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          NIR_PASS_V(state.ir.nir, nir_lower_global_vars_to_local);
 730          finalize = true;
 731       }
 732
 733       if (finalize || !st->allow_st_finalize_nir_twice) {
 734          st_finalize_nir(st, &stvp->Base, stvp->shader_program, state.ir.nir,
 735                          true);
 736
 737          /* Some of the lowering above may have introduced new varyings */
 738          nir_shader_gather_info(state.ir.nir,
 739                                 nir_shader_get_entrypoint(state.ir.nir));
 740       }
 741
 742       vpv->driver_shader = pipe->create_vs_state(pipe, &state);
 743
 744       /* When generating a NIR program, we usually don't have TGSI tokens.
 745        * However, we do create them for ARB_vertex_program / fixed-function VS
 746        * programs which we may need to use with the draw module for legacy
 747        * feedback/select emulation.  If they exist, copy them.
 748        *
 749        * TODO: Lowering for shader variants is not applied to TGSI when
 750        * generating a NIR shader.
 751        */
 752       if (stvp->state.tokens)
 753          vpv->tokens = tgsi_dup_tokens(stvp->state.tokens);
 754
 755       return vpv;
 756    }
 757
 758    state.type = PIPE_SHADER_IR_TGSI;
 759    state.tokens = tgsi_dup_tokens(stvp->state.tokens);
 760
 761    /* Emulate features. */
 762    if (key->clamp_color || key->passthrough_edgeflags) {
 763       const struct tgsi_token *tokens;
 764       unsigned flags =
 765          (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
 766          (key->passthrough_edgeflags ? TGSI_EMU_PASSTHROUGH_EDGEFLAG : 0);
 767
 768       tokens = tgsi_emulate(state.tokens, flags);
 769
 770       if (tokens) {
 771          tgsi_free_tokens(state.tokens);
 772          state.tokens = tokens;
 773
 774          if (key->passthrough_edgeflags)
 775             vpv->num_inputs++;
 776       } else
 777          fprintf(stderr, "mesa: cannot emulate deprecated features\n");
 778    }
 779
 780    if (key->lower_depth_clamp) {
 781       unsigned depth_range_const =
 782             _mesa_add_state_reference(params, depth_range_state);
 783
 784       const struct tgsi_token *tokens;
 785       tokens = st_tgsi_lower_depth_clamp(state.tokens, depth_range_const,
 786                                          key->clip_negative_one_to_one);
 787       if (tokens != state.tokens)
 788          tgsi_free_tokens(state.tokens);
 789       state.tokens = tokens;
 790    }
 791
 792    if (ST_DEBUG & DEBUG_TGSI) {
 793       tgsi_dump(state.tokens, 0);
 794       debug_printf("\n");
 795    }
 796
 797    vpv->driver_shader = pipe->create_vs_state(pipe, &state);
 798    /* Save this for selection/feedback/rasterpos. */
 799    vpv->tokens = state.tokens;
 800    return vpv;
 801 }
 802
 803
 804 /**
 805  * Find/create a vertex program variant.
 806  */
 807 struct st_vp_variant *
 808 st_get_vp_variant(struct st_context *st,
 809                   struct st_vertex_program *stvp,
 810                   const struct st_common_variant_key *key)
 811 {
 812    struct st_vp_variant *vpv;
 813
 814    /* Search for existing variant */
 815    for (vpv = stvp->variants; vpv; vpv = vpv->next) {
 816       if (memcmp(&vpv->key, key, sizeof(*key)) == 0) {
 817          break;
 818       }
 819    }
 820
 821    if (!vpv) {
 822       /* create now */
 823       vpv = st_create_vp_variant(st, stvp, key);
 824       if (vpv) {
 825           for (unsigned index = 0; index < vpv->num_inputs; ++index) {
 826              unsigned attr = stvp->index_to_input[index];
 827              if (attr == ST_DOUBLE_ATTRIB_PLACEHOLDER)
 828                 continue;
 829              vpv->vert_attrib_mask |= 1u << attr;
 830           }
 831
 832          /* insert into list */
 833          vpv->next = stvp->variants;
 834          stvp->variants = vpv;
 835       }
 836    }
 837
 838    return vpv;
 839 }
 840
 841
 842 /**
 843  * Translate a Mesa fragment shader into a TGSI shader.
 844  */
 845 bool
 846 st_translate_fragment_program(struct st_context *st,
 847                               struct st_common_program *stfp)
 848 {
 849    /* Non-GLSL programs: */
 850    if (!stfp->glsl_to_tgsi) {
 851       _mesa_remove_output_reads(&stfp->Base, PROGRAM_OUTPUT);
 852       if (st->ctx->Const.GLSLFragCoordIsSysVal)
 853          _mesa_program_fragment_position_to_sysval(&stfp->Base);
 854
 855       /* This determines which states will be updated when the assembly
 856        * shader is bound.
 857        *
 858        * fragment.position and glDrawPixels always use constants.
 859        */
 860       stfp->affected_states = ST_NEW_FS_STATE |
 861                               ST_NEW_SAMPLE_SHADING |
 862                               ST_NEW_FS_CONSTANTS;
 863
 864       if (stfp->ati_fs) {
 865          /* Just set them for ATI_fs unconditionally. */
 866          stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
 867                                   ST_NEW_FS_SAMPLERS;
 868       } else {
 869          /* ARB_fp */
 870          if (stfp->Base.SamplersUsed)
 871             stfp->affected_states |= ST_NEW_FS_SAMPLER_VIEWS |
 872                                      ST_NEW_FS_SAMPLERS;
 873       }
 874
 875       /* Translate to NIR. */
 876       if (!stfp->ati_fs &&
 877           st->pipe->screen->get_shader_param(st->pipe->screen,
 878                                              PIPE_SHADER_FRAGMENT,
 879                                              PIPE_SHADER_CAP_PREFERRED_IR)) {
 880          nir_shader *nir =
 881             st_translate_prog_to_nir(st, &stfp->Base, MESA_SHADER_FRAGMENT);
 882
 883          if (stfp->state.ir.nir)
 884             ralloc_free(stfp->state.ir.nir);
 885          stfp->state.type = PIPE_SHADER_IR_NIR;
 886          stfp->state.ir.nir = nir;
 887          stfp->Base.nir = nir;
 888          return true;
 889       }
 890    }
 891
 892    ubyte outputMapping[2 * FRAG_RESULT_MAX];
 893    ubyte inputMapping[VARYING_SLOT_MAX];
 894    ubyte inputSlotToAttr[VARYING_SLOT_MAX];
 895    ubyte interpMode[PIPE_MAX_SHADER_INPUTS];  /* XXX size? */
 896    GLuint attr;
 897    GLbitfield64 inputsRead;
 898    struct ureg_program *ureg;
 899
 900    GLboolean write_all = GL_FALSE;
 901
 902    ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
 903    ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
 904    uint fs_num_inputs = 0;
 905
 906    ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
 907    ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
 908    uint fs_num_outputs = 0;
 909
 910    memset(inputSlotToAttr, ~0, sizeof(inputSlotToAttr));
 911
 912    /*
 913     * Convert Mesa program inputs to TGSI input register semantics.
