src/compiler/nir/nir_lower_clip.c

   1 /*
   2  * Copyright © 2015 Red Hat
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #include "nir.h"
  28 #include "nir_builder.h"
  29
  30 #define MAX_CLIP_PLANES 8
  31
  32 /* Generates the lowering code for user-clip-planes, generating CLIPDIST
  33  * from UCP[n] + CLIPVERTEX or POSITION.  Additionally, an optional pass
  34  * for fragment shaders to insert conditional kills based on the inter-
  35  * polated CLIPDIST
  36  *
  37  * NOTE: should be run after nir_lower_outputs_to_temporaries() (or at
  38  * least in scenarios where you can count on each output written once
  39  * and only once).
  40  */
  41
  42
  43 static nir_variable *
  44 create_clipdist_var(nir_shader *shader,
  45                     bool output, gl_varying_slot slot, unsigned array_size)
  46 {
  47    nir_variable *var = rzalloc(shader, nir_variable);
  48
  49    /* TODO use type_size() for num_inputs/outputs */
  50    if (output) {
  51       var->data.driver_location = shader->num_outputs++;
  52       var->data.mode = nir_var_shader_out;
  53    } else {
  54       var->data.driver_location = shader->num_inputs++;
  55       var->data.mode = nir_var_shader_in;
  56    }
  57    var->name = ralloc_asprintf(var, "clipdist_%d", var->data.driver_location);
  58    var->data.index = 0;
  59    var->data.location = slot;
  60
  61    if (array_size > 0) {
  62       var->type = glsl_array_type(glsl_float_type(), array_size,
  63                                   sizeof(float));
  64       var->data.compact = 1;
  65    } else
  66       var->type = glsl_vec4_type();
  67
  68    if (output) {
  69       exec_list_push_tail(&shader->outputs, &var->node);
  70    }
  71    else {
  72       exec_list_push_tail(&shader->inputs, &var->node);
  73    }
  74    return var;
  75 }
  76
  77 static void
  78 create_clipdist_vars(nir_shader *shader, nir_variable **io_vars,
  79                      unsigned ucp_enables, bool output,
  80                      bool use_clipdist_array)
  81 {
  82    if (use_clipdist_array) {
  83       io_vars[0] =
  84          create_clipdist_var(shader, output,
  85                              VARYING_SLOT_CLIP_DIST0,
  86                              util_last_bit(ucp_enables));
  87    } else {
  88       if (ucp_enables & 0x0f)
  89          io_vars[0] =
  90             create_clipdist_var(shader, output,
  91                                 VARYING_SLOT_CLIP_DIST0, 0);
  92       if (ucp_enables & 0xf0)
  93          io_vars[1] =
  94             create_clipdist_var(shader, output,
  95                                 VARYING_SLOT_CLIP_DIST1, 0);
  96    }
  97 }
  98
  99 static void
 100 store_clipdist_output(nir_builder *b, nir_variable *out, nir_ssa_def **val)
 101 {
 102    nir_intrinsic_instr *store;
 103
 104    store = nir_intrinsic_instr_create(b->shader, nir_intrinsic_store_output);
 105    store->num_components = 4;
 106    nir_intrinsic_set_base(store, out->data.driver_location);
 107    nir_intrinsic_set_write_mask(store, 0xf);
 108    store->src[0].ssa = nir_vec4(b, val[0], val[1], val[2], val[3]);
 109    store->src[0].is_ssa = true;
 110    store->src[1] = nir_src_for_ssa(nir_imm_int(b, 0));
 111    nir_builder_instr_insert(b, &store->instr);
 112 }
 113
 114 static void
 115 load_clipdist_input(nir_builder *b, nir_variable *in, int location_offset,
 116                     nir_ssa_def **val)
 117 {
 118    nir_intrinsic_instr *load;
 119
 120    load = nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_input);
 121    load->num_components = 4;
 122    nir_intrinsic_set_base(load, in->data.driver_location + location_offset);
 123    load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
 124    nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
 125    nir_builder_instr_insert(b, &load->instr);
 126
 127    val[0] = nir_channel(b, &load->dest.ssa, 0);
 128    val[1] = nir_channel(b, &load->dest.ssa, 1);
 129    val[2] = nir_channel(b, &load->dest.ssa, 2);
 130    val[3] = nir_channel(b, &load->dest.ssa, 3);
 131 }
 132
 133 static nir_ssa_def *
 134 find_output_in_block(nir_block *block, unsigned drvloc)
 135 {
 136    nir_foreach_instr(instr, block) {
 137
 138       if (instr->type == nir_instr_type_intrinsic) {
 139          nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 140          if ((intr->intrinsic == nir_intrinsic_store_output) &&
 141              nir_intrinsic_base(intr) == drvloc) {
 142             assert(intr->src[0].is_ssa);
 143             assert(nir_src_is_const(intr->src[1]));
 144             return intr->src[0].ssa;
 145          }
 146       }
 147    }
 148
 149    return NULL;
 150 }
 151
 152 /* TODO: maybe this would be a useful helper?
