src/mesa/state_tracker/st_program.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 VMware, Inc.
   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 VMWARE 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   * Authors:
  29   *   Keith Whitwell <keithw@vmware.com>
  30   *   Brian Paul
  31   */
  32
  33
  34 #include "main/errors.h"
  35 #include "main/imports.h"
  36 #include "main/hash.h"
  37 #include "main/mtypes.h"
  38 #include "program/prog_parameter.h"
  39 #include "program/prog_print.h"
  40 #include "program/prog_to_nir.h"
  41 #include "program/programopt.h"
  42
  43 #include "compiler/nir/nir.h"
  44
  45 #include "pipe/p_context.h"
  46 #include "pipe/p_defines.h"
  47 #include "pipe/p_shader_tokens.h"
  48 #include "draw/draw_context.h"
  49 #include "tgsi/tgsi_dump.h"
  50 #include "tgsi/tgsi_emulate.h"
  51 #include "tgsi/tgsi_parse.h"
  52 #include "tgsi/tgsi_ureg.h"
  53
  54 #include "st_debug.h"
  55 #include "st_cb_bitmap.h"
  56 #include "st_cb_drawpixels.h"
  57 #include "st_context.h"
  58 #include "st_tgsi_lower_depth_clamp.h"
  59 #include "st_tgsi_lower_yuv.h"
  60 #include "st_program.h"
  61 #include "st_mesa_to_tgsi.h"
  62 #include "st_atifs_to_tgsi.h"
  63 #include "st_nir.h"
  64 #include "st_shader_cache.h"
  65 #include "cso_cache/cso_context.h"
  66
  67
  68
  69 static void
  70 set_affected_state_flags(uint64_t *states,
  71                          struct gl_program *prog,
  72                          uint64_t new_constants,
  73                          uint64_t new_sampler_views,
  74                          uint64_t new_samplers,
  75                          uint64_t new_images,
  76                          uint64_t new_ubos,
  77                          uint64_t new_ssbos,
  78                          uint64_t new_atomics)
  79 {
  80    if (prog->Parameters->NumParameters)
  81       *states |= new_constants;
  82
  83    if (prog->info.num_textures)
  84       *states |= new_sampler_views | new_samplers;
  85
  86    if (prog->info.num_images)
  87       *states |= new_images;
  88
  89    if (prog->info.num_ubos)
  90       *states |= new_ubos;
  91
  92    if (prog->info.num_ssbos)
  93       *states |= new_ssbos;
  94
  95    if (prog->info.num_abos)
  96       *states |= new_atomics;
  97 }
  98
  99 /**
 100  * This determines which states will be updated when the shader is bound.
 101  */
 102 void
 103 st_set_prog_affected_state_flags(struct gl_program *prog)
 104 {
 105    uint64_t *states;
 106
 107    switch (prog->info.stage) {
 108    case MESA_SHADER_VERTEX:
 109       states = &((struct st_vertex_program*)prog)->affected_states;
 110
 111       *states = ST_NEW_VS_STATE |
 112                 ST_NEW_RASTERIZER |
 113                 ST_NEW_VERTEX_ARRAYS;
 114
 115       set_affected_state_flags(states, prog,
 116                                ST_NEW_VS_CONSTANTS,
 117                                ST_NEW_VS_SAMPLER_VIEWS,
 118                                ST_NEW_VS_SAMPLERS,
 119                                ST_NEW_VS_IMAGES,
 120                                ST_NEW_VS_UBOS,
 121                                ST_NEW_VS_SSBOS,
 122                                ST_NEW_VS_ATOMICS);
 123       break;
 124
 125    case MESA_SHADER_TESS_CTRL:
 126       states = &(st_common_program(prog))->affected_states;
 127
 128       *states = ST_NEW_TCS_STATE;
 129
 130       set_affected_state_flags(states, prog,
 131                                ST_NEW_TCS_CONSTANTS,
 132                                ST_NEW_TCS_SAMPLER_VIEWS,
 133                                ST_NEW_TCS_SAMPLERS,
 134                                ST_NEW_TCS_IMAGES,
 135                                ST_NEW_TCS_UBOS,
 136                                ST_NEW_TCS_SSBOS,
 137                                ST_NEW_TCS_ATOMICS);
 138       break;
 139
 140    case MESA_SHADER_TESS_EVAL:
 141       states = &(st_common_program(prog))->affected_states;
 142
 143       *states = ST_NEW_TES_STATE |
 144                 ST_NEW_RASTERIZER;
 145
 146       set_affected_state_flags(states, prog,
 147                                ST_NEW_TES_CONSTANTS,
 148                                ST_NEW_TES_SAMPLER_VIEWS,
 149                                ST_NEW_TES_SAMPLERS,
 150                                ST_NEW_TES_IMAGES,
 151                                ST_NEW_TES_UBOS,
 152                                ST_NEW_TES_SSBOS,
 153                                ST_NEW_TES_ATOMICS);
 154       break;
 155
 156    case MESA_SHADER_GEOMETRY:
 157       states = &(st_common_program(prog))->affected_states;
 158
 159       *states = ST_NEW_GS_STATE |
 160                 ST_NEW_RASTERIZER;
 161
 162       set_affected_state_flags(states, prog,
 163                                ST_NEW_GS_CONSTANTS,
 164                                ST_NEW_GS_SAMPLER_VIEWS,
 165                                ST_NEW_GS_SAMPLERS,
 166                                ST_NEW_GS_IMAGES,
 167                                ST_NEW_GS_UBOS,
 168                                ST_NEW_GS_SSBOS,
 169                                ST_NEW_GS_ATOMICS);
 170       break;
 171
 172    case MESA_SHADER_FRAGMENT:
 173       states = &((struct st_common_program*)prog)->affected_states;
 174
 175       /* gl_FragCoord and glDrawPixels always use constants. */
 176       *states = ST_NEW_FS_STATE |
 177                 ST_NEW_SAMPLE_SHADING |
 178                 ST_NEW_FS_CONSTANTS;
 179
 180       set_affected_state_flags(states, prog,
 181                                ST_NEW_FS_CONSTANTS,
 182                                ST_NEW_FS_SAMPLER_VIEWS,
 183                                ST_NEW_FS_SAMPLERS,
 184                                ST_NEW_FS_IMAGES,
 185                                ST_NEW_FS_UBOS,
 186                                ST_NEW_FS_SSBOS,
 187                                ST_NEW_FS_ATOMICS);
 188       break;
 189
 190    case MESA_SHADER_COMPUTE:
 191       states = &((struct st_common_program*)prog)->affected_states;
 192
 193       *states = ST_NEW_CS_STATE;
 194
 195       set_affected_state_flags(states, prog,
 196                                ST_NEW_CS_CONSTANTS,
 197                                ST_NEW_CS_SAMPLER_VIEWS,
 198                                ST_NEW_CS_SAMPLERS,
 199                                ST_NEW_CS_IMAGES,
 200                                ST_NEW_CS_UBOS,
 201                                ST_NEW_CS_SSBOS,
 202                                ST_NEW_CS_ATOMICS);
 203       break;
 204
 205    default:
 206       unreachable("unhandled shader stage");
 207    }
 208 }
 209
 210 static void
 211 delete_ir(struct pipe_shader_state *ir)
 212 {
 213    if (ir->tokens) {
 214       ureg_free_tokens(ir->tokens);
 215       ir->tokens = NULL;
 216    }
 217
 218    /* Note: Any setup of ->ir.nir that has had pipe->create_*_state called on
 219     * it has resulted in the driver taking ownership of the NIR.  Those
 220     * callers should be NULLing out the nir field in any pipe_shader_state
 221     * that might have this called in order to indicate that.
 222     *
 223     * GLSL IR and ARB programs will have set gl_program->nir to the same
 224     * shader as ir->ir.nir, so it will be freed by _mesa_delete_program().
 225     */
 226 }
 227
 228 /**
 229  * Delete a vertex program variant.  Note the caller must unlink
 230  * the variant from the linked list.
 231  */
 232 static void
 233 delete_vp_variant(struct st_context *st, struct st_vp_variant *vpv)
 234 {
 235    if (vpv->driver_shader) {
 236       if (st->has_shareable_shaders || vpv->key.st == st) {
 237          cso_delete_vertex_shader(st->cso_context, vpv->driver_shader);
 238       } else {
 239          st_save_zombie_shader(vpv->key.st, PIPE_SHADER_VERTEX,
 240                                vpv->driver_shader);
 241       }
 242    }
 243
 244    if (vpv->draw_shader)
 245       draw_delete_vertex_shader( st->draw, vpv->draw_shader );
 246
 247    if (vpv->tokens)
 248       ureg_free_tokens(vpv->tokens);
 249
 250    free( vpv );
 251 }
 252
 253
 254
 255 /**
 256  * Clean out any old compilations:
 257  */
 258 void
 259 st_release_vp_variants( struct st_context *st,
 260                         struct st_vertex_program *stvp )
 261 {
 262    struct st_vp_variant *vpv;
 263
 264    for (vpv = stvp->variants; vpv; ) {
 265       struct st_vp_variant *next = vpv->next;
 266       delete_vp_variant(st, vpv);
 267       vpv = next;
 268    }
 269
 270    stvp->variants = NULL;
 271
 272    delete_ir(&stvp->state);
 273 }
 274
 275
 276
 277 /**
 278  * Delete a fragment program variant.  Note the caller must unlink
 279  * the variant from the linked list.