 914     */
 915    inputsRead = stfp->Base.info.inputs_read;
 916    for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 917       if ((inputsRead & BITFIELD64_BIT(attr)) != 0) {
 918          const GLuint slot = fs_num_inputs++;
 919
 920          inputMapping[attr] = slot;
 921          inputSlotToAttr[slot] = attr;
 922
 923          switch (attr) {
 924          case VARYING_SLOT_POS:
 925             input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 926             input_semantic_index[slot] = 0;
 927             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 928             break;
 929          case VARYING_SLOT_COL0:
 930             input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 931             input_semantic_index[slot] = 0;
 932             interpMode[slot] = stfp->glsl_to_tgsi ?
 933                TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
 934             break;
 935          case VARYING_SLOT_COL1:
 936             input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 937             input_semantic_index[slot] = 1;
 938             interpMode[slot] = stfp->glsl_to_tgsi ?
 939                TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_COLOR;
 940             break;
 941          case VARYING_SLOT_FOGC:
 942             input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 943             input_semantic_index[slot] = 0;
 944             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 945             break;
 946          case VARYING_SLOT_FACE:
 947             input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
 948             input_semantic_index[slot] = 0;
 949             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 950             break;
 951          case VARYING_SLOT_PRIMITIVE_ID:
 952             input_semantic_name[slot] = TGSI_SEMANTIC_PRIMID;
 953             input_semantic_index[slot] = 0;
 954             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 955             break;
 956          case VARYING_SLOT_LAYER:
 957             input_semantic_name[slot] = TGSI_SEMANTIC_LAYER;
 958             input_semantic_index[slot] = 0;
 959             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 960             break;
 961          case VARYING_SLOT_VIEWPORT:
 962             input_semantic_name[slot] = TGSI_SEMANTIC_VIEWPORT_INDEX;
 963             input_semantic_index[slot] = 0;
 964             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 965             break;
 966          case VARYING_SLOT_CLIP_DIST0:
 967             input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
 968             input_semantic_index[slot] = 0;
 969             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 970             break;
 971          case VARYING_SLOT_CLIP_DIST1:
 972             input_semantic_name[slot] = TGSI_SEMANTIC_CLIPDIST;
 973             input_semantic_index[slot] = 1;
 974             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 975             break;
 976          case VARYING_SLOT_CULL_DIST0:
 977          case VARYING_SLOT_CULL_DIST1:
 978             /* these should have been lowered by GLSL */
 979             assert(0);
 980             break;
 981             /* In most cases, there is nothing special about these
 982              * inputs, so adopt a convention to use the generic
 983              * semantic name and the mesa VARYING_SLOT_ number as the
 984              * index.
 985              *
 986              * All that is required is that the vertex shader labels
 987              * its own outputs similarly, and that the vertex shader
 988              * generates at least every output required by the
 989              * fragment shader plus fixed-function hardware (such as
 990              * BFC).
 991              *
 992              * However, some drivers may need us to identify the PNTC and TEXi
 993              * varyings if, for example, their capability to replace them with
 994              * sprite coordinates is limited.
 995              */
 996          case VARYING_SLOT_PNTC:
 997             if (st->needs_texcoord_semantic) {
 998                input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD;
 999                input_semantic_index[slot] = 0;
1000                interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
1001                break;
1002             }
1003             /* fall through */
1004          case VARYING_SLOT_TEX0:
1005          case VARYING_SLOT_TEX1:
1006          case VARYING_SLOT_TEX2:
1007          case VARYING_SLOT_TEX3:
1008          case VARYING_SLOT_TEX4:
1009          case VARYING_SLOT_TEX5:
1010          case VARYING_SLOT_TEX6:
1011          case VARYING_SLOT_TEX7:
1012             if (st->needs_texcoord_semantic) {
1013                input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
1014                input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
1015                interpMode[slot] = stfp->glsl_to_tgsi ?
1016                   TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
1017                break;
1018             }
1019             /* fall through */
1020          case VARYING_SLOT_VAR0:
1021          default:
1022             /* Semantic indices should be zero-based because drivers may choose
1023              * to assign a fixed slot determined by that index.
1024              * This is useful because ARB_separate_shader_objects uses location
1025              * qualifiers for linkage, and if the semantic index corresponds to
1026              * these locations, linkage passes in the driver become unecessary.
1027              *
1028              * If needs_texcoord_semantic is true, no semantic indices will be
1029              * consumed for the TEXi varyings, and we can base the locations of
1030              * the user varyings on VAR0.  Otherwise, we use TEX0 as base index.
1031              */
1032             assert(attr >= VARYING_SLOT_VAR0 || attr == VARYING_SLOT_PNTC ||
1033                    (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7));
1034             input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
1035             input_semantic_index[slot] = st_get_generic_varying_index(st, attr);
1036             if (attr == VARYING_SLOT_PNTC)
1037                interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
1038             else {
1039                interpMode[slot] = stfp->glsl_to_tgsi ?
