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 "shader/prog_instruction.h"
  33 #include "shader/prog_parameter.h"
  34 #include "shader/program.h"
  35 #include "shader/programopt.h"
  36
  37 #include "tnl/tnl.h"
  38 #include "tnl/t_context.h"
  39
  40 #include "intel_batchbuffer.h"
  41
  42 #include "i915_reg.h"
  43 #include "i915_context.h"
  44 #include "i915_program.h"
  45
  46 static const GLfloat sin_quad_constants[2][4] = {
  47    {
  48       2.0,
  49       -1.0,
  50       .5,
  51       .75
  52    },
  53    {
  54       4.0,
  55       -4.0,
  56       1.0 / (2.0 * M_PI),
  57       .2225
  58    }
  59 };
  60
  61 static const GLfloat sin_constants[4] = { 1.0,
  62    -1.0 / (3 * 2 * 1),
  63    1.0 / (5 * 4 * 3 * 2 * 1),
  64    -1.0 / (7 * 6 * 5 * 4 * 3 * 2 * 1)
  65 };
  66
  67 /* 1, -1/2!, 1/4!, -1/6! */
  68 static const GLfloat cos_constants[4] = { 1.0,
  69    -1.0 / (2 * 1),
  70    1.0 / (4 * 3 * 2 * 1),
  71    -1.0 / (6 * 5 * 4 * 3 * 2 * 1)
  72 };
  73
  74 /**
  75  * Retrieve a ureg for the given source register.  Will emit
  76  * constants, apply swizzling and negation as needed.
  77  */
  78 static GLuint
  79 src_vector(struct i915_fragment_program *p,
  80            const struct prog_src_register *source,
  81            const struct gl_fragment_program *program)
  82 {
  83    GLuint src;
  84
  85    switch (source->File) {
  86
  87       /* Registers:
  88        */
  89    case PROGRAM_TEMPORARY:
  90       if (source->Index >= I915_MAX_TEMPORARY) {
  91          i915_program_error(p, "Exceeded max temporary reg");
  92          return 0;
  93       }
  94       src = UREG(REG_TYPE_R, source->Index);
  95       break;
  96    case PROGRAM_INPUT:
  97       switch (source->Index) {
  98       case FRAG_ATTRIB_WPOS:
  99          src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
 100          break;
 101       case FRAG_ATTRIB_COL0:
 102          src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
 103          break;
 104       case FRAG_ATTRIB_COL1:
 105          src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
 106          src = swizzle(src, X, Y, Z, ONE);
 107          break;
 108       case FRAG_ATTRIB_FOGC:
 109          src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
 110          src = swizzle(src, W, ZERO, ZERO, ONE);
 111          break;
 112       case FRAG_ATTRIB_TEX0:
 113       case FRAG_ATTRIB_TEX1:
 114       case FRAG_ATTRIB_TEX2:
 115       case FRAG_ATTRIB_TEX3:
 116       case FRAG_ATTRIB_TEX4:
 117       case FRAG_ATTRIB_TEX5:
 118       case FRAG_ATTRIB_TEX6:
 119       case FRAG_ATTRIB_TEX7:
 120          src = i915_emit_decl(p, REG_TYPE_T,
 121                               T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
 122                               D0_CHANNEL_ALL);
 123          break;
 124
 125       default:
 126          i915_program_error(p, "Bad source->Index");
 127          return 0;
 128       }
 129       break;
 130
 131       /* Various paramters and env values.  All emitted to
 132        * hardware as program constants.