 280  */
 281 static void
 282 delete_fp_variant(struct st_context *st, struct st_fp_variant *fpv)
 283 {
 284    if (fpv->driver_shader) {
 285       if (st->has_shareable_shaders || fpv->key.st == st) {
 286          cso_delete_fragment_shader(st->cso_context, fpv->driver_shader);
 287       } else {
 288          st_save_zombie_shader(fpv->key.st, PIPE_SHADER_FRAGMENT,
 289                                fpv->driver_shader);
 290       }
 291    }
 292
 293    free(fpv);
 294 }
 295
 296
 297 /**
 298  * Free all variants of a fragment program.
 299  */
 300 void
 301 st_release_fp_variants(struct st_context *st, struct st_common_program *stfp)
 302 {
 303    struct st_fp_variant *fpv;
 304
 305    for (fpv = stfp->fp_variants; fpv; ) {
 306       struct st_fp_variant *next = fpv->next;
 307       delete_fp_variant(st, fpv);
 308       fpv = next;
 309    }
 310
 311    stfp->fp_variants = NULL;
 312
 313    delete_ir(&stfp->state);
 314 }
 315
 316
 317 /**
 318  * Delete a basic program variant.  Note the caller must unlink
 319  * the variant from the linked list.
 320  */
 321 static void
 322 delete_basic_variant(struct st_context *st, struct st_common_variant *v,
 323                      GLenum target)
 324 {
 325    if (v->driver_shader) {
 326       if (st->has_shareable_shaders || v->key.st == st) {
 327          /* The shader's context matches the calling context, or we
 328           * don't care.
 329           */
 330          switch (target) {
 331          case GL_TESS_CONTROL_PROGRAM_NV:
 332             cso_delete_tessctrl_shader(st->cso_context, v->driver_shader);
 333             break;
 334          case GL_TESS_EVALUATION_PROGRAM_NV:
 335             cso_delete_tesseval_shader(st->cso_context, v->driver_shader);
 336             break;
 337          case GL_GEOMETRY_PROGRAM_NV:
 338             cso_delete_geometry_shader(st->cso_context, v->driver_shader);
 339             break;
 340          case GL_COMPUTE_PROGRAM_NV:
 341             cso_delete_compute_shader(st->cso_context, v->driver_shader);
 342             break;
 343          default:
 344             unreachable("bad shader type in delete_basic_variant");
 345          }
 346       } else {
 347          /* We can't delete a shader with a context different from the one
 348           * that created it.  Add it to the creating context's zombie list.
 349           */
 350          enum pipe_shader_type type;
 351          switch (target) {
 352          case GL_TESS_CONTROL_PROGRAM_NV:
 353             type = PIPE_SHADER_TESS_CTRL;
 354             break;
 355          case GL_TESS_EVALUATION_PROGRAM_NV:
 356             type = PIPE_SHADER_TESS_EVAL;
 357             break;
 358          case GL_GEOMETRY_PROGRAM_NV:
 359             type = PIPE_SHADER_GEOMETRY;
 360             break;
 361          default:
 362             unreachable("");
 363          }
 364          st_save_zombie_shader(v->key.st, type, v->driver_shader);
 365       }
 366    }
 367
 368    free(v);
 369 }
 370
 371
 372 /**
 373  * Free all basic program variants.
 374  */
 375 void
 376 st_release_common_variants(struct st_context *st, struct st_common_program *p)
 377 {
 378    struct st_common_variant *v;
 379
 380    for (v = p->variants; v; ) {
 381       struct st_common_variant *next = v->next;
 382       delete_basic_variant(st, v, p->Base.Target);
 383       v = next;
 384    }
 385
 386    p->variants = NULL;
 387    delete_ir(&p->state);
 388 }
 389
 390 void
 391 st_finalize_nir_before_variants(struct nir_shader *nir)
 392 {
 393    NIR_PASS_V(nir, nir_split_var_copies);
 394    NIR_PASS_V(nir, nir_lower_var_copies);
 395    if (nir->options->lower_all_io_to_temps ||
 396        nir->options->lower_all_io_to_elements ||
 397        nir->info.stage == MESA_SHADER_VERTEX ||
 398        nir->info.stage == MESA_SHADER_GEOMETRY) {
 399       NIR_PASS_V(nir, nir_lower_io_arrays_to_elements_no_indirects, false);
 400    } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
 401       NIR_PASS_V(nir, nir_lower_io_arrays_to_elements_no_indirects, true);
 402    }
 403
 404    st_nir_assign_vs_in_locations(nir);
 405 }
 406
 407 /**
 408  * Translate ARB (asm) program to NIR
 409  */
 410 static nir_shader *
 411 st_translate_prog_to_nir(struct st_context *st, struct gl_program *prog,
 412                          gl_shader_stage stage)
 413 {
 414    struct pipe_screen *screen = st->pipe->screen;
 415    const struct gl_shader_compiler_options *options =
 416       &st->ctx->Const.ShaderCompilerOptions[stage];
 417
 418    /* Translate to NIR */
 419    nir_shader *nir = prog_to_nir(prog, options->NirOptions);
 420    NIR_PASS_V(nir, nir_lower_regs_to_ssa); /* turn registers into SSA */
 421    nir_validate_shader(nir, "after st/ptn lower_regs_to_ssa");
 422
 423    NIR_PASS_V(nir, st_nir_lower_wpos_ytransform, prog, screen);
 424    NIR_PASS_V(nir, nir_lower_system_values);
 425
 426    /* Optimise NIR */
 427    NIR_PASS_V(nir, nir_opt_constant_folding);
 428    st_nir_opts(nir);
 429    st_finalize_nir_before_variants(nir);
 430
 431    if (st->allow_st_finalize_nir_twice)
 432       st_finalize_nir(st, prog, NULL, nir, true);
 433
 434    nir_validate_shader(nir, "after st/glsl finalize_nir");
 435
 436    return nir;
 437 }
 438
 439 void
 440 st_prepare_vertex_program(struct st_vertex_program *stvp)
 441 {
 442    stvp->num_inputs = 0;
 443    memset(stvp->input_to_index, ~0, sizeof(stvp->input_to_index));
 444    memset(stvp->result_to_output, ~0, sizeof(stvp->result_to_output));
 445
 446    /* Determine number of inputs, the mappings between VERT_ATTRIB_x
 447     * and TGSI generic input indexes, plus input attrib semantic info.
 448     */
 449    for (unsigned attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
 450       if ((stvp->Base.info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
 451          stvp->input_to_index[attr] = stvp->num_inputs;
 452          stvp->index_to_input[stvp->num_inputs] = attr;
 453          stvp->num_inputs++;
 454
 455          if ((stvp->Base.DualSlotInputs & BITFIELD64_BIT(attr)) != 0) {
 456             /* add placeholder for second part of a double attribute */
 457             stvp->index_to_input[stvp->num_inputs] = ST_DOUBLE_ATTRIB_PLACEHOLDER;
 458             stvp->num_inputs++;
 459          }
 460       }
 461    }
 462    /* pre-setup potentially unused edgeflag input */
 463    stvp->input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
 464    stvp->index_to_input[stvp->num_inputs] = VERT_ATTRIB_EDGEFLAG;
 465
 466    /* Compute mapping of vertex program outputs to slots. */
 467    unsigned num_outputs = 0;
 468    for (unsigned attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 469       if (stvp->Base.info.outputs_written & BITFIELD64_BIT(attr))
 470          stvp->result_to_output[attr] = num_outputs++;
 471    }
 472    /* pre-setup potentially unused edgeflag output */
 473    stvp->result_to_output[VARYING_SLOT_EDGE] = num_outputs;
 474 }
 475
 476 void
 477 st_translate_stream_output_info(struct gl_program *prog)
 478 {
 479    struct gl_transform_feedback_info *info = prog->sh.LinkedTransformFeedback;
 480    if (!info)
 481       return;
 482
 483    /* Determine the (default) output register mapping for each output. */
 484    unsigned num_outputs = 0;
 485    ubyte output_mapping[VARYING_SLOT_TESS_MAX];
 486    memset(output_mapping, 0, sizeof(output_mapping));
 487
 488    for (unsigned attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 489       if (prog->info.outputs_written & BITFIELD64_BIT(attr))
 490          output_mapping[attr] = num_outputs++;
 491    }
 492
 493    /* Translate stream output info. */
 494    struct pipe_stream_output_info *so_info = NULL;
 495    if (prog->info.stage == MESA_SHADER_VERTEX)
 496       so_info = &((struct st_vertex_program*)prog)->state.stream_output;
 497    else
 498       so_info = &((struct st_common_program*)prog)->state.stream_output;
 499
 500    for (unsigned i = 0; i < info->NumOutputs; i++) {
 501       so_info->output[i].register_index =
 502          output_mapping[info->Outputs[i].OutputRegister];
 503       so_info->output[i].start_component = info->Outputs[i].ComponentOffset;
 504       so_info->output[i].num_components = info->Outputs[i].NumComponents;
 505       so_info->output[i].output_buffer = info->Outputs[i].OutputBuffer;
 506       so_info->output[i].dst_offset = info->Outputs[i].DstOffset;
 507       so_info->output[i].stream = info->Outputs[i].StreamId;
 508    }
 509
 510    for (unsigned i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
 511       so_info->stride[i] = info->Buffers[i].Stride;
 512    }
 513    so_info->num_outputs = info->NumOutputs;
 514 }
 515
 516 /**
 517  * Translate a vertex program.