1040                   TGSI_INTERPOLATE_COUNT : TGSI_INTERPOLATE_PERSPECTIVE;
1041             }
1042             break;
1043          }
1044       }
1045       else {
1046          inputMapping[attr] = -1;
1047       }
1048    }
1049
1050    /*
1051     * Semantics and mapping for outputs
1052     */
1053    GLbitfield64 outputsWritten = stfp->Base.info.outputs_written;
1054
1055    /* if z is written, emit that first */
1056    if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
1057       fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
1058       fs_output_semantic_index[fs_num_outputs] = 0;
1059       outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
1060       fs_num_outputs++;
1061       outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
1062    }
1063
1064    if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) {
1065       fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_STENCIL;
1066       fs_output_semantic_index[fs_num_outputs] = 0;
1067       outputMapping[FRAG_RESULT_STENCIL] = fs_num_outputs;
1068       fs_num_outputs++;
1069       outputsWritten &= ~(1 << FRAG_RESULT_STENCIL);
1070    }
1071
1072    if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)) {
1073       fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_SAMPLEMASK;
1074       fs_output_semantic_index[fs_num_outputs] = 0;
1075       outputMapping[FRAG_RESULT_SAMPLE_MASK] = fs_num_outputs;
1076       fs_num_outputs++;
1077       outputsWritten &= ~(1 << FRAG_RESULT_SAMPLE_MASK);
1078    }
1079
1080    /* handle remaining outputs (color) */
1081    for (attr = 0; attr < ARRAY_SIZE(outputMapping); attr++) {
1082       const GLbitfield64 written = attr < FRAG_RESULT_MAX ? outputsWritten :
1083          stfp->Base.SecondaryOutputsWritten;
1084       const unsigned loc = attr % FRAG_RESULT_MAX;
1085
1086       if (written & BITFIELD64_BIT(loc)) {
1087          switch (loc) {
1088          case FRAG_RESULT_DEPTH:
1089          case FRAG_RESULT_STENCIL:
1090          case FRAG_RESULT_SAMPLE_MASK:
1091             /* handled above */
1092             assert(0);
1093             break;
1094          case FRAG_RESULT_COLOR:
1095             write_all = GL_TRUE; /* fallthrough */
1096          default: {
1097             int index;
1098             assert(loc == FRAG_RESULT_COLOR ||
1099                    (FRAG_RESULT_DATA0 <= loc && loc < FRAG_RESULT_MAX));
1100
1101             index = (loc == FRAG_RESULT_COLOR) ? 0 : (loc - FRAG_RESULT_DATA0);
1102
1103             if (attr >= FRAG_RESULT_MAX) {
1104                /* Secondary color for dual source blending. */
1105                assert(index == 0);
1106                index++;
1107             }
1108
1109             fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
1110             fs_output_semantic_index[fs_num_outputs] = index;
1111             outputMapping[attr] = fs_num_outputs;
1112             break;
1113          }
1114          }
1115
1116          fs_num_outputs++;
1117       }
1118    }
1119
1120    ureg = ureg_create_with_screen(PIPE_SHADER_FRAGMENT, st->pipe->screen);
1121    if (ureg == NULL)
1122       return false;
1123
1124    if (ST_DEBUG & DEBUG_MESA) {
1125       _mesa_print_program(&stfp->Base);
1126       _mesa_print_program_parameters(st->ctx, &stfp->Base);
1127       debug_printf("\n");
1128    }
1129    if (write_all == GL_TRUE)
1130       ureg_property(ureg, TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS, 1);
1131
1132    if (stfp->Base.info.fs.depth_layout != FRAG_DEPTH_LAYOUT_NONE) {
1133       switch (stfp->Base.info.fs.depth_layout) {
1134       case FRAG_DEPTH_LAYOUT_ANY:
1135          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1136                        TGSI_FS_DEPTH_LAYOUT_ANY);
1137          break;
1138       case FRAG_DEPTH_LAYOUT_GREATER:
1139          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1140                        TGSI_FS_DEPTH_LAYOUT_GREATER);
1141          break;
1142       case FRAG_DEPTH_LAYOUT_LESS:
1143          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1144                        TGSI_FS_DEPTH_LAYOUT_LESS);
1145          break;
1146       case FRAG_DEPTH_LAYOUT_UNCHANGED:
1147          ureg_property(ureg, TGSI_PROPERTY_FS_DEPTH_LAYOUT,
1148                        TGSI_FS_DEPTH_LAYOUT_UNCHANGED);
1149          break;
1150       default:
1151          assert(0);
1152       }
1153    }
1154
1155    if (stfp->glsl_to_tgsi) {
1156       st_translate_program(st->ctx,
1157                            PIPE_SHADER_FRAGMENT,
1158                            ureg,
1159                            stfp->glsl_to_tgsi,
1160                            &stfp->Base,
1161                            /* inputs */
1162                            fs_num_inputs,
1163                            inputMapping,
1164                            inputSlotToAttr,
1165                            input_semantic_name,
1166                            input_semantic_index,
1167                            interpMode,
1168                            /* outputs */
1169                            fs_num_outputs,
1170                            outputMapping,
1171                            fs_output_semantic_name,
1172                            fs_output_semantic_index);
1173
1174       free_glsl_to_tgsi_visitor(stfp->glsl_to_tgsi);
1175    } else if (stfp->ati_fs)
1176       st_translate_atifs_program(ureg,
1177                                  stfp->ati_fs,
1178                                  &stfp->Base,
1179                                  /* inputs */
1180                                  fs_num_inputs,
1181                                  inputMapping,
1182                                  input_semantic_name,
1183                                  input_semantic_index,
1184                                  interpMode,
1185                                  /* outputs */
1186                                  fs_num_outputs,
1187                                  outputMapping,
1188                                  fs_output_semantic_name,
1189                                  fs_output_semantic_index);
1190    else
1191       st_translate_mesa_program(st->ctx,
1192                                 PIPE_SHADER_FRAGMENT,
1193                                 ureg,
1194                                 &stfp->Base,
1195                                 /* inputs */
1196                                 fs_num_inputs,
1197                                 inputMapping,
1198                                 input_semantic_name,
1199                                 input_semantic_index,
1200                                 interpMode,
1201                                 /* outputs */