 133        */
 134    case PROGRAM_LOCAL_PARAM:
 135       src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
 136       break;
 137
 138    case PROGRAM_ENV_PARAM:
 139       src =
 140          i915_emit_param4fv(p,
 141                             p->ctx->FragmentProgram.Parameters[source->
 142                                                                Index]);
 143       break;
 144
 145    case PROGRAM_CONSTANT:
 146    case PROGRAM_STATE_VAR:
 147    case PROGRAM_NAMED_PARAM:
 148       src =
 149          i915_emit_param4fv(p,
 150                             program->Base.Parameters->ParameterValues[source->
 151                                                                       Index]);
 152       break;
 153
 154    default:
 155       i915_program_error(p, "Bad source->File");
 156       return 0;
 157    }
 158
 159    src = swizzle(src,
 160                  GET_SWZ(source->Swizzle, 0),
 161                  GET_SWZ(source->Swizzle, 1),
 162                  GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
 163
 164    if (source->NegateBase)
 165       src = negate(src,
 166                    GET_BIT(source->NegateBase, 0),
 167                    GET_BIT(source->NegateBase, 1),
 168                    GET_BIT(source->NegateBase, 2),
 169                    GET_BIT(source->NegateBase, 3));
 170
 171    return src;
 172 }
 173
 174
 175 static GLuint
 176 get_result_vector(struct i915_fragment_program *p,
 177                   const struct prog_instruction *inst)
 178 {
 179    switch (inst->DstReg.File) {
 180    case PROGRAM_OUTPUT:
 181       switch (inst->DstReg.Index) {
 182       case FRAG_RESULT_COLR:
 183          return UREG(REG_TYPE_OC, 0);
 184       case FRAG_RESULT_DEPR:
 185          p->depth_written = 1;
 186          return UREG(REG_TYPE_OD, 0);
 187       default:
 188          i915_program_error(p, "Bad inst->DstReg.Index");
 189          return 0;
 190       }
 191    case PROGRAM_TEMPORARY:
 192       return UREG(REG_TYPE_R, inst->DstReg.Index);
 193    default:
 194       i915_program_error(p, "Bad inst->DstReg.File");
 195       return 0;
 196    }
 197 }
 198
 199 static GLuint
 200 get_result_flags(const struct prog_instruction *inst)
 201 {
 202    GLuint flags = 0;
 203
 204    if (inst->SaturateMode == SATURATE_ZERO_ONE)
 205       flags |= A0_DEST_SATURATE;
 206    if (inst->DstReg.WriteMask & WRITEMASK_X)
 207       flags |= A0_DEST_CHANNEL_X;
 208    if (inst->DstReg.WriteMask & WRITEMASK_Y)
 209       flags |= A0_DEST_CHANNEL_Y;
 210    if (inst->DstReg.WriteMask & WRITEMASK_Z)
 211       flags |= A0_DEST_CHANNEL_Z;
 212    if (inst->DstReg.WriteMask & WRITEMASK_W)
 213       flags |= A0_DEST_CHANNEL_W;
 214
 215    return flags;
 216 }
 217
 218 static GLuint
 219 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
 220 {
 221    switch (bit) {
 222    case TEXTURE_1D_INDEX:
 223       return D0_SAMPLE_TYPE_2D;
 224    case TEXTURE_2D_INDEX:
 225       return D0_SAMPLE_TYPE_2D;
 226    case TEXTURE_RECT_INDEX:
 227       return D0_SAMPLE_TYPE_2D;
 228    case TEXTURE_3D_INDEX:
 229       return D0_SAMPLE_TYPE_VOLUME;
 230    case TEXTURE_CUBE_INDEX:
 231       return D0_SAMPLE_TYPE_CUBE;
 232    default:
 233       i915_program_error(p, "TexSrcBit");
 234       return 0;
 235    }
 236 }
 237
 238 #define EMIT_TEX( OP )                                          \
 239 do {                                                            \
 240    GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget );      \
 241    GLuint sampler = i915_emit_decl(p, REG_TYPE_S,               \
 242                                   inst->TexSrcUnit, dim);       \
 243    GLuint coord = src_vector( p, &inst->SrcReg[0], program);    \
 244    /* Texel lookup */                                           \
 245                                                                 \
 246    i915_emit_texld( p, get_live_regs(p, inst),                                          \
 247                get_result_vector( p, inst ),                    \
 248                get_result_flags( inst ),                        \
 249                sampler,                                         \
 250                coord,                                           \
 251                OP);                                             \
 252 } while (0)
 253
 254 #define EMIT_ARITH( OP, N )                                             \
 255 do {                                                                    \
 256    i915_emit_arith( p,                                                  \
 257                OP,                                                      \
 258                get_result_vector( p, inst ),                            \
 259                get_result_flags( inst ), 0,                     \
 260                (N<1)?0:src_vector( p, &inst->SrcReg[0], program),       \
 261                (N<2)?0:src_vector( p, &inst->SrcReg[1], program),       \
 262                (N<3)?0:src_vector( p, &inst->SrcReg[2], program));      \
 263 } while (0)
 264
 265 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
 266 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
 267 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
 268
 269 /*
 270  * TODO: consider moving this into core
 271  */
 272 static void calc_live_regs( struct i915_fragment_program *p )
 273 {
 274     const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
 275     GLuint regsUsed = 0xffff0000;
 276     GLint i;
 277
 278     for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
 279         struct prog_instruction *inst = &program->Base.Instructions[i];
 280         int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
 281         int a;
 282
 283         /* Register is written to: unmark as live for this and preceeding ops */
 284         if (inst->DstReg.File == PROGRAM_TEMPORARY)
 285             regsUsed &= ~(1 << inst->DstReg.Index);
 286
 287         for (a = 0; a < opArgs; a++) {
 288             /* Register is read from: mark as live for this and preceeding ops */
 289             if (inst->SrcReg[a].File == PROGRAM_TEMPORARY)
 290                 regsUsed |= 1 << inst->SrcReg[a].Index;
 291         }
 292
 293         p->usedRegs[i] = regsUsed;
 294     }
 295 }
 296
 297 static GLuint get_live_regs( struct i915_fragment_program *p,
 298                              const struct prog_instruction *inst )
 299 {
 300     const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
 301     GLuint nr = inst - program->Base.Instructions;
 302
 303     return p->usedRegs[nr];
 304 }
 305
 306
 307 /* Possible concerns:
 308  *
 309  * SIN, COS -- could use another taylor step?
