y, W),
 781                             swizzle(tmp, X, Z, ONE, ONE), 0);
 782
 783             i915_emit_arith(p,
 784                             A0_DP4,
 785                             get_result_vector(p, inst),
 786                             A0_DEST_CHANNEL_Y, 0,
 787                             swizzle(tmp1, W, Z, Y, X),
 788                             i915_emit_const4fv(p, sin_constants), 0);
 789          }
 790
 791          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 792             i915_emit_arith(p,
 793                             A0_MUL,
 794                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 795                             swizzle(tmp, X, X, Z, ONE),
 796                             swizzle(tmp, Z, ONE, ONE, ONE), 0);
 797
 798             i915_emit_arith(p,
 799                             A0_DP4,
 800                             get_result_vector(p, inst),
 801                             A0_DEST_CHANNEL_X, 0,
 802                             swizzle(tmp, ONE, Z, Y, X),
 803                             i915_emit_const4fv(p, cos_constants), 0);
 804          }
 805          break;
 806
 807       case OPCODE_SEQ:
 808          tmp = i915_get_utemp(p);
 809          flags = get_result_flags(inst);
 810          dst = get_result_vector(p, inst);
 811
 812          /* tmp = src1 >= src2 */
 813          i915_emit_arith(p,
 814                          A0_SGE,
 815                          tmp,
 816                          flags, 0,
 817                          src_vector(p, &inst->SrcReg[0], program),
 818                          src_vector(p, &inst->SrcReg[1], program),
 819                          0);
 820          /* dst = src1 <= src2 */
 821          i915_emit_arith(p,
 822                          A0_SGE,
 823                          dst,
 824                          flags, 0,
 825                          negate(src_vector(p, &inst->SrcReg[0], program),
 826                                 1, 1, 1, 1),
 827                          negate(src_vector(p, &inst->SrcReg[1], program),
 828                                 1, 1, 1, 1),
 829                          0);
 830          /* dst = tmp && dst */
 831          i915_emit_arith(p,
 832                          A0_MUL,
 833                          dst,
 834                          flags, 0,
 835                          dst,
 836                          tmp,
 837                          0);
 838          break;
 839
 840       case OPCODE_SIN:
 841          src0 = src_vector(p, &inst->SrcReg[0], program);
 842          tmp = i915_get_utemp(p);
 843          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 844          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 845
 846          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 847          i915_emit_arith(p,
 848                          A0_MAD,
 849                          tmp, A0_DEST_CHANNEL_X, 0,
 850                          src0,
 851                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 852                          swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
 853
 854          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 855
 856          i915_emit_arith(p,
 857                          A0_MAD,
 858                          tmp, A0_DEST_CHANNEL_X, 0,
 859                          tmp,
 860                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 861                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 862
 863          /* Compute sin using a quadratic and quartic.  It gives continuity
 864           * that repeating the Taylor series lacks every 2*pi, and has
 865           * reduced error.
 866           *
 867           * The idea was described at:
 868           * http://www.devmaster.net/forums/showthread.php?t=5784
 869           */
 870
 871          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 872          i915_emit_arith(p,
 873                          A0_MAX,
 874                          tmp, A0_DEST_CHANNEL_Y, 0,
 875                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 876                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 877                          0);
 878
 879          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 880          i915_emit_arith(p,
 881                          A0_MUL,
 882                          tmp, A0_DEST_CHANNEL_Y, 0,
 883                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 884                          tmp,
 885                          0);
 886
 887          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 888          i915_emit_arith(p,
 889                          A0_DP3,
 890                          tmp, A0_DEST_CHANNEL_X, 0,
 891                          tmp,
 892                          swizzle(consts1, X, Y, ZERO, ZERO),
 893                          0);
 894
 895          /* tmp.x now contains a first approximation (y).  Now, weight it
 896           * against tmp.y**2 to get closer.
