src/compiler/nir/nir_lower_indirect_derefs.c

   1 /*
   2  * Copyright © 2016 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 #include "nir_deref.h"
  27
  28 static void
  29 emit_load_store_deref(nir_builder *b, nir_intrinsic_instr *orig_instr,
  30                       nir_deref_instr *parent,
  31                       nir_deref_instr **deref_arr,
  32                       nir_ssa_def **dest, nir_ssa_def *src);
  33
  34 static void
  35 emit_indirect_load_store_deref(nir_builder *b, nir_intrinsic_instr *orig_instr,
  36                                nir_deref_instr *parent,
  37                                nir_deref_instr **deref_arr,
  38                                int start, int end,
  39                                nir_ssa_def **dest, nir_ssa_def *src)
  40 {
  41    assert(start < end);
  42    if (start == end - 1) {
  43       nir_ssa_def *index = nir_imm_int(b, start);
  44       emit_load_store_deref(b, orig_instr,
  45                             nir_build_deref_array(b, parent, index),
  46                             deref_arr + 1, dest, src);
  47    } else {
  48       int mid = start + (end - start) / 2;
  49
  50       nir_ssa_def *then_dest, *else_dest;
  51
  52       nir_deref_instr *deref = *deref_arr;
  53       assert(deref->deref_type == nir_deref_type_array);
  54
  55       nir_push_if(b, nir_ilt(b, deref->arr.index.ssa, nir_imm_int(b, mid)));
  56       emit_indirect_load_store_deref(b, orig_instr, parent, deref_arr,
  57                                      start, mid, &then_dest, src);
  58       nir_push_else(b, NULL);
  59       emit_indirect_load_store_deref(b, orig_instr, parent, deref_arr,
  60                                      mid, end, &else_dest, src);
  61       nir_pop_if(b, NULL);
  62
  63       if (src == NULL)
  64          *dest = nir_if_phi(b, then_dest, else_dest);
  65    }
  66 }
  67
  68 static void
  69 emit_load_store_deref(nir_builder *b, nir_intrinsic_instr *orig_instr,
  70                       nir_deref_instr *parent,
  71                       nir_deref_instr **deref_arr,
  72                       nir_ssa_def **dest, nir_ssa_def *src)
  73 {
  74    for (; *deref_arr; deref_arr++) {
  75       nir_deref_instr *deref = *deref_arr;
  76       if (deref->deref_type == nir_deref_type_array &&
  77           !nir_src_is_const(deref->arr.index)) {
  78          int length = glsl_get_length(parent->type);
  79
  80          emit_indirect_load_store_deref(b, orig_instr, parent, deref_arr,
  81                                         0, length, dest, src);
  82          return;
  83       }
  84
  85       parent = nir_build_deref_follower(b, parent, deref);
  86    }
  87
  88    /* We reached the end of the deref chain.  Emit the instruction */
  89    assert(*deref_arr == NULL);
  90
  91    if (src == NULL) {
  92       /* This is a load instruction */
  93       nir_intrinsic_instr *load =
  94          nir_intrinsic_instr_create(b->shader, orig_instr->intrinsic);
  95       load->num_components = orig_instr->num_components;
  96
  97       load->src[0] = nir_src_for_ssa(&parent->dest.ssa);
  98
  99       /* Copy over any other sources.  This is needed for interp_deref_at */
 100       for (unsigned i = 1;
 101            i < nir_intrinsic_infos[orig_instr->intrinsic].num_srcs; i++)
 102          nir_src_copy(&load->src[i], &orig_instr->src[i], load);
 103
 104       nir_ssa_dest_init(&load->instr, &load->dest,
 105                         orig_instr->dest.ssa.num_components,
 106                         orig_instr->dest.ssa.bit_size, NULL);
 107       nir_builder_instr_insert(b, &load->instr);
 108       *dest = &load->dest.ssa;
 109    } else {
 110       assert(orig_instr->intrinsic == nir_intrinsic_store_deref);
 111       nir_store_deref(b, parent, src, nir_intrinsic_write_mask(orig_instr));
 112    }
 113 }
 114
 115 static bool
 116 lower_indirect_derefs_block(nir_block *block, nir_builder *b,
 117                             nir_variable_mode modes)
 118 {
 119    bool progress = false;
 120
 121    nir_foreach_instr_safe(instr, block) {
 122       if (instr->type != nir_instr_type_intrinsic)
 123          continue;
 124
 125       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 126       if (intrin->intrinsic != nir_intrinsic_load_deref &&
 127           intrin->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
 128           intrin->intrinsic != nir_intrinsic_interp_deref_at_sample &&
 129           intrin->intrinsic != nir_intrinsic_interp_deref_at_offset &&
 130           intrin->intrinsic != nir_intrinsic_store_deref)
 131          continue;
 132
 133       nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
 134
 135       /* Walk the deref chain back to the base and look for indirects */
 136       bool has_indirect = false;
 137       nir_deref_instr *base = deref;
 138       while (base->deref_type != nir_deref_type_var) {
 139          if (base->deref_type == nir_deref_type_array &&
 140              !nir_src_is_const(base->arr.index))
 141             has_indirect = true;
 142
 143          base = nir_deref_instr_parent(base);
 144       }
 145
 146       if (!has_indirect)
 147          continue;
 148
 149       /* Only lower variables whose mode is in the mask, or compact
 150        * array variables.  (We can't handle indirects on tightly packed
 151        * scalar arrays, so we need to lower them regardless.)
 152        */
 153       if (!(modes & base->var->data.mode) && !base->var->data.compact)
 154          continue;
 155
 156       b->cursor = nir_instr_remove(&intrin->instr);
 157
 158       nir_deref_path path;
 159       nir_deref_path_init(&path, deref, NULL);
 160       assert(path.path[0] == base);
 161
 162       if (intrin->intrinsic == nir_intrinsic_store_deref) {
 163          assert(intrin->src[1].is_ssa);
 164          emit_load_store_deref(b, intrin, base, &path.path[1],
 165                                NULL, intrin->src[1].ssa);
 166       } else {
 167          nir_ssa_def *result;
 168          emit_load_store_deref(b, intrin, base, &path.path[1],
 169                                &result, NULL);
 170          nir_ssa_def_rewrite_uses(&intrin->dest.ssa, nir_src_for_ssa(result));
 171       }
 172
 173       nir_deref_path_finish(&path);
 174
 175       progress = true;
 176    }
 177
 178    return progress;
 179 }
 180
 181 static bool
 182 lower_indirects_impl(nir_function_impl *impl, nir_variable_mode modes)
 183 {
 184    nir_builder builder;
 185    nir_builder_init(&builder, impl);
 186    bool progress = false;
 187
 188    nir_foreach_block_safe(block, impl) {
 189       progress |= lower_indirect_derefs_block(block, &builder, modes);
 190    }
 191
 192    if (progress)
 193       nir_metadata_preserve(impl, nir_metadata_none);
 194
 195    return progress;
 196 }
 197
 198 /** Lowers indirect variable loads/stores to direct loads/stores.
 199  *
 200  * The pass works by replacing any indirect load or store with an if-ladder
 201  * that does a binary search on the array index.
 202  */
 203 bool
 204 nir_lower_indirect_derefs(nir_shader *shader, nir_variable_mode modes)
 205 {
 206    bool progress = false;
 207
 208    if (modes == 0)
 209       return false;
 210
 211    nir_foreach_function(function, shader) {
 212       if (function->impl)
 213          progress = lower_indirects_impl(function->impl, modes) || progress;
 214    }
 215
 216    return progress;
 217 }