 310  * LIT      -- results seem a little different to sw mesa
 311  * LOG      -- different to mesa on negative numbers, but this is conformant.
 312  *
 313  * Parse failures -- Mesa doesn't currently give a good indication
 314  * internally whether a particular program string parsed or not.  This
 315  * can lead to confusion -- hopefully we cope with it ok now.
 316  *
 317  */
 318 static void
 319 upload_program(struct i915_fragment_program *p)
 320 {
 321    const struct gl_fragment_program *program =
 322       p->ctx->FragmentProgram._Current;
 323    const struct prog_instruction *inst = program->Base.Instructions;
 324
 325 /*    _mesa_debug_fp_inst(program->Base.NumInstructions, inst); */
 326
 327    /* Is this a parse-failed program?  Ensure a valid program is
 328     * loaded, as the flagging of an error isn't sufficient to stop
 329     * this being uploaded to hardware.
 330     */
 331    if (inst[0].Opcode == OPCODE_END) {
 332       GLuint tmp = i915_get_utemp(p);
 333       i915_emit_arith(p,
 334                       A0_MOV,
 335                       UREG(REG_TYPE_OC, 0),
 336                       A0_DEST_CHANNEL_ALL, 0,
 337                       swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
 338       return;
 339    }
 340
 341    if (program->Base.NumInstructions > I915_MAX_INSN) {
 342        i915_program_error( p, "Exceeded max instructions" );
 343        return;
 344     }
 345
 346    /* Not always needed:
 347     */
 348    calc_live_regs(p);
 349
 350    while (1) {
 351       GLuint src0, src1, src2, flags;
 352       GLuint tmp = 0, consts0 = 0, consts1 = 0;
 353
 354       switch (inst->Opcode) {
 355       case OPCODE_ABS:
 356          src0 = src_vector(p, &inst->SrcReg[0], program);
 357          i915_emit_arith(p,
 358                          A0_MAX,
 359                          get_result_vector(p, inst),
 360                          get_result_flags(inst), 0,
 361                          src0, negate(src0, 1, 1, 1, 1), 0);
 362          break;
 363
 364       case OPCODE_ADD:
 365          EMIT_2ARG_ARITH(A0_ADD);
 366          break;
 367
 368       case OPCODE_CMP:
 369          src0 = src_vector(p, &inst->SrcReg[0], program);
 370          src1 = src_vector(p, &inst->SrcReg[1], program);
 371          src2 = src_vector(p, &inst->SrcReg[2], program);
 372          i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1);   /* NOTE: order of src2, src1 */
 373          break;
 374
 375       case OPCODE_COS:
 376          src0 = src_vector(p, &inst->SrcReg[0], program);
 377          tmp = i915_get_utemp(p);
 378          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 379          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 380
 381          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 382          i915_emit_arith(p,
 383                          A0_MAD,
 384                          tmp, A0_DEST_CHANNEL_X, 0,
 385                          src0,
 386                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 387                          swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
 388
 389          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 390
 391          i915_emit_arith(p,
 392                          A0_MAD,
 393                          tmp, A0_DEST_CHANNEL_X, 0,
 394                          tmp,
 395                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 396                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 397
 398          /* Compute COS with the same calculation used for SIN, but a
 399           * different source range has been mapped to [-1,1] this time.
 400           */
 401
 402          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 403          i915_emit_arith(p,
 404                          A0_MAX,
 405                          tmp, A0_DEST_CHANNEL_Y, 0,
 406                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 407                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 408                          0);
 409
 410          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 411          i915_emit_arith(p,
 412                          A0_MUL,
 413                          tmp, A0_DEST_CHANNEL_Y, 0,
 414                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 415                          tmp,
 416                          0);
 417
 418          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 419          i915_emit_arith(p,
 420                          A0_DP3,
 421                          tmp, A0_DEST_CHANNEL_X, 0,
 422                          tmp,
 423                          swizzle(consts1, X, Y, ZERO, ZERO),
 424                          0);
 425
 426          /* tmp.x now contains a first approximation (y).  Now, weight it
 427           * against tmp.y**2 to get closer.
