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          return UREG(REG_TYPE_OC, 0);
 214       case FRAG_RESULT_DEPTH:
 215          p->depth_written = 1;
 216          return UREG(REG_TYPE_OD, 0);
 217       default:
 218          i915_program_error(p, "Bad inst->DstReg.Index: %d",
 219                             inst->DstReg.Index);
 220          return 0;
 221       }
 222    case PROGRAM_TEMPORARY:
 223       return UREG(REG_TYPE_R, inst->DstReg.Index);
 224    default:
 225       i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
 226       return 0;
 227    }
 228 }
 229
 230 static GLuint
 231 get_result_flags(const struct prog_instruction *inst)
 232 {
 233    GLuint flags = 0;
 234
 235    if (inst->SaturateMode == SATURATE_ZERO_ONE)
 236       flags |= A0_DEST_SATURATE;
 237    if (inst->DstReg.WriteMask & WRITEMASK_X)
 238       flags |= A0_DEST_CHANNEL_X;
 239    if (inst->DstReg.WriteMask & WRITEMASK_Y)
 240       flags |= A0_DEST_CHANNEL_Y;
 241    if (inst->DstReg.WriteMask & WRITEMASK_Z)
 242       flags |= A0_DEST_CHANNEL_Z;
 243    if (inst->DstReg.WriteMask & WRITEMASK_W)
 244       flags |= A0_DEST_CHANNEL_W;
 245
 246    return flags;
 247 }
 248
 249 static GLuint
 250 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 251 {
 252    switch (bit) {
 253    case TEXTURE_1D_INDEX:
 254       return D0_SAMPLE_TYPE_2D;
 255    case TEXTURE_2D_INDEX:
 256       return D0_SAMPLE_TYPE_2D;
 257    case TEXTURE_RECT_INDEX:
 258       return D0_SAMPLE_TYPE_2D;
 259    case TEXTURE_3D_INDEX:
 260       return D0_SAMPLE_TYPE_VOLUME;
 261    case TEXTURE_CUBE_INDEX:
 262       return D0_SAMPLE_TYPE_CUBE;
 263    default:
 264       i915_program_error(p, "TexSrcBit: %d", bit);
 265       return 0;
 266    }
 267 }
 268
 269 #define EMIT_TEX( OP )                                          \
 270 do {                                                            \
 271    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 272    const struct gl_fragment_program *program = &p->FragProg;    \
 273    GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit];  \
 274    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 275                                    unit, dim);                  \
 276    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 277    /* Texel lookup */                                           \
 278                                                                 \
 279    i915_emit_texld( p, get_live_regs(p, inst),                                          \
 280                get_result_vector( p, inst ),                    \
 281                get_result_flags( inst ),                        \
 282                sampler,                                         \
 283                coord,                                           \
 284                OP);                                             \
 285 } while (0)
 286
 287 #define EMIT_ARITH( OP, N )                                             \
 288 do {                                                                    \
 289    i915_emit_arith( p,                                                  \
 290                OP,                                                      \
 291                get_result_vector( p, inst ),                            \
 292                get_result_flags( inst ), 0,                     \
 293                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 294                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 295                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 296 } while (0)
 297
 298 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 299 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 300 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 301
 302 /*
 303  * TODO: consider moving this into core
 304  */
 305 static void calc_live_regs( struct i915_fragment_program *p )
 306 {
 307     const struct gl_fragment_program *program = &p->FragProg;
 308     GLuint regsUsed = 0xffff0000;
 309     uint8_t live_components[16] = { 0, };
 310     GLint i;
 311
 312     for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
 313         struct prog_instruction *inst = &program->Base.Instructions[i];
 314         int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
 315         int a;
 316
 317         /* Register is written to: unmark as live for this and preceeding ops */
 318         if (inst->DstReg.File == PROGRAM_TEMPORARY) {
 319             live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
 320             if (live_components[inst->DstReg.Index] == 0)
 321                 regsUsed &= ~(1 << inst->DstReg.Index);
 322         }
 323
 324         for (a = 0; a < opArgs; a++) {
 325             /* Register is read from: mark as live for this and preceeding ops */
 326             if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
 327                 unsigned c;
 328
 329                 regsUsed |= 1 << inst->SrcReg[a].Index;
 330
 331                 for (c = 0; c < 4; c++) {
 332                     const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
 333
 334                     if (field <= SWIZZLE_W)
 335                         live_components[inst->SrcReg[a].Index] |= (1U << field);
 336                 }
 337             }
 338         }
 339
 340         p->usedRegs[i] = regsUsed;
 341     }
 342 }
 343
 344 static GLuint get_live_regs( struct i915_fragment_program *p,
 345                              const struct prog_instruction *inst )
 346 {
 347     const struct gl_fragment_program *program = &p->FragProg;
 348     GLuint nr = inst - program->Base.Instructions;
 349
 350     return p->usedRegs[nr];
 351 }
 352
 353
 354 /* Possible concerns:
 355  *
 356  * SIN, COS -- could use another taylor step?