1202                                 fs_num_outputs,
1203                                 outputMapping,
1204                                 fs_output_semantic_name,
1205                                 fs_output_semantic_index);
1206
1207    stfp->state.tokens = ureg_get_tokens(ureg, NULL);
1208    ureg_destroy(ureg);
1209
1210    if (stfp->glsl_to_tgsi) {
1211       stfp->glsl_to_tgsi = NULL;
1212       st_store_ir_in_disk_cache(st, &stfp->Base, false);
1213    }
1214
1215    return stfp->state.tokens != NULL;
1216 }
1217
1218 static struct st_fp_variant *
1219 st_create_fp_variant(struct st_context *st,
1220                      struct st_common_program *stfp,
1221                      const struct st_fp_variant_key *key)
1222 {
1223    struct pipe_context *pipe = st->pipe;
1224    struct st_fp_variant *variant = CALLOC_STRUCT(st_fp_variant);
1225    struct pipe_shader_state state = {0};
1226    struct gl_program_parameter_list *params = stfp->Base.Parameters;
1227    static const gl_state_index16 texcoord_state[STATE_LENGTH] =
1228       { STATE_INTERNAL, STATE_CURRENT_ATTRIB, VERT_ATTRIB_TEX0 };
1229    static const gl_state_index16 scale_state[STATE_LENGTH] =
1230       { STATE_INTERNAL, STATE_PT_SCALE };
1231    static const gl_state_index16 bias_state[STATE_LENGTH] =
1232       { STATE_INTERNAL, STATE_PT_BIAS };
1233    static const gl_state_index16 alpha_ref_state[STATE_LENGTH] =
1234       { STATE_INTERNAL, STATE_ALPHA_REF };
1235
1236    if (!variant)
1237       return NULL;
1238
1239    if (stfp->state.type == PIPE_SHADER_IR_NIR) {
1240       bool finalize = false;
1241
1242       state.type = PIPE_SHADER_IR_NIR;
1243       state.ir.nir = nir_shader_clone(NULL, stfp->state.ir.nir);
1244
1245       if (key->clamp_color) {
1246          NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
1247          finalize = true;
1248       }
1249
1250       if (key->lower_flatshade) {
1251          NIR_PASS_V(state.ir.nir, nir_lower_flatshade);
1252          finalize = true;
1253       }
1254
1255       if (key->lower_alpha_func != COMPARE_FUNC_NEVER) {
1256          _mesa_add_state_reference(params, alpha_ref_state);
1257          NIR_PASS_V(state.ir.nir, nir_lower_alpha_test, key->lower_alpha_func,
1258                     false, alpha_ref_state);
1259          finalize = true;
1260       }
1261
1262       if (key->lower_two_sided_color) {
1263          NIR_PASS_V(state.ir.nir, nir_lower_two_sided_color);
1264          finalize = true;
1265       }
1266
1267       if (key->persample_shading) {
1268           nir_shader *shader = state.ir.nir;
1269           nir_foreach_variable(var, &shader->inputs)
1270              var->data.sample = true;
1271           finalize = true;
1272       }
1273
1274       assert(!(key->bitmap && key->drawpixels));
1275
1276       /* glBitmap */
1277       if (key->bitmap) {
1278          nir_lower_bitmap_options options = {0};
1279
1280          variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1281          options.sampler = variant->bitmap_sampler;
1282          options.swizzle_xxxx = st->bitmap.tex_format == PIPE_FORMAT_R8_UNORM;
1283
1284          NIR_PASS_V(state.ir.nir, nir_lower_bitmap, &options);
1285          finalize = true;
1286       }
1287
1288       /* glDrawPixels (color only) */
1289       if (key->drawpixels) {
1290          nir_lower_drawpixels_options options = {{0}};
1291          unsigned samplers_used = stfp->Base.SamplersUsed;
1292
1293          /* Find the first unused slot. */
1294          variant->drawpix_sampler = ffs(~samplers_used) - 1;
1295          options.drawpix_sampler = variant->drawpix_sampler;
1296          samplers_used |= (1 << variant->drawpix_sampler);
1297
1298          options.pixel_maps = key->pixelMaps;
1299          if (key->pixelMaps) {
1300             variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1301             options.pixelmap_sampler = variant->pixelmap_sampler;
1302          }
1303
1304          options.scale_and_bias = key->scaleAndBias;
1305          if (key->scaleAndBias) {
1306             _mesa_add_state_reference(params, scale_state);
1307             memcpy(options.scale_state_tokens, scale_state,
1308                    sizeof(options.scale_state_tokens));
1309             _mesa_add_state_reference(params, bias_state);
1310             memcpy(options.bias_state_tokens, bias_state,
1311                    sizeof(options.bias_state_tokens));
1312          }
1313
1314          _mesa_add_state_reference(params, texcoord_state);
1315          memcpy(options.texcoord_state_tokens, texcoord_state,
1316                 sizeof(options.texcoord_state_tokens));
1317
1318          NIR_PASS_V(state.ir.nir, nir_lower_drawpixels, &options);
1319          finalize = true;
1320       }
1321
1322       if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1323                    key->external.lower_xy_uxvx || key->external.lower_yx_xuxv ||
1324                    key->external.lower_ayuv || key->external.lower_xyuv)) {
1325          nir_lower_tex_options options = {0};
1326          options.lower_y_uv_external = key->external.lower_nv12;
1327          options.lower_y_u_v_external = key->external.lower_iyuv;
1328          options.lower_xy_uxvx_external = key->external.lower_xy_uxvx;
1329          options.lower_yx_xuxv_external = key->external.lower_yx_xuxv;
1330          options.lower_ayuv_external = key->external.lower_ayuv;
1331          options.lower_xyuv_external = key->external.lower_xyuv;
1332          NIR_PASS_V(state.ir.nir, nir_lower_tex, &options);
1333          finalize = true;
1334       }
1335
1336       if (finalize || !st->allow_st_finalize_nir_twice) {
1337          st_finalize_nir(st, &stfp->Base, stfp->shader_program, state.ir.nir,
1338                          false);
1339       }
1340
1341       /* This pass needs to happen *after* nir_lower_sampler */
1342       if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1343                    key->external.lower_xy_uxvx || key->external.lower_yx_xuxv)) {
1344          NIR_PASS_V(state.ir.nir, st_nir_lower_tex_src_plane,
1345                     ~stfp->Base.SamplersUsed,
1346                     key->external.lower_nv12 || key->external.lower_xy_uxvx ||
1347                        key->external.lower_yx_xuxv,
1348                     key->external.lower_iyuv);
1349          finalize = true;
1350       }
1351
1352       if (finalize || !st->allow_st_finalize_nir_twice) {
1353          /* Some of the lowering above may have introduced new varyings */
1354          nir_shader_gather_info(state.ir.nir,
1355                                 nir_shader_get_entrypoint(state.ir.nir));
1356
