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    assert(arr_parent->child &&
  39           arr_parent->child->deref_type == nir_deref_type_array);
  40    nir_deref_array *arr = nir_deref_as_array(arr_parent->child);
  41    assert(arr->deref_array_type == nir_deref_array_type_indirect);
  42    assert(arr->indirect.is_ssa);
  43
  44    assert(start < end);
  45    if (start == end - 1) {
  46       /* Base case.  Just emit the load/store op */
  47       nir_deref_array direct = *arr;
  48       direct.deref_array_type = nir_deref_array_type_direct;
  49       direct.base_offset += start;
  50       direct.indirect = NIR_SRC_INIT;
  51
  52       arr_parent->child = &direct.deref;
  53       emit_load_store(b, orig_instr, deref, &arr->deref, dest, src);
  54       arr_parent->child = &arr->deref;
  55    } else {
  56       int mid = start + (end - start) / 2;
  57
  58       nir_ssa_def *then_dest, *else_dest;
  59
  60       nir_if *if_stmt = nir_if_create(b->shader);
  61       if_stmt->condition = nir_src_for_ssa(nir_ilt(b, arr->indirect.ssa,
  62                                                       nir_imm_int(b, mid)));
  63       nir_cf_node_insert(b->cursor, &if_stmt->cf_node);
  64
  65       b->cursor = nir_after_cf_list(&if_stmt->then_list);
  66       emit_indirect_load_store(b, orig_instr, deref, arr_parent,
  67                                start, mid, &then_dest, src);
  68
  69       b->cursor = nir_after_cf_list(&if_stmt->else_list);
  70       emit_indirect_load_store(b, orig_instr, deref, arr_parent,
  71                                mid, end, &else_dest, src);
  72
  73       b->cursor = nir_after_cf_node(&if_stmt->cf_node);
  74
  75       if (src == NULL) {
  76          /* We're a load.  We need to insert a phi node */
  77          nir_phi_instr *phi = nir_phi_instr_create(b->shader);
  78          nir_ssa_dest_init(&phi->instr, &phi->dest,
  79                            then_dest->num_components, NULL);
  80
  81          nir_phi_src *src0 = ralloc(phi, nir_phi_src);
  82          src0->pred = nir_cf_node_as_block(nir_if_last_then_node(if_stmt));
  83          src0->src = nir_src_for_ssa(then_dest);
  84          exec_list_push_tail(&phi->srcs, &src0->node);
  85
  86          nir_phi_src *src1 = ralloc(phi, nir_phi_src);
  87          src1->pred = nir_cf_node_as_block(nir_if_last_else_node(if_stmt));
  88          src1->src = nir_src_for_ssa(else_dest);
  89          exec_list_push_tail(&phi->srcs, &src1->node);
  90
  91          nir_builder_instr_insert(b, &phi->instr);
  92          *dest = &phi->dest.ssa;
  93       }
  94    }
  95 }
  96
  97 static void
  98 emit_load_store(nir_builder *b, nir_intrinsic_instr *orig_instr,
  99                 nir_deref_var *deref, nir_deref *tail,
 100                 nir_ssa_def **dest, nir_ssa_def *src)
 101 {
 102    for (; tail->child; tail = tail->child) {
 103       if (tail->child->deref_type != nir_deref_type_array)
 104          continue;
 105
 106       nir_deref_array *arr = nir_deref_as_array(tail->child);
 107       if (arr->deref_array_type != nir_deref_array_type_indirect)
 108          continue;
 109
 110       int length = glsl_get_length(tail->type);
 111
 112       emit_indirect_load_store(b, orig_instr, deref, tail, -arr->base_offset,
 113                                length - arr->base_offset, dest, src);
 114       return;
 115    }
 116
 117    assert(tail && tail->child == NULL);
 118
 119    /* We reached the end of the deref chain.  Emit the instruction */
 120
 121    if (src == NULL) {
 122       /* This is a load instruction */
 123       nir_intrinsic_instr *load =
 124          nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_var);
 125       load->num_components = orig_instr->num_components;
 126       load->variables[0] =
 127          nir_deref_as_var(nir_copy_deref(load, &deref->deref));
 128       nir_ssa_dest_init(&load->instr, &load->dest,
 129                         load->num_components, 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       store->const_index[0] = orig_instr->const_index[0]; /* writemask */
 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 struct lower_indirect_state {
 161    nir_builder builder;
 162    uint32_t mode_mask;
 163    bool progress;
 164 };
 165
 166 static bool
 167 lower_indirect_block(nir_block *block, void *void_state)
 168 {
 169    struct lower_indirect_state *state = void_state;
 170
 171    nir_foreach_instr_safe(block, instr) {
 172       if (instr->type != nir_instr_type_intrinsic)
 173          continue;
 174
 175       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 176       if (intrin->intrinsic != nir_intrinsic_load_var &&
 177           intrin->intrinsic != nir_intrinsic_store_var)
 178          continue;
 179
 180       if (!deref_has_indirect(intrin->variables[0]))
 181          continue;
 182
 183       /* Only lower variables whose mode is in the mask */
 184       if (!(state->mode_mask & (1 << intrin->variables[0]->var->data.mode)))
 185          continue;
 186
 187       state->builder.cursor = nir_before_instr(&intrin->instr);
 188
 189       if (intrin->intrinsic == nir_intrinsic_load_var) {
 190          nir_ssa_def *result;
 191          emit_load_store(&state->builder, intrin, intrin->variables[0],
 192                          &intrin->variables[0]->deref, &result, NULL);
 193          nir_ssa_def_rewrite_uses(&intrin->dest.ssa, nir_src_for_ssa(result));
 194       } else {
 195          assert(intrin->src[0].is_ssa);
 196          emit_load_store(&state->builder, intrin, intrin->variables[0],
 197                          &intrin->variables[0]->deref, NULL, intrin->src[0].ssa);
 198       }
 199       nir_instr_remove(&intrin->instr);
 200       state->progress = true;
 201    }
 202
 203    return true;
 204 }
 205
 206 static bool
 207 lower_indirects_impl(nir_function_impl *impl, uint32_t mode_mask)
 208 {
 209    struct lower_indirect_state state;
 210
 211    state.progress = false;
 212    state.mode_mask = mode_mask;
 213    nir_builder_init(&state.builder, impl);
 214
 215    nir_foreach_block(impl, lower_indirect_block, &state);
 216
 217    if (state.progress)
 218       nir_metadata_preserve(impl, nir_metadata_none);
 219
 220    return state.progress;
 221 }
 222
 223 /** Lowers indirect variable loads/stores to direct loads/stores.
 224  *
 225  * The pass works by replacing any indirect load or store with an if-ladder
 226  * that does a binary search on the array index.
 227  */
 228 bool
 229 nir_lower_indirect_derefs(nir_shader *shader, uint32_t mode_mask)
 230 {
 231    bool progress = false;
 232
 233    nir_foreach_function(shader, function) {
 234       if (function->impl)
 235          progress = lower_indirects_impl(function->impl, mode_mask) || progress;
 236    }
 237
 238    return progress;
 239 }