 357  * LIT      -- results seem a little different to sw mesa
 358  * LOG      -- different to mesa on negative numbers, but this is conformant.
 359  *
 360  * Parse failures -- Mesa doesn't currently give a good indication
 361  * internally whether a particular program string parsed or not.  This
 362  * can lead to confusion -- hopefully we cope with it ok now.
 363  *
 364  */
 365 static void
 366 upload_program(struct i915_fragment_program *p)
 367 {
 368    const struct gl_fragment_program *program = &p->FragProg;
 369    const struct prog_instruction *inst = program->Base.Instructions;
 370
 371    if (INTEL_DEBUG & DEBUG_WM)
 372       _mesa_print_program(&program->Base);
 373
 374    /* Is this a parse-failed program?  Ensure a valid program is
 375     * loaded, as the flagging of an error isn't sufficient to stop
 376     * this being uploaded to hardware.
 377     */
 378    if (inst[0].Opcode == OPCODE_END) {
 379       GLuint tmp = i915_get_utemp(p);
 380       i915_emit_arith(p,
 381                       A0_MOV,
 382                       UREG(REG_TYPE_OC, 0),
 383                       A0_DEST_CHANNEL_ALL, 0,
 384                       swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
 385       return;
 386    }
 387
 388    if (program->Base.NumInstructions > I915_MAX_INSN) {
 389       i915_program_error(p, "Exceeded max instructions (%d out of %d)",
 390                          program->Base.NumInstructions, I915_MAX_INSN);
 391       return;
 392    }
 393
 394    /* Not always needed:
 395     */
 396    calc_live_regs(p);
 397
 398    while (1) {
 399       GLuint src0, src1, src2, flags;
 400       GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
 401
 402       switch (inst->Opcode) {
 403       case OPCODE_ABS:
 404          src0 = src_vector(p, &inst->SrcReg[0], program);
 405          i915_emit_arith(p,
 406                          A0_MAX,
 407                          get_result_vector(p, inst),
 408                          get_result_flags(inst), 0,
 409                          src0, negate(src0, 1, 1, 1, 1), 0);
 410          break;
 411
 412       case OPCODE_ADD:
 413          EMIT_2ARG_ARITH(A0_ADD);
 414          break;
 415
 416       case OPCODE_CMP:
 417          src0 = src_vector(p, &inst->SrcReg[0], program);
 418          src1 = src_vector(p, &inst->SrcReg[1], program);
 419          src2 = src_vector(p, &inst->SrcReg[2], program);
 420          i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
 421          break;
 422
 423       case OPCODE_COS:
 424          src0 = src_vector(p, &inst->SrcReg[0], program);
 425          tmp = i915_get_utemp(p);
 426          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 427          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 428
 429          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 430          i915_emit_arith(p,
 431                          A0_MAD,
 432                          tmp, A0_DEST_CHANNEL_X, 0,
 433                          src0,
 434                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 435                          swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
 436
 437          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 438
 439          i915_emit_arith(p,
 440                          A0_MAD,
 441                          tmp, A0_DEST_CHANNEL_X, 0,
 442                          tmp,
 443                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 444                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 445
 446          /* Compute COS with the same calculation used for SIN, but a
 447           * different source range has been mapped to [-1,1] this time.
 448           */
 449
 450          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 451          i915_emit_arith(p,
 452                          A0_MAX,
 453                          tmp, A0_DEST_CHANNEL_Y, 0,
 454                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 455                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 456                          0);
 457
 458          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 459          i915_emit_arith(p,
 460                          A0_MUL,
 461                          tmp, A0_DEST_CHANNEL_Y, 0,
 462                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 463                          tmp,
 464                          0);
 465
 466          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 467          i915_emit_arith(p,
 468                          A0_DP3,
 469                          tmp, A0_DEST_CHANNEL_X, 0,
 470                          tmp,
 471                          swizzle(consts1, X, Y, ZERO, ZERO),
 472                          0);
 473
 474          /* tmp.x now contains a first approximation (y).  Now, weight it
 475           * against tmp.y**2 to get closer.