 428           */
 429          i915_emit_arith(p,
 430                          A0_MAX,
 431                          tmp, A0_DEST_CHANNEL_Y, 0,
 432                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 433                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 434                          0);
 435
 436          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 437          i915_emit_arith(p,
 438                          A0_MAD,
 439                          tmp, A0_DEST_CHANNEL_Y, 0,
 440                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 441                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 442                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 443
 444          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 445          i915_emit_arith(p,
 446                          A0_MAD,
 447                          get_result_vector(p, inst),
 448                          get_result_flags(inst), 0,
 449                          swizzle(consts1, W, W, W, W),
 450                          swizzle(tmp, Y, Y, Y, Y),
 451                          swizzle(tmp, X, X, X, X));
 452          break;
 453
 454       case OPCODE_DP3:
 455          EMIT_2ARG_ARITH(A0_DP3);
 456          break;
 457
 458       case OPCODE_DP4:
 459          EMIT_2ARG_ARITH(A0_DP4);
 460          break;
 461
 462       case OPCODE_DPH:
 463          src0 = src_vector(p, &inst->SrcReg[0], program);
 464          src1 = src_vector(p, &inst->SrcReg[1], program);
 465
 466          i915_emit_arith(p,
 467                          A0_DP4,
 468                          get_result_vector(p, inst),
 469                          get_result_flags(inst), 0,
 470                          swizzle(src0, X, Y, Z, ONE), src1, 0);
 471          break;
 472
 473       case OPCODE_DST:
 474          src0 = src_vector(p, &inst->SrcReg[0], program);
 475          src1 = src_vector(p, &inst->SrcReg[1], program);
 476
 477          /* result[0] = 1    * 1;
 478           * result[1] = a[1] * b[1];
 479           * result[2] = a[2] * 1;
 480           * result[3] = 1    * b[3];
 481           */
 482          i915_emit_arith(p,
 483                          A0_MUL,
 484                          get_result_vector(p, inst),
 485                          get_result_flags(inst), 0,
 486                          swizzle(src0, ONE, Y, Z, ONE),
 487                          swizzle(src1, ONE, Y, ONE, W), 0);
 488          break;
 489
 490       case OPCODE_EX2:
 491          src0 = src_vector(p, &inst->SrcReg[0], program);
 492
 493          i915_emit_arith(p,
 494                          A0_EXP,
 495                          get_result_vector(p, inst),
 496                          get_result_flags(inst), 0,
 497                          swizzle(src0, X, X, X, X), 0, 0);
 498          break;
 499
 500       case OPCODE_FLR:
 501          EMIT_1ARG_ARITH(A0_FLR);
 502          break;
 503
 504       case OPCODE_FRC:
 505          EMIT_1ARG_ARITH(A0_FRC);
 506          break;
 507
 508       case OPCODE_KIL:
 509          src0 = src_vector(p, &inst->SrcReg[0], program);
 510          tmp = i915_get_utemp(p);
 511
 512          i915_emit_texld(p, get_live_regs(p, inst),
 513                          tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
 514                          0, src0, T0_TEXKILL);
 515          break;
 516
 517       case OPCODE_LG2:
 518          src0 = src_vector(p, &inst->SrcReg[0], program);
 519
 520          i915_emit_arith(p,
 521                          A0_LOG,
 522                          get_result_vector(p, inst),
 523                          get_result_flags(inst), 0,
 524                          swizzle(src0, X, X, X, X), 0, 0);
 525          break;
 526
 527       case OPCODE_LIT:
 528          src0 = src_vector(p, &inst->SrcReg[0], program);
 529          tmp = i915_get_utemp(p);
 530
 531          /* tmp = max( a.xyzw, a.00zw )
 532           * XXX: Clamp tmp.w to -128..128
 533           * tmp.y = log(tmp.y)
 534           * tmp.y = tmp.w * tmp.y
 535           * tmp.y = exp(tmp.y)
 536           * result = cmp (a.11-x1, a.1x01, a.1xy1 )
 537           */
 538          i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
 539                          src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
 540
 541          i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
 542                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 543
 544          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
 545                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 546                          swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
 547
 548          i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
 549                          swizzle(tmp, Y, Y, Y, Y), 0, 0);
 550
 551          i915_emit_arith(p, A0_CMP,
 552                          get_result_vector(p, inst),
 553                          get_result_flags(inst), 0,
 554                          negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
 555                          swizzle(tmp, ONE, X, ZERO, ONE),
 556                          swizzle(tmp, ONE, X, Y, ONE));
 557
 558          break;
 559
 560       case OPCODE_LRP:
 561          src0 = src_vector(p, &inst->SrcReg[0], program);
 562          src1 = src_vector(p, &inst->SrcReg[1], program);
 563          src2 = src_vector(p, &inst->SrcReg[2], program);
 564          flags = get_result_flags(inst);
 565          tmp = i915_get_utemp(p);
 566
 567          /* b*a + c*(1-a)
 568           *
 569           * b*a + c - ca
 570           *
 571           * tmp = b*a + c,
 572           * result = (-c)*a + tmp
 573           */
 574          i915_emit_arith(p, A0_MAD, tmp,
 575                          flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
 576
 577          i915_emit_arith(p, A0_MAD,
 578                          get_result_vector(p, inst),
 579                          flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
 580          break;
 581
 582       case OPCODE_MAD:
 583          EMIT_3ARG_ARITH(A0_MAD);
 584          break;
 585
 586       case OPCODE_MAX:
 587          EMIT_2ARG_ARITH(A0_MAX);
 588          break;
 589
 590       case OPCODE_MIN:
 591          src0 = src_vector(p, &inst->SrcReg[0], program);
 592          src1 = src_vector(p, &inst->SrcReg[1], program);
 593          tmp = i915_get_utemp(p);
 594          flags = get_result_flags(inst);
 595
 596          i915_emit_arith(p,
 597                          A0_MAX,
 598                          tmp, flags & A0_DEST_CHANNEL_ALL, 0,
 599                          negate(src0, 1, 1, 1, 1),
 600                          negate(src1, 1, 1, 1, 1), 0);
 601
 602          i915_emit_arith(p,
 603                          A0_MOV,
 604                          get_result_vector(p, inst),
 605                          flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
 606          break;
 607
 608       case OPCODE_MOV:
 609          EMIT_1ARG_ARITH(A0_MOV);
 610          break;
 611
 612       case OPCODE_MUL:
 613          EMIT_2ARG_ARITH(A0_MUL);
 614          break;
 615
 616       case OPCODE_POW:
 617          src0 = src_vector(p, &inst->SrcReg[0], program);
 618          src1 = src_vector(p, &inst->SrcReg[1], program);
 619          tmp = i915_get_utemp(p);
 620          flags = get_result_flags(inst);
 621
 622          /* XXX: masking on intermediate values, here and elsewhere.
