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