 153  * NOTE: assumes each output is written exactly once (and unconditionally)
 154  * so if needed nir_lower_outputs_to_temporaries()
 155  */
 156 static nir_ssa_def *
 157 find_output(nir_shader *shader, unsigned drvloc)
 158 {
 159    nir_ssa_def *def = NULL;
 160    nir_foreach_function(function, shader) {
 161       if (function->impl) {
 162          nir_foreach_block_reverse(block, function->impl) {
 163             nir_ssa_def *new_def = find_output_in_block(block, drvloc);
 164             assert(!(new_def && def));
 165             def = new_def;
 166 #if !defined(DEBUG)
 167             /* for debug builds, scan entire shader to assert
 168              * if output is written multiple times.  For release
 169              * builds just assume all is well and bail when we
 170              * find first:
 171              */
 172             if (def)
 173                break;
 174 #endif
 175          }
 176       }
 177    }
 178
 179    return def;
 180 }
 181
 182 static bool
 183 find_clipvertex_and_position_outputs(nir_shader *shader,
 184                                      nir_variable **clipvertex,
 185                                      nir_variable **position)
 186 {
 187    nir_foreach_shader_out_variable(var, shader) {
 188       switch (var->data.location) {
 189       case VARYING_SLOT_POS:
 190          *position = var;
 191          break;
 192       case VARYING_SLOT_CLIP_VERTEX:
 193          *clipvertex = var;
 194          break;
 195       case VARYING_SLOT_CLIP_DIST0:
 196       case VARYING_SLOT_CLIP_DIST1:
 197          /* if shader is already writing CLIPDIST, then
 198           * there should be no user-clip-planes to deal
 199           * with.
 200           *
 201           * We assume nir_remove_dead_variables has removed the clipdist
 202           * variables if they're not written.
 203           */
 204          return false;
 205       }
 206    }
 207
 208    return *clipvertex || *position;
 209 }
 210
 211 static nir_ssa_def *
 212 get_ucp(nir_builder *b, int plane,
 213         const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 214 {
 215    if (clipplane_state_tokens) {
 216       char tmp[100];
 217       snprintf(tmp, ARRAY_SIZE(tmp), "gl_ClipPlane%dMESA", plane);
 218       nir_variable *var = nir_variable_create(b->shader,
 219                                               nir_var_uniform,
 220                                               glsl_vec4_type(),
 221                                               tmp);
 222
 223       var->num_state_slots = 1;
 224       var->state_slots = ralloc_array(var, nir_state_slot, 1);
 225       memcpy(var->state_slots[0].tokens,
 226              clipplane_state_tokens[plane],
 227              sizeof(var->state_slots[0].tokens));
 228       return nir_load_var(b, var);
 229    } else
 230       return nir_load_user_clip_plane(b, plane);
 231 }
 232
 233
 234 static void
 235 lower_clip_outputs(nir_builder *b, nir_variable *position,
 236                    nir_variable *clipvertex, nir_variable **out,
 237                    unsigned ucp_enables, bool use_vars,
 238                    bool use_clipdist_array,
 239                    const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 240 {
 241    nir_ssa_def *clipdist[MAX_CLIP_PLANES];
 242    nir_ssa_def *cv;
 243
 244    if (use_vars) {
 245       cv = nir_load_var(b, clipvertex ? clipvertex : position);
 246
 247       if (clipvertex) {
 248          exec_node_remove(&clipvertex->node);
 249          clipvertex->data.mode = nir_var_shader_temp;
 250          exec_list_push_tail(&b->shader->globals, &clipvertex->node);
 251          nir_fixup_deref_modes(b->shader);
 252       }
 253    } else {
 254       if (clipvertex)
 255          cv = find_output(b->shader, clipvertex->data.driver_location);
 256       else {
 257          assert(position);
 258          cv = find_output(b->shader, position->data.driver_location);
 259       }
 260    }
 261
 262    for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
 263       if (ucp_enables & (1 << plane)) {
 264          nir_ssa_def *ucp = get_ucp(b, plane, clipplane_state_tokens);
 265
 266          /* calculate clipdist[plane] - dot(ucp, cv): */
 267          clipdist[plane] = nir_fdot4(b, ucp, cv);
 268       } else {
