src/gallium/auxiliary/tgsi/tgsi_point_sprite.c

   1 /*
   2  * Copyright 2014 VMware, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sub license, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the
  14  * next paragraph) shall be included in all copies or substantial portions
  15  * of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  20  * IN NO EVENT SHALL THE AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR
  21  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  22  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  23  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  */
  25
  26
  27 /**
  28  * This utility transforms the geometry shader to emulate point sprite by
  29  * drawing a quad. It also adds an extra output for the original point position
  30  * if the point position is to be written to a stream output buffer.
  31  * Note: It assumes the driver will add a constant for the inverse viewport
  32  *       after the user defined constants.
  33  */
  34
  35 #include "util/u_debug.h"
  36 #include "util/u_math.h"
  37 #include "tgsi_info.h"
  38 #include "tgsi_point_sprite.h"
  39 #include "tgsi_transform.h"
  40 #include "pipe/p_state.h"
  41
  42 #define INVALID_INDEX 9999
  43
  44 /* Set swizzle based on the immediates (0, 1, 0, -1) */
  45 static inline unsigned
  46 set_swizzle(int x, int y, int z, int w)
  47 {
  48    static const unsigned map[3] = {TGSI_SWIZZLE_W, TGSI_SWIZZLE_X,
  49                                    TGSI_SWIZZLE_Y};
  50    assert(x >= -1);
  51    assert(x <= 1);
  52    assert(y >= -1);
  53    assert(y <= 1);
  54    assert(z >= -1);
  55    assert(z <= 1);
  56    assert(w >= -1);
  57    assert(w <= 1);
  58
  59    return map[x+1] | (map[y+1] << 2) | (map[z+1] << 4) | (map[w+1] << 6);
  60 }
  61
  62 static inline unsigned
  63 get_swizzle(unsigned swizzle, unsigned component)
  64 {
  65    assert(component < 4);
  66    return (swizzle >> (component * 2)) & 0x3;
  67 }
  68
  69 struct psprite_transform_context
  70 {
  71    struct tgsi_transform_context base;
  72    unsigned num_tmp;
  73    unsigned num_out;
  74    unsigned num_orig_out;
  75    unsigned num_const;
  76    unsigned num_imm;
  77    unsigned point_size_in;          // point size input
  78    unsigned point_size_out;         // point size output
  79    unsigned point_size_tmp;         // point size temp
  80    unsigned point_pos_in;           // point pos input
  81    unsigned point_pos_out;          // point pos output
  82    unsigned point_pos_sout;         // original point pos for streamout
  83    unsigned point_pos_tmp;          // point pos temp
  84    unsigned point_scale_tmp;        // point scale temp
  85    unsigned point_color_out;        // point color output
  86    unsigned point_color_tmp;        // point color temp
  87    unsigned point_imm;              // point immediates
  88    unsigned point_ivp;              // point inverseViewport constant
  89    unsigned point_dir_swz[4];       // point direction swizzle
  90    unsigned point_coord_swz[4];     // point coord swizzle
  91    unsigned point_coord_enable;     // point coord enable mask
  92    unsigned point_coord_decl;       // point coord output declared mask
  93    unsigned point_coord_out;        // point coord output starting index
  94    unsigned point_coord_aa;         // aa point coord semantic index
  95    unsigned point_coord_k;          // aa point coord threshold distance
  96    unsigned stream_out_point_pos:1; // set if to stream out original point pos
  97    unsigned aa_point:1;             // set if doing aa point
  98    unsigned out_tmp_index[PIPE_MAX_SHADER_OUTPUTS];
  99    int max_generic;                 // max generic semantic index
 100 };
 101
 102 static inline struct psprite_transform_context *
 103 psprite_transform_context(struct tgsi_transform_context *ctx)
 104 {
 105    return (struct psprite_transform_context *) ctx;
 106 }
 107
 108
 109 /**
 110  * TGSI declaration transform callback.