 623           */
 624          i915_emit_arith(p,
 625                          A0_LOG,
 626                          tmp, A0_DEST_CHANNEL_X, 0,
 627                          swizzle(src0, X, X, X, X), 0, 0);
 628
 629          i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
 630
 631
 632          i915_emit_arith(p,
 633                          A0_EXP,
 634                          get_result_vector(p, inst),
 635                          flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
 636
 637          break;
 638
 639       case OPCODE_RCP:
 640          src0 = src_vector(p, &inst->SrcReg[0], program);
 641
 642          i915_emit_arith(p,
 643                          A0_RCP,
 644                          get_result_vector(p, inst),
 645                          get_result_flags(inst), 0,
 646                          swizzle(src0, X, X, X, X), 0, 0);
 647          break;
 648
 649       case OPCODE_RSQ:
 650
 651          src0 = src_vector(p, &inst->SrcReg[0], program);
 652
 653          i915_emit_arith(p,
 654                          A0_RSQ,
 655                          get_result_vector(p, inst),
 656                          get_result_flags(inst), 0,
 657                          swizzle(src0, X, X, X, X), 0, 0);
 658          break;
 659
 660       case OPCODE_SCS:
 661          src0 = src_vector(p, &inst->SrcReg[0], program);
 662          tmp = i915_get_utemp(p);
 663
 664          /*
 665           * t0.xy = MUL x.xx11, x.x1111  ; x^2, x, 1, 1
 666           * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
 667           * t1 = MUL t0.xyyw t0.yz11    ; x^7 x^5 x^3 x
 668           * scs.x = DP4 t1, sin_constants
 669           * t1 = MUL t0.xxz1 t0.z111    ; x^6 x^4 x^2 1
 670           * scs.y = DP4 t1, cos_constants
 671           */
 672          i915_emit_arith(p,
 673                          A0_MUL,
 674                          tmp, A0_DEST_CHANNEL_XY, 0,
 675                          swizzle(src0, X, X, ONE, ONE),
 676                          swizzle(src0, X, ONE, ONE, ONE), 0);
 677
 678          i915_emit_arith(p,
 679                          A0_MUL,
 680                          tmp, A0_DEST_CHANNEL_ALL, 0,
 681                          swizzle(tmp, X, Y, X, Y),
 682                          swizzle(tmp, X, X, ONE, ONE), 0);
 683
 684          if (inst->DstReg.WriteMask & WRITEMASK_Y) {
 685             GLuint tmp1;
 686
 687             if (inst->DstReg.WriteMask & WRITEMASK_X)
 688                tmp1 = i915_get_utemp(p);
 689             else
 690                tmp1 = tmp;
 691
 692             i915_emit_arith(p,
 693                             A0_MUL,
 694                             tmp1, A0_DEST_CHANNEL_ALL, 0,
 695                             swizzle(tmp, X, Y, Y, W),
 696                             swizzle(tmp, X, Z, ONE, ONE), 0);
 697
 698             i915_emit_arith(p,
 699                             A0_DP4,
 700                             get_result_vector(p, inst),
 701                             A0_DEST_CHANNEL_Y, 0,
 702                             swizzle(tmp1, W, Z, Y, X),
 703                             i915_emit_const4fv(p, sin_constants), 0);
 704          }
 705
 706          if (inst->DstReg.WriteMask & WRITEMASK_X) {
 707             i915_emit_arith(p,
 708                             A0_MUL,
 709                             tmp, A0_DEST_CHANNEL_XYZ, 0,
 710                             swizzle(tmp, X, X, Z, ONE),
 711                             swizzle(tmp, Z, ONE, ONE, ONE), 0);
 712
 713             i915_emit_arith(p,
 714                             A0_DP4,
 715                             get_result_vector(p, inst),
 716                             A0_DEST_CHANNEL_X, 0,
 717                             swizzle(tmp, ONE, Z, Y, X),
 718                             i915_emit_const4fv(p, cos_constants), 0);
 719          }
 720          break;
 721
 722       case OPCODE_SGE:
 723          EMIT_2ARG_ARITH(A0_SGE);
 724          break;
 725
 726       case OPCODE_SIN:
 727          src0 = src_vector(p, &inst->SrcReg[0], program);
 728          tmp = i915_get_utemp(p);
 729          consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
 730          consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
 731
 732          /* Reduce range from repeating about [-pi,pi] to [-1,1] */
 733          i915_emit_arith(p,
 734                          A0_MAD,
 735                          tmp, A0_DEST_CHANNEL_X, 0,
 736                          src0,
 737                          swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
 738                          swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
 739
 740          i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
 741
 742          i915_emit_arith(p,
 743                          A0_MAD,
 744                          tmp, A0_DEST_CHANNEL_X, 0,
 745                          tmp,
 746                          swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
 747                          swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
 748
 749          /* Compute sin using a quadratic and quartic.  It gives continuity
 750           * that repeating the Taylor series lacks every 2*pi, and has
 751           * reduced error.