 897           */
 898          i915_emit_arith(p,
 899                          A0_MAX,
 900                          tmp, A0_DEST_CHANNEL_Y, 0,
 901                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 902                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 903                          0);
 904
 905          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 906          i915_emit_arith(p,
 907                          A0_MAD,
 908                          tmp, A0_DEST_CHANNEL_Y, 0,
 909                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 910                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 911                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 912
 913          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 914          i915_emit_arith(p,
 915                          A0_MAD,
 916                          get_result_vector(p, inst),
 917                          get_result_flags(inst), 0,
 918                          swizzle(consts1, W, W, W, W),
 919                          swizzle(tmp, Y, Y, Y, Y),
 920                          swizzle(tmp, X, X, X, X));
 921
 922          break;
 923
 924       case OPCODE_SGE:
 925          EMIT_2ARG_ARITH(A0_SGE);
 926          break;
 927
 928       case OPCODE_SGT:
 929          i915_emit_arith(p,
 930                          A0_SLT,
 931                          get_result_vector( p, inst ),
 932                          get_result_flags( inst ), 0,
 933                          negate(src_vector( p, &inst->SrcReg[0], program),
 934                                 1, 1, 1, 1),
 935                          negate(src_vector( p, &inst->SrcReg[1], program),
 936                                 1, 1, 1, 1),
 937                          0);
 938          break;
 939
 940       case OPCODE_SLE:
 941          i915_emit_arith(p,
 942                          A0_SGE,
 943                          get_result_vector( p, inst ),
 944                          get_result_flags( inst ), 0,
 945                          negate(src_vector( p, &inst->SrcReg[0], program),
 946                                 1, 1, 1, 1),
 947                          negate(src_vector( p, &inst->SrcReg[1], program),
 948                                 1, 1, 1, 1),
 949                          0);
 950          break;
 951
 952       case OPCODE_SLT:
 953          EMIT_2ARG_ARITH(A0_SLT);
 954          break;
 955
 956       case OPCODE_SNE:
 957          tmp = i915_get_utemp(p);
 958          flags = get_result_flags(inst);
 959          dst = get_result_vector(p, inst);
 960
 961          /* tmp = src1 < src2 */
 962          i915_emit_arith(p,
 963                          A0_SLT,
 964                          tmp,
 965                          flags, 0,
 966                          src_vector(p, &inst->SrcReg[0], program),
 967                          src_vector(p, &inst->SrcReg[1], program),
 968                          0);
 969          /* dst = src1 > src2 */
 970          i915_emit_arith(p,
 971                          A0_SLT,
 972                          dst,
 973                          flags, 0,
 974                          negate(src_vector(p, &inst->SrcReg[0], program),
 975                                 1, 1, 1, 1),
 976                          negate(src_vector(p, &inst->SrcReg[1], program),
 977                                 1, 1, 1, 1),
 978                          0);
 979          /* dst = tmp || dst */
 980          i915_emit_arith(p,
 981                          A0_ADD,
 982                          dst,
 983                          flags | A0_DEST_SATURATE, 0,
 984                          dst,
 985                          tmp,
 986                          0);
 987          break;
 988
 989       case OPCODE_SSG:
 990          dst = get_result_vector(p, inst);
 991          flags = get_result_flags(inst);
 992          src0 = src_vector(p, &inst->SrcReg[0], program);
 993          tmp = i915_get_utemp(p);
 994
 995          /* tmp = (src < 0.0) */