 111  */
 112 static void
 113 psprite_decl(struct tgsi_transform_context *ctx,
 114              struct tgsi_full_declaration *decl)
 115 {
 116    struct psprite_transform_context *ts = psprite_transform_context(ctx);
 117
 118    if (decl->Declaration.File == TGSI_FILE_INPUT) {
 119       if (decl->Semantic.Name == TGSI_SEMANTIC_PSIZE) {
 120          ts->point_size_in = decl->Range.First;
 121       }
 122       else if (decl->Semantic.Name == TGSI_SEMANTIC_POSITION) {
 123          ts->point_pos_in = decl->Range.First;
 124       }
 125    }
 126    else if (decl->Declaration.File == TGSI_FILE_OUTPUT) {
 127       if (decl->Semantic.Name == TGSI_SEMANTIC_PSIZE) {
 128          ts->point_size_out = decl->Range.First;
 129       }
 130       else if (decl->Semantic.Name == TGSI_SEMANTIC_POSITION) {
 131          ts->point_pos_out = decl->Range.First;
 132       }
 133       else if (decl->Semantic.Name == TGSI_SEMANTIC_GENERIC &&
 134                decl->Semantic.Index < 32) {
 135          ts->point_coord_decl |= 1 << decl->Semantic.Index;
 136          ts->max_generic = MAX2(ts->max_generic, (int)decl->Semantic.Index);
 137       }
 138       ts->num_out = MAX2(ts->num_out, decl->Range.Last + 1);
 139    }
 140    else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
 141       ts->num_tmp = MAX2(ts->num_tmp, decl->Range.Last + 1);
 142    }
 143    else if (decl->Declaration.File == TGSI_FILE_CONSTANT) {
 144       ts->num_const = MAX2(ts->num_const, decl->Range.Last + 1);
 145    }
 146
 147    ctx->emit_declaration(ctx, decl);
 148 }
 149
 150 /**
 151  * TGSI immediate declaration transform callback.
 152  */
 153 static void
 154 psprite_immediate(struct tgsi_transform_context *ctx,
 155                   struct tgsi_full_immediate *imm)
 156 {
 157    struct psprite_transform_context *ts = psprite_transform_context(ctx);
 158
 159    ctx->emit_immediate(ctx, imm);
 160    ts->num_imm++;
 161 }
 162
 163
 164 /**
 165  * TGSI transform prolog callback.
 166  */
 167 static void
 168 psprite_prolog(struct tgsi_transform_context *ctx)
 169 {
 170    struct psprite_transform_context *ts = psprite_transform_context(ctx);
 171    unsigned point_coord_enable, en;
 172    int i;
 173
 174    /* Replace output registers with temporary registers */
 175    for (i = 0; i < ts->num_out; i++) {
 176       ts->out_tmp_index[i] = ts->num_tmp++;
 177    }
 178    ts->num_orig_out = ts->num_out;
 179
 180    /* Declare a tmp register for point scale */
 181    ts->point_scale_tmp = ts->num_tmp++;
 182
 183    if (ts->point_size_out != INVALID_INDEX)
 184       ts->point_size_tmp = ts->out_tmp_index[ts->point_size_out];
 185    else
 186       ts->point_size_tmp = ts->num_tmp++;
 187
 188    assert(ts->point_pos_out != INVALID_INDEX);
 189    ts->point_pos_tmp = ts->out_tmp_index[ts->point_pos_out];
 190    ts->out_tmp_index[ts->point_pos_out] = INVALID_INDEX;
 191
 192    /* Declare one more tmp register for point coord threshold distance
 193     * if we are generating anti-aliased point.
