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    nir_shader_add_variable(shader, var);
  69    return var;
  70 }
  71
  72 static void
  73 create_clipdist_vars(nir_shader *shader, nir_variable **io_vars,
  74                      unsigned ucp_enables, bool output,
  75                      bool use_clipdist_array)
  76 {
  77    if (use_clipdist_array) {
  78       io_vars[0] =
  79          create_clipdist_var(shader, output,
  80                              VARYING_SLOT_CLIP_DIST0,
  81                              util_last_bit(ucp_enables));
  82    } else {
  83       if (ucp_enables & 0x0f)
  84          io_vars[0] =
  85             create_clipdist_var(shader, output,
  86                                 VARYING_SLOT_CLIP_DIST0, 0);
  87       if (ucp_enables & 0xf0)
  88          io_vars[1] =
  89             create_clipdist_var(shader, output,
  90                                 VARYING_SLOT_CLIP_DIST1, 0);
  91    }
  92 }
  93
  94 static void
  95 store_clipdist_output(nir_builder *b, nir_variable *out, nir_ssa_def **val)
  96 {
  97    nir_intrinsic_instr *store;
  98
  99    store = nir_intrinsic_instr_create(b->shader, nir_intrinsic_store_output);
 100    store->num_components = 4;
 101    nir_intrinsic_set_base(store, out->data.driver_location);
 102    nir_intrinsic_set_write_mask(store, 0xf);
 103    store->src[0].ssa = nir_vec4(b, val[0], val[1], val[2], val[3]);
 104    store->src[0].is_ssa = true;
 105    store->src[1] = nir_src_for_ssa(nir_imm_int(b, 0));
 106    nir_builder_instr_insert(b, &store->instr);
 107 }
 108
 109 static void
 110 load_clipdist_input(nir_builder *b, nir_variable *in, int location_offset,
 111                     nir_ssa_def **val)
 112 {
 113    nir_intrinsic_instr *load;
 114
 115    load = nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_input);
 116    load->num_components = 4;
 117    nir_intrinsic_set_base(load, in->data.driver_location + location_offset);
 118    load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
 119    nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
 120    nir_builder_instr_insert(b, &load->instr);
 121
 122    val[0] = nir_channel(b, &load->dest.ssa, 0);
 123    val[1] = nir_channel(b, &load->dest.ssa, 1);
 124    val[2] = nir_channel(b, &load->dest.ssa, 2);
 125    val[3] = nir_channel(b, &load->dest.ssa, 3);
 126 }
 127
 128 static nir_ssa_def *
 129 find_output_in_block(nir_block *block, unsigned drvloc)
 130 {
 131    nir_foreach_instr(instr, block) {
 132
 133       if (instr->type == nir_instr_type_intrinsic) {
 134          nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 135          if ((intr->intrinsic == nir_intrinsic_store_output) &&
 136              nir_intrinsic_base(intr) == drvloc) {
 137             assert(intr->src[0].is_ssa);
 138             assert(nir_src_is_const(intr->src[1]));
 139             return intr->src[0].ssa;
 140          }
 141       }
 142    }
 143
 144    return NULL;
 145 }
 146
 147 /* TODO: maybe this would be a useful helper?