 752           *
 753           * The idea was described at:
 754           * http://www.devmaster.net/forums/showthread.php?t=5784
 755           */
 756
 757          /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
 758          i915_emit_arith(p,
 759                          A0_MAX,
 760                          tmp, A0_DEST_CHANNEL_Y, 0,
 761                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 762                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 763                          0);
 764
 765          /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
 766          i915_emit_arith(p,
 767                          A0_MUL,
 768                          tmp, A0_DEST_CHANNEL_Y, 0,
 769                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 770                          tmp,
 771                          0);
 772
 773          /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
 774          i915_emit_arith(p,
 775                          A0_DP3,
 776                          tmp, A0_DEST_CHANNEL_X, 0,
 777                          tmp,
 778                          swizzle(consts1, X, Y, ZERO, ZERO),
 779                          0);
 780
 781          /* tmp.x now contains a first approximation (y).  Now, weight it
 782           * against tmp.y**2 to get closer.
 783           */
 784          i915_emit_arith(p,
 785                          A0_MAX,
 786                          tmp, A0_DEST_CHANNEL_Y, 0,
 787                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 788                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
 789                          0);
 790
 791          /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
 792          i915_emit_arith(p,
 793                          A0_MAD,
 794                          tmp, A0_DEST_CHANNEL_Y, 0,
 795                          swizzle(tmp, ZERO, X, ZERO, ZERO),
 796                          swizzle(tmp, ZERO, Y, ZERO, ZERO),
 797                          negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
 798
 799          /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
 800          i915_emit_arith(p,
 801                          A0_MAD,
 802                          get_result_vector(p, inst),
 803                          get_result_flags(inst), 0,
 804                          swizzle(consts1, W, W, W, W),
 805                          swizzle(tmp, Y, Y, Y, Y),
 806                          swizzle(tmp, X, X, X, X));
 807
 808          break;
 809
 810       case OPCODE_SLT:
 811          EMIT_2ARG_ARITH(A0_SLT);
 812          break;
 813
 814       case OPCODE_SUB:
 815          src0 = src_vector(p, &inst->SrcReg[0], program);
 816          src1 = src_vector(p, &inst->SrcReg[1], program);
 817
 818          i915_emit_arith(p,
 819                          A0_ADD,
 820                          get_result_vector(p, inst),
 821                          get_result_flags(inst), 0,
 822                          src0, negate(src1, 1, 1, 1, 1), 0);
 823          break;
 824
 825       case OPCODE_SWZ:
 826          EMIT_1ARG_ARITH(A0_MOV);       /* extended swizzle handled natively */
 827          break;
 828
 829       case OPCODE_TEX:
 830          EMIT_TEX(T0_TEXLD);
 831          break;
 832
 833       case OPCODE_TXB:
 834          EMIT_TEX(T0_TEXLDB);
 835          break;
 836
 837       case OPCODE_TXP:
 838          EMIT_TEX(T0_TEXLDP);
 839          break;
 840
 841       case OPCODE_XPD:
 842          /* Cross product:
 843           *      result.x = src0.y * src1.z - src0.z * src1.y;
 844           *      result.y = src0.z * src1.x - src0.x * src1.z;
 845           *      result.z = src0.x * src1.y - src0.y * src1.x;
 846           *      result.w = undef;
 847           */
 848          src0 = src_vector(p, &inst->SrcReg[0], program);
 849          src1 = src_vector(p, &inst->SrcReg[1], program);
 850          tmp = i915_get_utemp(p);
 851
 852          i915_emit_arith(p,
 853                          A0_MUL,
 854                          tmp, A0_DEST_CHANNEL_ALL, 0,
 855                          swizzle(src0, Z, X, Y, ONE),
 856                          swizzle(src1, Y, Z, X, ONE), 0);
 857
 858          i915_emit_arith(p,
 859                          A0_MAD,
 860                          get_result_vector(p, inst),
 861                          get_result_flags(inst), 0,
 862                          swizzle(src0, Y, Z, X, ONE),
 863                          swizzle(src1, Z, X, Y, ONE),
 864                          negate(tmp, 1, 1, 1, 0));
 865          break;
 866
 867       case OPCODE_END:
 868          return;
 869
 870       default:
 871          i915_program_error(p, "bad opcode");
 872          return;
 873       }
 874
