src/compiler/nir/nir_lower_clip_cull_distance_arrays.c

   1 /*
   2  * Copyright © 2015 Intel Corporation
   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
  24 #include "nir.h"
  25 #include "nir_builder.h"
  26
  27 /**
  28  * @file
  29  *
  30  * This pass combines separate clip and cull distance arrays into a
  31  * single array that contains both.  Clip distances come first, then
  32  * cull distances.  It also populates nir_shader_info with the size
  33  * of the original arrays so the driver knows which are which.
  34  */
  35
  36 /**
  37  * Get the length of the clip/cull distance array, looking past
  38  * any interface block arrays.
  39  */
  40 static unsigned
  41 get_unwrapped_array_length(nir_shader *nir, nir_variable *var)
  42 {
  43    if (!var)
  44       return 0;
  45
  46    /* Unwrap GS input and TCS input/output interfaces.  We want the
  47     * underlying clip/cull distance array length, not the per-vertex
  48     * array length.
  49     */
  50    const struct glsl_type *type = var->type;
  51    if (nir_is_per_vertex_io(var, nir->info.stage))
  52       type = glsl_get_array_element(type);
  53
  54    assert(glsl_type_is_array(type));
  55
  56    return glsl_get_length(type);
  57 }
  58
  59 /**
  60  * Update the type of the combined array (including interface block nesting).
  61  */
  62 static void
  63 update_type(nir_variable *var, gl_shader_stage stage, unsigned length)
  64 {
  65    const struct glsl_type *type = glsl_array_type(glsl_float_type(), length);
  66
  67    if (nir_is_per_vertex_io(var, stage))
  68       type = glsl_array_type(type, glsl_get_length(var->type));
  69
  70    var->type = type;
  71 }
  72
  73 /**
  74  * Rewrite any clip/cull distances to refer to the new combined array.
  75  */
  76 static void
  77 rewrite_var_references(nir_instr *instr,
  78                        nir_variable *combined,
  79                        unsigned cull_offset)
  80 {
  81    if (instr->type != nir_instr_type_intrinsic)
  82       return;
  83
  84    nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
  85
  86    /* copy_var needs to be lowered to load/store before calling this pass */
  87    assert(intrin->intrinsic != nir_intrinsic_copy_var);
  88
  89    if (intrin->intrinsic != nir_intrinsic_load_var &&
  90        intrin->intrinsic != nir_intrinsic_store_var)
  91       return;
  92
  93    nir_deref_var *var_ref = intrin->variables[0];
  94    if (var_ref->var->data.mode != combined->data.mode)
  95       return;
  96
  97    if (var_ref->var->data.location != VARYING_SLOT_CLIP_DIST0 &&
  98        var_ref->var->data.location != VARYING_SLOT_CULL_DIST0)
  99       return;
 100
 101    /* Update types along the deref chain */
 102    const struct glsl_type *type = combined->type;
 103    nir_deref *deref = &var_ref->deref;
 104    while (deref) {
 105       deref->type = type;
 106       deref = deref->child;
 107       type = glsl_get_array_element(type);
 108    }
 109
 110    /* For cull distances, add an offset to the array index */
 111    if (var_ref->var->data.location == VARYING_SLOT_CULL_DIST0) {
 112       nir_deref *tail = nir_deref_tail(&intrin->variables[0]->deref);
 113       nir_deref_array *array_ref = nir_deref_as_array(tail);
 114
 115       array_ref->base_offset += cull_offset;
 116    }
 117
 118    /* Point the deref at the combined array */
 119    var_ref->var = combined;
 120
 121    /* There's no need to update writemasks; it's a scalar array. */
 122 }
 123
 124 static void
 125 rewrite_clip_cull_deref(nir_builder *b,
 126                         nir_deref_instr *deref,
 127                         const struct glsl_type *type,
 128                         unsigned tail_offset)
 129 {
 130    deref->type = type;
 131
 132    if (glsl_type_is_array(type)) {
 133       const struct glsl_type *child_type = glsl_get_array_element(type);
 134       nir_foreach_use(src, &deref->dest.ssa) {
 135          rewrite_clip_cull_deref(b, nir_instr_as_deref(src->parent_instr),
 136                                  child_type, tail_offset);
 137       }
 138    } else {
 139       assert(glsl_type_is_scalar(type));
 140
 141       /* This is the end of the line.  Add the tail offset if needed */
