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 VARYING_SLOT_POS:
 101          src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
 102          break;
 103       case VARYING_SLOT_COL0:
 104          src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
 105          break;
 106       case VARYING_SLOT_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 VARYING_SLOT_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 VARYING_SLOT_TEX0:
 115       case VARYING_SLOT_TEX1:
 116       case VARYING_SLOT_TEX2:
 117       case VARYING_SLOT_TEX3:
 118       case VARYING_SLOT_TEX4:
 119       case VARYING_SLOT_TEX5:
 120       case VARYING_SLOT_TEX6:
 121       case VARYING_SLOT_TEX7:
 122          src = i915_emit_decl(p, REG_TYPE_T,
 123                               T_TEX0 + (source->Index - VARYING_SLOT_TEX0),
 124                               D0_CHANNEL_ALL);
 125          break;
 126
 127       case VARYING_SLOT_VAR0:
 128       case VARYING_SLOT_VAR0 + 1:
 129       case VARYING_SLOT_VAR0 + 2:
 130       case VARYING_SLOT_VAR0 + 3:
 131       case VARYING_SLOT_VAR0 + 4:
 132       case VARYING_SLOT_VAR0 + 5:
 133       case VARYING_SLOT_VAR0 + 6:
 134       case VARYING_SLOT_VAR0 + 7:
 135          src = i915_emit_decl(p, REG_TYPE_T,
 136                               T_TEX0 + (source->Index - VARYING_SLOT_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_UNIFORM:
 177       src = i915_emit_param4fv(p,
 178          &program->Base.Parameters->ParameterValues[source->Index][0].f);
 179       break;
 180
 181    default:
 182       i915_program_error(p, "Bad source->File: %d", source->File);
 183       return 0;
 184    }
 185
 186    src = swizzle(src,
 187                  GET_SWZ(source->Swizzle, 0),
 188                  GET_SWZ(source->Swizzle, 1),
 189                  GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
 190
 191    if (source->Negate)
 192       src = negate(src,
 193                    GET_BIT(source->Negate, 0),
 194                    GET_BIT(source->Negate, 1),
 195                    GET_BIT(source->Negate, 2),
 196                    GET_BIT(source->Negate, 3));
 197
 198    return src;
 199 }
 200
 201
 202 static GLuint
 203 get_result_vector(struct i915_fragment_program *p,
 204                   const struct prog_instruction *inst)
 205 {
 206    switch (inst->DstReg.File) {
 207    case PROGRAM_OUTPUT:
 208       switch (inst->DstReg.Index) {
 209       case FRAG_RESULT_COLOR:
 210       case FRAG_RESULT_DATA0:
 211          return UREG(REG_TYPE_OC, 0);
 212       case FRAG_RESULT_DEPTH:
 213          p->depth_written = 1;
 214          return UREG(REG_TYPE_OD, 0);
 215       default:
 216          i915_program_error(p, "Bad inst->DstReg.Index: %d",
 217                             inst->DstReg.Index);
 218          return 0;
 219       }
 220    case PROGRAM_TEMPORARY:
 221       return UREG(REG_TYPE_R, inst->DstReg.Index);
 222    default:
 223       i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
 224       return 0;
 225    }
 226 }
 227
 228 static GLuint
 229 get_result_flags(const struct prog_instruction *inst)
 230 {
 231    GLuint flags = 0;
 232
 233    if (inst->SaturateMode == SATURATE_ZERO_ONE)
 234       flags |= A0_DEST_SATURATE;
 235    if (inst->DstReg.WriteMask & WRITEMASK_X)
 236       flags |= A0_DEST_CHANNEL_X;
 237    if (inst->DstReg.WriteMask & WRITEMASK_Y)
 238       flags |= A0_DEST_CHANNEL_Y;
 239    if (inst->DstReg.WriteMask & WRITEMASK_Z)
 240       flags |= A0_DEST_CHANNEL_Z;
 241    if (inst->DstReg.WriteMask & WRITEMASK_W)
 242       flags |= A0_DEST_CHANNEL_W;
 243
 244    return flags;
 245 }
 246
 247 static GLuint
 248 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 249 {
 250    switch (bit) {
 251    case TEXTURE_1D_INDEX:
 252       return D0_SAMPLE_TYPE_2D;
 253    case TEXTURE_2D_INDEX:
 254       return D0_SAMPLE_TYPE_2D;
 255    case TEXTURE_RECT_INDEX:
 256       return D0_SAMPLE_TYPE_2D;
 257    case TEXTURE_3D_INDEX:
 258       return D0_SAMPLE_TYPE_VOLUME;
 259    case TEXTURE_CUBE_INDEX:
 260       return D0_SAMPLE_TYPE_CUBE;
 261    default:
 262       i915_program_error(p, "TexSrcBit: %d", bit);
 263       return 0;
 264    }
 265 }
 266
 267 #define EMIT_TEX( OP )                                          \
 268 do {                                                            \
 269    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 270    const struct gl_fragment_program *program = &p->FragProg;    \
 271    GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit];  \
 272    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 273                                    unit, dim);                  \
 274    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 275    /* Texel lookup */                                           \
 276                                                                 \
 277    i915_emit_texld( p, get_live_regs(p, inst),                                          \
 278                get_result_vector( p, inst ),                    \
 279                get_result_flags( inst ),                        \
 280                sampler,                                         \
 281                coord,                                           \
 282                OP);                                             \
 283 } while (0)
 284
 285 #define EMIT_ARITH( OP, N )                                             \
 286 do {                                                                    \
 287    i915_emit_arith( p,                                                  \
 288                OP,                                                      \
 289                get_result_vector( p, inst ),                            \
 290                get_result_flags( inst ), 0,                     \
 291                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 292                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 293                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 294 } while (0)