 518  */
 519 bool
 520 st_translate_vertex_program(struct st_context *st,
 521                             struct st_vertex_program *stvp)
 522 {
 523    struct ureg_program *ureg;
 524    enum pipe_error error;
 525    unsigned num_outputs = 0;
 526    unsigned attr;
 527    ubyte output_semantic_name[VARYING_SLOT_MAX] = {0};
 528    ubyte output_semantic_index[VARYING_SLOT_MAX] = {0};
 529
 530    if (stvp->Base.arb.IsPositionInvariant)
 531       _mesa_insert_mvp_code(st->ctx, &stvp->Base);
 532
 533    st_prepare_vertex_program(stvp);
 534
 535    /* ARB_vp: */
 536    if (!stvp->glsl_to_tgsi) {
 537       _mesa_remove_output_reads(&stvp->Base, PROGRAM_OUTPUT);
 538
 539       /* This determines which states will be updated when the assembly
 540        * shader is bound.
 541        */
 542       stvp->affected_states = ST_NEW_VS_STATE |
 543                               ST_NEW_RASTERIZER |
 544                               ST_NEW_VERTEX_ARRAYS;
 545
 546       if (stvp->Base.Parameters->NumParameters)
 547          stvp->affected_states |= ST_NEW_VS_CONSTANTS;
 548
 549       /* No samplers are allowed in ARB_vp. */
 550    }
 551
 552    /* Get semantic names and indices. */
 553    for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 554       if (stvp->Base.info.outputs_written & BITFIELD64_BIT(attr)) {
 555          unsigned slot = num_outputs++;
 556          unsigned semantic_name, semantic_index;
 557          tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
 558                                       &semantic_name, &semantic_index);
 559          output_semantic_name[slot] = semantic_name;
 560          output_semantic_index[slot] = semantic_index;
 561       }
 562    }
 563    /* pre-setup potentially unused edgeflag output */
 564    output_semantic_name[num_outputs] = TGSI_SEMANTIC_EDGEFLAG;
 565    output_semantic_index[num_outputs] = 0;
 566
 567    ureg = ureg_create_with_screen(PIPE_SHADER_VERTEX, st->pipe->screen);
 568    if (ureg == NULL)
 569       return false;
 570
 571    if (stvp->Base.info.clip_distance_array_size)
 572       ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
 573                     stvp->Base.info.clip_distance_array_size);
 574    if (stvp->Base.info.cull_distance_array_size)
 575       ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
 576                     stvp->Base.info.cull_distance_array_size);
 577
 578    if (ST_DEBUG & DEBUG_MESA) {
 579       _mesa_print_program(&stvp->Base);
 580       _mesa_print_program_parameters(st->ctx, &stvp->Base);
 581       debug_printf("\n");
 582    }
 583
 584    if (stvp->glsl_to_tgsi) {
 585       error = st_translate_program(st->ctx,
 586                                    PIPE_SHADER_VERTEX,
 587                                    ureg,
 588                                    stvp->glsl_to_tgsi,
 589                                    &stvp->Base,
 590                                    /* inputs */
 591                                    stvp->num_inputs,
 592                                    stvp->input_to_index,
 593                                    NULL, /* inputSlotToAttr */
 594                                    NULL, /* input semantic name */
 595                                    NULL, /* input semantic index */
 596                                    NULL, /* interp mode */
 597                                    /* outputs */
 598                                    num_outputs,
 599                                    stvp->result_to_output,
 600                                    output_semantic_name,
 601                                    output_semantic_index);
 602
 603       st_translate_stream_output_info(&stvp->Base);
 604
 605       free_glsl_to_tgsi_visitor(stvp->glsl_to_tgsi);
 606    } else
 607       error = st_translate_mesa_program(st->ctx,
 608                                         PIPE_SHADER_VERTEX,
 609                                         ureg,
 610                                         &stvp->Base,
 611                                         /* inputs */
 612                                         stvp->num_inputs,
 613                                         stvp->input_to_index,
 614                                         NULL, /* input semantic name */
 615                                         NULL, /* input semantic index */
 616                                         NULL,
 617                                         /* outputs */
 618                                         num_outputs,
 619                                         stvp->result_to_output,
 620                                         output_semantic_name,
 621                                         output_semantic_index);
 622
 623    if (error) {
 624       debug_printf("%s: failed to translate Mesa program:\n", __func__);
 625       _mesa_print_program(&stvp->Base);
 626       debug_assert(0);
 627       return false;
 628    }
 629
 630    stvp->state.tokens = ureg_get_tokens(ureg, NULL);
 631    ureg_destroy(ureg);
 632
 633    if (stvp->glsl_to_tgsi) {
 634       stvp->glsl_to_tgsi = NULL;
 635       st_store_ir_in_disk_cache(st, &stvp->Base, false);
 636    }
 637
 638    /* Translate to NIR.
 639     *
 640     * This must be done after the translation to TGSI is done, because
 641     * we'll pass the NIR shader to the driver and the TGSI version to
 642     * the draw module for the select/feedback/rasterpos code.
 643     */
 644    if (st->pipe->screen->get_shader_param(st->pipe->screen,
 645                                           PIPE_SHADER_VERTEX,
 646                                           PIPE_SHADER_CAP_PREFERRED_IR)) {
 647       assert(!stvp->glsl_to_tgsi);
 648
 649       nir_shader *nir =
 650          st_translate_prog_to_nir(st, &stvp->Base, MESA_SHADER_VERTEX);
 651
 652       if (stvp->state.ir.nir)
 653          ralloc_free(stvp->state.ir.nir);
 654       stvp->state.type = PIPE_SHADER_IR_NIR;
 655       stvp->state.ir.nir = nir;
 656       stvp->Base.nir = nir;
 657       return true;
 658    }
 659
 660    return stvp->state.tokens != NULL;
 661 }
 662
 663 static const gl_state_index16 depth_range_state[STATE_LENGTH] =
 664    { STATE_DEPTH_RANGE };
 665
 666 static struct st_vp_variant *
 667 st_create_vp_variant(struct st_context *st,
 668                      struct st_vertex_program *stvp,
 669                      const struct st_common_variant_key *key)
 670 {
 671    struct st_vp_variant *vpv = CALLOC_STRUCT(st_vp_variant);
 672    struct pipe_context *pipe = st->pipe;
 673    struct pipe_screen *screen = pipe->screen;
 674    struct pipe_shader_state state = {0};
 675
 676    static const gl_state_index16 point_size_state[STATE_LENGTH] =
 677       { STATE_INTERNAL, STATE_POINT_SIZE_CLAMPED, 0 };
 678    struct gl_program_parameter_list *params = stvp->Base.Parameters;
 679
 680    vpv->key = *key;
 681    vpv->num_inputs = stvp->num_inputs;
 682
 683    state.stream_output = stvp->state.stream_output;
 684
 685    if (stvp->state.type == PIPE_SHADER_IR_NIR) {
 686       bool finalize = false;
 687
 688       state.type = PIPE_SHADER_IR_NIR;
 689       state.ir.nir = nir_shader_clone(NULL, stvp->state.ir.nir);
 690       if (key->clamp_color) {
 691          NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
 692          finalize = true;
 693       }
 694       if (key->passthrough_edgeflags) {
 695          NIR_PASS_V(state.ir.nir, nir_lower_passthrough_edgeflags);
 696          vpv->num_inputs++;
 697          finalize = true;
 698       }
 699
 700       if (key->lower_point_size) {
 701          _mesa_add_state_reference(params, point_size_state);
 702          NIR_PASS_V(state.ir.nir, nir_lower_point_size_mov,
 703                     point_size_state);
 704          finalize = true;
 705       }
 706
 707       if (key->lower_ucp) {
 708          bool can_compact = screen->get_param(screen,
 709                                               PIPE_CAP_NIR_COMPACT_ARRAYS);
 710
 711          bool use_eye = st->ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX] != NULL;
 712          gl_state_index16 clipplane_state[MAX_CLIP_PLANES][STATE_LENGTH];
 713          for (int i = 0; i < MAX_CLIP_PLANES; ++i) {
 714             if (use_eye) {
 715                clipplane_state[i][0] = STATE_CLIPPLANE;
 716                clipplane_state[i][1] = i;
 717             } else {
 718                clipplane_state[i][0] = STATE_INTERNAL;
 719                clipplane_state[i][1] = STATE_CLIP_INTERNAL;
 720                clipplane_state[i][2] = i;
 721             }
 722             _mesa_add_state_reference(params, clipplane_state[i]);
 723          }
 724
 725          NIR_PASS_V(state.ir.nir, nir_lower_clip_vs, key->lower_ucp,
 726                     true, can_compact, clipplane_state);
 727          NIR_PASS_V(state.ir.nir, nir_lower_io_to_temporaries,
 728                     nir_shader_get_entrypoint(state.ir.nir), true, false);
 729          finalize = true;
 730       }
 731
 732       if (finalize || !st->allow_st_finalize_nir_twice) {
 733          st_finalize_nir(st, &stvp->Base, stvp->shader_program, state.ir.nir,
 734                          true);
 735
 736          /* Some of the lowering above may have introduced new varyings */
 737          nir_shader_gather_info(state.ir.nir,
 738                                 nir_shader_get_entrypoint(state.ir.nir));
 739       }
 740
 741       vpv->driver_shader = pipe->create_vs_state(pipe, &state);
 742
 743       /* When generating a NIR program, we usually don't have TGSI tokens.
 744        * However, we do create them for ARB_vertex_program / fixed-function VS
 745        * programs which we may need to use with the draw module for legacy
 746        * feedback/select emulation.  If they exist, copy them.
 747        *
 748        * TODO: Lowering for shader variants is not applied to TGSI when
 749        * generating a NIR shader.
