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, unsigned drvloc,
  45                     bool output, gl_varying_slot slot)
  46 {
  47    nir_variable *var = rzalloc(shader, nir_variable);
  48
  49    var->data.driver_location = drvloc;
  50    var->type = glsl_vec4_type();
  51    var->data.mode = output ? nir_var_shader_out : nir_var_shader_in;
  52    var->name = ralloc_asprintf(var, "clipdist_%d", drvloc);
  53    var->data.index = 0;
  54    var->data.location = slot;
  55
  56    if (output) {
  57       exec_list_push_tail(&shader->outputs, &var->node);
  58       shader->num_outputs++; /* TODO use type_size() */
  59    }
  60    else {
  61       exec_list_push_tail(&shader->inputs, &var->node);
  62       shader->num_inputs++;  /* TODO use type_size() */
  63    }
  64    return var;
  65 }
  66
  67 static void
  68 create_clipdist_vars(nir_shader *shader, nir_variable **io_vars,
  69                      unsigned ucp_enables, int *drvloc, bool output)
  70 {
  71    if (ucp_enables & 0x0f)
  72       io_vars[0] =
  73          create_clipdist_var(shader, ++(*drvloc), output,
  74                              VARYING_SLOT_CLIP_DIST0);
  75    if (ucp_enables & 0xf0)
  76       io_vars[1] =
  77          create_clipdist_var(shader, ++(*drvloc), output,
  78                              VARYING_SLOT_CLIP_DIST1);
  79 }
  80
  81 static void
  82 store_clipdist_output(nir_builder *b, nir_variable *out, nir_ssa_def **val)
  83 {
  84    nir_intrinsic_instr *store;
  85
  86    store = nir_intrinsic_instr_create(b->shader, nir_intrinsic_store_output);
  87    store->num_components = 4;
  88    nir_intrinsic_set_base(store, out->data.driver_location);
  89    nir_intrinsic_set_write_mask(store, 0xf);
  90    store->src[0].ssa = nir_vec4(b, val[0], val[1], val[2], val[3]);
  91    store->src[0].is_ssa = true;
  92    store->src[1] = nir_src_for_ssa(nir_imm_int(b, 0));
  93    nir_builder_instr_insert(b, &store->instr);
  94 }
  95
  96 static void
  97 load_clipdist_input(nir_builder *b, nir_variable *in, nir_ssa_def **val)
  98 {
  99    nir_intrinsic_instr *load;
 100
 101    load = nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_input);
 102    load->num_components = 4;
 103    nir_intrinsic_set_base(load, in->data.driver_location);
 104    load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
 105    nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
 106    nir_builder_instr_insert(b, &load->instr);
 107
 108    val[0] = nir_channel(b, &load->dest.ssa, 0);
 109    val[1] = nir_channel(b, &load->dest.ssa, 1);
 110    val[2] = nir_channel(b, &load->dest.ssa, 2);
 111    val[3] = nir_channel(b, &load->dest.ssa, 3);
 112 }
 113
 114 static nir_ssa_def *
 115 find_output_in_block(nir_block *block, unsigned drvloc)
 116 {
 117    nir_foreach_instr(instr, block) {
 118
 119       if (instr->type == nir_instr_type_intrinsic) {
 120          nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
 121          if ((intr->intrinsic == nir_intrinsic_store_output) &&
 122              nir_intrinsic_base(intr) == drvloc) {
 123             assert(intr->src[0].is_ssa);
 124             assert(nir_src_is_const(intr->src[1]));
 125             return intr->src[0].ssa;
 126          }
 127       }
 128    }
 129
 130    return NULL;
 131 }
 132
 133 /* TODO: maybe this would be a useful helper?