 194     */
 195    if (ts->aa_point)
 196       ts->point_coord_k = ts->num_tmp++;
 197
 198    tgsi_transform_temps_decl(ctx, ts->point_size_tmp, ts->num_tmp-1);
 199
 200    /* Declare an extra output for the original point position for stream out */
 201    if (ts->stream_out_point_pos) {
 202       ts->point_pos_sout = ts->num_out++;
 203       tgsi_transform_output_decl(ctx, ts->point_pos_sout,
 204                                  TGSI_SEMANTIC_GENERIC, 0, 0);
 205    }
 206
 207    /* point coord outputs to be declared */
 208    point_coord_enable = ts->point_coord_enable & ~ts->point_coord_decl;
 209
 210    /* Declare outputs for those point coord that are enabled but are not
 211     * already declared in this shader.
 212     */
 213    ts->point_coord_out = ts->num_out;
 214    if (point_coord_enable) {
 215       for (i = 0, en = point_coord_enable; en; en>>=1, i++) {
 216          if (en & 0x1) {
 217             tgsi_transform_output_decl(ctx, ts->num_out++,
 218                                        TGSI_SEMANTIC_GENERIC, i, 0);
 219             ts->max_generic = MAX2(ts->max_generic, (int)i);
 220          }
 221       }
 222    }
 223
 224    /* add an extra generic output for aa point texcoord */
 225    if (ts->aa_point) {
 226       ts->point_coord_aa = ts->max_generic + 1;
 227       assert((ts->point_coord_enable & (1 << ts->point_coord_aa)) == 0);
 228       ts->point_coord_enable |= 1 << (ts->point_coord_aa);
 229       tgsi_transform_output_decl(ctx, ts->num_out++, TGSI_SEMANTIC_GENERIC,
 230                                  ts->point_coord_aa, 0);
 231    }
 232
 233    /* Declare extra immediates */
 234    ts->point_imm = ts->num_imm;
 235    tgsi_transform_immediate_decl(ctx, 0, 1, 0.5, -1);
 236
 237    /* Declare point constant -
 238     * constant.xy -- inverseViewport
 239     * constant.z -- current point size
 240     * constant.w -- max point size
 241     * The driver needs to add this constant to the constant buffer
 242     */
 243    ts->point_ivp = ts->num_const++;
 244    tgsi_transform_const_decl(ctx, ts->point_ivp, ts->point_ivp);
 245
 246    /* If this geometry shader does not specify point size,
 247     * get the current point size from the point constant.
 248     */
 249    if (ts->point_size_out == INVALID_INDEX) {
 250       struct tgsi_full_instruction inst;
 251
 252       inst = tgsi_default_full_instruction();
 253       inst.Instruction.Opcode = TGSI_OPCODE_MOV;
 254       inst.Instruction.NumDstRegs = 1;
 255       tgsi_transform_dst_reg(&inst.Dst[0], TGSI_FILE_TEMPORARY,
 256                              ts->point_size_tmp, TGSI_WRITEMASK_XYZW);
 257       inst.Instruction.NumSrcRegs = 1;
 258       tgsi_transform_src_reg(&inst.Src[0], TGSI_FILE_CONSTANT,
 259                              ts->point_ivp, TGSI_SWIZZLE_Z,
 260                              TGSI_SWIZZLE_Z, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_Z);
 261       ctx->emit_instruction(ctx, &inst);
 262    }
 263 }
 264
 265
 266 /**
 267  * Add the point sprite emulation instructions at the emit vertex instruction
 268  */
 269 static void
 270 psprite_emit_vertex_inst(struct tgsi_transform_context *ctx,
 271                          struct tgsi_full_instruction *vert_inst)
 272 {
 273    struct psprite_transform_context *ts = psprite_transform_context(ctx);
 274    struct tgsi_full_instruction inst;
 275    unsigned point_coord_enable, en;
 276    unsigned i, j, s;
 277