 476           */
 477          i915_emit_arith(p,
 478                          A0_MAX,
 479                          tmp, A0_DEST_CHANNEL_Y, 0,
 480                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 481                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 482                          0);
 483
 484          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 485          i915_emit_arith(p,
 486                          A0_MAD,
 487                          tmp, A0_DEST_CHANNEL_Y, 0,
 488                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 489                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 490                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 491
 492          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 493          i915_emit_arith(p,
 494                          A0_MAD,
 495                          get_result_vector(p, inst),
 496                          get_result_flags(inst), 0,
 497                          swizzle(consts1, W, W, W, W),
 498                          swizzle(tmp, Y, Y, Y, Y),
 499                          swizzle(tmp, X, X, X, X));
 500          break;
 501
 502       case OPCODE_DP2:
 503          src0 = src_vector(p, &inst->SrcReg[0], program);
 504          src1 = src_vector(p, &inst->SrcReg[1], program);
 505          i915_emit_arith(p,
 506                          A0_DP3,
 507                          get_result_vector(p, inst),
 508                          get_result_flags(inst), 0,
 509                          swizzle(src0, X, Y, ZERO, ZERO),
 510                          swizzle(src1, X, Y, ZERO, ZERO),
 511                          0);
 512          break;
 513
 514       case OPCODE_DP3:
 515          EMIT_2ARG_ARITH(A0_DP3);
 516          break;
 517
 518       case OPCODE_DP4:
 519          EMIT_2ARG_ARITH(A0_DP4);
 520          break;
 521
 522       case OPCODE_DPH:
 523          src0 = src_vector(p, &inst->SrcReg[0], program);
 524          src1 = src_vector(p, &inst->SrcReg[1], program);
 525
 526          i915_emit_arith(p,
 527                          A0_DP4,
 528                          get_result_vector(p, inst),
 529                          get_result_flags(inst), 0,
 530                          swizzle(src0, X, Y, Z, ONE), src1, 0);
 531          break;
 532
 533       case OPCODE_DST:
 534          src0 = src_vector(p, &inst->SrcReg[0], program);
 535          src1 = src_vector(p, &inst->SrcReg[1], program);
 536
 537          /* result[0] = 1    * 1;
 538           * result[1] = a[1] * b[1];
 539           * result[2] = a[2] * 1;
 540           * result[3] = 1    * b[3];
 541           */
 542          i915_emit_arith(p,
 543                          A0_MUL,
 544                          get_result_vector(p, inst),
 545                          get_result_flags(inst), 0,
 546                          swizzle(src0, ONE, Y, Z, ONE),
 547                          swizzle(src1, ONE, Y, ONE, W), 0);
 548          break;
 549
 550       case OPCODE_EX2:
 551          src0 = src_vector(p, &inst->SrcReg[0], program);
 552
 553          i915_emit_arith(p,
 554                          A0_EXP,
 555                          get_result_vector(p, inst),
 556                          get_result_flags(inst), 0,
 557                          swizzle(src0, X, X, X, X), 0, 0);
 558          break;
 559
 560       case OPCODE_FLR:
 561          EMIT_1ARG_ARITH(A0_FLR);
 562          break;
 563
 564       case OPCODE_TRUNC:
 565          EMIT_1ARG_ARITH(A0_TRC);
 566          break;
 567
 568       case OPCODE_FRC:
 569          EMIT_1ARG_ARITH(A0_FRC);
 570          break;
 571
 572       case OPCODE_KIL:
 573          src0 = src_vector(p, &inst->SrcReg[0], program);
 574          tmp = i915_get_utemp(p);
 575
 576          i915_emit_texld(p, get_live_regs(p, inst),
 577                          tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
 578                          0, src0, T0_TEXKILL);
 579          break;
 580
 581       case OPCODE_KIL_NV:
 582          if (inst->DstReg.CondMask == COND_TR) {
 583             tmp = i915_get_utemp(p);
 584
 585             /* The KIL instruction discards the fragment if any component of
 586              * the source is < 0.  Emit an immediate operand of {-1}.xywz.
 587              */
 588             i915_emit_texld(p, get_live_regs(p, inst),
 589                             tmp, A0_DEST_CHANNEL_ALL,
 590                             0, /* use a dummy dest reg */
 591                             negate(swizzle(tmp, ONE, ONE, ONE, ONE),
 592                                    1, 1, 1, 1),
 593                             T0_TEXKILL);
 594          } else {
 595             p->error = 1;
 596             i915_program_error(p, "Unsupported KIL_NV condition code: %d",
 597                                inst->DstReg.CondMask);
 598          }
 599          break;
 600
 601       case OPCODE_LG2:
 602          src0 = src_vector(p, &inst->SrcReg[0], program);
 603
 604          i915_emit_arith(p,
 605                          A0_LOG,
 606                          get_result_vector(p, inst),
 607                          get_result_flags(inst), 0,
 608                          swizzle(src0, X, X, X, X), 0, 0);
 609          break;
 610
 611       case OPCODE_LIT:
 612          src0 = src_vector(p, &inst->SrcReg[0], program);
 613          tmp = i915_get_utemp(p);
 614
 615          /* tmp = max( a.xyzw, a.00zw )
 616           * XXX: Clamp tmp.w to -128..128
 617           * tmp.y = log(tmp.y)
 618           * tmp.y = tmp.w * tmp.y
 619           * tmp.y = exp(tmp.y)
 620           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 621           */
 622          i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 623                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
 624
 625          i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 626                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 627
 628          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 629                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 630                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