 750        */
 751       if (stvp->state.tokens)
 752          vpv->tokens = tgsi_dup_tokens(stvp->state.tokens);
 753
 754       return vpv;
 755    }
 756
 757    state.type = PIPE_SHADER_IR_TGSI;
 758    state.tokens = tgsi_dup_tokens(stvp->state.tokens);
 759
 760    /* Emulate features. */
 761    if (key->clamp_color || key->passthrough_edgeflags) {
 762       const struct tgsi_token *tokens;
 763       unsigned flags =
 764          (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
 765          (key->passthrough_edgeflags ? TGSI_EMU_PASSTHROUGH_EDGEFLAG : 0);
 766
 767       tokens = tgsi_emulate(state.tokens, flags);
 768
 769       if (tokens) {
 770          tgsi_free_tokens(state.tokens);
 771          state.tokens = tokens;
 772
 773          if (key->passthrough_edgeflags)
 774             vpv->num_inputs++;
 775       } else
 776          fprintf(stderr, "mesa: cannot emulate deprecated features\n");
 777    }
 778
 779    if (key->lower_depth_clamp) {
 780       unsigned depth_range_const =
 781             _mesa_add_state_reference(params, depth_range_state);
 782
 783       const struct tgsi_token *tokens;
 784       tokens = st_tgsi_lower_depth_clamp(state.tokens, depth_range_const,
 785                                          key->clip_negative_one_to_one);
 786       if (tokens != state.tokens)
 787          tgsi_free_tokens(state.tokens);
 788       state.tokens = tokens;
 789    }
 790
 791    if (ST_DEBUG & DEBUG_TGSI) {
 792       tgsi_dump(state.tokens, 0);
 793       debug_printf("\n");
 794    }
 795
 796    vpv->driver_shader = pipe->create_vs_state(pipe, &state);
 797    /* Save this for selection/feedback/rasterpos. */
 798    vpv->tokens = state.tokens;
 799    return vpv;
 800 }
 801
 802
 803 /**
 804  * Find/create a vertex program variant.
 805  */
 806 struct st_vp_variant *
 807 st_get_vp_variant(struct st_context *st,
 808                   struct st_vertex_program *stvp,
 809                   const struct st_common_variant_key *key)
 810 {
 811    struct st_vp_variant *vpv;
 812
 813    /* Search for existing variant */
 814    for (vpv = stvp->variants; vpv; vpv = vpv->next) {
 815       if (memcmp(&vpv->key, key, sizeof(*key)) == 0) {
 816          break;
 817       }
 818    }
 819
 820    if (!vpv) {
 821       /* create now */
 822       vpv = st_create_vp_variant(st, stvp, key);
 823       if (vpv) {
 824           for (unsigned index = 0; index < vpv->num_inputs; ++index) {
 825              unsigned attr = stvp->index_to_input[index];
 826              if (attr == ST_DOUBLE_ATTRIB_PLACEHOLDER)
 827                 continue;
 828              vpv->vert_attrib_mask |= 1u << attr;
 829           }
 830
 831          /* insert into list */
 832          vpv->next = stvp->variants;
 833          stvp->variants = vpv;
 834       }
 835    }
 836
 837    return vpv;
 838 }
 839
 840
 841 /**
 842  * Translate a Mesa fragment shader into a TGSI shader.
 843  */
 844 bool
 845 st_translate_fragment_program(struct st_context *st,
 846                               struct st_common_program *stfp)
 847 {
 848    /* Non-GLSL programs: */
 849    if (!stfp->glsl_to_tgsi) {
 850       _mesa_remove_output_reads(&stfp->Base, PROGRAM_OUTPUT);
 851       if (st->ctx->Const.GLSLFragCoordIsSysVal)
 852          _mesa_program_fragment_position_to_sysval(&stfp->Base);
 853
 854       /* This determines which states will be updated when the assembly
 855        * shader is bound.
 856        *
 857        * fragment.position and glDrawPixels always use constants.
 858        */
 859       stfp->affected_states = ST_NEW_FS_STATE |
 860                               ST_NEW_SAMPLE_SHADING |
 861                               ST_NEW_FS_CONSTANTS;
 862
 863       if (stfp->ati_fs) {
 864          /* Just set them for ATI_fs unconditionally. */
 865          stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
 866                                   ST_NEW_FS_SAMPLERS;
 867       } else {
 868          /* ARB_fp */
 869          if (stfp->Base.SamplersUsed)
 870             stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
 871                                      ST_NEW_FS_SAMPLERS;
 872       }
 873
 874       /* Translate to NIR. */
 875       if (!stfp->ati_fs &&
 876           st->pipe->screen->get_shader_param(st->pipe->screen,
 877                                              PIPE_SHADER_FRAGMENT,
 878                                              PIPE_SHADER_CAP_PREFERRED_IR)) {
 879          nir_shader *nir =
 880             st_translate_prog_to_nir(st, &stfp->Base, MESA_SHADER_FRAGMENT);
 881
 882          if (stfp->state.ir.nir)
 883             ralloc_free(stfp->state.ir.nir);
 884          stfp->state.type = PIPE_SHADER_IR_NIR;
 885          stfp->state.ir.nir = nir;
 886          stfp->Base.nir = nir;
 887          return true;
 888       }
 889    }
 890
 891    ubyte outputMapping[2 * FRAG_RESULT_MAX];
 892    ubyte inputMapping[VARYING_SLOT_MAX];
 893    ubyte inputSlotToAttr[VARYING_SLOT_MAX];
 894    ubyte interpMode[PIPE_MAX_SHADER_INPUTS];  /* XXX size? */
 895    GLuint attr;
 896    GLbitfield64 inputsRead;
 897    struct ureg_program *ureg;
 898
 899    GLboolean write_all = GL_FALSE;
 900
 901    ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
 902    ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
 903    uint fs_num_inputs = 0;
 904
 905    ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
 906    ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
 907    uint fs_num_outputs = 0;
 908
 909    memset(inputSlotToAttr, ~0, sizeof(inputSlotToAttr));
 910
 911    /*
 912     * Convert Mesa program inputs to TGSI input register semantics.
 913     */
 914    inputsRead = stfp->Base.info.inputs_read;
 915    for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 916       if ((inputsRead & BITFIELD64_BIT(attr)) != 0) {
 917          const GLuint slot = fs_num_inputs++;
 918
 919          inputMapping[attr] = slot;
 920          inputSlotToAttr[slot] = attr;
 921
 922          switch (attr) {
 923          case VARYING_SLOT_POS:
 924             input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 925             input_semantic_index[slot] = 0;
 926             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 927             break;
 928          case VARYING_SLOT_COL0:
 929             input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 930             input_semantic_index[slot] = 0;
 931             interpMode[slot] = stfp->glsl_to_tgsi ?
 932                TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
 933             break;
 934          case VARYING_SLOT_COL1:
 935             input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 936             input_semantic_index[slot] = 1;
 937             interpMode[slot] = stfp->glsl_to_tgsi ?
 938                TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
 939             break;
 940          case VARYING_SLOT_FOGC:
 941             input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 942             input_semantic_index[slot] = 0;
 943             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 944             break;
 945          case VARYING_SLOT_FACE:
 946             input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
 947             input_semantic_index[slot] = 0;
 948             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 949             break;
 950          case VARYING_SLOT_PRIMITIVE_ID:
 951             input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
 952             input_semantic_index[slot] = 0;
 953             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 954             break;
 955          case VARYING_SLOT_LAYER:
 956             input_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
 957             input_semantic_index[slot] = 0;
 958             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 959             break;
 960          case VARYING_SLOT_VIEWPORT:
 961             input_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
 962             input_semantic_index[slot] = 0;
 963             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 964             break;
 965          case VARYING_SLOT_CLIP_DIST0:
 966             input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
 967             input_semantic_index[slot] = 0;
 968             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 969             break;
 970          case VARYING_SLOT_CLIP_DIST1:
 971             input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
 972             input_semantic_index[slot] = 1;
 973             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 974             break;
 975          case VARYING_SLOT_CULL_DIST0:
 976          case VARYING_SLOT_CULL_DIST1:
 977             /* these should have been lowered by GLSL */
 978             assert(0);
 979             break;
 980             /* In most cases, there is nothing special about these
 981              * inputs, so adopt a convention to use the generic
 982              * semantic name and the mesa VARYING_SLOT_ number as the
 983              * index.
 984              *
 985              * All that is required is that the vertex shader labels
 986              * its own outputs similarly, and that the vertex shader
 987              * generates at least every output required by the
 988              * fragment shader plus fixed-function hardware (such as
 989              * BFC).
 990              *
 991              * However, some drivers may need us to identify the PNTC and TEXi
 992              * varyings if, for example, their capability to replace them with
 993              * sprite coordinates is limited.
 994              */
 995          case VARYING_SLOT_PNTC:
 996             if (st->needs_texcoord_semantic) {
 997                input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD;
 998                input_semantic_index[slot] = 0;
 999                interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
1000                break;
1001             }
1002             /* fall through */
1003          case VARYING_SLOT_TEX0:
1004          case VARYING_SLOT_TEX1:
1005          case VARYING_SLOT_TEX2:
1006          case VARYING_SLOT_TEX3:
1007          case VARYING_SLOT_TEX4:
1008          case VARYING_SLOT_TEX5:
1009          case VARYING_SLOT_TEX6:
1010          case VARYING_SLOT_TEX7:
1011             if (st->needs_texcoord_semantic) {
1012                input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
1013                input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
1014                interpMode[slot] = stfp->glsl_to_tgsi ?
