src/compiler/nir/nir_lower_var_copies.c

   1 /*
   2  * Copyright © 2014 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  * Authors:
  24  *    Jason Ekstrand (jason@jlekstrand.net)
  25  *
  26  */
  27
  28 #include "nir.h"
  29 #include "compiler/nir_types.h"
  30
  31 /*
  32  * Lowers all copy intrinsics to sequences of load/store intrinsics.
  33  */
  34
  35 /* Walks down the deref chain and returns the next deref in the chain whose
  36  * child is a wildcard.  In other words, given the chain  a[1].foo[*].bar,
  37  * this function will return the deref to foo.  Calling it a second time
  38  * with the [*].bar, it will return NULL.
  39  */
  40 static nir_deref *
  41 deref_next_wildcard_parent(nir_deref *deref)
  42 {
  43    for (nir_deref *tail = deref; tail->child; tail = tail->child) {
  44       if (tail->child->deref_type != nir_deref_type_array)
  45          continue;
  46
  47       nir_deref_array *arr = nir_deref_as_array(tail->child);
  48
  49       if (arr->deref_array_type == nir_deref_array_type_wildcard)
  50          return tail;
  51    }
  52
  53    return NULL;
  54 }
  55
  56 /* This function recursively walks the given deref chain and replaces the
  57  * given copy instruction with an equivalent sequence load/store
  58  * operations.
  59  *
  60  * @copy_instr    The copy instruction to replace; new instructions will be
  61  *                inserted before this one
  62  *
  63  * @dest_head     The head of the destination variable deref chain
  64  *
  65  * @src_head      The head of the source variable deref chain
  66  *
  67  * @dest_tail     The current tail of the destination variable deref chain;
  68  *                this is used for recursion and external callers of this
  69  *                function should call it with tail == head
  70  *
  71  * @src_tail      The current tail of the source variable deref chain;
  72  *                this is used for recursion and external callers of this
  73  *                function should call it with tail == head
  74  *
  75  * @state         The current variable lowering state
  76  */
  77 static void
  78 emit_copy_load_store(nir_intrinsic_instr *copy_instr,
  79                      nir_deref_var *dest_head, nir_deref_var *src_head,
  80                      nir_deref *dest_tail, nir_deref *src_tail,
  81                      nir_shader *shader)
  82 {
  83    /* Find the next pair of wildcards */
  84    nir_deref *src_arr_parent = deref_next_wildcard_parent(src_tail);
  85    nir_deref *dest_arr_parent = deref_next_wildcard_parent(dest_tail);
  86
  87    if (src_arr_parent || dest_arr_parent) {
  88       /* Wildcards had better come in matched pairs */
  89       assert(src_arr_parent && dest_arr_parent);
  90
  91       nir_deref_array *src_arr = nir_deref_as_array(src_arr_parent->child);
  92       nir_deref_array *dest_arr = nir_deref_as_array(dest_arr_parent->child);
  93
  94       unsigned length = glsl_get_length(src_arr_parent->type);
  95       /* The wildcards should represent the same number of elements */
  96       assert(length == glsl_get_length(dest_arr_parent->type));
  97       assert(length > 0);
  98
  99       /* Walk over all of the elements that this wildcard refers to and
 100        * call emit_copy_load_store on each one of them */
 101       src_arr->deref_array_type = nir_deref_array_type_direct;
 102       dest_arr->deref_array_type = nir_deref_array_type_direct;
 103       for (unsigned i = 0; i < length; i++) {
 104          src_arr->base_offset = i;
 105          dest_arr->base_offset = i;
 106          emit_copy_load_store(copy_instr, dest_head, src_head,
 107                               &dest_arr->deref, &src_arr->deref, shader);
 108       }
 109       src_arr->deref_array_type = nir_deref_array_type_wildcard;
 110       dest_arr->deref_array_type = nir_deref_array_type_wildcard;
 111    } else {
 112       /* In this case, we have no wildcards anymore, so all we have to do
 113        * is just emit the load and store operations. */
 114       src_tail = nir_deref_tail(src_tail);
 115       dest_tail = nir_deref_tail(dest_tail);
 116
 117       assert(src_tail->type == dest_tail->type);
 118
 119       unsigned num_components = glsl_get_vector_elements(src_tail->type);
 120       unsigned bit_size = glsl_get_bit_size(src_tail->type);
 121
 122       nir_intrinsic_instr *load =
 123          nir_intrinsic_instr_create(shader, nir_intrinsic_load_var);
 124       load->num_components = num_components;
 125       load->variables[0] = nir_deref_var_clone(src_head, load);
 126       nir_ssa_dest_init(&load->instr, &load->dest, num_components, bit_size,
 127                         NULL);
 128
 129       nir_instr_insert_before(&copy_instr->instr, &load->instr);
 130
 131       nir_intrinsic_instr *store =
 132          nir_intrinsic_instr_create(shader, nir_intrinsic_store_var);
 133       store->num_components = num_components;
 134       nir_intrinsic_set_write_mask(store, (1 << num_components) - 1);
 135       store->variables[0] = nir_deref_var_clone(dest_head, store);
 136
 137       store->src[0].is_ssa = true;
 138       store->src[0].ssa = &load->dest.ssa;
 139
 140       nir_instr_insert_before(&copy_instr->instr, &store->instr);
 141    }
 142 }
 143
 144 /* Lowers a copy instruction to a sequence of load/store instructions
 145  *
 146  * The new instructions are placed before the copy instruction in the IR.
 147  */
 148 void
 149 nir_lower_var_copy_instr(nir_intrinsic_instr *copy, nir_shader *shader)
 150 {
 151    assert(copy->intrinsic == nir_intrinsic_copy_var);
 152    emit_copy_load_store(copy, copy->variables[0], copy->variables[1],
 153                         &copy->variables[0]->deref,
 154                         &copy->variables[1]->deref, shader);
 155 }
 156
 157 static bool
 158 lower_var_copies_impl(nir_function_impl *impl)
 159 {
 160    nir_shader *shader = impl->function->shader;
 161    bool progress = false;
 162
 163    nir_foreach_block(block, impl) {
 164       nir_foreach_instr_safe(instr, block) {
 165          if (instr->type != nir_instr_type_intrinsic)
 166             continue;
 167
 168          nir_intrinsic_instr *copy = nir_instr_as_intrinsic(instr);
 169          if (copy->intrinsic != nir_intrinsic_copy_var)
 170             continue;
 171
 172          nir_lower_var_copy_instr(copy, shader);
 173
 174          nir_instr_remove(&copy->instr);
 175          progress = true;
 176          ralloc_free(copy);
 177       }
 178    }
 179
 180    if (progress)
 181       nir_metadata_preserve(impl, nir_metadata_block_index |
 182                                   nir_metadata_dominance);
 183
 184    return progress;
 185 }
 186
 187 /* Lowers every copy_var instruction in the program to a sequence of
 188  * load/store instructions.
 189  */
 190 bool
 191 nir_lower_var_copies(nir_shader *shader)
 192 {
 193    bool progress = false;
 194
 195    nir_foreach_function(function, shader) {
 196       if (function->impl)
 197          progress |= lower_var_copies_impl(function->impl);
 198    }
 199
 200    return progress;
 201 }