 631
 632          i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 633                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 634
 635          i915_emit_arith(p, A0_CMP,
 636                          get_result_vector(p, inst),
 637                          get_result_flags(inst), 0,
 638                          negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
 639                          swizzle(tmp, ONE, X, ZERO, ONE),
 640                          swizzle(tmp, ONE, X, Y, ONE));
 641
 642          break;
 643
 644       case OPCODE_LRP:
 645          src0 = src_vector(p, &inst->SrcReg[0], program);
 646          src1 = src_vector(p, &inst->SrcReg[1], program);
 647          src2 = src_vector(p, &inst->SrcReg[2], program);
 648          flags = get_result_flags(inst);
 649          tmp = i915_get_utemp(p);
 650
 651          /* b*a + c*(1-a)
 652           *
 653           * b*a + c - ca
 654           *
 655           * tmp = b*a + c,
 656           * result = (-c)*a + tmp
 657           */
 658          i915_emit_arith(p, A0_MAD, tmp,
 659                          flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
 660
 661          i915_emit_arith(p, A0_MAD,
 662                          get_result_vector(p, inst),
 663                          flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
 664          break;
 665
 666       case OPCODE_MAD:
 667          EMIT_3ARG_ARITH(A0_MAD);
 668          break;
 669
 670       case OPCODE_MAX:
 671          EMIT_2ARG_ARITH(A0_MAX);
 672          break;
 673
 674       case OPCODE_MIN:
 675          src0 = src_vector(p, &inst->SrcReg[0], program);
 676          src1 = src_vector(p, &inst->SrcReg[1], program);
 677          tmp = i915_get_utemp(p);
 678          flags = get_result_flags(inst);
 679
 680          i915_emit_arith(p,
 681                          A0_MAX,
 682                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 683                          negate(src0, 1, 1, 1, 1),
 684                          negate(src1, 1, 1, 1, 1), 0);
 685
 686          i915_emit_arith(p,
 687                          A0_MOV,
 688                          get_result_vector(p, inst),
 689                          flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
 690          break;
 691
 692       case OPCODE_MOV:
 693          EMIT_1ARG_ARITH(A0_MOV);
 694          break;
 695
 696       case OPCODE_MUL:
 697          EMIT_2ARG_ARITH(A0_MUL);
 698          break;
 699
 700       case OPCODE_POW:
 701          src0 = src_vector(p, &inst->SrcReg[0], program);
 702          src1 = src_vector(p, &inst->SrcReg[1], program);
 703          tmp = i915_get_utemp(p);
 704          flags = get_result_flags(inst);
 705
 706          /* XXX: masking on intermediate values, here and elsewhere.
 707           */
 708          i915_emit_arith(p,
 709                          A0_LOG,
 710                          tmp, A0_DEST_CHANNEL_X, 0,
 711                          swizzle(src0, X, X, X, X), 0, 0);
 712
 713          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
 714
 715
 716          i915_emit_arith(p,
 717                          A0_EXP,
 718                          get_result_vector(p, inst),
 719                          flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
 720
 721          break;
 722
 723       case OPCODE_RCP:
 724          src0 = src_vector(p, &inst->SrcReg[0], program);
 725
 726          i915_emit_arith(p,
 727                          A0_RCP,
 728                          get_result_vector(p, inst),
 729                          get_result_flags(inst), 0,
 730                          swizzle(src0, X, X, X, X), 0, 0);
 731          break;
 732
 733       case OPCODE_RSQ:
 734
 735          src0 = src_vector(p, &inst->SrcReg[0], program);
 736
 737          i915_emit_arith(p,
 738                          A0_RSQ,
 739                          get_result_vector(p, inst),
 740                          get_result_flags(inst), 0,
 741                          swizzle(src0, X, X, X, X), 0, 0);
 742          break;
 743
 744       case OPCODE_SCS:
 745          src0 = src_vector(p, &inst->SrcReg[0], program);
 746          tmp = i915_get_utemp(p);
 747
 748          /*
 749           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 750           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 751           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 752           * scs.x = DP4 t1, sin_constants
 753           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 754           * scs.y = DP4 t1, cos_constants
 755           */
 756          i915_emit_arith(p,
 757                          A0_MUL,
 758                          tmp, A0_DEST_CHANNEL_XY, 0,
 759                          swizzle(src0, X, X, ONE, ONE),
 760                          swizzle(src0, X, ONE, ONE, ONE), 0);
 761
 762          i915_emit_arith(p,
 763                          A0_MUL,
 764                          tmp, A0_DEST_CHANNEL_ALL, 0,
 765                          swizzle(tmp, X, Y, X, Y),
 766                          swizzle(tmp, X, X, ONE, ONE), 0);
 767
 768          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 769             GLuint tmp1;
 770
 771             if (inst->DstReg.WriteMask & WRITEMASK_X)
 772                tmp1 = i915_get_utemp(p);
 773             else
 774                tmp1 = tmp;
 775
 776             i915_emit_arith(p,
 777                             A0_MUL,
 778                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 779                             swizzle(tmp, X, Y, Y, W),
 780                             swizzle(tmp, X, Z, ONE, ONE), 0);
 781
 782             i915_emit_arith(p,
 783                             A0_DP4,
 784                             get_result_vector(p, inst),
 785                             A0_DEST_CHANNEL_Y, 0,
 786                             swizzle(tmp1, W, Z, Y, X),
 787                             i915_emit_const4fv(p, sin_constants), 0);
 788          }
 789
 790          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 791             i915_emit_arith(p,