1357          struct pipe_screen *screen = pipe->screen;
1358          if (screen->finalize_nir)
1359             screen->finalize_nir(screen, state.ir.nir, false);
1360       }
1361
1362       variant->driver_shader = pipe->create_fs_state(pipe, &state);
1363       variant->key = *key;
1364
1365       return variant;
1366    }
1367
1368    state.tokens = stfp->state.tokens;
1369
1370    assert(!(key->bitmap && key->drawpixels));
1371
1372    /* Fix texture targets and add fog for ATI_fs */
1373    if (stfp->ati_fs) {
1374       const struct tgsi_token *tokens = st_fixup_atifs(state.tokens, key);
1375
1376       if (tokens)
1377          state.tokens = tokens;
1378       else
1379          fprintf(stderr, "mesa: cannot post-process ATI_fs\n");
1380    }
1381
1382    /* Emulate features. */
1383    if (key->clamp_color || key->persample_shading) {
1384       const struct tgsi_token *tokens;
1385       unsigned flags =
1386          (key->clamp_color ? TGSI_EMU_CLAMP_COLOR_OUTPUTS : 0) |
1387          (key->persample_shading ? TGSI_EMU_FORCE_PERSAMPLE_INTERP : 0);
1388
1389       tokens = tgsi_emulate(state.tokens, flags);
1390
1391       if (tokens) {
1392          if (state.tokens != stfp->state.tokens)
1393             tgsi_free_tokens(state.tokens);
1394          state.tokens = tokens;
1395       } else
1396          fprintf(stderr, "mesa: cannot emulate deprecated features\n");
1397    }
1398
1399    /* glBitmap */
1400    if (key->bitmap) {
1401       const struct tgsi_token *tokens;
1402
1403       variant->bitmap_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1404
1405       tokens = st_get_bitmap_shader(state.tokens,
1406                                     st->internal_target,
1407                                     variant->bitmap_sampler,
1408                                     st->needs_texcoord_semantic,
1409                                     st->bitmap.tex_format ==
1410                                     PIPE_FORMAT_R8_UNORM);
1411
1412       if (tokens) {
1413          if (state.tokens != stfp->state.tokens)
1414             tgsi_free_tokens(state.tokens);
1415          state.tokens = tokens;
1416       } else
1417          fprintf(stderr, "mesa: cannot create a shader for glBitmap\n");
1418    }
1419
1420    /* glDrawPixels (color only) */
1421    if (key->drawpixels) {
1422       const struct tgsi_token *tokens;
1423       unsigned scale_const = 0, bias_const = 0, texcoord_const = 0;
1424
1425       /* Find the first unused slot. */
1426       variant->drawpix_sampler = ffs(~stfp->Base.SamplersUsed) - 1;
1427
1428       if (key->pixelMaps) {
1429          unsigned samplers_used = stfp->Base.SamplersUsed |
1430                                   (1 << variant->drawpix_sampler);
1431
1432          variant->pixelmap_sampler = ffs(~samplers_used) - 1;
1433       }
1434
1435       if (key->scaleAndBias) {
1436          scale_const = _mesa_add_state_reference(params, scale_state);
1437          bias_const = _mesa_add_state_reference(params, bias_state);
1438       }
1439
1440       texcoord_const = _mesa_add_state_reference(params, texcoord_state);
1441
1442       tokens = st_get_drawpix_shader(state.tokens,
1443                                      st->needs_texcoord_semantic,
1444                                      key->scaleAndBias, scale_const,
1445                                      bias_const, key->pixelMaps,
1446                                      variant->drawpix_sampler,
1447                                      variant->pixelmap_sampler,
1448                                      texcoord_const, st->internal_target);
1449
1450       if (tokens) {
1451          if (state.tokens != stfp->state.tokens)
1452             tgsi_free_tokens(state.tokens);
1453          state.tokens = tokens;
1454       } else
1455          fprintf(stderr, "mesa: cannot create a shader for glDrawPixels\n");
1456    }
1457
1458    if (unlikely(key->external.lower_nv12 || key->external.lower_iyuv ||
1459                 key->external.lower_xy_uxvx || key->external.lower_yx_xuxv)) {
1460       const struct tgsi_token *tokens;
1461
1462       /* samplers inserted would conflict, but this should be unpossible: */
1463       assert(!(key->bitmap || key->drawpixels));
1464
1465       tokens = st_tgsi_lower_yuv(state.tokens,
1466                                  ~stfp->Base.SamplersUsed,
1467                                  key->external.lower_nv12 ||
1468                                     key->external.lower_xy_uxvx ||
1469                                     key->external.lower_yx_xuxv,
1470                                  key->external.lower_iyuv);
1471       if (tokens) {
1472          if (state.tokens != stfp->state.tokens)
1473             tgsi_free_tokens(state.tokens);
1474          state.tokens = tokens;
1475       } else {
1476          fprintf(stderr, "mesa: cannot create a shader for samplerExternalOES\n");
1477       }
1478    }
1479
1480    if (key->lower_depth_clamp) {
1481       unsigned depth_range_const = _mesa_add_state_reference(params, depth_range_state);
1482
1483       const struct tgsi_token *tokens;
1484       tokens = st_tgsi_lower_depth_clamp_fs(state.tokens, depth_range_const);
1485       if (state.tokens != stfp->state.tokens)
1486          tgsi_free_tokens(state.tokens);
1487       state.tokens = tokens;
1488    }
1489
1490    if (ST_DEBUG & DEBUG_TGSI) {
1491       tgsi_dump(state.tokens, 0);
1492       debug_printf("\n");
1493    }
1494
1495    /* fill in variant */
1496    variant->driver_shader = pipe->create_fs_state(pipe, &state);
1497    variant->key = *key;
1498
1499    if (state.tokens != stfp->state.tokens)
1500       tgsi_free_tokens(state.tokens);
1501    return variant;
1502 }
1503
1504 /**
1505  * Translate fragment program if needed.
1506  */
1507 struct st_fp_variant *
1508 st_get_fp_variant(struct st_context *st,
1509                   struct st_common_program *stfp,
1510                   const struct st_fp_variant_key *key)
1511 {
1512    struct st_fp_variant *fpv;
1513
1514    /* Search for existing variant */
1515    for (fpv = stfp->fp_variants; fpv; fpv = fpv->next) {
1516       if (memcmp(&fpv->key, key, sizeof(*key)) == 0) {
1517          break;
1518       }
1519    }
1520
1521    if (!fpv) {
1522       /* create new */
1523       fpv = st_create_fp_variant(st, stfp, key);
1524       if (fpv) {
1525          if (key->bitmap || key->drawpixels) {
1526             /* Regular variants should always come before the
1527              * bitmap & drawpixels variants, (unless there
1528              * are no regular variants) so that
1529              * st_update_fp can take a fast path when
1530              * shader_has_one_variant is set.