 278    /* new point coord outputs */
 279    point_coord_enable = ts->point_coord_enable & ~ts->point_coord_decl;
 280
 281    /* OUTPUT[pos_sout] = TEMP[pos] */
 282    if (ts->point_pos_sout != INVALID_INDEX) {
 283       tgsi_transform_op1_inst(ctx, TGSI_OPCODE_MOV,
 284                               TGSI_FILE_OUTPUT, ts->point_pos_sout,
 285                               TGSI_WRITEMASK_XYZW,
 286                               TGSI_FILE_TEMPORARY, ts->point_pos_tmp);
 287    }
 288
 289    /**
 290     * Set up the point scale vector
 291     * scale = pointSize * pos.w * inverseViewport
 292     */
 293
 294    /* MUL point_scale.x, point_size.x, point_pos.w */
 295    tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MUL,
 296                   TGSI_FILE_TEMPORARY, ts->point_scale_tmp, TGSI_WRITEMASK_X,
 297                   TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_SWIZZLE_X,
 298                   TGSI_FILE_TEMPORARY, ts->point_pos_tmp, TGSI_SWIZZLE_W, false);
 299
 300    /* MUL point_scale.xy, point_scale.xx, inverseViewport.xy */
 301    inst = tgsi_default_full_instruction();
 302    inst.Instruction.Opcode = TGSI_OPCODE_MUL;
 303    inst.Instruction.NumDstRegs = 1;
 304    tgsi_transform_dst_reg(&inst.Dst[0], TGSI_FILE_TEMPORARY,
 305                           ts->point_scale_tmp, TGSI_WRITEMASK_XY);
 306    inst.Instruction.NumSrcRegs = 2;
 307    tgsi_transform_src_reg(&inst.Src[0], TGSI_FILE_TEMPORARY,
 308                           ts->point_scale_tmp, TGSI_SWIZZLE_X,
 309                           TGSI_SWIZZLE_X, TGSI_SWIZZLE_X, TGSI_SWIZZLE_X);
 310    tgsi_transform_src_reg(&inst.Src[1], TGSI_FILE_CONSTANT,
 311                           ts->point_ivp, TGSI_SWIZZLE_X,
 312                           TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_Z);
 313    ctx->emit_instruction(ctx, &inst);
 314
 315    /**
 316     * Set up the point coord threshold distance
 317     * k = 0.5 - 1 / pointsize
 318     */
 319    if (ts->aa_point) {
 320       tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_DIV,
 321                                   TGSI_FILE_TEMPORARY, ts->point_coord_k,
 322                                   TGSI_WRITEMASK_X,
 323                                   TGSI_FILE_IMMEDIATE, ts->point_imm,
 324                                   TGSI_SWIZZLE_Y,
 325                                   TGSI_FILE_TEMPORARY, ts->point_size_tmp,
 326                                   TGSI_SWIZZLE_X, false);
 327
 328       tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_ADD,
 329                                   TGSI_FILE_TEMPORARY, ts->point_coord_k,
 330                                   TGSI_WRITEMASK_X,
 331                                   TGSI_FILE_IMMEDIATE, ts->point_imm,
 332                                   TGSI_SWIZZLE_Z,
 333                                   TGSI_FILE_TEMPORARY, ts->point_coord_k,
 334                                   TGSI_SWIZZLE_X, true);
 335    }
 336
 337
 338    for (i = 0; i < 4; i++) {
 339       unsigned point_dir_swz = ts->point_dir_swz[i];
 340       unsigned point_coord_swz = ts->point_coord_swz[i];
 341
 342       /* All outputs need to be emitted for each vertex */
 343       for (j = 0; j < ts->num_orig_out; j++) {
 344          if (ts->out_tmp_index[j] != INVALID_INDEX) {
 345             tgsi_transform_op1_inst(ctx, TGSI_OPCODE_MOV,
 346                                     TGSI_FILE_OUTPUT, j,
 347                                     TGSI_WRITEMASK_XYZW,