 295
 296 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 297 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 298 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 299
 300 /*
 301  * TODO: consider moving this into core
 302  */
 303 static bool calc_live_regs( struct i915_fragment_program *p )
 304 {
 305     const struct gl_fragment_program *program = &p->FragProg;
 306     GLuint regsUsed = ~((1 << I915_MAX_TEMPORARY) - 1);
 307     uint8_t live_components[I915_MAX_TEMPORARY] = { 0, };
 308     GLint i;
 309
 310     for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
 311         struct prog_instruction *inst = &program->Base.Instructions[i];
 312         int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
 313         int a;
 314
 315         /* Register is written to: unmark as live for this and preceeding ops */
 316         if (inst->DstReg.File == PROGRAM_TEMPORARY) {
 317             if (inst->DstReg.Index >= I915_MAX_TEMPORARY)
 318                return false;
 319
 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                 if (inst->SrcReg[a].Index >= I915_MAX_TEMPORARY)
 331                    return false;
 332
 333                 regsUsed |= 1 << inst->SrcReg[a].Index;
 334
 335                 for (c = 0; c < 4; c++) {
 336                     const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
 337
 338                     if (field <= SWIZZLE_W)
 339                         live_components[inst->SrcReg[a].Index] |= (1U << field);
 340                 }
 341             }
 342         }
 343
 344         p->usedRegs[i] = regsUsed;
 345     }
 346
 347     return true;
 348 }
 349
 350 static GLuint get_live_regs( struct i915_fragment_program *p,
 351                              const struct prog_instruction *inst )
 352 {
 353     const struct gl_fragment_program *program = &p->FragProg;
 354     GLuint nr = inst - program->Base.Instructions;
 355
 356     return p->usedRegs[nr];
 357 }
 358
 359
 360 /* Possible concerns:
 361  *
 362  * SIN, COS -- could use another taylor step?
 363  * LIT      -- results seem a little different to sw mesa
 364  * LOG      -- different to mesa on negative numbers, but this is conformant.
 365  *
 366  * Parse failures -- Mesa doesn't currently give a good indication
 367  * internally whether a particular program string parsed or not.  This
 368  * can lead to confusion -- hopefully we cope with it ok now.
 369  *
 370  */
 371 static void
 372 upload_program(struct i915_fragment_program *p)
 373 {
 374    const struct gl_fragment_program *program = &p->FragProg;
 375    const struct prog_instruction *inst = program->Base.Instructions;
 376
 377    if (INTEL_DEBUG & DEBUG_WM)
 378       _mesa_print_program(&program->Base);
 379
 380    /* Is this a parse-failed program?  Ensure a valid program is
 381     * loaded, as the flagging of an error isn't sufficient to stop
 382     * this being uploaded to hardware.
 383     */
 384    if (inst[0].Opcode == OPCODE_END) {
 385       GLuint tmp = i915_get_utemp(p);
 386       i915_emit_arith(p,
 387                       A0_MOV,
 388                       UREG(REG_TYPE_OC, 0),
 389                       A0_DEST_CHANNEL_ALL, 0,
 390                       swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
 391       return;
 392    }
 393
 394    if (program->Base.NumInstructions > I915_MAX_INSN) {
 395       i915_program_error(p, "Exceeded max instructions (%d out of %d)",
 396                          program->Base.NumInstructions, I915_MAX_INSN);
 397       return;
 398    }
 399
 400    /* Not always needed:
 401     */
 402    if (!calc_live_regs(p)) {
 403       i915_program_error(p, "Could not allocate registers");
 404       return;
 405    }
 406
 407    while (1) {
 408       GLuint src0, src1, src2, flags;
 409       GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
 410
 411       switch (inst->Opcode) {
 412       case OPCODE_ABS:
 413          src0 = src_vector(p, &inst->SrcReg[0], program);
 414          i915_emit_arith(p,
 415                          A0_MAX,
 416                          get_result_vector(p, inst),
 417                          get_result_flags(inst), 0,
 418                          src0, negate(src0, 1, 1, 1, 1), 0);
 419          break;
 420
 421       case OPCODE_ADD:
 422          EMIT_2ARG_ARITH(A0_ADD);
 423          break;
 424
 425       case OPCODE_CMP:
 426          src0 = src_vector(p, &inst->SrcReg[0], program);
 427          src1 = src_vector(p, &inst->SrcReg[1], program);
 428          src2 = src_vector(p, &inst->SrcReg[2], program);
 429          i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
 430          break;
 431
 432       case OPCODE_COS:
 433          src0 = src_vector(p, &inst->SrcReg[0], program);
 434          tmp = i915_get_utemp(p);
 435          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 436          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 437
 438          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 439          i915_emit_arith(p,
 440                          A0_MAD,
 441                          tmp, A0_DEST_CHANNEL_X, 0,
 442                          src0,
 443                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 444                          swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
 445
 446          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 447
 448          i915_emit_arith(p,
 449                          A0_MAD,
 450                          tmp, A0_DEST_CHANNEL_X, 0,
 451                          tmp,
 452                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 453                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 454
 455          /* Compute COS with the same calculation used for SIN, but a
 456           * different source range has been mapped to [-1,1] this time.