 134  * NOTE: assumes each output is written exactly once (and unconditionally)
 135  * so if needed nir_lower_outputs_to_temporaries()
 136  */
 137 static nir_ssa_def *
 138 find_output(nir_shader *shader, unsigned drvloc)
 139 {
 140    nir_ssa_def *def = NULL;
 141    nir_foreach_function(function, shader) {
 142       if (function->impl) {
 143          nir_foreach_block_reverse(block, function->impl) {
 144             nir_ssa_def *new_def = find_output_in_block(block, drvloc);
 145             assert(!(new_def && def));
 146             def = new_def;
 147 #if !defined(DEBUG)
 148             /* for debug builds, scan entire shader to assert
 149              * if output is written multiple times.  For release
 150              * builds just assume all is well and bail when we
 151              * find first:
 152              */
 153             if (def)
 154                break;
 155 #endif
 156          }
 157       }
 158    }
 159
 160    return def;
 161 }
 162
 163 static bool
 164 find_clipvertex_and_position_outputs(nir_shader *shader,
 165                                      nir_variable **clipvertex,
 166                                      nir_variable **position)
 167 {
 168    nir_foreach_variable(var, &shader->outputs) {
 169       switch (var->data.location) {
 170       case VARYING_SLOT_POS:
 171          *position = var;
 172          break;
 173       case VARYING_SLOT_CLIP_VERTEX:
 174          *clipvertex = var;
 175          break;
 176       case VARYING_SLOT_CLIP_DIST0:
 177       case VARYING_SLOT_CLIP_DIST1:
 178          /* if shader is already writing CLIPDIST, then
 179           * there should be no user-clip-planes to deal
 180           * with.
 181           *
 182           * We assume nir_remove_dead_variables has removed the clipdist
 183           * variables if they're not written.
 184           */
 185          return false;
 186       }
 187    }
 188
 189    return *clipvertex || *position;
 190 }
 191
 192 static void
 193 lower_clip_outputs(nir_builder *b, nir_variable *position,
 194                    nir_variable *clipvertex, nir_variable **out,
 195                    unsigned ucp_enables, bool use_vars)
 196 {
 197    nir_ssa_def *clipdist[MAX_CLIP_PLANES];
 198    nir_ssa_def *cv;
 199
 200    if (use_vars) {
 201       cv = nir_load_var(b, clipvertex ? clipvertex : position);
 202
 203       if (clipvertex) {
 204          exec_node_remove(&clipvertex->node);
 205          clipvertex->data.mode = nir_var_shader_temp;
 206          exec_list_push_tail(&b->shader->globals, &clipvertex->node);
 207       }
 208    } else {
 209       if (clipvertex)
 210          cv = find_output(b->shader, clipvertex->data.driver_location);
 211       else {
 212          assert(position);
 213          cv = find_output(b->shader, position->data.driver_location);
 214       }
 215    }
 216
 217    for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
 218       if (ucp_enables & (1 << plane)) {
 219          nir_ssa_def *ucp = nir_load_user_clip_plane(b, plane);
 220
 221          /* calculate clipdist[plane] - dot(ucp, cv): */
 222          clipdist[plane] = nir_fdot4(b, ucp, cv);
 223       } else {
 224          /* 0.0 == don't-clip == disabled: */
 225          clipdist[plane] = nir_imm_float(b, 0.0);
 226       }
 227    }
 228
 229    if (use_vars) {
 230       if (ucp_enables & 0x0f)
 231          nir_store_var(b, out[0], nir_vec(b, clipdist, 4), 0xf);
 232       if (ucp_enables & 0xf0)
 233          nir_store_var(b, out[1], nir_vec(b, &clipdist[4], 4), 0xf);
 234    } else {
 235       if (ucp_enables & 0x0f)
 236          store_clipdist_output(b, out[0], &clipdist[0]);
 237       if (ucp_enables & 0xf0)
 238          store_clipdist_output(b, out[1], &clipdist[4]);
 239    }
 240 }
 241
 242 /*
 243  * VS lowering
 244  */
 245
 246 /* ucp_enables is bitmask of enabled ucps.  Actual ucp values are
 247  * passed in to shader via user_clip_plane system-values
 248  *
 249  * If use_vars is true, the pass will use variable loads and stores instead
 250  * of working with store_output intrinsics.
