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