1015                   TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
1016                break;
1017             }
1018             /* fall through */
1019          case VARYING_SLOT_VAR0:
1020          default:
1021             /* Semantic indices should be zero-based because drivers may choose
1022              * to assign a fixed slot determined by that index.
1023              * This is useful because ARB_separate_shader_objects uses location
1024              * qualifiers for linkage, and if the semantic index corresponds to
1025              * these locations, linkage passes in the driver become unecessary.
1026              *
1027              * If needs_texcoord_semantic is true, no semantic indices will be
1028              * consumed for the TEXi varyings, and we can base the locations of
1029              * the user varyings on VAR0.  Otherwise, we use TEX0 as base index.
1030              */
1031             assert(attr >= VARYING_SLOT_VAR0 || attr == VARYING_SLOT_PNTC ||
1032                    (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
1033             input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
1034             input_semantic_index[slot] = st_get_generic_varying_index(st, attr);
1035             if (attr == VARYING_SLOT_PNTC)
1036                interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
1037             else {
1038                interpMode[slot] = stfp->glsl_to_tgsi ?
1039                   TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
1040             }
1041             break;
1042          }
1043       }
1044       else {
1045          inputMapping[attr] = -1;
1046       }
1047    }
1048
1049    /*
1050     * Semantics and mapping for outputs
1051     */
1052    GLbitfield64 outputsWritten = stfp->Base.info.outputs_written;
1053
1054    /* if z is written, emit that first */
1055    if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
1056       fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
1057       fs_output_semantic_index[fs_num_outputs] = 0;
1058       outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
1059       fs_num_outputs++;
1060       outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
1061    }
1062
1063    if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
1064       fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
1065       fs_output_semantic_index[fs_num_outputs] = 0;
1066       outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
1067       fs_num_outputs++;
1068       outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
1069    }
1070
1071    if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)) {
1072       fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_SAMPLEMASK;
1073       fs_output_semantic_index[fs_num_outputs] = 0;
1074       outputMapping[FRAG_RESULT_SAMPLE_MASK] = fs_num_outputs;
1075       fs_num_outputs++;
1076       outputsWritten &= ~(1 << FRAG_RESULT_SAMPLE_MASK);
1077    }
1078
1079    /* handle remaining outputs (color) */
1080    for (attr = 0; attr < ARRAY_SIZE(outputMapping); attr++) {
1081       const GLbitfield64 written = attr < FRAG_RESULT_MAX ? outputsWritten :
1082          stfp->Base.SecondaryOutputsWritten;
1083       const unsigned loc = attr % FRAG_RESULT_MAX;
1084
1085       if (written & BITFIELD64_BIT(loc)) {
1086          switch (loc) {
1087          case FRAG_RESULT_DEPTH:
1088          case FRAG_RESULT_STENCIL:
1089          case FRAG_RESULT_SAMPLE_MASK:
1090             /* handled above */
1091             assert(0);
1092             break;
1093          case FRAG_RESULT_COLOR:
1094             write_all = GL_TRUE; /* fallthrough */
1095          default: {
1096             int index;
1097             assert(loc == FRAG_RESULT_COLOR ||
1098                    (FRAG_RESULT_DATA0 <= loc && loc < FRAG_RESULT_MAX));
1099
1100             index = (loc == FRAG_RESULT_COLOR) ? 0 : (loc - FRAG_RESULT_DATA0);
1101
1102             if (attr >= FRAG_RESULT_MAX) {
1103                /* Secondary color for dual source blending. */
1104                assert(index == 0);
1105                index++;
1106             }
1107
1108             fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
1109             fs_output_semantic_index[fs_num_outputs] = index;
1110             outputMapping[attr] = fs_num_outputs;
1111             break;
1112          }
1113          }
1114
1115          fs_num_outputs++;
1116       }
1117    }
1118
1119    ureg = ureg_create_with_screen(PIPE_SHADER_FRAGMENT, st->pipe->screen);
1120    if (ureg == NULL)
1121       return false;
1122
1123    if (ST_DEBUG & DEBUG_MESA) {
1124       _mesa_print_program(&stfp->Base);
1125       _mesa_print_program_parameters(st->ctx, &stfp->Base);
1126       debug_printf("\n");
1127    }
1128    if (write_all == GL_TRUE)
1129       ureg_property(ureg, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, 1);
1130
1131    if (stfp->Base.info.fs.depth_layout != FRAG_DEPTH_LAYOUT_NONE) {
1132       switch (stfp->Base.info.fs.depth_layout) {
1133       case FRAG_DEPTH_LAYOUT_ANY:
1134          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1135                        TGSI_FS_DEPTH_LAYOUT_ANY);
1136          break;
1137       case FRAG_DEPTH_LAYOUT_GREATER:
1138          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1139                        TGSI_FS_DEPTH_LAYOUT_GREATER);
1140          break;
1141       case FRAG_DEPTH_LAYOUT_LESS:
1142          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1143                        TGSI_FS_DEPTH_LAYOUT_LESS);
1144          break;
1145       case FRAG_DEPTH_LAYOUT_UNCHANGED:
1146          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1147                        TGSI_FS_DEPTH_LAYOUT_UNCHANGED);
1148          break;
1149       default:
1150          assert(0);
1151       }
1152    }
1153
1154    if (stfp->glsl_to_tgsi) {
1155       st_translate_program(st->ctx,
1156                            PIPE_SHADER_FRAGMENT,
1157                            ureg,
1158                            stfp->glsl_to_tgsi,
1159                            &stfp->Base,
1160                            /* inputs */
1161                            fs_num_inputs,
1162                            inputMapping,
1163                            inputSlotToAttr,
1164                            input_semantic_name,
1165                            input_semantic_index,
1166                            interpMode,
1167                            /* outputs */
1168                            fs_num_outputs,
1169                            outputMapping,
1170                            fs_output_semantic_name,
1171                            fs_output_semantic_index);
1172
1173       free_glsl_to_tgsi_visitor(stfp->glsl_to_tgsi);
1174    } else if (stfp->ati_fs)
1175       st_translate_atifs_program(ureg,
1176                                  stfp->ati_fs,
1177                                  &stfp->Base,
1178                                  /* inputs */
1179                                  fs_num_inputs,
1180                                  inputMapping,
1181                                  input_semantic_name,
1182                                  input_semantic_index,
1183                                  interpMode,
1184                                  /* outputs */
1185                                  fs_num_outputs,
1186                                  outputMapping,
1187                                  fs_output_semantic_name,
1188                                  fs_output_semantic_index);
1189    else
1190       st_translate_mesa_program(st->ctx,
1191                                 PIPE_SHADER_FRAGMENT,
1192                                 ureg,
1193                                 &stfp->Base,
1194                                 /* inputs */
1195                                 fs_num_inputs,
1196                                 inputMapping,
1197                                 input_semantic_name,
1198                                 input_semantic_index,
1199                                 interpMode,
1200                                 /* outputs */
1201                                 fs_num_outputs,
1202                                 outputMapping,
1203                                 fs_output_semantic_name,
1204                                 fs_output_semantic_index);
1205
1206    stfp->state.tokens = ureg_get_tokens(ureg, NULL);
1207    ureg_destroy(ureg);
1208
1209    if (stfp->glsl_to_tgsi) {
1210       stfp->glsl_to_tgsi = NULL;
1211       st_store_ir_in_disk_cache(st, &stfp->Base, false);
1212    }
1213
1214    return stfp->state.tokens != NULL;
1215 }
1216
1217 static struct st_fp_variant *
1218 st_create_fp_variant(struct st_context *st,
1219                      struct st_common_program *stfp,
1220                      const struct st_fp_variant_key *key)
1221 {
1222    struct pipe_context *pipe = st->pipe;
1223    struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
1224    struct pipe_shader_state state = {0};
1225    struct gl_program_parameter_list *params = stfp->Base.Parameters;
1226    static const gl_state_index16 texcoord_state[STATE_LENGTH] =
1227       { STATE_INTERNAL, STATE_CURRENT_ATTRIB, VERT_ATTRIB_TEX0 };
1228    static const gl_state_index16 scale_state[STATE_LENGTH] =
1229       { STATE_INTERNAL, STATE_PT_SCALE };
1230    static const gl_state_index16 bias_state[STATE_LENGTH] =
1231       { STATE_INTERNAL, STATE_PT_BIAS };
1232    static const gl_state_index16 alpha_ref_state[STATE_LENGTH] =
1233       { STATE_INTERNAL, STATE_ALPHA_REF };
1234
1235    if (!variant)
1236       return NULL;
1237
1238    if (stfp->state.type == PIPE_SHADER_IR_NIR) {
1239       bool finalize = false;
1240
1241       state.type = PIPE_SHADER_IR_NIR;
1242       state.ir.nir = nir_shader_clone(NULL, stfp->state.ir.nir);
1243
1244       if (key->clamp_color) {
1245          NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
1246          finalize = true;
1247       }
1248
1249       if (key->lower_flatshade) {
1250          NIR_PASS_V(state.ir.nir, nir_lower_flatshade);
1251          finalize = true;
1252       }
1253
1254       if (key->lower_alpha_func != COMPARE_FUNC_NEVER) {
1255          _mesa_add_state_reference(params, alpha_ref_state);