 792                             A0_MUL,
 793                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 794                             swizzle(tmp, X, X, Z, ONE),
 795                             swizzle(tmp, Z, ONE, ONE, ONE), 0);
 796
 797             i915_emit_arith(p,
 798                             A0_DP4,
 799                             get_result_vector(p, inst),
 800                             A0_DEST_CHANNEL_X, 0,
 801                             swizzle(tmp, ONE, Z, Y, X),
 802                             i915_emit_const4fv(p, cos_constants), 0);
 803          }
 804          break;
 805
 806       case OPCODE_SEQ:
 807          tmp = i915_get_utemp(p);
 808          flags = get_result_flags(inst);
 809          dst = get_result_vector(p, inst);
 810
 811          /* tmp = src1 >= src2 */
 812          i915_emit_arith(p,
 813                          A0_SGE,
 814                          tmp,
 815                          flags, 0,
 816                          src_vector(p, &inst->SrcReg[0], program),
 817                          src_vector(p, &inst->SrcReg[1], program),
 818                          0);
 819          /* dst = src1 <= src2 */
 820          i915_emit_arith(p,
 821                          A0_SGE,
 822                          dst,
 823                          flags, 0,
 824                          negate(src_vector(p, &inst->SrcReg[0], program),
 825                                 1, 1, 1, 1),
 826                          negate(src_vector(p, &inst->SrcReg[1], program),
 827                                 1, 1, 1, 1),
 828                          0);
 829          /* dst = tmp && dst */
 830          i915_emit_arith(p,
 831                          A0_MUL,
 832                          dst,
 833                          flags, 0,
 834                          dst,
 835                          tmp,
 836                          0);
 837          break;
 838
 839       case OPCODE_SIN:
 840          src0 = src_vector(p, &inst->SrcReg[0], program);
 841          tmp = i915_get_utemp(p);
 842          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 843          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 844
 845          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 846          i915_emit_arith(p,
 847                          A0_MAD,
 848                          tmp, A0_DEST_CHANNEL_X, 0,
 849                          src0,
 850                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 851                          swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
 852
 853          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 854
 855          i915_emit_arith(p,
 856                          A0_MAD,
 857                          tmp, A0_DEST_CHANNEL_X, 0,
 858                          tmp,
 859                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 860                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 861
 862          /* Compute sin using a quadratic and quartic.  It gives continuity
 863           * that repeating the Taylor series lacks every 2*pi, and has
 864           * reduced error.
 865           *
 866           * The idea was described at:
 867           * http://www.devmaster.net/forums/showthread.php?t=5784
 868           */
 869
 870          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 871          i915_emit_arith(p,
 872                          A0_MAX,
 873                          tmp, A0_DEST_CHANNEL_Y, 0,
 874                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 875                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 876                          0);
 877
 878          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 879          i915_emit_arith(p,
 880                          A0_MUL,
 881                          tmp, A0_DEST_CHANNEL_Y, 0,
 882                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 883                          tmp,
 884                          0);
 885
 886          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 887          i915_emit_arith(p,
 888                          A0_DP3,
 889                          tmp, A0_DEST_CHANNEL_X, 0,
 890                          tmp,
 891                          swizzle(consts1, X, Y, ZERO, ZERO),
 892                          0);
 893
 894          /* tmp.x now contains a first approximation (y).  Now, weight it
 895           * against tmp.y**2 to get closer.
 896           */
 897          i915_emit_arith(p,
 898                          A0_MAX,
 899                          tmp, A0_DEST_CHANNEL_Y, 0,
 900                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 901                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 902                          0);
 903
 904          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 905          i915_emit_arith(p,
 906                          A0_MAD,
 907                          tmp, A0_DEST_CHANNEL_Y, 0,
 908                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 909                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 910                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 911
 912          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 913          i915_emit_arith(p,
 914                          A0_MAD,
 915                          get_result_vector(p, inst),
 916                          get_result_flags(inst), 0,
 917                          swizzle(consts1, W, W, W, W),
 918                          swizzle(tmp, Y, Y, Y, Y),
 919                          swizzle(tmp, X, X, X, X));
 920
 921          break;
 922
 923       case OPCODE_SGE:
 924          EMIT_2ARG_ARITH(A0_SGE);
 925          break;
 926
 927       case OPCODE_SGT:
 928          i915_emit_arith(p,
 929                          A0_SLT,
 930                          get_result_vector( p, inst ),
 931                          get_result_flags( inst ), 0,
 932                          negate(src_vector( p, &inst->SrcReg[0], program),
 933                                 1, 1, 1, 1),
 934                          negate(src_vector( p, &inst->SrcReg[1], program),
