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
  27 static void
  28 emit_load_store(nir_builder *b, nir_intrinsic_instr *orig_instr,
  29                 nir_deref_var *deref, nir_deref *tail,
  30                 nir_ssa_def **dest, nir_ssa_def *src);
  31
  32 static void
  33 emit_indirect_load_store(nir_builder *b, nir_intrinsic_instr *orig_instr,
  34                          nir_deref_var *deref, nir_deref *arr_parent,
  35                          int start, int end,
  36                          nir_ssa_def **dest, nir_ssa_def *src)
  37 {
  38    nir_deref_array *arr = nir_deref_as_array(arr_parent->child);
  39    assert(arr->deref_array_type == nir_deref_array_type_indirect);
  40    assert(arr->indirect.is_ssa);
  41
  42    assert(start < end);
  43    if (start == end - 1) {
  44       /* Base case.  Just emit the load/store op */
  45       nir_deref_array direct = *arr;
  46       direct.deref_array_type = nir_deref_array_type_direct;
  47       direct.base_offset += start;
  48       direct.indirect = NIR_SRC_INIT;
  49
  50       arr_parent->child = &direct.deref;
  51       emit_load_store(b, orig_instr, deref, &direct.deref, dest, src);
  52       arr_parent->child = &arr->deref;
  53    } else {
  54       int mid = start + (end - start) / 2;
  55
  56       nir_ssa_def *then_dest, *else_dest;
  57
  58       nir_if *if_stmt = nir_if_create(b->shader);
  59       if_stmt->condition = nir_src_for_ssa(nir_ilt(b, arr->indirect.ssa,
  60                                                       nir_imm_int(b, mid)));
  61       nir_cf_node_insert(b->cursor, &if_stmt->cf_node);
  62
  63       b->cursor = nir_after_cf_list(&if_stmt->then_list);
  64       emit_indirect_load_store(b, orig_instr, deref, arr_parent,
  65                                start, mid, &then_dest, src);
  66
  67       b->cursor = nir_after_cf_list(&if_stmt->else_list);
  68       emit_indirect_load_store(b, orig_instr, deref, arr_parent,
  69                                mid, end, &else_dest, src);
  70
  71       b->cursor = nir_after_cf_node(&if_stmt->cf_node);
  72
  73       if (src == NULL) {
  74          /* We're a load.  We need to insert a phi node */
  75          nir_phi_instr *phi = nir_phi_instr_create(b->shader);
  76          unsigned bit_size = then_dest->bit_size;
  77          nir_ssa_dest_init(&phi->instr, &phi->dest,
  78                            then_dest->num_components, bit_size, NULL);
  79
  80          nir_phi_src *src0 = ralloc(phi, nir_phi_src);
  81          src0->pred = nir_if_last_then_block(if_stmt);
  82          src0->src = nir_src_for_ssa(then_dest);
  83          exec_list_push_tail(&phi->srcs, &src0->node);
  84
  85          nir_phi_src *src1 = ralloc(phi, nir_phi_src);
  86          src1->pred = nir_if_last_else_block(if_stmt);
  87          src1->src = nir_src_for_ssa(else_dest);
  88          exec_list_push_tail(&phi->srcs, &src1->node);
  89
  90          nir_builder_instr_insert(b, &phi->instr);
  91          *dest = &phi->dest.ssa;
  92       }
  93    }
  94 }
  95
  96 static void
  97 emit_load_store(nir_builder *b, nir_intrinsic_instr *orig_instr,
  98                 nir_deref_var *deref, nir_deref *tail,
  99                 nir_ssa_def **dest, nir_ssa_def *src)
 100 {
 101    for (; tail->child; tail = tail->child) {
 102       if (tail->child->deref_type != nir_deref_type_array)
 103          continue;
 104
 105       nir_deref_array *arr = nir_deref_as_array(tail->child);
 106       if (arr->deref_array_type != nir_deref_array_type_indirect)
 107          continue;
 108
 109       int length = glsl_get_length(tail->type);
 110
 111       emit_indirect_load_store(b, orig_instr, deref, tail, -arr->base_offset,
 112                                length - arr->base_offset, dest, src);
 113       return;
 114    }
 115
 116    assert(tail && tail->child == NULL);
 117
 118    /* We reached the end of the deref chain.  Emit the instruction */
 119
 120    if (src == NULL) {
 121       /* This is a load instruction */