 996          i915_emit_arith(p,
 997                          A0_SLT,
 998                          tmp,
 999                          flags, 0,
1000                          src0,
1001                          swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1002                          0);
1003
1004          /* dst = (0.0 < src) */
1005          i915_emit_arith(p,
1006                          A0_SLT,
1007                          dst,
1008                          flags, 0,
1009                          swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1010                          src0,
1011                          0);
1012
1013          /* dst = (src > 0.0) - (src < 0.0) */
1014          i915_emit_arith(p,
1015                          A0_ADD,
1016                          dst,
1017                          flags, 0,
1018                          dst,
1019                          negate(tmp, 1, 1, 1, 1),
1020                          0);
1021
1022          break;
1023
1024       case OPCODE_SUB:
1025          src0 = src_vector(p, &inst->SrcReg[0], program);
1026          src1 = src_vector(p, &inst->SrcReg[1], program);
1027
1028          i915_emit_arith(p,
1029                          A0_ADD,
1030                          get_result_vector(p, inst),
1031                          get_result_flags(inst), 0,
1032                          src0, negate(src1, 1, 1, 1, 1), 0);
1033          break;
1034
1035       case OPCODE_SWZ:
1036          EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
1037          break;
1038
1039       case OPCODE_TEX:
1040          EMIT_TEX(T0_TEXLD);
1041          break;
1042
1043       case OPCODE_TXB:
1044          EMIT_TEX(T0_TEXLDB);
1045          break;
1046
1047       case OPCODE_TXP:
1048          EMIT_TEX(T0_TEXLDP);
1049          break;
1050
1051       case OPCODE_XPD:
1052          /* Cross product:
1053           *      result.x = src0.y * src1.z - src0.z * src1.y;
1054           *      result.y = src0.z * src1.x - src0.x * src1.z;
1055           *      result.z = src0.x * src1.y - src0.y * src1.x;
1056           *      result.w = undef;
1057           */
1058          src0 = src_vector(p, &inst->SrcReg[0], program);
1059          src1 = src_vector(p, &inst->SrcReg[1], program);
1060          tmp = i915_get_utemp(p);
1061
1062          i915_emit_arith(p,
1063                          A0_MUL,
1064                          tmp, A0_DEST_CHANNEL_ALL, 0,
1065                          swizzle(src0, Z, X, Y, ONE),
1066                          swizzle(src1, Y, Z, X, ONE), 0);
1067
1068          i915_emit_arith(p,
1069                          A0_MAD,
1070                          get_result_vector(p, inst),
1071                          get_result_flags(inst), 0,
1072                          swizzle(src0, Y, Z, X, ONE),
1073                          swizzle(src1, Z, X, Y, ONE),
1074                          negate(tmp, 1, 1, 1, 0));
1075          break;
1076
1077       case OPCODE_END:
1078          return;
1079
1080       case OPCODE_BGNLOOP:
1081       case OPCODE_BGNSUB:
1082       case OPCODE_BRA:
1083       case OPCODE_BRK:
1084       case OPCODE_CAL:
1085       case OPCODE_CONT:
1086       case OPCODE_DDX:
1087       case OPCODE_DDY:
1088       case OPCODE_ELSE:
1089       case OPCODE_ENDIF:
1090       case OPCODE_ENDLOOP:
1091       case OPCODE_ENDSUB:
1092       case OPCODE_IF:
1093       case OPCODE_RET:
1094          p->error = 1;
1095          i915_program_error(p, "Unsupported opcode: %s",
1096                             _mesa_opcode_string(inst->Opcode));
1097          return;
1098
1099       case OPCODE_EXP:
1100       case OPCODE_LOG:
1101          /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1102           * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1103           * Instead, it translates to EX2 or LG2.
1104           */
1105       case OPCODE_TXD:
1106       case OPCODE_TXL:
1107          /* These opcodes are claimed by GLSL in prog_instruction.h, but
1108           * only NV_vp/fp appears to emit them.