 269          /* 0.0 == don't-clip == disabled: */
 270          clipdist[plane] = nir_imm_float(b, 0.0);
 271       }
 272       if (use_clipdist_array && plane < util_last_bit(ucp_enables)) {
 273          assert(use_vars);
 274          nir_deref_instr *deref;
 275          deref = nir_build_deref_array_imm(b,
 276                                            nir_build_deref_var(b, out[0]),
 277                                            plane);
 278          nir_store_deref(b, deref, clipdist[plane], 1);
 279       }
 280    }
 281
 282    if (!use_clipdist_array) {
 283       if (use_vars) {
 284          if (ucp_enables & 0x0f)
 285             nir_store_var(b, out[0], nir_vec(b, clipdist, 4), 0xf);
 286          if (ucp_enables & 0xf0)
 287             nir_store_var(b, out[1], nir_vec(b, &clipdist[4], 4), 0xf);
 288       } else {
 289          if (ucp_enables & 0x0f)
 290             store_clipdist_output(b, out[0], &clipdist[0]);
 291          if (ucp_enables & 0xf0)
 292             store_clipdist_output(b, out[1], &clipdist[4]);
 293       }
 294    }
 295 }
 296
 297 /*
 298  * VS lowering
 299  */
 300
 301 /* ucp_enables is bitmask of enabled ucps.  Actual ucp values are
 302  * passed in to shader via user_clip_plane system-values
 303  *
 304  * If use_vars is true, the pass will use variable loads and stores instead
 305  * of working with store_output intrinsics.
 306  *
 307  * If use_clipdist_array is true, the pass will use compact arrays for the
 308  * clipdist output instead of two vec4s.
 309  */
 310 bool
 311 nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars,
 312                   bool use_clipdist_array,
 313                   const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 314 {
 315    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 316    nir_builder b;
 317    nir_variable *position = NULL;
 318    nir_variable *clipvertex = NULL;
 319    nir_variable *out[2] = { NULL };
 320
 321    if (!ucp_enables)
 322       return false;
 323
 324    nir_builder_init(&b, impl);
 325
 326    /* NIR should ensure that, even in case of loops/if-else, there
 327     * should be only a single predecessor block to end_block, which
 328     * makes the perfect place to insert the clipdist calculations.
 329     *
 330     * NOTE: in case of early returns, these would have to be lowered
 331     * to jumps to end_block predecessor in a previous pass.  Not sure
 332     * if there is a good way to sanity check this, but for now the
 333     * users of this pass don't support sub-routines.
 334     */
 335    assert(impl->end_block->predecessors->entries == 1);
 336    b.cursor = nir_after_cf_list(&impl->body);
 337
 338    /* find clipvertex/position outputs */
 339    if (!find_clipvertex_and_position_outputs(shader, &clipvertex, &position))
 340       return false;
 341
 342    /* insert CLIPDIST outputs */
 343    create_clipdist_vars(shader, out, ucp_enables, true,
 344                         use_clipdist_array);
 345
 346    lower_clip_outputs(&b, position, clipvertex, out, ucp_enables, use_vars,
 347                       use_clipdist_array, clipplane_state_tokens);
 348
 349    nir_metadata_preserve(impl, nir_metadata_dominance);
 350
 351    return true;
 352 }
 353
 354 static void
 355 lower_clip_in_gs_block(nir_builder *b, nir_block *block, nir_variable *position,
 356                        nir_variable *clipvertex, nir_variable **out,
 357                        unsigned ucp_enables, bool use_clipdist_array,
 358                        const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 359 {
 360    nir_foreach_instr_safe(instr, block) {
 361       if (instr->type != nir_instr_type_intrinsic)
 362          continue;
 363
 364       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 365       switch (intrin->intrinsic) {
 366       case nir_intrinsic_emit_vertex_with_counter:
 367       case nir_intrinsic_emit_vertex:
 368          b->cursor = nir_before_instr(instr);
 369          lower_clip_outputs(b, position, clipvertex, out, ucp_enables, true,
 370                             use_clipdist_array, clipplane_state_tokens);
 371          break;
 372       default:
 373          /* not interesting; skip this */
 374          break;
 375       }
 376    }
 377 }
 378
 379 /*