 148  * NOTE: assumes each output is written exactly once (and unconditionally)
 149  * so if needed nir_lower_outputs_to_temporaries()
 150  */
 151 static nir_ssa_def *
 152 find_output(nir_shader *shader, unsigned drvloc)
 153 {
 154    nir_ssa_def *def = NULL;
 155    nir_foreach_function(function, shader) {
 156       if (function->impl) {
 157          nir_foreach_block_reverse(block, function->impl) {
 158             nir_ssa_def *new_def = find_output_in_block(block, drvloc);
 159             assert(!(new_def && def));
 160             def = new_def;
 161 #if !defined(DEBUG)
 162             /* for debug builds, scan entire shader to assert
 163              * if output is written multiple times.  For release
 164              * builds just assume all is well and bail when we
 165              * find first:
 166              */
 167             if (def)
 168                break;
 169 #endif
 170          }
 171       }
 172    }
 173
 174    return def;
 175 }
 176
 177 static bool
 178 find_clipvertex_and_position_outputs(nir_shader *shader,
 179                                      nir_variable **clipvertex,
 180                                      nir_variable **position)
 181 {
 182    nir_foreach_shader_out_variable(var, shader) {
 183       switch (var->data.location) {
 184       case VARYING_SLOT_POS:
 185          *position = var;
 186          break;
 187       case VARYING_SLOT_CLIP_VERTEX:
 188          *clipvertex = var;
 189          break;
 190       case VARYING_SLOT_CLIP_DIST0:
 191       case VARYING_SLOT_CLIP_DIST1:
 192          /* if shader is already writing CLIPDIST, then
 193           * there should be no user-clip-planes to deal
 194           * with.
 195           *
 196           * We assume nir_remove_dead_variables has removed the clipdist
 197           * variables if they're not written.
 198           */
 199          return false;
 200       }
 201    }
 202
 203    return *clipvertex || *position;
 204 }
 205
 206 static nir_ssa_def *
 207 get_ucp(nir_builder *b, int plane,
 208         const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 209 {
 210    if (clipplane_state_tokens) {
 211       char tmp[100];
 212       snprintf(tmp, ARRAY_SIZE(tmp), "gl_ClipPlane%dMESA", plane);
 213       nir_variable *var = nir_variable_create(b->shader,
 214                                               nir_var_uniform,
 215                                               glsl_vec4_type(),
 216                                               tmp);
 217
 218       var->num_state_slots = 1;
 219       var->state_slots = ralloc_array(var, nir_state_slot, 1);
 220       memcpy(var->state_slots[0].tokens,
 221              clipplane_state_tokens[plane],
 222              sizeof(var->state_slots[0].tokens));
 223       return nir_load_var(b, var);
 224    } else
 225       return nir_load_user_clip_plane(b, plane);
 226 }
 227
 228
 229 static void
 230 lower_clip_outputs(nir_builder *b, nir_variable *position,
 231                    nir_variable *clipvertex, nir_variable **out,
 232                    unsigned ucp_enables, bool use_vars,
 233                    bool use_clipdist_array,
 234                    const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 235 {
 236    nir_ssa_def *clipdist[MAX_CLIP_PLANES];
 237    nir_ssa_def *cv;
 238
 239    if (use_vars) {
 240       cv = nir_load_var(b, clipvertex ? clipvertex : position);
 241
 242       if (clipvertex) {
 243          clipvertex->data.mode = nir_var_shader_temp;
 244          nir_fixup_deref_modes(b->shader);
 245       }
 246    } else {
 247       if (clipvertex)
 248          cv = find_output(b->shader, clipvertex->data.driver_location);
 249       else {
 250          assert(position);
 251          cv = find_output(b->shader, position->data.driver_location);
 252       }
 253    }
 254
 255    for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
 256       if (ucp_enables & (1 << plane)) {
 257          nir_ssa_def *ucp = get_ucp(b, plane, clipplane_state_tokens);
 258
 259          /* calculate clipdist[plane] - dot(ucp, cv): */
 260          clipdist[plane] = nir_fdot4(b, ucp, cv);
 261       } else {
 262          /* 0.0 == don't-clip == disabled: */
 263          clipdist[plane] = nir_imm_float(b, 0.0);
 264       }
 265       if (use_clipdist_array && plane < util_last_bit(ucp_enables)) {
 266          assert(use_vars);
 267          nir_deref_instr *deref;
 268          deref = nir_build_deref_array_imm(b,
 269                                            nir_build_deref_var(b, out[0]),
 270                                            plane);
 271          nir_store_deref(b, deref, clipdist[plane], 1);
 272       }
 273    }
 274
 275    if (!use_clipdist_array) {
 276       if (use_vars) {
 277          if (ucp_enables & 0x0f)
 278             nir_store_var(b, out[0], nir_vec(b, clipdist, 4), 0xf);
 279          if (ucp_enables & 0xf0)
 280             nir_store_var(b, out[1], nir_vec(b, &clipdist[4], 4), 0xf);
 281       } else {
 282          if (ucp_enables & 0x0f)
 283             store_clipdist_output(b, out[0], &clipdist[0]);
 284          if (ucp_enables & 0xf0)
 285             store_clipdist_output(b, out[1], &clipdist[4]);
 286       }
 287    }
 288 }
 289
 290 /*
 291  * VS lowering
 292  */
 293
 294 /* ucp_enables is bitmask of enabled ucps.  Actual ucp values are
 295  * passed in to shader via user_clip_plane system-values
 296  *
 297  * If use_vars is true, the pass will use variable loads and stores instead
 298  * of working with store_output intrinsics.