1109           */
1110       default:
1111          i915_program_error(p, "bad opcode: %s",
1112                             _mesa_opcode_string(inst->Opcode));
1113          return;
1114       }
1115
1116       inst++;
1117       i915_release_utemps(p);
1118    }
1119 }
1120
1121 /* Rather than trying to intercept and jiggle depth writes during
1122  * emit, just move the value into its correct position at the end of
1123  * the program:
1124  */
1125 static void
1126 fixup_depth_write(struct i915_fragment_program *p)
1127 {
1128    if (p->depth_written) {
1129       GLuint depth = UREG(REG_TYPE_OD, 0);
1130
1131       i915_emit_arith(p,
1132                       A0_MOV,
1133                       depth, A0_DEST_CHANNEL_W, 0,
1134                       swizzle(depth, X, Y, Z, Z), 0, 0);
1135    }
1136 }
1137
1138
1139 static void
1140 check_wpos(struct i915_fragment_program *p)
1141 {
1142    GLuint inputs = p->FragProg.Base.InputsRead;
1143    GLint i;
1144
1145    p->wpos_tex = -1;
1146
1147    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1148       if (inputs & (FRAG_BIT_TEX(i) | FRAG_BIT_VAR(i)))
1149          continue;
1150       else if (inputs & FRAG_BIT_WPOS) {
1151          p->wpos_tex = i;
1152          inputs &= ~FRAG_BIT_WPOS;
1153       }
1154    }
1155
1156    if (inputs & FRAG_BIT_WPOS) {
1157       i915_program_error(p, "No free texcoord for wpos value");
1158    }
1159 }
1160
1161
1162 static void
1163 translate_program(struct i915_fragment_program *p)
1164 {
1165    struct i915_context *i915 = I915_CONTEXT(p->ctx);
1166
1167    if (INTEL_DEBUG & DEBUG_WM) {
1168       printf("fp:\n");
1169       _mesa_print_program(&p->FragProg.Base);
1170       printf("\n");
1171    }
1172
1173    i915_init_program(i915, p);
1174    check_wpos(p);
1175    upload_program(p);
1176    fixup_depth_write(p);
1177    i915_fini_program(p);
1178
1179    p->translated = 1;
1180 }
1181
1182
1183 static void
1184 track_params(struct i915_fragment_program *p)
1185 {
1186    GLint i;
1187
1188    if (p->nr_params)
1189       _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1190
1191    for (i = 0; i < p->nr_params; i++) {
1192       GLint reg = p->param[i].reg;
1193       COPY_4V(p->constant[reg], p->param[i].values);
1194    }
1195
1196    p->params_uptodate = 1;
1197    p->on_hardware = 0;          /* overkill */
1198 }
1199
1200
1201 static void
1202 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1203 {
1204    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1205       struct i915_context *i915 = I915_CONTEXT(ctx);
1206       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1207
1208       if (i915->current_program == p)
1209          return;
1210
1211       if (i915->current_program) {
1212          i915->current_program->on_hardware = 0;
1213          i915->current_program->params_uptodate = 0;
1214       }
1215
1216       i915->current_program = p;
1217
1218       assert(p->on_hardware == 0);
1219       assert(p->params_uptodate == 0);
1220
1221    }
1222 }
1223
1224 static struct gl_program *
1225 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1226 {
1227    switch (target) {
1228    case GL_VERTEX_PROGRAM_ARB:
1229       return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1230                                        target, id);
1231
1232    case GL_FRAGMENT_PROGRAM_ARB:{
1233          struct i915_fragment_program *prog =
1234             CALLOC_STRUCT(i915_fragment_program);
1235          if (prog) {
1236             i915_init_program(I915_CONTEXT(ctx), prog);
1237
1238             return _mesa_init_fragment_program(ctx, &prog->FragProg,
1239                                                target, id);
1240          }
1241          else
1242             return NULL;
1243       }
1244
1245    default:
1246       /* Just fallback:
1247        */
1248       return _mesa_new_program(ctx, target, id);
1249    }
1250 }
1251
1252 static void
1253 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1254 {
1255    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1256       struct i915_context *i915 = I915_CONTEXT(ctx);
1257       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1258
1259       if (i915->current_program == p)
1260          i915->current_program = 0;
1261    }
1262
1263    _mesa_delete_program(ctx, prog);
1264 }
1265
1266
1267 static GLboolean
1268 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1269 {
1270    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1271       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1272
1273       if (!p->translated)
1274          translate_program(p);
1275
1276       return !p->error;
1277    }
1278    else
1279       return GL_TRUE;
1280 }
1281
1282 static GLboolean
1283 i915ProgramStringNotify(struct gl_context * ctx,
1284                         GLenum target, struct gl_program *prog)
1285 {
1286    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1287       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1288       p->translated = 0;
1289    }
1290
1291    (void) _tnl_program_string(ctx, target, prog);
1292
1293    /* XXX check if program is legal, within limits */
1294    return GL_TRUE;
1295 }
1296
1297 void
1298 i915_update_program(struct gl_context *ctx)
1299 {