 935                                 1, 1, 1, 1),
 936                          0);
 937          break;
 938
 939       case OPCODE_SLE:
 940          i915_emit_arith(p,
 941                          A0_SGE,
 942                          get_result_vector( p, inst ),
 943                          get_result_flags( inst ), 0,
 944                          negate(src_vector( p, &inst->SrcReg[0], program),
 945                                 1, 1, 1, 1),
 946                          negate(src_vector( p, &inst->SrcReg[1], program),
 947                                 1, 1, 1, 1),
 948                          0);
 949          break;
 950
 951       case OPCODE_SLT:
 952          EMIT_2ARG_ARITH(A0_SLT);
 953          break;
 954
 955       case OPCODE_SNE:
 956          tmp = i915_get_utemp(p);
 957          flags = get_result_flags(inst);
 958          dst = get_result_vector(p, inst);
 959
 960          /* tmp = src1 < src2 */
 961          i915_emit_arith(p,
 962                          A0_SLT,
 963                          tmp,
 964                          flags, 0,
 965                          src_vector(p, &inst->SrcReg[0], program),
 966                          src_vector(p, &inst->SrcReg[1], program),
 967                          0);
 968          /* dst = src1 > src2 */
 969          i915_emit_arith(p,
 970                          A0_SLT,
 971                          dst,
 972                          flags, 0,
 973                          negate(src_vector(p, &inst->SrcReg[0], program),
 974                                 1, 1, 1, 1),
 975                          negate(src_vector(p, &inst->SrcReg[1], program),
 976                                 1, 1, 1, 1),
 977                          0);
 978          /* dst = tmp || dst */
 979          i915_emit_arith(p,
 980                          A0_ADD,
 981                          dst,
 982                          flags | A0_DEST_SATURATE, 0,
 983                          dst,
 984                          tmp,
 985                          0);
 986          break;
 987
 988       case OPCODE_SSG:
 989          dst = get_result_vector(p, inst);
 990          flags = get_result_flags(inst);
 991          src0 = src_vector(p, &inst->SrcReg[0], program);
 992          tmp = i915_get_utemp(p);
 993
 994          /* tmp = (src < 0.0) */
 995          i915_emit_arith(p,
 996                          A0_SLT,
 997                          tmp,
 998                          flags, 0,
 999                          src0,
1000                          swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1001                          0);
1002
1003          /* dst = (0.0 < src) */
1004          i915_emit_arith(p,
1005                          A0_SLT,
1006                          dst,
1007                          flags, 0,
1008                          swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1009                          src0,
1010                          0);
1011
1012          /* dst = (src > 0.0) - (src < 0.0) */
1013          i915_emit_arith(p,
1014                          A0_ADD,
1015                          dst,
1016                          flags, 0,
1017                          dst,
1018                          negate(tmp, 1, 1, 1, 1),
1019                          0);
1020
1021          break;
1022
1023       case OPCODE_SUB:
1024          src0 = src_vector(p, &inst->SrcReg[0], program);
1025          src1 = src_vector(p, &inst->SrcReg[1], program);
1026
1027          i915_emit_arith(p,
1028                          A0_ADD,
1029                          get_result_vector(p, inst),
1030                          get_result_flags(inst), 0,
1031                          src0, negate(src1, 1, 1, 1, 1), 0);
1032          break;
1033
1034       case OPCODE_SWZ:
1035          EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
1036          break;
1037
1038       case OPCODE_TEX:
1039          EMIT_TEX(T0_TEXLD);
1040          break;
1041
1042       case OPCODE_TXB:
1043          EMIT_TEX(T0_TEXLDB);
1044          break;
1045
1046       case OPCODE_TXP:
1047          EMIT_TEX(T0_TEXLDP);
1048          break;
1049
1050       case OPCODE_XPD:
1051          /* Cross product:
1052           *      result.x = src0.y * src1.z - src0.z * src1.y;
1053           *      result.y = src0.z * src1.x - src0.x * src1.z;
1054           *      result.z = src0.x * src1.y - src0.y * src1.x;
1055           *      result.w = undef;
1056           */
1057          src0 = src_vector(p, &inst->SrcReg[0], program);
1058          src1 = src_vector(p, &inst->SrcReg[1], program);
1059          tmp = i915_get_utemp(p);
1060
1061          i915_emit_arith(p,
1062                          A0_MUL,
1063                          tmp, A0_DEST_CHANNEL_ALL, 0,
1064                          swizzle(src0, Z, X, Y, ONE),
1065                          swizzle(src1, Y, Z, X, ONE), 0);
1066
1067          i915_emit_arith(p,
1068                          A0_MAD,
1069                          get_result_vector(p, inst),
1070                          get_result_flags(inst), 0,
1071                          swizzle(src0, Y, Z, X, ONE),
1072                          swizzle(src1, Z, X, Y, ONE),
1073                          negate(tmp, 1, 1, 1, 0));
1074          break;
1075
1076       case OPCODE_END:
1077          return;
1078
1079       case OPCODE_BGNLOOP:
1080       case OPCODE_BGNSUB:
1081       case OPCODE_BRA:
1082       case OPCODE_BRK:
1083       case OPCODE_CAL:
1084       case OPCODE_CONT:
1085       case OPCODE_DDX:
1086       case OPCODE_DDY:
1087       case OPCODE_ELSE:
1088       case OPCODE_ENDIF:
1089       case OPCODE_ENDLOOP:
1090       case OPCODE_ENDSUB:
1091       case OPCODE_IF:
1092       case OPCODE_RET:
1093          p->error = 1;
1094          i915_program_error(p, "Unsupported opcode: %s",
1095                             _mesa_opcode_string(inst->Opcode));
1096          return;
1097
1098       case OPCODE_EXP:
1099       case OPCODE_LOG:
1100          /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1101           * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1102           * Instead, it translates to EX2 or LG2.