 380  * GS lowering
 381  */
 382
 383 bool
 384 nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables,
 385                   bool use_clipdist_array,
 386                   const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 387 {
 388    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 389    nir_builder b;
 390    nir_variable *position = NULL;
 391    nir_variable *clipvertex = NULL;
 392    nir_variable *out[2] = { NULL };
 393
 394    if (!ucp_enables)
 395       return false;
 396
 397    /* find clipvertex/position outputs */
 398    if (!find_clipvertex_and_position_outputs(shader, &clipvertex, &position))
 399       return false;
 400
 401    /* insert CLIPDIST outputs */
 402    create_clipdist_vars(shader, out, ucp_enables, true,
 403                         use_clipdist_array);
 404
 405    nir_builder_init(&b, impl);
 406
 407    nir_foreach_block(block, impl)
 408       lower_clip_in_gs_block(&b, block, position, clipvertex, out,
 409                              ucp_enables, use_clipdist_array,
 410                              clipplane_state_tokens);
 411
 412    nir_metadata_preserve(impl, nir_metadata_dominance);
 413
 414    return true;
 415 }
 416
 417 /*
 418  * FS lowering
 419  */
 420
 421 static void
 422 lower_clip_fs(nir_function_impl *impl, unsigned ucp_enables,
 423               nir_variable **in, bool use_clipdist_array)
 424 {
 425    nir_ssa_def *clipdist[MAX_CLIP_PLANES];
 426    nir_builder b;
 427
 428    nir_builder_init(&b, impl);
 429    b.cursor = nir_before_cf_list(&impl->body);
 430
 431    if (!use_clipdist_array) {
 432       if (ucp_enables & 0x0f)
 433          load_clipdist_input(&b, in[0], 0, &clipdist[0]);
 434       if (ucp_enables & 0xf0)
 435          load_clipdist_input(&b, in[1], 0, &clipdist[4]);
 436    } else {
 437       if (ucp_enables & 0x0f)
 438          load_clipdist_input(&b, in[0], 0, &clipdist[0]);
 439       if (ucp_enables & 0xf0)
 440          load_clipdist_input(&b, in[0], 1, &clipdist[4]);
 441    }
 442
 443    for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
 444       if (ucp_enables & (1 << plane)) {
 445          nir_intrinsic_instr *discard;
 446          nir_ssa_def *cond;
 447
 448          cond = nir_flt(&b, clipdist[plane], nir_imm_float(&b, 0.0));
 449
 450          discard = nir_intrinsic_instr_create(b.shader,
 451                                               nir_intrinsic_discard_if);
 452          discard->src[0] = nir_src_for_ssa(cond);
 453          nir_builder_instr_insert(&b, &discard->instr);
 454
 455          b.shader->info.fs.uses_discard = true;
 456       }
 457    }
 458
 459    nir_metadata_preserve(impl, nir_metadata_dominance);
 460 }
 461
 462 static bool
 463 fs_has_clip_dist_input_var(nir_shader *shader, nir_variable **io_vars,
 464                             unsigned *ucp_enables)
 465 {
 466    assert(shader->info.stage == MESA_SHADER_FRAGMENT);
 467    nir_foreach_shader_in_variable(var, shader) {
 468       switch (var->data.location) {
 469       case VARYING_SLOT_CLIP_DIST0:
 470          assert(var->data.compact);
 471          io_vars[0] = var;
 472          *ucp_enables &= (1 << glsl_get_length(var->type)) - 1;
 473          return true;
 474       default:
 475          break;
 476       }
 477    }
 478    return false;
 479 }
 480
 481 /* insert conditional kill based on interpolated CLIPDIST
 482  */
 483 bool
 484 nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables,
 485                   bool use_clipdist_array)
 486 {
 487    nir_variable *in[2] = {0};
 488
 489    if (!ucp_enables)
 490       return false;
 491
 492    /* Fragment shaders can't read gl_ClipDistance[] in OpenGL so it will not
 493     * have the variable defined, but Vulkan allows this, in which case the
 494     * SPIR-V compiler would have already added it as a compact array.
 495     */
 496    if (!fs_has_clip_dist_input_var(shader, in, &ucp_enables))
 497       create_clipdist_vars(shader, in, ucp_enables, false, use_clipdist_array);
 498    else
 499       assert(use_clipdist_array);
 500
 501    nir_foreach_function(function, shader) {
 502       if (!strcmp(function->name, "main"))
 503          lower_clip_fs(function->impl, ucp_enables, in, use_clipdist_array);
 504    }
 505
 506    return true;
 507 }