 348                                     TGSI_FILE_TEMPORARY, ts->out_tmp_index[j]);
 349          }
 350       }
 351
 352       /* pos = point_scale * point_dir + point_pos */
 353       inst = tgsi_default_full_instruction();
 354       inst.Instruction.Opcode = TGSI_OPCODE_MAD;
 355       inst.Instruction.NumDstRegs = 1;
 356       tgsi_transform_dst_reg(&inst.Dst[0], TGSI_FILE_OUTPUT, ts->point_pos_out,
 357                              TGSI_WRITEMASK_XYZW);
 358       inst.Instruction.NumSrcRegs = 3;
 359       tgsi_transform_src_reg(&inst.Src[0], TGSI_FILE_TEMPORARY, ts->point_scale_tmp,
 360                              TGSI_SWIZZLE_X, TGSI_SWIZZLE_Y, TGSI_SWIZZLE_X,
 361                              TGSI_SWIZZLE_X);
 362       tgsi_transform_src_reg(&inst.Src[1], TGSI_FILE_IMMEDIATE, ts->point_imm,
 363                              get_swizzle(point_dir_swz, 0),
 364                              get_swizzle(point_dir_swz, 1),
 365                              get_swizzle(point_dir_swz, 2),
 366                              get_swizzle(point_dir_swz, 3));
 367       tgsi_transform_src_reg(&inst.Src[2], TGSI_FILE_TEMPORARY, ts->point_pos_tmp,
 368                              TGSI_SWIZZLE_X, TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Z,
 369                              TGSI_SWIZZLE_W);
 370       ctx->emit_instruction(ctx, &inst);
 371
 372       /* point coord */
 373       for (j = 0, s = 0, en = point_coord_enable; en; en>>=1, s++) {
 374          unsigned dstReg;
 375
 376          if (en & 0x1) {
 377             dstReg = ts->point_coord_out + j;
 378
 379             inst = tgsi_default_full_instruction();
 380             inst.Instruction.Opcode = TGSI_OPCODE_MOV;
 381             inst.Instruction.NumDstRegs = 1;
 382             tgsi_transform_dst_reg(&inst.Dst[0], TGSI_FILE_OUTPUT,
 383                                    dstReg, TGSI_WRITEMASK_XYZW);
 384             inst.Instruction.NumSrcRegs = 1;
 385             tgsi_transform_src_reg(&inst.Src[0], TGSI_FILE_IMMEDIATE, ts->point_imm,
 386                                    get_swizzle(point_coord_swz, 0),
 387                                    get_swizzle(point_coord_swz, 1),
 388                                    get_swizzle(point_coord_swz, 2),
 389                                    get_swizzle(point_coord_swz, 3));
 390             ctx->emit_instruction(ctx, &inst);
 391
 392             /* MOV point_coord.z  point_coord_k.x */
 393             if (s == ts->point_coord_aa) {
 394                tgsi_transform_op1_swz_inst(ctx, TGSI_OPCODE_MOV,
 395                                            TGSI_FILE_OUTPUT, dstReg, TGSI_WRITEMASK_Z,
 396                                            TGSI_FILE_TEMPORARY, ts->point_coord_k,
 397                                            TGSI_SWIZZLE_X);
 398             }
 399             j++;  /* the next point coord output offset */
 400          }
 401       }
 402
 403       /* Emit the EMIT instruction for each vertex of the quad */
 404       ctx->emit_instruction(ctx, vert_inst);
 405    }
 406
 407    /* Emit the ENDPRIM instruction for the quad */
 408    inst = tgsi_default_full_instruction();
 409    inst.Instruction.Opcode = TGSI_OPCODE_ENDPRIM;
 410    inst.Instruction.NumDstRegs = 0;
 411    inst.Instruction.NumSrcRegs = 1;
 412    inst.Src[0] = vert_inst->Src[0];
 413    ctx->emit_instruction(ctx, &inst);
 414 }
 415
 416
 417 /**
 418  * TGSI instruction transform callback.