1256          NIR_PASS_V(state.ir.nir, nir_lower_alpha_test, key->lower_alpha_func,
1257                     false, alpha_ref_state);
1258          finalize = true;
1259       }
1260
1261       if (key->lower_two_sided_color) {
1262          NIR_PASS_V(state.ir.nir, nir_lower_two_sided_color);
1263          finalize = true;
1264       }
1265
1266       if (key->persample_shading) {
1267           nir_shader *shader = state.ir.nir;
1268           nir_foreach_variable(var, &shader->inputs)
1269              var->data.sample = true;
1270           finalize = true;
1271       }
1272
1273       assert(!(key->bitmap && key->drawpixels));
1274
1275       /* glBitmap */
1276       if (key->bitmap) {
1277          nir_lower_bitmap_options options = {0};
1278
1279          variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1280          options.sampler = variant->bitmap_sampler;
1281          options.swizzle_xxxx = st->bitmap.tex_format == PIPE_FORMAT_R8_UNORM;
1282
1283          NIR_PASS_V(state.ir.nir, nir_lower_bitmap, &options);
1284          finalize = true;
1285       }
1286
1287       /* glDrawPixels (color only) */
1288       if (key->drawpixels) {
1289          nir_lower_drawpixels_options options = {{0}};
1290          unsigned samplers_used = stfp->Base.SamplersUsed;
1291
1292          /* Find the first unused slot. */
1293          variant->drawpix_sampler = ffs(~samplers_used) - 1;
1294          options.drawpix_sampler = variant->drawpix_sampler;
1295          samplers_used |= (1 << variant->drawpix_sampler);
1296
1297          options.pixel_maps = key->pixelMaps;
1298          if (key->pixelMaps) {
1299             variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1300             options.pixelmap_sampler = variant->pixelmap_sampler;
1301          }
1302
1303          options.scale_and_bias = key->scaleAndBias;
1304          if (key->scaleAndBias) {
1305             _mesa_add_state_reference(params, scale_state);
1306             memcpy(options.scale_state_tokens, scale_state,
1307                    sizeof(options.scale_state_tokens));
1308             _mesa_add_state_reference(params, bias_state);
1309             memcpy(options.bias_state_tokens, bias_state,
1310                    sizeof(options.bias_state_tokens));
1311          }
1312
1313          _mesa_add_state_reference(params, texcoord_state);
1314          memcpy(options.texcoord_state_tokens, texcoord_state,
1315                 sizeof(options.texcoord_state_tokens));
1316
1317          NIR_PASS_V(state.ir.nir, nir_lower_drawpixels, &options);
1318          finalize = true;
1319       }
1320
1321       if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1322                    key->external.lower_xy_uxvx || key->external.lower_yx_xuxv ||
1323                    key->external.lower_ayuv || key->external.lower_xyuv)) {
1324          nir_lower_tex_options options = {0};
1325          options.lower_y_uv_external = key->external.lower_nv12;
1326          options.lower_y_u_v_external = key->external.lower_iyuv;
1327          options.lower_xy_uxvx_external = key->external.lower_xy_uxvx;
1328          options.lower_yx_xuxv_external = key->external.lower_yx_xuxv;
1329          options.lower_ayuv_external = key->external.lower_ayuv;
1330          options.lower_xyuv_external = key->external.lower_xyuv;
1331          NIR_PASS_V(state.ir.nir, nir_lower_tex, &options);
1332          finalize = true;
1333       }
1334
1335       if (finalize || !st->allow_st_finalize_nir_twice) {
1336          st_finalize_nir(st, &stfp->Base, stfp->shader_program, state.ir.nir,
1337                          false);
1338       }
1339
1340       /* This pass needs to happen *after* nir_lower_sampler */
1341       if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1342                    key->external.lower_xy_uxvx || key->external.lower_yx_xuxv)) {
1343          NIR_PASS_V(state.ir.nir, st_nir_lower_tex_src_plane,
1344                     ~stfp->Base.SamplersUsed,
1345                     key->external.lower_nv12 || key->external.lower_xy_uxvx ||
1346                        key->external.lower_yx_xuxv,
1347                     key->external.lower_iyuv);
1348          finalize = true;
1349       }
1350
1351       if (finalize || !st->allow_st_finalize_nir_twice) {
1352          /* Some of the lowering above may have introduced new varyings */
1353          nir_shader_gather_info(state.ir.nir,
1354                                 nir_shader_get_entrypoint(state.ir.nir));
1355
1356          struct pipe_screen *screen = pipe->screen;
1357          if (screen->finalize_nir)
1358             screen->finalize_nir(screen, state.ir.nir, false);
1359       }
1360
1361       variant->driver_shader = pipe->create_fs_state(pipe, &state);
1362       variant->key = *key;
1363
1364       return variant;
1365    }
1366
1367    state.tokens = stfp->state.tokens;
1368
1369    assert(!(key->bitmap && key->drawpixels));
1370
1371    /* Fix texture targets and add fog for ATI_fs */
1372    if (stfp->ati_fs) {
1373       const struct tgsi_token *tokens = st_fixup_atifs(state.tokens, key);
1374
1375       if (tokens)
1376          state.tokens = tokens;
1377       else
1378          fprintf(stderr, "mesa: cannot post-process ATI_fs\n");
1379    }
1380
1381    /* Emulate features. */
1382    if (key->clamp_color || key->persample_shading) {
1383       const struct tgsi_token *tokens;
1384       unsigned flags =
1385          (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
1386          (key->persample_shading ? TGSI_EMU_FORCE_PERSAMPLE_INTERP : 0);
1387
1388       tokens = tgsi_emulate(state.tokens, flags);
1389
1390       if (tokens) {
1391          if (state.tokens != stfp->state.tokens)
1392             tgsi_free_tokens(state.tokens);
1393          state.tokens = tokens;
1394       } else
1395          fprintf(stderr, "mesa: cannot emulate deprecated features\n");
1396    }
1397
1398    /* glBitmap */
1399    if (key->bitmap) {
1400       const struct tgsi_token *tokens;
1401
1402       variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1403
1404       tokens = st_get_bitmap_shader(state.tokens,
1405                                     st->internal_target,
1406                                     variant->bitmap_sampler,
1407                                     st->needs_texcoord_semantic,
1408                                     st->bitmap.tex_format ==
1409                                     PIPE_FORMAT_R8_UNORM);
1410
1411       if (tokens) {
1412          if (state.tokens != stfp->state.tokens)
1413             tgsi_free_tokens(state.tokens);
1414          state.tokens = tokens;
1415       } else
1416          fprintf(stderr, "mesa: cannot create a shader for glBitmap\n");
1417    }
1418
1419    /* glDrawPixels (color only) */
1420    if (key->drawpixels) {
1421       const struct tgsi_token *tokens;
1422       unsigned scale_const = 0, bias_const = 0, texcoord_const = 0;
1423
1424       /* Find the first unused slot. */
1425       variant->drawpix_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1426
1427       if (key->pixelMaps) {
1428          unsigned samplers_used = stfp->Base.SamplersUsed |
1429                                   (1 << variant->drawpix_sampler);
1430
1431          variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1432       }
1433
1434       if (key->scaleAndBias) {
1435          scale_const = _mesa_add_state_reference(params, scale_state);
1436          bias_const = _mesa_add_state_reference(params, bias_state);
1437       }
1438
1439       texcoord_const = _mesa_add_state_reference(params, texcoord_state);
1440
1441       tokens = st_get_drawpix_shader(state.tokens,
1442                                      st->needs_texcoord_semantic,
1443                                      key->scaleAndBias, scale_const,
1444                                      bias_const, key->pixelMaps,
1445                                      variant->drawpix_sampler,
1446                                      variant->pixelmap_sampler,
1447                                      texcoord_const, st->internal_target);
1448
1449       if (tokens) {
1450          if (state.tokens != stfp->state.tokens)
1451             tgsi_free_tokens(state.tokens);
1452          state.tokens = tokens;
1453       } else
1454          fprintf(stderr, "mesa: cannot create a shader for glDrawPixels\n");
1455    }
1456
1457    if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1458                 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv)) {
1459       const struct tgsi_token *tokens;
1460
1461       /* samplers inserted would conflict, but this should be unpossible: */
1462       assert(!(key->bitmap || key->drawpixels));
1463
1464       tokens = st_tgsi_lower_yuv(state.tokens,
1465                                  ~stfp->Base.SamplersUsed,
1466                                  key->external.lower_nv12 ||
1467                                     key->external.lower_xy_uxvx ||
1468                                     key->external.lower_yx_xuxv,
1469                                  key->external.lower_iyuv);
1470       if (tokens) {
1471          if (state.tokens != stfp->state.tokens)
1472             tgsi_free_tokens(state.tokens);
1473          state.tokens = tokens;
1474       } else {
1475          fprintf(stderr, "mesa: cannot create a shader for samplerExternalOES\n");
1476       }
1477    }
1478
1479    if (key->lower_depth_clamp) {
1480       unsigned depth_range_const = _mesa_add_state_reference(params, depth_range_state);
1481
1482       const struct tgsi_token *tokens;
1483       tokens = st_tgsi_lower_depth_clamp_fs(state.tokens, depth_range_const);
1484       if (state.tokens != stfp->state.tokens)
1485          tgsi_free_tokens(state.tokens);
1486       state.tokens = tokens;
1487    }
1488
1489    if (ST_DEBUG & DEBUG_TGSI) {
1490       tgsi_dump(state.tokens, 0);
1491       debug_printf("\n");
1492    }
1493
1494    /* fill in variant */
1495    variant->driver_shader = pipe->create_fs_state(pipe, &state);
1496    variant->key = *key;
1497
1498    if (state.tokens != stfp->state.tokens)
1499       tgsi_free_tokens(state.tokens);
1500    return variant;
1501 }
1502
1503 /**
1504  * Translate fragment program if needed.