 299  *
 300  * If use_clipdist_array is true, the pass will use compact arrays for the
 301  * clipdist output instead of two vec4s.
 302  */
 303 bool
 304 nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars,
 305                   bool use_clipdist_array,
 306                   const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 307 {
 308    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 309    nir_builder b;
 310    nir_variable *position = NULL;
 311    nir_variable *clipvertex = NULL;
 312    nir_variable *out[2] = { NULL };
 313
 314    if (!ucp_enables)
 315       return false;
 316
 317    nir_builder_init(&b, impl);
 318
 319    /* NIR should ensure that, even in case of loops/if-else, there
 320     * should be only a single predecessor block to end_block, which
 321     * makes the perfect place to insert the clipdist calculations.
 322     *
 323     * NOTE: in case of early returns, these would have to be lowered
 324     * to jumps to end_block predecessor in a previous pass.  Not sure
 325     * if there is a good way to sanity check this, but for now the
 326     * users of this pass don't support sub-routines.
 327     */
 328    assert(impl->end_block->predecessors->entries == 1);
 329    b.cursor = nir_after_cf_list(&impl->body);
 330
 331    /* find clipvertex/position outputs */
 332    if (!find_clipvertex_and_position_outputs(shader, &clipvertex, &position))
 333       return false;
 334
 335    /* insert CLIPDIST outputs */
 336    create_clipdist_vars(shader, out, ucp_enables, true,
 337                         use_clipdist_array);
 338
 339    lower_clip_outputs(&b, position, clipvertex, out, ucp_enables, use_vars,
 340                       use_clipdist_array, clipplane_state_tokens);
 341
 342    nir_metadata_preserve(impl, nir_metadata_dominance);
 343
 344    return true;
 345 }
 346
 347 static void
 348 lower_clip_in_gs_block(nir_builder *b, nir_block *block, nir_variable *position,
 349                        nir_variable *clipvertex, nir_variable **out,
 350                        unsigned ucp_enables, bool use_clipdist_array,
 351                        const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 352 {
 353    nir_foreach_instr_safe(instr, block) {
 354       if (instr->type != nir_instr_type_intrinsic)
 355          continue;
 356
 357       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 358       switch (intrin->intrinsic) {
 359       case nir_intrinsic_emit_vertex_with_counter:
 360       case nir_intrinsic_emit_vertex:
 361          b->cursor = nir_before_instr(instr);
 362          lower_clip_outputs(b, position, clipvertex, out, ucp_enables, true,
 363                             use_clipdist_array, clipplane_state_tokens);
 364          break;
 365       default:
 366          /* not interesting; skip this */
 367          break;
 368       }
 369    }
 370 }
 371
 372 /*
 373  * GS lowering
 374  */
 375
 376 bool
 377 nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables,
 378                   bool use_clipdist_array,
 379                   const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
 380 {
 381    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 382    nir_builder b;
 383    nir_variable *position = NULL;
 384    nir_variable *clipvertex = NULL;
 385    nir_variable *out[2] = { NULL };
 386
 387    if (!ucp_enables)
 388       return false;
 389
 390    /* find clipvertex/position outputs */
 391    if (!find_clipvertex_and_position_outputs(shader, &clipvertex, &position))
 392       return false;
 393
 394    /* insert CLIPDIST outputs */