 419  */
 420 static void
 421 psprite_inst(struct tgsi_transform_context *ctx,
 422              struct tgsi_full_instruction *inst)
 423 {
 424    struct psprite_transform_context *ts = psprite_transform_context(ctx);
 425
 426    if (inst->Instruction.Opcode == TGSI_OPCODE_EMIT) {
 427       psprite_emit_vertex_inst(ctx, inst);
 428    }
 429    else if (inst->Dst[0].Register.File == TGSI_FILE_OUTPUT &&
 430             inst->Dst[0].Register.Index == ts->point_size_out) {
 431       /**
 432        * Replace point size output reg with tmp reg.
 433        * The tmp reg will be later used as a src reg for computing
 434        * the point scale factor.
 435        */
 436       inst->Dst[0].Register.File = TGSI_FILE_TEMPORARY;
 437       inst->Dst[0].Register.Index = ts->point_size_tmp;
 438       ctx->emit_instruction(ctx, inst);
 439
 440       /* Clamp the point size */
 441       /* MAX point_size_tmp.x, point_size_tmp.x, point_imm.y */
 442       tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MAX,
 443                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_WRITEMASK_X,
 444                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_SWIZZLE_X,
 445                  TGSI_FILE_IMMEDIATE, ts->point_imm, TGSI_SWIZZLE_Y, false);
 446
 447       /* MIN point_size_tmp.x, point_size_tmp.x, point_ivp.w */
 448       tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MIN,
 449                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_WRITEMASK_X,
 450                  TGSI_FILE_TEMPORARY, ts->point_size_tmp, TGSI_SWIZZLE_X,
 451                  TGSI_FILE_CONSTANT, ts->point_ivp, TGSI_SWIZZLE_W, false);
 452    }
 453    else if (inst->Dst[0].Register.File == TGSI_FILE_OUTPUT &&
 454             inst->Dst[0].Register.Index == ts->point_pos_out) {
 455       /**
 456        * Replace point pos output reg with tmp reg.
 457        */
 458       inst->Dst[0].Register.File = TGSI_FILE_TEMPORARY;
 459       inst->Dst[0].Register.Index = ts->point_pos_tmp;
 460       ctx->emit_instruction(ctx, inst);
 461    }
 462    else if (inst->Dst[0].Register.File == TGSI_FILE_OUTPUT) {
 463       /**
 464        * Replace output reg with tmp reg.
 465        */
 466       inst->Dst[0].Register.File = TGSI_FILE_TEMPORARY;
 467       inst->Dst[0].Register.Index = ts->out_tmp_index[inst->Dst[0].Register.Index];
 468       ctx->emit_instruction(ctx, inst);
 469    }
 470    else {
 471       ctx->emit_instruction(ctx, inst);
 472    }
 473 }
 474
 475
 476 /**
 477  * TGSI property instruction transform callback.
 478  * Transforms a point into a 4-vertex triangle strip.
 479  */
 480 static void
 481 psprite_property(struct tgsi_transform_context *ctx,
 482                  struct tgsi_full_property *prop)
 483 {
 484    switch (prop->Property.PropertyName) {
 485    case TGSI_PROPERTY_GS_OUTPUT_PRIM:
 486        prop->u[0].Data = PIPE_PRIM_TRIANGLE_STRIP;
 487        break;
 488    case TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES:
 489        prop->u[0].Data *= 4;
 490        break;
 491    default:
 492        break;
 493    }
 494    ctx->emit_property(ctx, prop);
 495 }
 496
 497 /**
 498  * TGSI utility to transform a geometry shader to support point sprite.