1103           */
1104       case OPCODE_TXD:
1105       case OPCODE_TXL:
1106          /* These opcodes are claimed by GLSL in prog_instruction.h, but
1107           * only NV_vp/fp appears to emit them.
1108           */
1109       default:
1110          i915_program_error(p, "bad opcode: %s",
1111                             _mesa_opcode_string(inst->Opcode));
1112          return;
1113       }
1114
1115       inst++;
1116       i915_release_utemps(p);
1117    }
1118 }
1119
1120 /* Rather than trying to intercept and jiggle depth writes during
1121  * emit, just move the value into its correct position at the end of
1122  * the program:
1123  */
1124 static void
1125 fixup_depth_write(struct i915_fragment_program *p)
1126 {
1127    if (p->depth_written) {
1128       GLuint depth = UREG(REG_TYPE_OD, 0);
1129
1130       i915_emit_arith(p,
1131                       A0_MOV,
1132                       depth, A0_DEST_CHANNEL_W, 0,
1133                       swizzle(depth, X, Y, Z, Z), 0, 0);
1134    }
1135 }
1136
1137
1138 static void
1139 check_wpos(struct i915_fragment_program *p)
1140 {
1141    GLuint inputs = p->FragProg.Base.InputsRead;
1142    GLint i;
1143
1144    p->wpos_tex = -1;
1145
1146    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1147       if (inputs & (FRAG_BIT_TEX(i) | FRAG_BIT_VAR(i)))
1148          continue;
1149       else if (inputs & FRAG_BIT_WPOS) {
1150          p->wpos_tex = i;
1151          inputs &= ~FRAG_BIT_WPOS;
1152       }
1153    }
1154
1155    if (inputs & FRAG_BIT_WPOS) {
1156       i915_program_error(p, "No free texcoord for wpos value");
1157    }
1158 }
1159
1160
1161 static void
1162 translate_program(struct i915_fragment_program *p)
1163 {
1164    struct i915_context *i915 = I915_CONTEXT(p->ctx);
1165
1166    if (INTEL_DEBUG & DEBUG_WM) {
1167       printf("fp:\n");
1168       _mesa_print_program(&p->FragProg.Base);
1169       printf("\n");
1170    }
1171
1172    i915_init_program(i915, p);
1173    check_wpos(p);
1174    upload_program(p);
1175    fixup_depth_write(p);
1176    i915_fini_program(p);
1177
1178    p->translated = 1;
1179 }
1180
1181
1182 static void
1183 track_params(struct i915_fragment_program *p)
1184 {
1185    GLint i;
1186
1187    if (p->nr_params)
1188       _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1189
1190    for (i = 0; i < p->nr_params; i++) {
1191       GLint reg = p->param[i].reg;
1192       COPY_4V(p->constant[reg], p->param[i].values);
1193    }
1194
1195    p->params_uptodate = 1;
1196    p->on_hardware = 0;          /* overkill */
1197 }
1198
1199
1200 static void
1201 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1202 {
1203    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1204       struct i915_context *i915 = I915_CONTEXT(ctx);
1205       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1206
1207       if (i915->current_program == p)
1208          return;
1209
1210       if (i915->current_program) {
1211          i915->current_program->on_hardware = 0;
1212          i915->current_program->params_uptodate = 0;
1213       }
1214
1215       i915->current_program = p;
1216
1217       assert(p->on_hardware == 0);
1218       assert(p->params_uptodate == 0);
1219
1220    }
1221 }
1222
1223 static struct gl_program *
1224 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1225 {
1226    switch (target) {
1227    case GL_VERTEX_PROGRAM_ARB:
1228       return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1229                                        target, id);
1230
1231    case GL_FRAGMENT_PROGRAM_ARB:{
1232          struct i915_fragment_program *prog =
1233             CALLOC_STRUCT(i915_fragment_program);
1234          if (prog) {
1235             i915_init_program(I915_CONTEXT(ctx), prog);
1236
1237             return _mesa_init_fragment_program(ctx, &prog->FragProg,
1238                                                target, id);
1239          }
1240          else
1241             return NULL;
1242       }
1243
1244    default:
1245       /* Just fallback:
1246        */
1247       return _mesa_new_program(ctx, target, id);
1248    }
1249 }
1250
1251 static void
1252 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1253 {
1254    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1255       struct i915_context *i915 = I915_CONTEXT(ctx);
1256       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1257
1258       if (i915->current_program == p)
1259          i915->current_program = 0;
1260    }
1261
1262    _mesa_delete_program(ctx, prog);
1263 }
1264
1265
1266 static GLboolean
1267 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1268 {
1269    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1270       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1271
1272       if (!p->translated)
1273          translate_program(p);
1274
1275       return !p->error;
1276    }
1277    else
1278       return GL_TRUE;
1279 }
1280
1281 static GLboolean
1282 i915ProgramStringNotify(struct gl_context * ctx,
1283                         GLenum target, struct gl_program *prog)
1284 {
1285    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1286       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1287       p->translated = 0;
1288    }
1289
1290    (void) _tnl_program_string(ctx, target, prog);
1291
1292    /* XXX check if program is legal, within limits */