 395    create_clipdist_vars(shader, out, ucp_enables, true,
 396                         use_clipdist_array);
 397
 398    nir_builder_init(&b, impl);
 399
 400    nir_foreach_block(block, impl)
 401       lower_clip_in_gs_block(&b, block, position, clipvertex, out,
 402                              ucp_enables, use_clipdist_array,
 403                              clipplane_state_tokens);
 404
 405    nir_metadata_preserve(impl, nir_metadata_dominance);
 406
 407    return true;
 408 }
 409
 410 /*
 411  * FS lowering
 412  */
 413
 414 static void
 415 lower_clip_fs(nir_function_impl *impl, unsigned ucp_enables,
 416               nir_variable **in, bool use_clipdist_array)
 417 {
 418    nir_ssa_def *clipdist[MAX_CLIP_PLANES];
 419    nir_builder b;
 420
 421    nir_builder_init(&b, impl);
 422    b.cursor = nir_before_cf_list(&impl->body);
 423
 424    if (!use_clipdist_array) {
 425       if (ucp_enables & 0x0f)
 426          load_clipdist_input(&b, in[0], 0, &clipdist[0]);
 427       if (ucp_enables & 0xf0)
 428          load_clipdist_input(&b, in[1], 0, &clipdist[4]);
 429    } else {
 430       if (ucp_enables & 0x0f)
 431          load_clipdist_input(&b, in[0], 0, &clipdist[0]);
 432       if (ucp_enables & 0xf0)
 433          load_clipdist_input(&b, in[0], 1, &clipdist[4]);
 434    }
 435
 436    for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
 437       if (ucp_enables & (1 << plane)) {
 438          nir_intrinsic_instr *discard;
 439          nir_ssa_def *cond;
 440
 441          cond = nir_flt(&b, clipdist[plane], nir_imm_float(&b, 0.0));
 442
 443          discard = nir_intrinsic_instr_create(b.shader,
 444                                               nir_intrinsic_discard_if);
 445          discard->src[0] = nir_src_for_ssa(cond);
 446          nir_builder_instr_insert(&b, &discard->instr);
 447
 448          b.shader->info.fs.uses_discard = true;
 449       }
 450    }
 451
 452    nir_metadata_preserve(impl, nir_metadata_dominance);
 453 }
 454
 455 static bool
 456 fs_has_clip_dist_input_var(nir_shader *shader, nir_variable **io_vars,
 457                             unsigned *ucp_enables)
 458 {
 459    assert(shader->info.stage == MESA_SHADER_FRAGMENT);
 460    nir_foreach_shader_in_variable(var, shader) {
 461       switch (var->data.location) {
 462       case VARYING_SLOT_CLIP_DIST0:
 463          assert(var->data.compact);
 464          io_vars[0] = var;
 465          *ucp_enables &= (1 << glsl_get_length(var->type)) - 1;
 466          return true;
 467       default:
 468          break;
 469       }
 470    }
 471    return false;
 472 }
 473
 474 /* insert conditional kill based on interpolated CLIPDIST
 475  */
 476 bool
 477 nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables,
 478                   bool use_clipdist_array)
 479 {
 480    nir_variable *in[2] = {0};
 481
 482    if (!ucp_enables)
 483       return false;
 484
 485    /* Fragment shaders can't read gl_ClipDistance[] in OpenGL so it will not
 486     * have the variable defined, but Vulkan allows this, in which case the
 487     * SPIR-V compiler would have already added it as a compact array.
 488     */
 489    if (!fs_has_clip_dist_input_var(shader, in, &ucp_enables))
 490       create_clipdist_vars(shader, in, ucp_enables, false, use_clipdist_array);
 491    else
 492       assert(use_clipdist_array);
 493
 494    nir_foreach_function(function, shader) {
 495       if (!strcmp(function->name, "main"))
 496          lower_clip_fs(function->impl, ucp_enables, in, use_clipdist_array);
 497    }
 498
 499    return true;
 500 }