 251  */
 252 bool
 253 nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars)
 254 {
 255    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 256    nir_builder b;
 257    int maxloc = -1;
 258    nir_variable *position = NULL;
 259    nir_variable *clipvertex = NULL;
 260    nir_variable *out[2] = { NULL };
 261
 262    if (!ucp_enables)
 263       return false;
 264
 265    nir_builder_init(&b, impl);
 266
 267    /* NIR should ensure that, even in case of loops/if-else, there
 268     * should be only a single predecessor block to end_block, which
 269     * makes the perfect place to insert the clipdist calculations.
 270     *
 271     * NOTE: in case of early returns, these would have to be lowered
 272     * to jumps to end_block predecessor in a previous pass.  Not sure
 273     * if there is a good way to sanity check this, but for now the
 274     * users of this pass don't support sub-routines.
 275     */
 276    assert(impl->end_block->predecessors->entries == 1);
 277    b.cursor = nir_after_cf_list(&impl->body);
 278
 279    /* find clipvertex/position outputs */
 280    if (!find_clipvertex_and_position_outputs(shader, &clipvertex, &position))
 281       return false;
 282
 283    /* insert CLIPDIST outputs */
 284    create_clipdist_vars(shader, out, ucp_enables, &maxloc, true);
 285
 286    lower_clip_outputs(&b, position, clipvertex, out, ucp_enables, use_vars);
 287
 288    nir_metadata_preserve(impl, nir_metadata_dominance);
 289
 290    return true;
 291 }
 292
 293 /*
 294  * FS lowering
 295  */
 296
 297 static void
 298 lower_clip_fs(nir_function_impl *impl, unsigned ucp_enables,
 299               nir_variable **in)
 300 {
 301    nir_ssa_def *clipdist[MAX_CLIP_PLANES];
 302    nir_builder b;
 303
 304    nir_builder_init(&b, impl);
 305    b.cursor = nir_before_cf_list(&impl->body);
 306
 307    if (ucp_enables & 0x0f)
 308       load_clipdist_input(&b, in[0], &clipdist[0]);
 309    if (ucp_enables & 0xf0)
 310       load_clipdist_input(&b, in[1], &clipdist[4]);
 311
 312    for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
 313       if (ucp_enables & (1 << plane)) {
 314          nir_intrinsic_instr *discard;
 315          nir_ssa_def *cond;
 316
 317          cond = nir_flt(&b, clipdist[plane], nir_imm_float(&b, 0.0));
 318
 319          discard = nir_intrinsic_instr_create(b.shader,
 320                                               nir_intrinsic_discard_if);
 321          discard->src[0] = nir_src_for_ssa(cond);
 322          nir_builder_instr_insert(&b, &discard->instr);
 323
 324          b.shader->info.fs.uses_discard = true;
 325       }
 326    }
 327
 328    nir_metadata_preserve(impl, nir_metadata_dominance);
 329 }
 330
 331 /* insert conditional kill based on interpolated CLIPDIST
 332  */
 333 bool
 334 nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables)
 335 {
 336    nir_variable *in[2];
 337    int maxloc = -1;
 338
 339    if (!ucp_enables)
 340       return false;
 341
 342    nir_foreach_variable(var, &shader->inputs) {
 343       int loc = var->data.driver_location;
 344
 345       /* keep track of last used driver-location.. we'll be
 346        * appending CLIP_DIST0/CLIP_DIST1 after last existing
 347        * input:
 348        */
 349       maxloc = MAX2(maxloc, loc);
 350    }
 351
 352    /* The shader won't normally have CLIPDIST inputs, so we
 353     * must add our own:
 354     */
 355    /* insert CLIPDIST inputs */
 356    create_clipdist_vars(shader, in, ucp_enables, &maxloc, false);
 357
 358    nir_foreach_function(function, shader) {
 359       if (!strcmp(function->name, "main"))
 360          lower_clip_fs(function->impl, ucp_enables, in);
 361    }
 362
 363    return true;
 364 }