 457           */
 458
 459          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 460          i915_emit_arith(p,
 461                          A0_MAX,
 462                          tmp, A0_DEST_CHANNEL_Y, 0,
 463                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 464                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 465                          0);
 466
 467          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 468          i915_emit_arith(p,
 469                          A0_MUL,
 470                          tmp, A0_DEST_CHANNEL_Y, 0,
 471                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 472                          tmp,
 473                          0);
 474
 475          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 476          i915_emit_arith(p,
 477                          A0_DP3,
 478                          tmp, A0_DEST_CHANNEL_X, 0,
 479                          tmp,
 480                          swizzle(consts1, X, Y, ZERO, ZERO),
 481                          0);
 482
 483          /* tmp.x now contains a first approximation (y).  Now, weight it
 484           * against tmp.y**2 to get closer.
 485           */
 486          i915_emit_arith(p,
 487                          A0_MAX,
 488                          tmp, A0_DEST_CHANNEL_Y, 0,
 489                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 490                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 491                          0);
 492
 493          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 494          i915_emit_arith(p,
 495                          A0_MAD,
 496                          tmp, A0_DEST_CHANNEL_Y, 0,
 497                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 498                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 499                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 500
 501          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 502          i915_emit_arith(p,
 503                          A0_MAD,
 504                          get_result_vector(p, inst),
 505                          get_result_flags(inst), 0,
 506                          swizzle(consts1, W, W, W, W),
 507                          swizzle(tmp, Y, Y, Y, Y),
 508                          swizzle(tmp, X, X, X, X));
 509          break;
 510
 511       case OPCODE_DP2:
 512          src0 = src_vector(p, &inst->SrcReg[0], program);
 513          src1 = src_vector(p, &inst->SrcReg[1], program);
 514          i915_emit_arith(p,
 515                          A0_DP3,
 516                          get_result_vector(p, inst),
 517                          get_result_flags(inst), 0,
 518                          swizzle(src0, X, Y, ZERO, ZERO),
 519                          swizzle(src1, X, Y, ZERO, ZERO),
 520                          0);
 521          break;
 522
 523       case OPCODE_DP3:
 524          EMIT_2ARG_ARITH(A0_DP3);
 525          break;
 526
 527       case OPCODE_DP4:
 528          EMIT_2ARG_ARITH(A0_DP4);
 529          break;
 530
 531       case OPCODE_DPH:
 532          src0 = src_vector(p, &inst->SrcReg[0], program);
 533          src1 = src_vector(p, &inst->SrcReg[1], program);
 534
 535          i915_emit_arith(p,
 536                          A0_DP4,
 537                          get_result_vector(p, inst),
 538                          get_result_flags(inst), 0,
 539                          swizzle(src0, X, Y, Z, ONE), src1, 0);
 540          break;
 541
 542       case OPCODE_DST:
 543          src0 = src_vector(p, &inst->SrcReg[0], program);
 544          src1 = src_vector(p, &inst->SrcReg[1], program);
 545
 546          /* result[0] = 1    * 1;
 547           * result[1] = a[1] * b[1];
 548           * result[2] = a[2] * 1;
 549           * result[3] = 1    * b[3];
 550           */
 551          i915_emit_arith(p,
 552                          A0_MUL,
 553                          get_result_vector(p, inst),
 554                          get_result_flags(inst), 0,
 555                          swizzle(src0, ONE, Y, Z, ONE),
 556                          swizzle(src1, ONE, Y, ONE, W), 0);
 557          break;
 558
 559       case OPCODE_EX2:
 560          src0 = src_vector(p, &inst->SrcReg[0], program);
 561
 562          i915_emit_arith(p,
 563                          A0_EXP,
 564                          get_result_vector(p, inst),
 565                          get_result_flags(inst), 0,
 566                          swizzle(src0, X, X, X, X), 0, 0);
 567          break;
 568
 569       case OPCODE_FLR:
 570          EMIT_1ARG_ARITH(A0_FLR);
 571          break;
 572
 573       case OPCODE_TRUNC:
 574          EMIT_1ARG_ARITH(A0_TRC);
 575          break;
 576
 577       case OPCODE_FRC:
 578          EMIT_1ARG_ARITH(A0_FRC);
 579          break;
 580
 581       case OPCODE_KIL:
 582          src0 = src_vector(p, &inst->SrcReg[0], program);
 583          tmp = i915_get_utemp(p);
 584
 585          i915_emit_texld(p, get_live_regs(p, inst),
 586                          tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
 587                          0, src0, T0_TEXKILL);
 588          break;
 589
 590       case OPCODE_KIL_NV:
 591          if (inst->DstReg.CondMask == COND_TR) {
 592             tmp = i915_get_utemp(p);
 593
 594             /* The KIL instruction discards the fragment if any component of
 595              * the source is < 0.  Emit an immediate operand of {-1}.xywz.