 499  */
 500 struct tgsi_token *
 501 tgsi_add_point_sprite(const struct tgsi_token *tokens_in,
 502                       const unsigned point_coord_enable,
 503                       const bool sprite_origin_lower_left,
 504                       const bool stream_out_point_pos,
 505                       int *aa_point_coord_index)
 506 {
 507    struct psprite_transform_context transform;
 508    const uint num_new_tokens = 200; /* should be enough */
 509    const uint new_len = tgsi_num_tokens(tokens_in) + num_new_tokens;
 510    struct tgsi_token *new_tokens;
 511
 512    /* setup transformation context */
 513    memset(&transform, 0, sizeof(transform));
 514    transform.base.transform_declaration = psprite_decl;
 515    transform.base.transform_instruction = psprite_inst;
 516    transform.base.transform_property = psprite_property;
 517    transform.base.transform_immediate = psprite_immediate;
 518    transform.base.prolog = psprite_prolog;
 519
 520    transform.point_size_in = INVALID_INDEX;
 521    transform.point_size_out = INVALID_INDEX;
 522    transform.point_size_tmp = INVALID_INDEX;
 523    transform.point_pos_in = INVALID_INDEX;
 524    transform.point_pos_out = INVALID_INDEX;
 525    transform.point_pos_sout = INVALID_INDEX;
 526    transform.point_pos_tmp = INVALID_INDEX;
 527    transform.point_scale_tmp = INVALID_INDEX;
 528    transform.point_imm = INVALID_INDEX;
 529    transform.point_coord_aa = INVALID_INDEX;
 530    transform.point_coord_k = INVALID_INDEX;
 531
 532    transform.stream_out_point_pos = stream_out_point_pos;
 533    transform.point_coord_enable = point_coord_enable;
 534    transform.aa_point = aa_point_coord_index != NULL;
 535    transform.max_generic = -1;
 536
 537    /* point sprite directions based on the immediates (0, 1, 0.5, -1) */
 538    /* (-1, -1, 0, 0) */
 539    transform.point_dir_swz[0] = set_swizzle(-1, -1, 0, 0);
 540    /* (-1, 1, 0, 0) */
 541    transform.point_dir_swz[1] = set_swizzle(-1, 1, 0, 0);
 542    /* (1, -1, 0, 0) */
 543    transform.point_dir_swz[2] = set_swizzle(1, -1, 0, 0);
 544    /* (1, 1, 0, 0) */
 545    transform.point_dir_swz[3] = set_swizzle(1, 1, 0, 0);
 546
 547    /* point coord based on the immediates (0, 1, 0, -1) */
 548    if (sprite_origin_lower_left) {
 549       /* (0, 0, 0, 1) */
 550       transform.point_coord_swz[0] = set_swizzle(0, 0, 0, 1);
 551       /* (0, 1, 0, 1) */
 552       transform.point_coord_swz[1] = set_swizzle(0, 1, 0, 1);
 553       /* (1, 0, 0, 1) */
 554       transform.point_coord_swz[2] = set_swizzle(1, 0, 0, 1);
 555       /* (1, 1, 0, 1) */
 556       transform.point_coord_swz[3] = set_swizzle(1, 1, 0, 1);
 557    }
 558    else {
 559       /* (0, 1, 0, 1) */
 560       transform.point_coord_swz[0] = set_swizzle(0, 1, 0, 1);
 561       /* (0, 0, 0, 1) */
 562       transform.point_coord_swz[1] = set_swizzle(0, 0, 0, 1);
 563       /* (1, 1, 0, 1) */
 564       transform.point_coord_swz[2] = set_swizzle(1, 1, 0, 1);
 565       /* (1, 0, 0, 1) */
 566       transform.point_coord_swz[3] = set_swizzle(1, 0, 0, 1);
 567    }
 568
 569
 570    /* allocate new tokens buffer */
 571    new_tokens = tgsi_alloc_tokens(new_len);
 572    if (!new_tokens)
 573       return NULL;
 574
 575    /* transform the shader */
 576    tgsi_transform_shader(tokens_in, new_tokens, new_len, &transform.base);
 577
 578    if (aa_point_coord_index)
 579       *aa_point_coord_index = transform.point_coord_aa;
 580
 581    return new_tokens;
 582 }