 122       nir_intrinsic_instr *load =
 123          nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_var);
 124       load->num_components = orig_instr->num_components;
 125       load->variables[0] =
 126          nir_deref_as_var(nir_copy_deref(load, &deref->deref));
 127       unsigned bit_size = orig_instr->dest.ssa.bit_size;
 128       nir_ssa_dest_init(&load->instr, &load->dest,
 129                         load->num_components, bit_size, NULL);
 130       nir_builder_instr_insert(b, &load->instr);
 131       *dest = &load->dest.ssa;
 132    } else {
 133       /* This is a store instruction */
 134       nir_intrinsic_instr *store =
 135          nir_intrinsic_instr_create(b->shader, nir_intrinsic_store_var);
 136       store->num_components = orig_instr->num_components;
 137       nir_intrinsic_set_write_mask(store, nir_intrinsic_write_mask(orig_instr));
 138       store->variables[0] =
 139          nir_deref_as_var(nir_copy_deref(store, &deref->deref));
 140       store->src[0] = nir_src_for_ssa(src);
 141       nir_builder_instr_insert(b, &store->instr);
 142    }
 143 }
 144
 145 static bool
 146 deref_has_indirect(nir_deref_var *deref)
 147 {
 148    for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
 149       if (tail->deref_type != nir_deref_type_array)
 150          continue;
 151
 152       nir_deref_array *arr = nir_deref_as_array(tail);
 153       if (arr->deref_array_type == nir_deref_array_type_indirect)
 154          return true;
 155    }
 156
 157    return false;
 158 }
 159
 160 static bool
 161 lower_indirect_block(nir_block *block, nir_builder *b,
 162                      nir_variable_mode modes)
 163 {
 164    bool progress = false;
 165
 166    nir_foreach_instr_safe(instr, block) {
 167       if (instr->type != nir_instr_type_intrinsic)
 168          continue;
 169
 170       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 171       if (intrin->intrinsic != nir_intrinsic_load_var &&
 172           intrin->intrinsic != nir_intrinsic_store_var)
 173          continue;
 174
 175       if (!deref_has_indirect(intrin->variables[0]))
 176          continue;
 177
 178       /* Only lower variables whose mode is in the mask */
 179       if (!(modes & intrin->variables[0]->var->data.mode))
 180          continue;
 181
 182       b->cursor = nir_before_instr(&intrin->instr);
 183
 184       if (intrin->intrinsic == nir_intrinsic_load_var) {
 185          nir_ssa_def *result;
 186          emit_load_store(b, intrin, intrin->variables[0],
 187                          &intrin->variables[0]->deref, &result, NULL);
 188          nir_ssa_def_rewrite_uses(&intrin->dest.ssa, nir_src_for_ssa(result));
 189       } else {
 190          assert(intrin->src[0].is_ssa);
 191          emit_load_store(b, intrin, intrin->variables[0],
 192                          &intrin->variables[0]->deref, NULL, intrin->src[0].ssa);
 193       }
 194       nir_instr_remove(&intrin->instr);
 195       progress = true;
 196    }
 197
 198    return progress;
 199 }
 200
 201 static bool
 202 lower_indirects_impl(nir_function_impl *impl, nir_variable_mode modes)
 203 {
 204    nir_builder builder;
 205    nir_builder_init(&builder, impl);
 206    bool progress = false;
 207
 208    nir_foreach_block_safe(block, impl) {
 209       progress |= lower_indirect_block(block, &builder, modes);
 210    }
 211
 212    if (progress)
 213       nir_metadata_preserve(impl, nir_metadata_none);
 214
 215    return progress;
 216 }
 217
 218 /** Lowers indirect variable loads/stores to direct loads/stores.
 219  *
 220  * The pass works by replacing any indirect load or store with an if-ladder
 221  * that does a binary search on the array index.
 222  */
 223 bool
 224 nir_lower_indirect_derefs(nir_shader *shader, nir_variable_mode modes)
 225 {
 226    bool progress = false;
 227
 228    nir_foreach_function(function, shader) {
 229       if (function->impl)
 230          progress = lower_indirects_impl(function->impl, modes) || progress;
 231    }
 232
 233    return progress;
 234 }