 596              */
 597             i915_emit_texld(p, get_live_regs(p, inst),
 598                             tmp, A0_DEST_CHANNEL_ALL,
 599                             0, /* use a dummy dest reg */
 600                             negate(swizzle(tmp, ONE, ONE, ONE, ONE),
 601                                    1, 1, 1, 1),
 602                             T0_TEXKILL);
 603          } else {
 604             p->error = 1;
 605             i915_program_error(p, "Unsupported KIL_NV condition code: %d",
 606                                inst->DstReg.CondMask);
 607          }
 608          break;
 609
 610       case OPCODE_LG2:
 611          src0 = src_vector(p, &inst->SrcReg[0], program);
 612
 613          i915_emit_arith(p,
 614                          A0_LOG,
 615                          get_result_vector(p, inst),
 616                          get_result_flags(inst), 0,
 617                          swizzle(src0, X, X, X, X), 0, 0);
 618          break;
 619
 620       case OPCODE_LIT:
 621          src0 = src_vector(p, &inst->SrcReg[0], program);
 622          tmp = i915_get_utemp(p);
 623
 624          /* tmp = max( a.xyzw, a.00zw )
 625           * XXX: Clamp tmp.w to -128..128
 626           * tmp.y = log(tmp.y)
 627           * tmp.y = tmp.w * tmp.y
 628           * tmp.y = exp(tmp.y)
 629           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 630           */
 631          i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 632                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
 633
 634          i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 635                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 636
 637          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 638                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 639                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
 640
 641          i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 642                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 643
 644          i915_emit_arith(p, A0_CMP,
 645                          get_result_vector(p, inst),
 646                          get_result_flags(inst), 0,
 647                          negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
 648                          swizzle(tmp, ONE, X, ZERO, ONE),
 649                          swizzle(tmp, ONE, X, Y, ONE));
 650
 651          break;
 652
 653       case OPCODE_LRP:
 654          src0 = src_vector(p, &inst->SrcReg[0], program);
 655          src1 = src_vector(p, &inst->SrcReg[1], program);
 656          src2 = src_vector(p, &inst->SrcReg[2], program);
 657          flags = get_result_flags(inst);
 658          tmp = i915_get_utemp(p);
 659
 660          /* b*a + c*(1-a)
 661           *
 662           * b*a + c - ca
 663           *
 664           * tmp = b*a + c,
 665           * result = (-c)*a + tmp
 666           */
 667          i915_emit_arith(p, A0_MAD, tmp,
 668                          flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
 669
 670          i915_emit_arith(p, A0_MAD,
 671                          get_result_vector(p, inst),
 672                          flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
 673          break;
 674
 675       case OPCODE_MAD:
 676          EMIT_3ARG_ARITH(A0_MAD);
 677          break;
 678
 679       case OPCODE_MAX:
 680          EMIT_2ARG_ARITH(A0_MAX);
 681          break;
 682
 683       case OPCODE_MIN:
 684          src0 = src_vector(p, &inst->SrcReg[0], program);
 685          src1 = src_vector(p, &inst->SrcReg[1], program);
 686          tmp = i915_get_utemp(p);
 687          flags = get_result_flags(inst);
 688
 689          i915_emit_arith(p,
 690                          A0_MAX,
 691                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 692                          negate(src0, 1, 1, 1, 1),
 693                          negate(src1, 1, 1, 1, 1), 0);
 694
 695          i915_emit_arith(p,
 696                          A0_MOV,
 697                          get_result_vector(p, inst),
 698                          flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
 699          break;
 700
 701       case OPCODE_MOV:
 702          EMIT_1ARG_ARITH(A0_MOV);
 703          break;
 704
 705       case OPCODE_MUL:
 706          EMIT_2ARG_ARITH(A0_MUL);
 707          break;
 708
 709       case OPCODE_POW:
 710          src0 = src_vector(p, &inst->SrcReg[0], program);
 711          src1 = src_vector(p, &inst->SrcReg[1], program);
 712          tmp = i915_get_utemp(p);
 713          flags = get_result_flags(inst);
 714
 715          /* XXX: masking on intermediate values, here and elsewhere.
 716           */
 717          i915_emit_arith(p,
 718                          A0_LOG,
 719                          tmp, A0_DEST_CHANNEL_X, 0,
 720                          swizzle(src0, X, X, X, X), 0, 0);
 721
 722          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
 723
 724
 725          i915_emit_arith(p,
 726                          A0_EXP,
 727                          get_result_vector(p, inst),
 728                          flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
 729
 730          break;
 731
 732       case OPCODE_RCP:
 733          src0 = src_vector(p, &inst->SrcReg[0], program);
 734
 735          i915_emit_arith(p,
 736                          A0_RCP,
 737                          get_result_vector(p, inst),
 738                          get_result_flags(inst), 0,
 739                          swizzle(src0, X, X, X, X), 0, 0);
 740          break;
 741
 742       case OPCODE_RSQ:
 743
 744          src0 = src_vector(p, &inst->SrcReg[0], program);
 745
 746          i915_emit_arith(p,
 747                          A0_RSQ,
 748                          get_result_vector(p, inst),
 749                          get_result_flags(inst), 0,
 750                          swizzle(src0, X, X, X, X), 0, 0);
 751          break;
 752
 753       case OPCODE_SCS:
 754          src0 = src_vector(p, &inst->SrcReg[0], program);
 755          tmp = i915_get_utemp(p);
 756
 757          /*