1531              */
1532             if (!stfp->fp_variants) {
1533                stfp->fp_variants = fpv;
1534             } else {
1535                /* insert into list after the first one */
1536                fpv->next = stfp->fp_variants->next;
1537                stfp->fp_variants->next = fpv;
1538             }
1539          } else {
1540             /* insert into list */
1541             fpv->next = stfp->fp_variants;
1542             stfp->fp_variants = fpv;
1543          }
1544       }
1545    }
1546
1547    return fpv;
1548 }
1549
1550 /**
1551  * Translate a program. This is common code for geometry and tessellation
1552  * shaders.
1553  */
1554 bool
1555 st_translate_common_program(struct st_context *st,
1556                             struct st_common_program *stcp)
1557 {
1558    struct gl_program *prog = &stcp->Base;
1559    enum pipe_shader_type stage =
1560       pipe_shader_type_from_mesa(stcp->Base.info.stage);
1561    struct ureg_program *ureg = ureg_create_with_screen(stage, st->pipe->screen);
1562
1563    if (ureg == NULL)
1564       return false;
1565
1566    switch (stage) {
1567    case PIPE_SHADER_TESS_CTRL:
1568       ureg_property(ureg, TGSI_PROPERTY_TCS_VERTICES_OUT,
1569                     stcp->Base.info.tess.tcs_vertices_out);
1570       break;
1571
1572    case PIPE_SHADER_TESS_EVAL:
1573       if (stcp->Base.info.tess.primitive_mode == GL_ISOLINES)
1574          ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE, GL_LINES);
1575       else
1576          ureg_property(ureg, TGSI_PROPERTY_TES_PRIM_MODE,
1577                        stcp->Base.info.tess.primitive_mode);
1578
1579       STATIC_ASSERT((TESS_SPACING_EQUAL + 1) % 3 == PIPE_TESS_SPACING_EQUAL);
1580       STATIC_ASSERT((TESS_SPACING_FRACTIONAL_ODD + 1) % 3 ==
1581                     PIPE_TESS_SPACING_FRACTIONAL_ODD);
1582       STATIC_ASSERT((TESS_SPACING_FRACTIONAL_EVEN + 1) % 3 ==
1583                     PIPE_TESS_SPACING_FRACTIONAL_EVEN);
1584
1585       ureg_property(ureg, TGSI_PROPERTY_TES_SPACING,
1586                     (stcp->Base.info.tess.spacing + 1) % 3);
1587
1588       ureg_property(ureg, TGSI_PROPERTY_TES_VERTEX_ORDER_CW,
1589                     !stcp->Base.info.tess.ccw);
1590       ureg_property(ureg, TGSI_PROPERTY_TES_POINT_MODE,
1591                     stcp->Base.info.tess.point_mode);
1592       break;
1593
1594    case PIPE_SHADER_GEOMETRY:
1595       ureg_property(ureg, TGSI_PROPERTY_GS_INPUT_PRIM,
1596                     stcp->Base.info.gs.input_primitive);
1597       ureg_property(ureg, TGSI_PROPERTY_GS_OUTPUT_PRIM,
1598                     stcp->Base.info.gs.output_primitive);
1599       ureg_property(ureg, TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES,
1600                     stcp->Base.info.gs.vertices_out);
1601       ureg_property(ureg, TGSI_PROPERTY_GS_INVOCATIONS,
1602                     stcp->Base.info.gs.invocations);
1603       break;
1604
1605    default:
1606       break;
1607    }
1608
1609    ubyte inputSlotToAttr[VARYING_SLOT_TESS_MAX];
1610    ubyte inputMapping[VARYING_SLOT_TESS_MAX];
1611    ubyte outputMapping[VARYING_SLOT_TESS_MAX];
1612    GLuint attr;
1613
1614    ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
1615    ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
1616    uint num_inputs = 0;
1617
1618    ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
1619    ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
1620    uint num_outputs = 0;
1621
1622    GLint i;
1623
1624    memset(inputSlotToAttr, 0, sizeof(inputSlotToAttr));
1625    memset(inputMapping, 0, sizeof(inputMapping));
1626    memset(outputMapping, 0, sizeof(outputMapping));
1627    memset(&stcp->state, 0, sizeof(stcp->state));
1628
1629    if (prog->info.clip_distance_array_size)
1630       ureg_property(ureg, TGSI_PROPERTY_NUM_CLIPDIST_ENABLED,
1631                     prog->info.clip_distance_array_size);
1632    if (prog->info.cull_distance_array_size)
1633       ureg_property(ureg, TGSI_PROPERTY_NUM_CULLDIST_ENABLED,
1634                     prog->info.cull_distance_array_size);
1635
1636    /*
1637     * Convert Mesa program inputs to TGSI input register semantics.
1638     */
1639    for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1640       if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) == 0)
1641          continue;
1642
1643       unsigned slot = num_inputs++;
1644
1645       inputMapping[attr] = slot;
1646       inputSlotToAttr[slot] = attr;
1647
1648       unsigned semantic_name, semantic_index;
1649       tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1650                                    &semantic_name, &semantic_index);
1651       input_semantic_name[slot] = semantic_name;
1652       input_semantic_index[slot] = semantic_index;
1653    }
1654
1655    /* Also add patch inputs. */
1656    for (attr = 0; attr < 32; attr++) {
1657       if (prog->info.patch_inputs_read & (1u << attr)) {
1658          GLuint slot = num_inputs++;
1659          GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1660
1661          inputMapping[patch_attr] = slot;
1662          inputSlotToAttr[slot] = patch_attr;
1663          input_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1664          input_semantic_index[slot] = attr;
1665       }
1666    }
1667
1668    /* initialize output semantics to defaults */
1669    for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
1670       output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
1671       output_semantic_index[i] = 0;
1672    }
1673
1674    /*
1675     * Determine number of outputs, the (default) output register
1676     * mapping and the semantic information for each output.