1505  */
1506 struct st_fp_variant *
1507 st_get_fp_variant(struct st_context *st,
1508                   struct st_common_program *stfp,
1509                   const struct st_fp_variant_key *key)
1510 {
1511    struct st_fp_variant *fpv;
1512
1513    /* Search for existing variant */
1514    for (fpv = stfp->fp_variants; fpv; fpv = fpv->next) {
1515       if (memcmp(&fpv->key, key, sizeof(*key)) == 0) {
1516          break;
1517       }
1518    }
1519
1520    if (!fpv) {
1521       /* create new */
1522       fpv = st_create_fp_variant(st, stfp, key);
1523       if (fpv) {
1524          if (key->bitmap || key->drawpixels) {
1525             /* Regular variants should always come before the
1526              * bitmap & drawpixels variants, (unless there
1527              * are no regular variants) so that
1528              * st_update_fp can take a fast path when
1529              * shader_has_one_variant is set.
1530              */
1531             if (!stfp->fp_variants) {
1532                stfp->fp_variants = fpv;
1533             } else {
1534                /* insert into list after the first one */
1535                fpv->next = stfp->fp_variants->next;
1536                stfp->fp_variants->next = fpv;
1537             }
1538          } else {
1539             /* insert into list */
1540             fpv->next = stfp->fp_variants;
1541             stfp->fp_variants = fpv;
1542          }
1543       }
1544    }
1545
1546    return fpv;
1547 }
1548
1549 /**
1550  * Translate a program. This is common code for geometry and tessellation
1551  * shaders.
1552  */
1553 bool
1554 st_translate_common_program(struct st_context *st,
1555                             struct st_common_program *stcp)
1556 {
1557    struct gl_program *prog = &stcp->Base;
1558    enum pipe_shader_type stage =
1559       pipe_shader_type_from_mesa(stcp->Base.info.stage);
1560    struct ureg_program *ureg = ureg_create_with_screen(stage, st->pipe->screen);
1561
1562    if (ureg == NULL)
1563       return false;
1564
1565    switch (stage) {
1566    case PIPE_SHADER_TESS_CTRL:
1567       ureg_property(ureg, TGSI_PROPERTY_TCS_VERTICES_OUT,
1568                     stcp->Base.info.tess.tcs_vertices_out);
1569       break;
1570
1571    case PIPE_SHADER_TESS_EVAL:
1572       if (stcp->Base.info.tess.primitive_mode == GL_ISOLINES)
1573          ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE, GL_LINES);
1574       else
1575          ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE,
1576                        stcp->Base.info.tess.primitive_mode);
1577
1578       STATIC_ASSERT((TESS_SPACING_EQUAL + 1) % 3 == PIPE_TESS_SPACING_EQUAL);
1579       STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD + 1) % 3 ==
1580                     PIPE_TESS_SPACING_FRACTIONAL_ODD);
1581       STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN + 1) % 3 ==
1582                     PIPE_TESS_SPACING_FRACTIONAL_EVEN);
1583
1584       ureg_property(ureg, TGSI_PROPERTY_TES_SPACING,
1585                     (stcp->Base.info.tess.spacing + 1) % 3);
1586
1587       ureg_property(ureg, TGSI_PROPERTY_TES_VERTEX_ORDER_CW,
1588                     !stcp->Base.info.tess.ccw);
1589       ureg_property(ureg, TGSI_PROPERTY_TES_POINT_MODE,
1590                     stcp->Base.info.tess.point_mode);
1591       break;
1592
1593    case PIPE_SHADER_GEOMETRY:
1594       ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM,
1595                     stcp->Base.info.gs.input_primitive);
1596       ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM,
1597                     stcp->Base.info.gs.output_primitive);
1598       ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES,
1599                     stcp->Base.info.gs.vertices_out);
1600       ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS,
1601                     stcp->Base.info.gs.invocations);
1602       break;
1603
1604    default:
1605       break;
1606    }
1607
1608    ubyte inputSlotToAttr[VARYING_SLOT_TESS_MAX];
1609    ubyte inputMapping[VARYING_SLOT_TESS_MAX];
1610    ubyte outputMapping[VARYING_SLOT_TESS_MAX];
1611    GLuint attr;
1612
1613    ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
1614    ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
1615    uint num_inputs = 0;
1616
1617    ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
1618    ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
1619    uint num_outputs = 0;
1620
1621    GLint i;
1622
1623    memset(inputSlotToAttr, 0, sizeof(inputSlotToAttr));
1624    memset(inputMapping, 0, sizeof(inputMapping));
1625    memset(outputMapping, 0, sizeof(outputMapping));
1626    memset(&stcp->state, 0, sizeof(stcp->state));
1627
1628    if (prog->info.clip_distance_array_size)
1629       ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
1630                     prog->info.clip_distance_array_size);
1631    if (prog->info.cull_distance_array_size)
1632       ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
1633                     prog->info.cull_distance_array_size);
1634
1635    /*
1636     * Convert Mesa program inputs to TGSI input register semantics.
1637     */
1638    for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1639       if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) == 0)
1640          continue;
1641
1642       unsigned slot = num_inputs++;
1643
1644       inputMapping[attr] = slot;
1645       inputSlotToAttr[slot] = attr;
1646
1647       unsigned semantic_name, semantic_index;
1648       tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1649                                    &semantic_name, &semantic_index);
1650       input_semantic_name[slot] = semantic_name;
1651       input_semantic_index[slot] = semantic_index;
1652    }
1653
1654    /* Also add patch inputs. */
1655    for (attr = 0; attr < 32; attr++) {
1656       if (prog->info.patch_inputs_read & (1u << attr)) {
1657          GLuint slot = num_inputs++;
1658          GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1659
1660          inputMapping[patch_attr] = slot;
1661          inputSlotToAttr[slot] = patch_attr;
1662          input_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1663          input_semantic_index[slot] = attr;
1664       }
1665    }
1666
1667    /* initialize output semantics to defaults */
1668    for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
1669       output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
1670       output_semantic_index[i] = 0;
1671    }
1672
1673    /*
1674     * Determine number of outputs, the (default) output register
1675     * mapping and the semantic information for each output.
1676     */
1677    for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1678       if (prog->info.outputs_written & BITFIELD64_BIT(attr)) {
1679          GLuint slot = num_outputs++;
1680
1681          outputMapping[attr] = slot;
1682
1683          unsigned semantic_name, semantic_index;
1684          tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1685                                       &semantic_name, &semantic_index);
1686          output_semantic_name[slot] = semantic_name;
1687          output_semantic_index[slot] = semantic_index;
1688       }
1689    }
1690
1691    /* Also add patch outputs. */
1692    for (attr = 0; attr < 32; attr++) {
1693       if (prog->info.patch_outputs_written & (1u << attr)) {
1694          GLuint slot = num_outputs++;
1695          GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1696
1697          outputMapping[patch_attr] = slot;
1698          output_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1699          output_semantic_index[slot] = attr;
1700       }
1701    }
1702
1703    st_translate_program(st->ctx,
1704                         stage,
1705                         ureg,
1706                         stcp->glsl_to_tgsi,
1707                         prog,
1708                         /* inputs */
1709                         num_inputs,
1710                         inputMapping,
1711                         inputSlotToAttr,
1712                         input_semantic_name,
1713                         input_semantic_index,
1714                         NULL,
1715                         /* outputs */
1716                         num_outputs,
1717                         outputMapping,
1718                         output_semantic_name,
1719                         output_semantic_index);
1720
1721    stcp->state.tokens = ureg_get_tokens(ureg, NULL);
1722
1723    ureg_destroy(ureg);
1724
1725    st_translate_stream_output_info(prog);
1726
1727    st_store_ir_in_disk_cache(st, prog, false);
1728
1729    if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
1730       _mesa_print_program(prog);
1731       debug_printf("\n");
1732    }
1733
1734    if (ST_DEBUG & DEBUG_TGSI) {
1735       tgsi_dump(stcp->state.tokens, 0);
1736       debug_printf("\n");
1737    }
1738
1739    free_glsl_to_tgsi_visitor(stcp->glsl_to_tgsi);
1740    stcp->glsl_to_tgsi = NULL;
1741    return true;
1742 }
1743
1744
1745 /**
1746  * Get/create a basic program variant.