 758           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 759           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 760           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 761           * scs.x = DP4 t1, sin_constants
 762           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 763           * scs.y = DP4 t1, cos_constants
 764           */
 765          i915_emit_arith(p,
 766                          A0_MUL,
 767                          tmp, A0_DEST_CHANNEL_XY, 0,
 768                          swizzle(src0, X, X, ONE, ONE),
 769                          swizzle(src0, X, ONE, ONE, ONE), 0);
 770
 771          i915_emit_arith(p,
 772                          A0_MUL,
 773                          tmp, A0_DEST_CHANNEL_ALL, 0,
 774                          swizzle(tmp, X, Y, X, Y),
 775                          swizzle(tmp, X, X, ONE, ONE), 0);
 776
 777          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 778             GLuint tmp1;
 779
 780             if (inst->DstReg.WriteMask & WRITEMASK_X)
 781                tmp1 = i915_get_utemp(p);
 782             else
 783                tmp1 = tmp;
 784
 785             i915_emit_arith(p,
 786                             A0_MUL,
 787                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 788                             swizzle(tmp, X, Y, Y, W),
 789                             swizzle(tmp, X, Z, ONE, ONE), 0);
 790
 791             i915_emit_arith(p,
 792                             A0_DP4,
 793                             get_result_vector(p, inst),
 794                             A0_DEST_CHANNEL_Y, 0,
 795                             swizzle(tmp1, W, Z, Y, X),
 796                             i915_emit_const4fv(p, sin_constants), 0);
 797          }
 798
 799          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 800             i915_emit_arith(p,
 801                             A0_MUL,
 802                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 803                             swizzle(tmp, X, X, Z, ONE),
 804                             swizzle(tmp, Z, ONE, ONE, ONE), 0);
 805
 806             i915_emit_arith(p,
 807                             A0_DP4,
 808                             get_result_vector(p, inst),
 809                             A0_DEST_CHANNEL_X, 0,
 810                             swizzle(tmp, ONE, Z, Y, X),
 811                             i915_emit_const4fv(p, cos_constants), 0);
 812          }
 813          break;
 814
 815       case OPCODE_SEQ:
 816          tmp = i915_get_utemp(p);
 817          flags = get_result_flags(inst);
 818          dst = get_result_vector(p, inst);
 819
 820          /* tmp = src1 >= src2 */
 821          i915_emit_arith(p,
 822                          A0_SGE,
 823                          tmp,
 824                          flags, 0,
 825                          src_vector(p, &inst->SrcReg[0], program),
 826                          src_vector(p, &inst->SrcReg[1], program),
 827                          0);
 828          /* dst = src1 <= src2 */
 829          i915_emit_arith(p,
 830                          A0_SGE,
 831                          dst,
 832                          flags, 0,
 833                          negate(src_vector(p, &inst->SrcReg[0], program),
 834                                 1, 1, 1, 1),
 835                          negate(src_vector(p, &inst->SrcReg[1], program),
 836                                 1, 1, 1, 1),
 837                          0);
 838          /* dst = tmp && dst */
 839          i915_emit_arith(p,
 840                          A0_MUL,
 841                          dst,
 842                          flags, 0,
 843                          dst,
 844                          tmp,
 845                          0);
 846          break;
 847
 848       case OPCODE_SIN:
 849          src0 = src_vector(p, &inst->SrcReg[0], program);
 850          tmp = i915_get_utemp(p);
 851          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 852          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 853
 854          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 855          i915_emit_arith(p,
 856                          A0_MAD,
 857                          tmp, A0_DEST_CHANNEL_X, 0,
 858                          src0,
 859                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 860                          swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
 861
 862          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 863
 864          i915_emit_arith(p,
 865                          A0_MAD,
 866                          tmp, A0_DEST_CHANNEL_X, 0,
 867                          tmp,
 868                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 869                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 870
 871          /* Compute sin using a quadratic and quartic.  It gives continuity
 872           * that repeating the Taylor series lacks every 2*pi, and has
 873           * reduced error.
 874           *
 875           * The idea was described at:
 876           * http://www.devmaster.net/forums/showthread.php?t=5784
 877           */
 878
 879          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 880          i915_emit_arith(p,
 881                          A0_MAX,
 882                          tmp, A0_DEST_CHANNEL_Y, 0,
 883                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 884                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 885                          0);
 886
 887          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 888          i915_emit_arith(p,
 889                          A0_MUL,
 890                          tmp, A0_DEST_CHANNEL_Y, 0,
 891                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 892                          tmp,
 893                          0);
 894
 895          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 896          i915_emit_arith(p,
 897                          A0_DP3,
 898                          tmp, A0_DEST_CHANNEL_X, 0,
 899                          tmp,
 900                          swizzle(consts1, X, Y, ZERO, ZERO),
 901                          0);
 902
 903          /* tmp.x now contains a first approximation (y).  Now, weight it
 904           * against tmp.y**2 to get closer.