1300    struct intel_context *intel = intel_context(ctx);
1301    struct i915_context *i915 = i915_context(&intel->ctx);
1302    struct i915_fragment_program *fp =
1303       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1304
1305    if (i915->current_program != fp) {
1306       if (i915->current_program) {
1307          i915->current_program->on_hardware = 0;
1308          i915->current_program->params_uptodate = 0;
1309       }
1310
1311       i915->current_program = fp;
1312    }
1313
1314    if (!fp->translated)
1315       translate_program(fp);
1316
1317    FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1318 }
1319
1320 void
1321 i915ValidateFragmentProgram(struct i915_context *i915)
1322 {
1323    struct gl_context *ctx = &i915->intel.ctx;
1324    struct intel_context *intel = intel_context(ctx);
1325    TNLcontext *tnl = TNL_CONTEXT(ctx);
1326    struct vertex_buffer *VB = &tnl->vb;
1327
1328    struct i915_fragment_program *p =
1329       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1330
1331    const GLuint inputsRead = p->FragProg.Base.InputsRead;
1332    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1333    GLuint s2 = S2_TEXCOORD_NONE;
1334    int i, offset = 0;
1335
1336    /* Important:
1337     */
1338    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1339
1340    if (!p->translated)
1341       translate_program(p);
1342
1343    intel->vertex_attr_count = 0;
1344    intel->wpos_offset = 0;
1345    intel->wpos_size = 0;
1346    intel->coloroffset = 0;
1347    intel->specoffset = 0;
1348
1349    if (inputsRead & FRAG_BITS_TEX_ANY) {
1350       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1351    }
1352    else {
1353       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1354    }
1355
1356    if (inputsRead & FRAG_BIT_COL0) {
1357       intel->coloroffset = offset / 4;
1358       EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1359    }
1360
1361    if (inputsRead & FRAG_BIT_COL1) {
1362        intel->specoffset = offset / 4;
1363        EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1364    }
1365
1366    if ((inputsRead & FRAG_BIT_FOGC)) {
1367       EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1368    }
1369
1370    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1371       if (inputsRead & FRAG_BIT_TEX(i)) {
1372          int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1373
1374          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1375          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1376
1377          EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1378       }
1379       else if (inputsRead & FRAG_BIT_VAR(i)) {
1380          int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1381
1382          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1383          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1384
1385          EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1386       }
1387       else if (i == p->wpos_tex) {
1388
1389          /* If WPOS is required, duplicate the XYZ position data in an
1390           * unused texture coordinate:
1391           */
1392          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1393          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1394
1395          intel->wpos_offset = offset;
1396          intel->wpos_size = 3 * sizeof(GLuint);
1397
1398          EMIT_PAD(intel->wpos_size);
1399       }
1400    }
1401
1402    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1403        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1404       int k;
1405
1406       I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1407
1408       /* Must do this *after* statechange, so as not to affect
1409        * buffered vertices reliant on the old state:
1410        */
1411       intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1412                                               intel->vertex_attrs,
1413                                               intel->vertex_attr_count,
1414                                               intel->ViewportMatrix.m, 0);
1415
1416       assert(intel->prim.current_offset == intel->prim.start_offset);
1417       intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1418       intel->prim.current_offset = intel->prim.start_offset;
1419
1420       intel->vertex_size >>= 2;
1421
1422       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1423       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1424
1425       k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1426       assert(k);
1427    }
1428
1429    if (!p->params_uptodate)
1430       track_params(p);
1431
1432    if (!p->on_hardware)
1433       i915_upload_program(i915, p);
1434
1435    if (INTEL_DEBUG & DEBUG_WM) {
1436       printf("i915:\n");
1437       i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1438    }
1439 }
1440
1441 void
1442 i915InitFragProgFuncs(struct dd_function_table *functions)
1443 {
1444    functions->BindProgram = i915BindProgram;
1445    functions->NewProgram = i915NewProgram;
1446    functions->DeleteProgram = i915DeleteProgram;
1447    functions->IsProgramNative = i915IsProgramNative;
1448    functions->ProgramStringNotify = i915ProgramStringNotify;
1449 }