1677     */
1678    for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
1679       if (prog->info.outputs_written & BITFIELD64_BIT(attr)) {
1680          GLuint slot = num_outputs++;
1681
1682          outputMapping[attr] = slot;
1683
1684          unsigned semantic_name, semantic_index;
1685          tgsi_get_gl_varying_semantic(attr, st->needs_texcoord_semantic,
1686                                       &semantic_name, &semantic_index);
1687          output_semantic_name[slot] = semantic_name;
1688          output_semantic_index[slot] = semantic_index;
1689       }
1690    }
1691
1692    /* Also add patch outputs. */
1693    for (attr = 0; attr < 32; attr++) {
1694       if (prog->info.patch_outputs_written & (1u << attr)) {
1695          GLuint slot = num_outputs++;
1696          GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
1697
1698          outputMapping[patch_attr] = slot;
1699          output_semantic_name[slot] = TGSI_SEMANTIC_PATCH;
1700          output_semantic_index[slot] = attr;
1701       }
1702    }
1703
1704    st_translate_program(st->ctx,
1705                         stage,
1706                         ureg,
1707                         stcp->glsl_to_tgsi,
1708                         prog,
1709                         /* inputs */
1710                         num_inputs,
1711                         inputMapping,
1712                         inputSlotToAttr,
1713                         input_semantic_name,
1714                         input_semantic_index,
1715                         NULL,
1716                         /* outputs */
1717                         num_outputs,
1718                         outputMapping,
1719                         output_semantic_name,
1720                         output_semantic_index);
1721
1722    stcp->state.tokens = ureg_get_tokens(ureg, NULL);
1723
1724    ureg_destroy(ureg);
1725
1726    st_translate_stream_output_info(prog);
1727
1728    st_store_ir_in_disk_cache(st, prog, false);
1729
1730    if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
1731       _mesa_print_program(prog);
1732       debug_printf("\n");
1733    }
1734
1735    if (ST_DEBUG & DEBUG_TGSI) {
1736       tgsi_dump(stcp->state.tokens, 0);
1737       debug_printf("\n");
1738    }
1739
1740    free_glsl_to_tgsi_visitor(stcp->glsl_to_tgsi);
1741    stcp->glsl_to_tgsi = NULL;
1742    return true;
1743 }
1744
1745
1746 /**
1747  * Get/create a basic program variant.
1748  */
1749 struct st_common_variant *
1750 st_get_common_variant(struct st_context *st,
1751                       struct st_common_program *prog,
1752                       const struct st_common_variant_key *key)
1753 {
1754    struct pipe_context *pipe = st->pipe;
1755    struct st_common_variant *v;
1756    struct pipe_shader_state state = {0};
1757
1758    /* Search for existing variant */
1759    for (v = prog->variants; v; v = v->next) {
1760       if (memcmp(&v->key, key, sizeof(*key)) == 0) {
1761          break;
1762       }
1763    }
1764
1765    if (!v) {
1766       /* create new */
1767       v = CALLOC_STRUCT(st_common_variant);
1768       if (v) {
1769          if (prog->state.type == PIPE_SHADER_IR_NIR) {
1770             bool finalize = false;
1771
1772             state.type = PIPE_SHADER_IR_NIR;
1773             state.ir.nir = nir_shader_clone(NULL, prog->state.ir.nir);
1774
1775             if (key->clamp_color) {
1776                NIR_PASS_V(state.ir.nir, nir_lower_clamp_color_outputs);
1777                finalize = true;
1778             }
1779
1780             state.stream_output = prog->state.stream_output;
1781
1782             if (finalize || !st->allow_st_finalize_nir_twice) {
1783                st_finalize_nir(st, &prog->Base, prog->shader_program,
1784                                state.ir.nir, true);
1785             }
1786          } else {
1787             if (key->lower_depth_clamp) {
1788                struct gl_program_parameter_list *params = prog->Base.Parameters;
1789
1790                unsigned depth_range_const =
1791                      _mesa_add_state_reference(params, depth_range_state);
1792
1793                const struct tgsi_token *tokens;
1794                tokens =
1795                      st_tgsi_lower_depth_clamp(prog->state.tokens,
1796                                                depth_range_const,
1797                                                key->clip_negative_one_to_one);
1798
1799                if (tokens != prog->state.tokens)
1800                   tgsi_free_tokens(prog->state.tokens);
1801
1802                prog->state.tokens = tokens;
1803             }
1804             state = prog->state;
1805          }
1806          /* fill in new variant */
1807          switch (prog->Base.info.stage) {
1808          case MESA_SHADER_TESS_CTRL:
1809             v->driver_shader = pipe->create_tcs_state(pipe, &state);
1810             break;
1811          case MESA_SHADER_TESS_EVAL:
1812             v->driver_shader = pipe->create_tes_state(pipe, &state);
1813             break;
1814          case MESA_SHADER_GEOMETRY:
1815             v->driver_shader = pipe->create_gs_state(pipe, &state);
1816             break;
1817          case MESA_SHADER_COMPUTE: {
1818             struct pipe_compute_state cs = {0};
1819             cs.ir_type = state.type;
1820             cs.req_local_mem = prog->Base.info.cs.shared_size;
1821
1822             if (state.type == PIPE_SHADER_IR_NIR)
1823                cs.prog = state.ir.nir;
1824             else
1825                cs.prog = state.tokens;
1826
1827             v->driver_shader = pipe->create_compute_state(pipe, &cs);
1828             break;
1829          }
1830          default:
1831             assert(!"unhandled shader type");
1832             free(v);
1833             return NULL;
1834          }
1835
1836          v->key = *key;
1837
1838          /* insert into list */
1839          v->next = prog->variants;
1840          prog->variants = v;
1841       }
1842    }
1843
1844    return v;
1845 }
1846
1847
1848 /**
1849  * Vert/Geom/Frag programs have per-context variants.  Free all the
1850  * variants attached to the given program which match the given context.