 905           */
 906          i915_emit_arith(p,
 907                          A0_MAX,
 908                          tmp, A0_DEST_CHANNEL_Y, 0,
 909                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 910                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 911                          0);
 912
 913          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 914          i915_emit_arith(p,
 915                          A0_MAD,
 916                          tmp, A0_DEST_CHANNEL_Y, 0,
 917                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 918                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 919                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 920
 921          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 922          i915_emit_arith(p,
 923                          A0_MAD,
 924                          get_result_vector(p, inst),
 925                          get_result_flags(inst), 0,
 926                          swizzle(consts1, W, W, W, W),
 927                          swizzle(tmp, Y, Y, Y, Y),
 928                          swizzle(tmp, X, X, X, X));
 929
 930          break;
 931
 932       case OPCODE_SGE:
 933          EMIT_2ARG_ARITH(A0_SGE);
 934          break;
 935
 936       case OPCODE_SGT:
 937          i915_emit_arith(p,
 938                          A0_SLT,
 939                          get_result_vector( p, inst ),
 940                          get_result_flags( inst ), 0,
 941                          negate(src_vector( p, &inst->SrcReg[0], program),
 942                                 1, 1, 1, 1),
 943                          negate(src_vector( p, &inst->SrcReg[1], program),
 944                                 1, 1, 1, 1),
 945                          0);
 946          break;
 947
 948       case OPCODE_SLE:
 949          i915_emit_arith(p,
 950                          A0_SGE,
 951                          get_result_vector( p, inst ),
 952                          get_result_flags( inst ), 0,
 953                          negate(src_vector( p, &inst->SrcReg[0], program),
 954                                 1, 1, 1, 1),
 955                          negate(src_vector( p, &inst->SrcReg[1], program),
 956                                 1, 1, 1, 1),
 957                          0);
 958          break;
 959
 960       case OPCODE_SLT:
 961          EMIT_2ARG_ARITH(A0_SLT);
 962          break;
 963
 964       case OPCODE_SNE:
 965          tmp = i915_get_utemp(p);
 966          flags = get_result_flags(inst);
 967          dst = get_result_vector(p, inst);
 968
 969          /* tmp = src1 < src2 */
 970          i915_emit_arith(p,
 971                          A0_SLT,
 972                          tmp,
 973                          flags, 0,
 974                          src_vector(p, &inst->SrcReg[0], program),
 975                          src_vector(p, &inst->SrcReg[1], program),
 976                          0);
 977          /* dst = src1 > src2 */
 978          i915_emit_arith(p,
 979                          A0_SLT,
 980                          dst,
 981                          flags, 0,
 982                          negate(src_vector(p, &inst->SrcReg[0], program),
 983                                 1, 1, 1, 1),
 984                          negate(src_vector(p, &inst->SrcReg[1], program),
 985                                 1, 1, 1, 1),
 986                          0);
 987          /* dst = tmp || dst */
 988          i915_emit_arith(p,
 989                          A0_ADD,
 990                          dst,
 991                          flags | A0_DEST_SATURATE, 0,
 992                          dst,
 993                          tmp,
 994                          0);
 995          break;
 996
 997       case OPCODE_SSG:
 998          dst = get_result_vector(p, inst);
 999          flags = get_result_flags(inst);
1000          src0 = src_vector(p, &inst->SrcReg[0], program);
1001          tmp = i915_get_utemp(p);
1002
1003          /* tmp = (src < 0.0) */
1004          i915_emit_arith(p,
1005                          A0_SLT,
1006                          tmp,
1007                          flags, 0,
1008                          src0,
1009                          swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1010                          0);
1011
1012          /* dst = (0.0 < src) */
1013          i915_emit_arith(p,
1014                          A0_SLT,
1015                          dst,
1016                          flags, 0,
1017                          swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1018                          src0,
1019                          0);
1020
1021          /* dst = (src > 0.0) - (src < 0.0) */
1022          i915_emit_arith(p,
1023                          A0_ADD,
1024                          dst,
1025                          flags, 0,
1026                          dst,
1027                          negate(tmp, 1, 1, 1, 1),
1028                          0);
1029
1030          break;
1031
1032       case OPCODE_SUB:
1033          src0 = src_vector(p, &inst->SrcReg[0], program);
1034          src1 = src_vector(p, &inst->SrcReg[1], program);
1035
1036          i915_emit_arith(p,
1037                          A0_ADD,
1038                          get_result_vector(p, inst),
1039                          get_result_flags(inst), 0,
1040                          src0, negate(src1, 1, 1, 1, 1), 0);
1041          break;
1042
1043       case OPCODE_SWZ:
1044          EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
1045          break;
1046
1047       case OPCODE_TEX:
1048          EMIT_TEX(T0_TEXLD);
1049          break;
1050
1051       case OPCODE_TXB:
1052          EMIT_TEX(T0_TEXLDB);
1053          break;
1054
1055       case OPCODE_TXP:
1056          EMIT_TEX(T0_TEXLDP);
1057          break;
1058
1059       case OPCODE_XPD:
1060          /* Cross product:
1061           *      result.x = src0.y * src1.z - src0.z * src1.y;
1062           *      result.y = src0.z * src1.x - src0.x * src1.z;
1063           *      result.z = src0.x * src1.y - src0.y * src1.x;
1064           *      result.w = undef;
1065           */
1066          src0 = src_vector(p, &inst->SrcReg[0], program);
1067          src1 = src_vector(p, &inst->SrcReg[1], program);
1068          tmp = i915_get_utemp(p);
1069
1070          i915_emit_arith(p,
1071                          A0_MUL,
1072                          tmp, A0_DEST_CHANNEL_ALL, 0,
1073                          swizzle(src0, Z, X, Y, ONE),
1074                          swizzle(src1, Y, Z, X, ONE), 0);
1075
1076          i915_emit_arith(p,
1077                          A0_MAD,
1078                          get_result_vector(p, inst),
1079                          get_result_flags(inst), 0,
1080                          swizzle(src0, Y, Z, X, ONE),
1081                          swizzle(src1, Z, X, Y, ONE),
1082                          negate(tmp, 1, 1, 1, 0));
1083          break;
1084
1085       case OPCODE_END:
1086          return;
1087
1088       case OPCODE_BGNLOOP:
1089       case OPCODE_BGNSUB:
1090       case OPCODE_BRK:
1091       case OPCODE_CAL:
1092       case OPCODE_CONT:
1093       case OPCODE_DDX:
1094       case OPCODE_DDY:
1095       case OPCODE_ELSE:
1096       case OPCODE_ENDIF:
1097       case OPCODE_ENDLOOP:
1098       case OPCODE_ENDSUB:
1099       case OPCODE_IF:
1100       case OPCODE_RET:
1101          p->error = 1;
1102          i915_program_error(p, "Unsupported opcode: %s",
1103                             _mesa_opcode_string(inst->Opcode));
1104          return;
1105
1106       case OPCODE_EXP:
1107       case OPCODE_LOG:
1108          /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1109           * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1110           * Instead, it translates to EX2 or LG2.