 875       inst++;
 876       i915_release_utemps(p);
 877    }
 878 }
 879
 880 /* Rather than trying to intercept and jiggle depth writes during
 881  * emit, just move the value into its correct position at the end of
 882  * the program:
 883  */
 884 static void
 885 fixup_depth_write(struct i915_fragment_program *p)
 886 {
 887    if (p->depth_written) {
 888       GLuint depth = UREG(REG_TYPE_OD, 0);
 889
 890       i915_emit_arith(p,
 891                       A0_MOV,
 892                       depth, A0_DEST_CHANNEL_W, 0,
 893                       swizzle(depth, X, Y, Z, Z), 0, 0);
 894    }
 895 }
 896
 897
 898 static void
 899 check_wpos(struct i915_fragment_program *p)
 900 {
 901    GLuint inputs = p->FragProg.Base.InputsRead;
 902    GLint i;
 903
 904    p->wpos_tex = -1;
 905
 906    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
 907       if (inputs & FRAG_BIT_TEX(i))
 908          continue;
 909       else if (inputs & FRAG_BIT_WPOS) {
 910          p->wpos_tex = i;
 911          inputs &= ~FRAG_BIT_WPOS;
 912       }
 913    }
 914
 915    if (inputs & FRAG_BIT_WPOS) {
 916       i915_program_error(p, "No free texcoord for wpos value");
 917    }
 918 }
 919
 920
 921 static void
 922 translate_program(struct i915_fragment_program *p)
 923 {
 924    struct i915_context *i915 = I915_CONTEXT(p->ctx);
 925
 926    i915_init_program(i915, p);
 927    check_wpos(p);
 928    upload_program(p);
 929    fixup_depth_write(p);
 930    i915_fini_program(p);
 931
 932    p->translated = 1;
 933 }
 934
 935
 936 static void
 937 track_params(struct i915_fragment_program *p)
 938 {
 939    GLint i;
 940
 941    if (p->nr_params)
 942       _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
 943
 944    for (i = 0; i < p->nr_params; i++) {
 945       GLint reg = p->param[i].reg;
 946       COPY_4V(p->constant[reg], p->param[i].values);
 947    }
 948
 949    p->params_uptodate = 1;
 950    p->on_hardware = 0;          /* overkill */
 951 }
 952
 953
 954 static void
 955 i915BindProgram(GLcontext * ctx, GLenum target, struct gl_program *prog)
 956 {
 957    if (target == GL_FRAGMENT_PROGRAM_ARB) {
 958       struct i915_context *i915 = I915_CONTEXT(ctx);
 959       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
 960
 961       if (i915->current_program == p)
 962          return;
 963
 964       if (i915->current_program) {
 965          i915->current_program->on_hardware = 0;
 966          i915->current_program->params_uptodate = 0;
 967       }
 968
 969       i915->current_program = p;
 970
 971       assert(p->on_hardware == 0);
 972       assert(p->params_uptodate == 0);
 973
 974    }
 975 }
 976
 977 static struct gl_program *
 978 i915NewProgram(GLcontext * ctx, GLenum target, GLuint id)
 979 {
 980    switch (target) {
 981    case GL_VERTEX_PROGRAM_ARB:
 982       return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
 983                                        target, id);
 984
 985    case GL_FRAGMENT_PROGRAM_ARB:{
 986          struct i915_fragment_program *prog =
 987             CALLOC_STRUCT(i915_fragment_program);
 988          if (prog) {
 989             i915_init_program(I915_CONTEXT(ctx), prog);
 990
 991             return _mesa_init_fragment_program(ctx, &prog->FragProg,
 992                                                target, id);
 993          }
 994          else
 995             return NULL;
 996       }
 997
 998    default:
 999       /* Just fallback:
1000        */
1001       return _mesa_new_program(ctx, target, id);
1002    }
1003 }
1004
1005 static void
1006 i915DeleteProgram(GLcontext * ctx, struct gl_program *prog)
1007 {
1008    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1009       struct i915_context *i915 = I915_CONTEXT(ctx);
1010       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1011
1012       if (i915->current_program == p)
1013          i915->current_program = 0;
1014    }
1015
1016    _mesa_delete_program(ctx, prog);
1017 }
1018
1019
1020 static GLboolean
1021 i915IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
1022 {
1023    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1024       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1025
1026       if (!p->translated)
1027          translate_program(p);
1028
1029       return !p->error;
1030    }
1031    else
1032       return GL_TRUE;
1033 }
1034
1035 static void
1036 i915ProgramStringNotify(GLcontext * ctx,
1037                         GLenum target, struct gl_program *prog)
1038 {
1039    if (target == GL_FRAGMENT_PROGRAM_ARB) {
1040       struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1041       p->translated = 0;
1042
1043       /* Hack: make sure fog is correctly enabled according to this
1044        * fragment program's fog options.