1851  */
1852 static void
1853 destroy_program_variants(struct st_context *st, struct gl_program *target)
1854 {
1855    if (!target || target == &_mesa_DummyProgram)
1856       return;
1857
1858    switch (target->Target) {
1859    case GL_VERTEX_PROGRAM_ARB:
1860       {
1861          struct st_vertex_program *stvp = (struct st_vertex_program *) target;
1862          struct st_vp_variant *vpv, **prevPtr = &stvp->variants;
1863
1864          for (vpv = stvp->variants; vpv; ) {
1865             struct st_vp_variant *next = vpv->next;
1866             if (vpv->key.st == st) {
1867                /* unlink from list */
1868                *prevPtr = next;
1869                /* destroy this variant */
1870                delete_vp_variant(st, vpv);
1871             }
1872             else {
1873                prevPtr = &vpv->next;
1874             }
1875             vpv = next;
1876          }
1877       }
1878       break;
1879    case GL_FRAGMENT_PROGRAM_ARB:
1880       {
1881          struct st_common_program *stfp =
1882             (struct st_common_program *) target;
1883          struct st_fp_variant *fpv, **prevPtr = &stfp->fp_variants;
1884
1885          for (fpv = stfp->fp_variants; fpv; ) {
1886             struct st_fp_variant *next = fpv->next;
1887             if (fpv->key.st == st) {
1888                /* unlink from list */
1889                *prevPtr = next;
1890                /* destroy this variant */
1891                delete_fp_variant(st, fpv);
1892             }
1893             else {
1894                prevPtr = &fpv->next;
1895             }
1896             fpv = next;
1897          }
1898       }
1899       break;
1900    case GL_GEOMETRY_PROGRAM_NV:
1901    case GL_TESS_CONTROL_PROGRAM_NV:
1902    case GL_TESS_EVALUATION_PROGRAM_NV:
1903    case GL_COMPUTE_PROGRAM_NV:
1904       {
1905          struct st_common_program *p = st_common_program(target);
1906          struct st_common_variant *v, **prevPtr = &p->variants;
1907
1908          for (v = p->variants; v; ) {
1909             struct st_common_variant *next = v->next;
1910             if (v->key.st == st) {
1911                /* unlink from list */
1912                *prevPtr = next;
1913                /* destroy this variant */
1914                delete_basic_variant(st, v, target->Target);
1915             }
1916             else {
1917                prevPtr = &v->next;
1918             }
1919             v = next;
1920          }
1921       }
1922       break;
1923    default:
1924       _mesa_problem(NULL, "Unexpected program target 0x%x in "
1925                     "destroy_program_variants_cb()", target->Target);
1926    }
1927 }
1928
1929
1930 /**
1931  * Callback for _mesa_HashWalk.  Free all the shader's program variants
1932  * which match the given context.
1933  */
1934 static void
1935 destroy_shader_program_variants_cb(GLuint key, void *data, void *userData)
1936 {
1937    struct st_context *st = (struct st_context *) userData;
1938    struct gl_shader *shader = (struct gl_shader *) data;
1939
1940    switch (shader->Type) {
1941    case GL_SHADER_PROGRAM_MESA:
1942       {
1943          struct gl_shader_program *shProg = (struct gl_shader_program *) data;
1944          GLuint i;
1945
1946          for (i = 0; i < ARRAY_SIZE(shProg->_LinkedShaders); i++) {
1947             if (shProg->_LinkedShaders[i])
1948                destroy_program_variants(st, shProg->_LinkedShaders[i]->Program);
1949          }
1950       }
1951       break;
1952    case GL_VERTEX_SHADER:
1953    case GL_FRAGMENT_SHADER:
1954    case GL_GEOMETRY_SHADER:
1955    case GL_TESS_CONTROL_SHADER:
1956    case GL_TESS_EVALUATION_SHADER:
1957    case GL_COMPUTE_SHADER:
1958       break;
1959    default:
1960       assert(0);
1961    }
1962 }
1963
1964
1965 /**
1966  * Callback for _mesa_HashWalk.  Free all the program variants which match
1967  * the given context.
1968  */
1969 static void
1970 destroy_program_variants_cb(GLuint key, void *data, void *userData)
1971 {
1972    struct st_context *st = (struct st_context *) userData;
1973    struct gl_program *program = (struct gl_program *) data;
1974    destroy_program_variants(st, program);
1975 }
1976
1977
1978 /**
1979  * Walk over all shaders and programs to delete any variants which
1980  * belong to the given context.
1981  * This is called during context tear-down.
1982  */
1983 void
1984 st_destroy_program_variants(struct st_context *st)
1985 {
1986    /* If shaders can be shared with other contexts, the last context will
1987     * call DeleteProgram on all shaders, releasing everything.
1988     */
1989    if (st->has_shareable_shaders)
1990       return;
1991
1992    /* ARB vert/frag program */
1993    _mesa_HashWalk(st->ctx->Shared->Programs,
1994                   destroy_program_variants_cb, st);
1995
1996    /* GLSL vert/frag/geom shaders */
1997    _mesa_HashWalk(st->ctx->Shared->ShaderObjects,
1998                   destroy_shader_program_variants_cb, st);
1999 }
2000
2001
2002 /**
2003  * For debugging, print/dump the current vertex program.
2004  */
2005 void
2006 st_print_current_vertex_program(void)
2007 {
2008    GET_CURRENT_CONTEXT(ctx);
2009
2010    if (ctx->VertexProgram._Current) {
2011       struct st_vertex_program *stvp =
2012          (struct st_vertex_program *) ctx->VertexProgram._Current;
2013       struct st_vp_variant *stv;
2014
2015       debug_printf("Vertex program %u\n", stvp->Base.Id);
2016
2017       for (stv = stvp->variants; stv; stv = stv->next) {
2018          debug_printf("variant %p\n", stv);
2019          tgsi_dump(stv->tokens, 0);
2020       }
2021    }
2022 }
2023
2024
2025 /**
2026  * Compile one shader variant.
2027  */
2028 void
2029 st_precompile_shader_variant(struct st_context *st,
2030                              struct gl_program *prog)
2031 {
2032    switch (prog->Target) {
2033    case GL_VERTEX_PROGRAM_ARB: {
2034       struct st_vertex_program *p = (struct st_vertex_program *)prog;
2035       struct st_common_variant_key key;
2036
2037       memset(&key, 0, sizeof(key));
2038
2039       key.st = st->has_shareable_shaders ? NULL : st;
2040       st_get_vp_variant(st, p, &key);
2041       break;
2042    }
2043
2044    case GL_FRAGMENT_PROGRAM_ARB: {
2045       struct st_common_program *p = (struct st_common_program *)prog;
2046       struct st_fp_variant_key key;
2047
2048       memset(&key, 0, sizeof(key));
2049
2050       key.st = st->has_shareable_shaders ? NULL : st;
2051       st_get_fp_variant(st, p, &key);
2052       break;
2053    }
2054
2055    case GL_TESS_CONTROL_PROGRAM_NV:
2056    case GL_TESS_EVALUATION_PROGRAM_NV:
2057    case GL_GEOMETRY_PROGRAM_NV:
2058    case GL_COMPUTE_PROGRAM_NV: {
2059       struct st_common_program *p = st_common_program(prog);
2060       struct st_common_variant_key key;
2061
2062       memset(&key, 0, sizeof(key));
2063
2064       key.st = st->has_shareable_shaders ? NULL : st;
2065       st_get_common_variant(st, p, &key);
2066       break;
2067    }
2068
2069    default:
2070       assert(0);
2071    }
2072 }