1111           */
1112       case OPCODE_TXD:
1113       case OPCODE_TXL:
1114          /* These opcodes are claimed by GLSL in prog_instruction.h, but
1115           * only NV_vp/fp appears to emit them.
1116           */
1117       default:
1118          i915_program_error(p, "bad opcode: %s",
1119                             _mesa_opcode_string(inst->Opcode));
1120          return;
1121       }
1122
1123       inst++;
1124       i915_release_utemps(p);
1125    }
1126 }
1127
1128 /* Rather than trying to intercept and jiggle depth writes during
1129  * emit, just move the value into its correct position at the end of
1130  * the program:
1131  */
1132 static void
1133 fixup_depth_write(struct i915_fragment_program *p)
1134 {
1135    if (p->depth_written) {
1136       GLuint depth = UREG(REG_TYPE_OD, 0);
1137
1138       i915_emit_arith(p,
1139                       A0_MOV,
1140                       depth, A0_DEST_CHANNEL_W, 0,
1141                       swizzle(depth, X, Y, Z, Z), 0, 0);
1142    }
1143 }
1144
1145
1146 static void
1147 check_wpos(struct i915_fragment_program *p)
1148 {
1149    GLbitfield64 inputs = p->FragProg.Base.InputsRead;
1150    GLint i;
1151
1152    p->wpos_tex = -1;
1153
1154    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1155       if (inputs & (VARYING_BIT_TEX(i) | VARYING_BIT_VAR(i)))
1156          continue;
1157       else if (inputs & VARYING_BIT_POS) {
1158          p->wpos_tex = i;
1159          inputs &= ~VARYING_BIT_POS;
1160       }
1161    }
1162
1163    if (inputs & VARYING_BIT_POS) {
1164       i915_program_error(p, "No free texcoord for wpos value");
1165    }
1166 }
1167
1168
1169 static void
1170 translate_program(struct i915_fragment_program *p)
1171 {
1172    struct i915_context *i915 = I915_CONTEXT(p->ctx);
1173
1174    if (INTEL_DEBUG & DEBUG_WM) {
1175       printf("fp:\n");
1176       _mesa_print_program(&p->FragProg.Base);
1177       printf("\n");
1178    }
1179
1180    i915_init_program(i915, p);
1181    check_wpos(p);
1182    upload_program(p);
1183    fixup_depth_write(p);
1184    i915_fini_program(p);
1185
1186    p->translated = 1;
1187 }
1188
1189
1190 static void
1191 track_params(struct i915_fragment_program *p)
1192 {
1193    GLint i;
1194
1195    if (p->nr_params)
1196       _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1197
1198    for (i = 0; i < p->nr_params; i++) {
1199       GLint reg = p->param[i].reg;
1200       COPY_4V(p->constant[reg], p->param[i].values);
1201    }
1202
1203    p->params_uptodate = 1;
1204    p->on_hardware = 0;          /* overkill */
1205 }
1206
1207
1208 static void
1209 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1210 {
1211    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1212       struct i915_context *i915 = I915_CONTEXT(ctx);
1213       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1214
1215       if (i915->current_program == p)
1216          return;
1217
1218       if (i915->current_program) {
1219          i915->current_program->on_hardware = 0;
1220          i915->current_program->params_uptodate = 0;
1221       }
1222
1223       i915->current_program = p;
1224
1225       assert(p->on_hardware == 0);
1226       assert(p->params_uptodate == 0);
1227
1228    }
1229 }
1230
1231 static struct gl_program *
1232 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1233 {
1234    switch (target) {
1235    case GL_VERTEX_PROGRAM_ARB:
1236       return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1237                                        target, id);
1238
1239    case GL_FRAGMENT_PROGRAM_ARB:{
1240          struct i915_fragment_program *prog =
1241             CALLOC_STRUCT(i915_fragment_program);
1242          if (prog) {
1243             i915_init_program(I915_CONTEXT(ctx), prog);
1244
1245             return _mesa_init_fragment_program(ctx, &prog->FragProg,
1246                                                target, id);
1247          }
1248          else
1249             return NULL;
1250       }
1251
1252    default:
1253       /* Just fallback:
1254        */
1255       return _mesa_new_program(ctx, target, id);
1256    }
1257 }
1258
1259 static void
1260 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1261 {
1262    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1263       struct i915_context *i915 = I915_CONTEXT(ctx);
1264       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1265
1266       if (i915->current_program == p)
1267          i915->current_program = 0;
1268    }
1269
1270    _mesa_delete_program(ctx, prog);
1271 }
1272
1273
1274 static GLboolean
1275 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1276 {
1277    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1278       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1279
1280       if (!p->translated)
1281          translate_program(p);
1282
1283       return !p->error;
1284    }
1285    else
1286       return true;
1287 }
1288
1289 static GLboolean
1290 i915ProgramStringNotify(struct gl_context * ctx,
1291                         GLenum target, struct gl_program *prog)
1292 {
1293    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1294       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1295       p->translated = 0;
1296    }
1297
1298    (void) _tnl_program_string(ctx, target, prog);
1299
1300    /* XXX check if program is legal, within limits */
1301    return true;
1302 }
1303
1304 static void
1305 i915SamplerUniformChange(struct gl_context *ctx,
1306                          GLenum target, struct gl_program *prog)
1307 {
1308    i915ProgramStringNotify(ctx, target, prog);
1309 }