1045        */
1046       if (p->FragProg.FogOption) {
1047          /* add extra instructions to do fog, then turn off FogOption field */
1048          _mesa_append_fog_code(ctx, &p->FragProg);
1049          p->FragProg.FogOption = GL_NONE;
1050       }
1051    }
1052
1053    _tnl_program_string(ctx, target, prog);
1054 }
1055
1056
1057 void
1058 i915ValidateFragmentProgram(struct i915_context *i915)
1059 {
1060    GLcontext *ctx = &i915->intel.ctx;
1061    struct intel_context *intel = intel_context(ctx);
1062    TNLcontext *tnl = TNL_CONTEXT(ctx);
1063    struct vertex_buffer *VB = &tnl->vb;
1064
1065    struct i915_fragment_program *p =
1066       (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1067
1068    const GLuint inputsRead = p->FragProg.Base.InputsRead;
1069    GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1070    GLuint s2 = S2_TEXCOORD_NONE;
1071    int i, offset = 0;
1072
1073    if (i915->current_program != p) {
1074       if (i915->current_program) {
1075          i915->current_program->on_hardware = 0;
1076          i915->current_program->params_uptodate = 0;
1077       }
1078
1079       i915->current_program = p;
1080    }
1081
1082
1083    /* Important:
1084     */
1085    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1086
1087    if (!p->translated)
1088       translate_program(p);
1089
1090    intel->vertex_attr_count = 0;
1091    intel->wpos_offset = 0;
1092    intel->wpos_size = 0;
1093    intel->coloroffset = 0;
1094    intel->specoffset = 0;
1095
1096    if (inputsRead & FRAG_BITS_TEX_ANY) {
1097       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1098    }
1099    else {
1100       EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1101    }
1102
1103    if (inputsRead & FRAG_BIT_COL0) {
1104       intel->coloroffset = offset / 4;
1105       EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1106    }
1107
1108    if ((inputsRead & (FRAG_BIT_COL1 | FRAG_BIT_FOGC)) ||
1109        i915->vertex_fog != I915_FOG_NONE) {
1110
1111       if (inputsRead & FRAG_BIT_COL1) {
1112          intel->specoffset = offset / 4;
1113          EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, S4_VFMT_SPEC_FOG, 3);
1114       }
1115       else
1116          EMIT_PAD(3);
1117
1118       if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE)
1119          EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1UB_1F, S4_VFMT_SPEC_FOG, 1);
1120       else
1121          EMIT_PAD(1);
1122    }
1123
1124    /* XXX this was disabled, but enabling this code helped fix the Glean
1125     * tfragprog1 fog tests.
1126     */
1127 #if 1
1128    if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1129       EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1130    }
1131 #endif
1132
1133    for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1134       if (inputsRead & FRAG_BIT_TEX(i)) {
1135          int sz = VB->TexCoordPtr[i]->size;
1136
1137          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1138          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1139
1140          EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1141       }
1142       else if (i == p->wpos_tex) {
1143
1144          /* If WPOS is required, duplicate the XYZ position data in an
1145           * unused texture coordinate:
1146           */
1147          s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1148          s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1149
1150          intel->wpos_offset = offset;
1151          intel->wpos_size = 3 * sizeof(GLuint);
1152
1153          EMIT_PAD(intel->wpos_size);
1154       }
1155    }
1156
1157    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1158        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1159       int k;
1160
1161       I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1162
1163       /* Must do this *after* statechange, so as not to affect
1164        * buffered vertices reliant on the old state:
1165        */
1166       intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1167                                               intel->vertex_attrs,
1168                                               intel->vertex_attr_count,
1169                                               intel->ViewportMatrix.m, 0);
1170
1171       intel->vertex_size >>= 2;
1172
1173       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1174       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1175
1176       k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1177       assert(k);
1178    }
1179
1180    if (!p->params_uptodate)
1181       track_params(p);
1182
1183    if (!p->on_hardware)
1184       i915_upload_program(i915, p);
1185 }
1186
1187 void
1188 i915InitFragProgFuncs(struct dd_function_table *functions)
1189 {
1190    functions->BindProgram = i915BindProgram;
1191    functions->NewProgram = i915NewProgram;
1192    functions->DeleteProgram = i915DeleteProgram;
1193    functions->IsProgramNative = i915IsProgramNative;
1194    functions->ProgramStringNotify = i915ProgramStringNotify;
1195 }