1747  */
1748 struct st_common_variant *
1749 st_get_common_variant(struct st_context *st,
1750                       struct st_common_program *prog,
1751                       const struct st_common_variant_key *key)
1752 {
1753    struct pipe_context *pipe = st->pipe;
1754    struct st_common_variant *v;
1755    struct pipe_shader_state state = {0};
1756
1757    /* Search for existing variant */
1758    for (v = prog->variants; v; v = v->next) {
1759       if (memcmp(&v->key, key, sizeof(*key)) == 0) {
1760          break;
1761       }
1762    }
1763
1764    if (!v) {
1765       /* create new */
1766       v = CALLOC_STRUCT(st_common_variant);
1767       if (v) {
1768          if (prog->state.type == PIPE_SHADER_IR_NIR) {
1769             bool finalize = false;
1770
1771             state.type = PIPE_SHADER_IR_NIR;
1772             state.ir.nir = nir_shader_clone(NULL, prog->state.ir.nir);
1773
1774             if (key->clamp_color) {
1775                NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
1776                finalize = true;
1777             }
1778
1779             state.stream_output = prog->state.stream_output;
1780
1781             if (finalize || !st->allow_st_finalize_nir_twice) {
1782                st_finalize_nir(st, &prog->Base, prog->shader_program,
1783                                state.ir.nir, true);
1784             }
1785          } else {
1786             if (key->lower_depth_clamp) {
1787                struct gl_program_parameter_list *params = prog->Base.Parameters;
1788
1789                unsigned depth_range_const =
1790                      _mesa_add_state_reference(params, depth_range_state);
1791
1792                const struct tgsi_token *tokens;
1793                tokens =
1794                      st_tgsi_lower_depth_clamp(prog->state.tokens,
1795                                                depth_range_const,
1796                                                key->clip_negative_one_to_one);
1797
1798                if (tokens != prog->state.tokens)
1799                   tgsi_free_tokens(prog->state.tokens);
1800
1801                prog->state.tokens = tokens;
1802             }
1803             state = prog->state;
1804          }
1805          /* fill in new variant */
1806          switch (prog->Base.info.stage) {
1807          case MESA_SHADER_TESS_CTRL:
1808             v->driver_shader = pipe->create_tcs_state(pipe, &state);
1809             break;
1810          case MESA_SHADER_TESS_EVAL:
1811             v->driver_shader = pipe->create_tes_state(pipe, &state);
1812             break;
1813          case MESA_SHADER_GEOMETRY:
1814             v->driver_shader = pipe->create_gs_state(pipe, &state);
1815             break;
1816          case MESA_SHADER_COMPUTE: {
1817             struct pipe_compute_state cs = {0};
1818             cs.ir_type = state.type;
1819             cs.req_local_mem = prog->Base.info.cs.shared_size;
1820
1821             if (state.type == PIPE_SHADER_IR_NIR)
1822                cs.prog = state.ir.nir;
1823             else
1824                cs.prog = state.tokens;
1825
1826             v->driver_shader = pipe->create_compute_state(pipe, &cs);
1827             break;
1828          }
1829          default:
1830             assert(!"unhandled shader type");
1831             free(v);
1832             return NULL;
1833          }
1834
1835          v->key = *key;
1836
1837          /* insert into list */
1838          v->next = prog->variants;
1839          prog->variants = v;
1840       }
1841    }
1842
1843    return v;
1844 }
1845
1846
1847 /**
1848  * Vert/Geom/Frag programs have per-context variants.  Free all the
1849  * variants attached to the given program which match the given context.
1850  */
1851 static void
1852 destroy_program_variants(struct st_context *st, struct gl_program *target)
1853 {
1854    if (!target || target == &_mesa_DummyProgram)
1855       return;
1856
1857    switch (target->Target) {
1858    case GL_VERTEX_PROGRAM_ARB:
1859       {
1860          struct st_vertex_program *stvp = (struct st_vertex_program *) target;
1861          struct st_vp_variant *vpv, **prevPtr = &stvp->variants;
1862
1863          for (vpv = stvp->variants; vpv; ) {
1864             struct st_vp_variant *next = vpv->next;
1865             if (vpv->key.st == st) {
1866                /* unlink from list */
1867                *prevPtr = next;
1868                /* destroy this variant */
1869                delete_vp_variant(st, vpv);
1870             }
1871             else {
1872                prevPtr = &vpv->next;
1873             }
1874             vpv = next;
1875          }
1876       }
1877       break;
1878    case GL_FRAGMENT_PROGRAM_ARB:
1879       {
1880          struct st_common_program *stfp =
1881             (struct st_common_program *) target;
1882          struct st_fp_variant *fpv, **prevPtr = &stfp->fp_variants;
1883
1884          for (fpv = stfp->fp_variants; fpv; ) {
1885             struct st_fp_variant *next = fpv->next;
1886             if (fpv->key.st == st) {
1887                /* unlink from list */
1888                *prevPtr = next;
1889                /* destroy this variant */
1890                delete_fp_variant(st, fpv);
1891             }
1892             else {
1893                prevPtr = &fpv->next;
1894             }
1895             fpv = next;
1896          }
1897       }
1898       break;
1899    case GL_GEOMETRY_PROGRAM_NV:
1900    case GL_TESS_CONTROL_PROGRAM_NV:
1901    case GL_TESS_EVALUATION_PROGRAM_NV:
1902    case GL_COMPUTE_PROGRAM_NV:
1903       {
1904          struct st_common_program *p = st_common_program(target);
1905          struct st_common_variant *v, **prevPtr = &p->variants;
1906
1907          for (v = p->variants; v; ) {
1908             struct st_common_variant *next = v->next;
1909             if (v->key.st == st) {
1910                /* unlink from list */
1911                *prevPtr = next;
1912                /* destroy this variant */
1913                delete_basic_variant(st, v, target->Target);
1914             }
1915             else {
1916                prevPtr = &v->next;
1917             }
1918             v = next;
1919          }
1920       }
1921       break;
1922    default:
1923       _mesa_problem(NULL, "Unexpected program target 0x%x in "
1924                     "destroy_program_variants_cb()", target->Target);
1925    }
1926 }
1927
1928
1929 /**
1930  * Callback for _mesa_HashWalk.  Free all the shader's program variants
1931  * which match the given context.
1932  */
1933 static void
1934 destroy_shader_program_variants_cb(GLuint key, void *data, void *userData)
1935 {
1936    struct st_context *st = (struct st_context *) userData;
1937    struct gl_shader *shader = (struct gl_shader *) data;
1938
1939    switch (shader->Type) {
1940    case GL_SHADER_PROGRAM_MESA:
1941       {
1942          struct gl_shader_program *shProg = (struct gl_shader_program *) data;
1943          GLuint i;
1944
1945          for (i = 0; i < ARRAY_SIZE(shProg->_LinkedShaders); i++) {
1946             if (shProg->_LinkedShaders[i])
1947                destroy_program_variants(st, shProg->_LinkedShaders[i]->Program);
1948          }
1949       }
1950       break;
1951    case GL_VERTEX_SHADER:
1952    case GL_FRAGMENT_SHADER:
1953    case GL_GEOMETRY_SHADER:
1954    case GL_TESS_CONTROL_SHADER:
1955    case GL_TESS_EVALUATION_SHADER:
1956    case GL_COMPUTE_SHADER:
1957       break;
1958    default:
1959       assert(0);
1960    }
1961 }
1962
1963
1964 /**
1965  * Callback for _mesa_HashWalk.  Free all the program variants which match
1966  * the given context.
1967  */
1968 static void
1969 destroy_program_variants_cb(GLuint key, void *data, void *userData)
1970 {
1971    struct st_context *st = (struct st_context *) userData;
1972    struct gl_program *program = (struct gl_program *) data;
1973    destroy_program_variants(st, program);
1974 }
1975
1976
1977 /**
1978  * Walk over all shaders and programs to delete any variants which
1979  * belong to the given context.
1980  * This is called during context tear-down.
1981  */
1982 void
1983 st_destroy_program_variants(struct st_context *st)
1984 {
1985    /* If shaders can be shared with other contexts, the last context will
1986     * call DeleteProgram on all shaders, releasing everything.
1987     */
1988    if (st->has_shareable_shaders)
1989       return;
1990
1991    /* ARB vert/frag program */
1992    _mesa_HashWalk(st->ctx->Shared->Programs,
1993                   destroy_program_variants_cb, st);
1994
1995    /* GLSL vert/frag/geom shaders */
1996    _mesa_HashWalk(st->ctx->Shared->ShaderObjects,
1997                   destroy_shader_program_variants_cb, st);
1998 }
1999
2000
2001 /**
2002  * For debugging, print/dump the current vertex program.
2003  */
2004 void
2005 st_print_current_vertex_program(void)
2006 {
2007    GET_CURRENT_CONTEXT(ctx);
2008
2009    if (ctx->VertexProgram._Current) {
2010       struct st_vertex_program *stvp =
2011          (struct st_vertex_program *) ctx->VertexProgram._Current;
2012       struct st_vp_variant *stv;
2013
2014       debug_printf("Vertex program %u\n", stvp->Base.Id);
2015
2016       for (stv = stvp->variants; stv; stv = stv->next) {
2017          debug_printf("variant %p\n", stv);
2018          tgsi_dump(stv->tokens, 0);
2019       }
2020    }
2021 }
2022
2023
2024 /**
2025  * Compile one shader variant.
2026  */
2027 void
2028 st_precompile_shader_variant(struct st_context *st,
2029                              struct gl_program *prog)
2030 {
2031    switch (prog->Target) {
2032    case GL_VERTEX_PROGRAM_ARB: {
2033       struct st_vertex_program *p = (struct st_vertex_program *)prog;
2034       struct st_common_variant_key key;
2035
2036       memset(&key, 0, sizeof(key));
2037
2038       key.st = st->has_shareable_shaders ? NULL : st;
2039       st_get_vp_variant(st, p, &key);
2040       break;
2041    }
2042
2043    case GL_FRAGMENT_PROGRAM_ARB: {
2044       struct st_common_program *p = (struct st_common_program *)prog;
2045       struct st_fp_variant_key key;
2046
2047       memset(&key, 0, sizeof(key));
2048
2049       key.st = st->has_shareable_shaders ? NULL : st;
2050       st_get_fp_variant(st, p, &key);
2051       break;
2052    }
2053
2054    case GL_TESS_CONTROL_PROGRAM_NV:
2055    case GL_TESS_EVALUATION_PROGRAM_NV:
2056    case GL_GEOMETRY_PROGRAM_NV:
2057    case GL_COMPUTE_PROGRAM_NV: {
2058       struct st_common_program *p = st_common_program(prog);
2059       struct st_common_variant_key key;
2060
2061       memset(&key, 0, sizeof(key));
2062
2063       key.st = st->has_shareable_shaders ? NULL : st;
2064       st_get_common_variant(st, p, &key);
2065       break;
2066    }
2067
2068    default:
2069       assert(0);
2070    }
2071 }