1310
1311 void
1312 i915_update_program(struct gl_context *ctx)
1313 {
1314    struct intel_context *intel = intel_context(ctx);
1315    struct i915_context *i915 = i915_context(&intel->ctx);
1316    struct i915_fragment_program *fp =
1317       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1318
1319    if (i915->current_program != fp) {
1320       if (i915->current_program) {
1321          i915->current_program->on_hardware = 0;
1322          i915->current_program->params_uptodate = 0;
1323       }
1324
1325       i915->current_program = fp;
1326    }
1327
1328    if (!fp->translated)
1329       translate_program(fp);
1330
1331    FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1332 }
1333
1334 void
1335 i915ValidateFragmentProgram(struct i915_context *i915)
1336 {
1337    struct gl_context *ctx = &i915->intel.ctx;
1338    struct intel_context *intel = intel_context(ctx);
1339    TNLcontext *tnl = TNL_CONTEXT(ctx);
1340    struct vertex_buffer *VB = &tnl->vb;
1341
1342    struct i915_fragment_program *p =
1343       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1344
1345    const GLbitfield64 inputsRead = p->FragProg.Base.InputsRead;
1346    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1347    GLuint s2 = S2_TEXCOORD_NONE;
1348    int i, offset = 0;
1349
1350    /* Important:
1351     */
1352    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1353
1354    if (!p->translated)
1355       translate_program(p);
1356
1357    intel->vertex_attr_count = 0;
1358    intel->wpos_offset = 0;
1359    intel->coloroffset = 0;
1360    intel->specoffset = 0;
1361
1362    if (inputsRead & VARYING_BITS_TEX_ANY || p->wpos_tex != -1) {
1363       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1364    }
1365    else {
1366       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1367    }
1368
1369    /* Handle gl_PointSize builtin var here */
1370    if (ctx->Point._Attenuated || ctx->VertexProgram.PointSizeEnabled)
1371       EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4);
1372
1373    if (inputsRead & VARYING_BIT_COL0) {
1374       intel->coloroffset = offset / 4;
1375       EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1376    }
1377
1378    if (inputsRead & VARYING_BIT_COL1) {
1379        intel->specoffset = offset / 4;
1380        EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1381    }
1382
1383    if ((inputsRead & VARYING_BIT_FOGC)) {
1384       EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1385    }
1386
1387    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1388       if (inputsRead & VARYING_BIT_TEX(i)) {
1389          int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1390
1391          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1392          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1393
1394          EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1395       }
1396       else if (inputsRead & VARYING_BIT_VAR(i)) {
1397          int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1398
1399          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1400          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1401
1402          EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1403       }
1404       else if (i == p->wpos_tex) {
1405          int wpos_size = 4 * sizeof(float);
1406          /* If WPOS is required, duplicate the XYZ position data in an
1407           * unused texture coordinate:
1408           */
1409          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1410          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(wpos_size));
1411
1412          intel->wpos_offset = offset;
1413          EMIT_PAD(wpos_size);
1414       }
1415    }
1416
1417    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1418        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1419       int k;
1420
1421       I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1422
1423       /* Must do this *after* statechange, so as not to affect
1424        * buffered vertices reliant on the old state:
1425        */
1426       intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1427                                               intel->vertex_attrs,
1428                                               intel->vertex_attr_count,
1429                                               intel->ViewportMatrix.m, 0);
1430
1431       assert(intel->prim.current_offset == intel->prim.start_offset);
1432       intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1433       intel->prim.current_offset = intel->prim.start_offset;
1434
1435       intel->vertex_size >>= 2;
1436
1437       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1438       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1439
1440       k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1441       assert(k);
1442    }
1443
1444    if (!p->params_uptodate)
1445       track_params(p);
1446
1447    if (!p->on_hardware)
1448       i915_upload_program(i915, p);
1449
1450    if (INTEL_DEBUG & DEBUG_WM) {
1451       printf("i915:\n");
1452       i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1453    }
1454 }
1455
1456 void
1457 i915InitFragProgFuncs(struct dd_function_table *functions)
1458 {
1459    functions->BindProgram = i915BindProgram;
1460    functions->NewProgram = i915NewProgram;
1461    functions->DeleteProgram = i915DeleteProgram;
1462    functions->IsProgramNative = i915IsProgramNative;
1463    functions->ProgramStringNotify = i915ProgramStringNotify;
1464    functions->SamplerUniformChange = i915SamplerUniformChange;
1465 }