 142       if (tail_offset > 0) {
 143          b->cursor = nir_before_instr(&deref->instr);
 144          assert(deref->deref_type == nir_deref_type_array);
 145          nir_ssa_def *index = nir_iadd(b, deref->arr.index.ssa,
 146                                           nir_imm_int(b, tail_offset));
 147          nir_instr_rewrite_src(&deref->instr, &deref->arr.index,
 148                                nir_src_for_ssa(index));
 149       }
 150    }
 151 }
 152
 153 static void
 154 rewrite_references(nir_builder *b,
 155                    nir_instr *instr,
 156                    nir_variable *combined,
 157                    unsigned cull_offset)
 158 {
 159    if (instr->type != nir_instr_type_deref)
 160       return;
 161
 162    nir_deref_instr *deref = nir_instr_as_deref(instr);
 163    if (deref->deref_type != nir_deref_type_var)
 164       return;
 165
 166    if (deref->var->data.mode != combined->data.mode)
 167       return;
 168
 169    const unsigned location = deref->var->data.location;
 170    if (location != VARYING_SLOT_CLIP_DIST0 &&
 171        location != VARYING_SLOT_CULL_DIST0)
 172       return;
 173
 174    deref->var = combined;
 175    if (location == VARYING_SLOT_CULL_DIST0)
 176       rewrite_clip_cull_deref(b, deref, combined->type, cull_offset);
 177    else
 178       rewrite_clip_cull_deref(b, deref, combined->type, 0);
 179 }
 180
 181 static bool
 182 combine_clip_cull(nir_shader *nir,
 183                   struct exec_list *vars,
 184                   bool store_info)
 185 {
 186    nir_variable *cull = NULL;
 187    nir_variable *clip = NULL;
 188    bool progress = false;
 189
 190    nir_foreach_variable(var, vars) {
 191       if (var->data.location == VARYING_SLOT_CLIP_DIST0)
 192          clip = var;
 193
 194       if (var->data.location == VARYING_SLOT_CULL_DIST0)
 195          cull = var;
 196    }
 197
 198    const unsigned clip_array_size = get_unwrapped_array_length(nir, clip);
 199    const unsigned cull_array_size = get_unwrapped_array_length(nir, cull);
 200
 201    if (store_info) {
 202       nir->info.clip_distance_array_size = clip_array_size;
 203       nir->info.cull_distance_array_size = cull_array_size;
 204    }
 205
 206    if (clip)
 207       clip->data.compact = true;
 208
 209    if (cull)
 210       cull->data.compact = true;
 211
 212    if (cull_array_size > 0) {
 213       if (clip_array_size == 0) {
 214          /* No clip distances, just change the cull distance location */
 215          cull->data.location = VARYING_SLOT_CLIP_DIST0;
 216       } else {
 217          /* Turn the ClipDistance array into a combined one */
 218          update_type(clip, nir->info.stage, clip_array_size + cull_array_size);
 219
 220          /* Rewrite CullDistance to reference the combined array */
 221          nir_foreach_function(function, nir) {
 222             if (function->impl) {
 223                nir_builder b;
 224                nir_builder_init(&b, function->impl);
 225
 226                nir_foreach_block(block, function->impl) {
 227                   nir_foreach_instr(instr, block) {
 228                      rewrite_var_references(instr, clip, clip_array_size);
 229                      rewrite_references(&b, instr, clip, clip_array_size);
 230                   }
 231                }
 232             }
 233          }
 234
 235          /* Delete the old CullDistance variable */
 236          exec_node_remove(&cull->node);
 237          ralloc_free(cull);
 238       }
 239
 240       nir_foreach_function(function, nir) {
 241          if (function->impl) {
 242             nir_metadata_preserve(function->impl,
 243                                   nir_metadata_block_index |
 244                                   nir_metadata_dominance);
 245          }
 246       }
 247       progress = true;
 248    }
 249
 250    return progress;
 251 }
 252
 253 bool
 254 nir_lower_clip_cull_distance_arrays(nir_shader *nir)
 255 {
 256    bool progress = false;
 257
 258    if (nir->info.stage <= MESA_SHADER_GEOMETRY)
 259       progress |= combine_clip_cull(nir, &nir->outputs, true);
 260
 261    if (nir->info.stage > MESA_SHADER_VERTEX)
 262       progress |= combine_clip_cull(nir, &nir->inputs, false);
 263
 264    return progress;
 265 }