1293    return GL_TRUE;
1294 }
1295
1296 void
1297 i915_update_program(struct gl_context *ctx)
1298 {
1299    struct intel_context *intel = intel_context(ctx);
1300    struct i915_context *i915 = i915_context(&intel->ctx);
1301    struct i915_fragment_program *fp =
1302       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1303
1304    if (i915->current_program != fp) {
1305       if (i915->current_program) {
1306          i915->current_program->on_hardware = 0;
1307          i915->current_program->params_uptodate = 0;
1308       }
1309
1310       i915->current_program = fp;
1311    }
1312
1313    if (!fp->translated)
1314       translate_program(fp);
1315
1316    FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1317 }
1318
1319 void
1320 i915ValidateFragmentProgram(struct i915_context *i915)
1321 {
1322    struct gl_context *ctx = &i915->intel.ctx;
1323    struct intel_context *intel = intel_context(ctx);
1324    TNLcontext *tnl = TNL_CONTEXT(ctx);
1325    struct vertex_buffer *VB = &tnl->vb;
1326
1327    struct i915_fragment_program *p =
1328       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1329
1330    const GLuint inputsRead = p->FragProg.Base.InputsRead;
1331    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1332    GLuint s2 = S2_TEXCOORD_NONE;
1333    int i, offset = 0;
1334
1335    /* Important:
1336     */
1337    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1338
1339    if (!p->translated)
1340       translate_program(p);
1341
1342    intel->vertex_attr_count = 0;
1343    intel->wpos_offset = 0;
1344    intel->wpos_size = 0;
1345    intel->coloroffset = 0;
1346    intel->specoffset = 0;
1347
1348    if (inputsRead & FRAG_BITS_TEX_ANY) {
1349       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1350    }
1351    else {
1352       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1353    }
1354
1355    if (inputsRead & FRAG_BIT_COL0) {
1356       intel->coloroffset = offset / 4;
1357       EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1358    }
1359
1360    if (inputsRead & FRAG_BIT_COL1) {
1361        intel->specoffset = offset / 4;
1362        EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1363    }
1364
1365    if ((inputsRead & FRAG_BIT_FOGC)) {
1366       EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1367    }
1368
1369    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1370       if (inputsRead & FRAG_BIT_TEX(i)) {
1371          int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1372
1373          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1374          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1375
1376          EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1377       }
1378       else if (inputsRead & FRAG_BIT_VAR(i)) {
1379          int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1380
1381          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1382          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1383
1384          EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1385       }
1386       else if (i == p->wpos_tex) {
1387
1388          /* If WPOS is required, duplicate the XYZ position data in an
1389           * unused texture coordinate:
1390           */
1391          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1392          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1393
1394          intel->wpos_offset = offset;
1395          intel->wpos_size = 3 * sizeof(GLuint);
1396
1397          EMIT_PAD(intel->wpos_size);
1398       }
1399    }
1400
1401    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1402        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1403       int k;
1404
1405       I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1406
1407       /* Must do this *after* statechange, so as not to affect
1408        * buffered vertices reliant on the old state:
1409        */
1410       intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1411                                               intel->vertex_attrs,
1412                                               intel->vertex_attr_count,
1413                                               intel->ViewportMatrix.m, 0);
1414
1415       assert(intel->prim.current_offset == intel->prim.start_offset);
1416       intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1417       intel->prim.current_offset = intel->prim.start_offset;
1418
1419       intel->vertex_size >>= 2;
1420
1421       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1422       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1423
1424       k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1425       assert(k);
1426    }
1427
1428    if (!p->params_uptodate)
1429       track_params(p);
1430
1431    if (!p->on_hardware)
1432       i915_upload_program(i915, p);
1433
1434    if (INTEL_DEBUG & DEBUG_WM) {
1435       printf("i915:\n");
1436       i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1437    }
1438 }
1439
1440 void
1441 i915InitFragProgFuncs(struct dd_function_table *functions)
1442 {
1443    functions->BindProgram = i915BindProgram;
1444    functions->NewProgram = i915NewProgram;
1445    functions->DeleteProgram = i915DeleteProgram;
1446    functions->IsProgramNative = i915IsProgramNative;
1447    functions->ProgramStringNotify = i915ProgramStringNotify;
1448 }