src/compiler/nir/nir_opt_dead_write_vars.c

   1 /*
   2  * Copyright © 2018 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 #include "util/u_dynarray.h"
  29
  30 /**
  31  * Elimination of dead writes based on derefs.
  32  *
  33  * Dead writes are stores and copies that write to a deref, which then gets
  34  * another write before it was used (read or sourced for a copy).  Those
  35  * writes can be removed since they don't affect anything.
  36  *
  37  * For derefs that refer to a memory area that can be read after the program,
  38  * the last write is considered used.  The presence of certain instructions
  39  * may also cause writes to be considered used, e.g. memory barrier (in this case
  40  * the value must be written as other thread might use it).
  41  *
  42  * The write mask for store instructions is considered, so it is possible that
  43  * a store is removed because of the combination of other stores overwritten
  44  * its value.
  45  */
  46
  47 /* Entry for unused_writes arrays. */
  48 struct write_entry {
  49    /* If NULL indicates the entry is free to be reused. */
  50    nir_intrinsic_instr *intrin;
  51    nir_component_mask_t mask;
  52    nir_deref_instr *dst;
  53 };
  54
  55 static void
  56 clear_unused_for_modes(struct util_dynarray *unused_writes, nir_variable_mode modes)
  57 {
  58    util_dynarray_foreach_reverse(unused_writes, struct write_entry, entry) {
  59       nir_variable *var = nir_deref_instr_get_variable(entry->dst);
  60       if (var->data.mode & modes)
  61          *entry = util_dynarray_pop(unused_writes, struct write_entry);
  62    }
  63 }
  64
  65 static void
  66 clear_unused_for_read(struct util_dynarray *unused_writes, nir_deref_instr *src)
  67 {
  68    util_dynarray_foreach_reverse(unused_writes, struct write_entry, entry) {
  69       if (nir_compare_derefs(src, entry->dst) & nir_derefs_may_alias_bit)
  70          *entry = util_dynarray_pop(unused_writes, struct write_entry);
  71    }
  72 }
  73
  74 static bool
  75 update_unused_writes(struct util_dynarray *unused_writes,
  76                      nir_intrinsic_instr *intrin,
  77                      nir_deref_instr *dst, nir_component_mask_t mask)
  78 {
  79    bool progress = false;
  80
  81    /* This pass assumes that destination of copies and stores are derefs that
  82     * end in a vector or scalar (it is OK to have wildcards or indirects for
  83     * arrays).
  84     */
  85    assert(glsl_type_is_vector_or_scalar(dst->type));
  86
  87    /* Find writes that are unused and can be removed. */
  88    util_dynarray_foreach_reverse(unused_writes, struct write_entry, entry) {
  89       nir_deref_compare_result comp = nir_compare_derefs(dst, entry->dst);
  90       if (comp & nir_derefs_a_contains_b_bit) {
  91          entry->mask &= ~mask;
  92          if (entry->mask == 0) {
  93             nir_instr_remove(&entry->intrin->instr);
  94             *entry = util_dynarray_pop(unused_writes, struct write_entry);
  95             progress = true;
  96          }
  97       }
  98    }
  99
 100    /* Add the new write to the unused array. */
 101    struct write_entry new_entry = {
 102       .intrin = intrin,
 103       .mask = mask,
 104       .dst = dst,
 105    };
 106
 107    util_dynarray_append(unused_writes, struct write_entry, new_entry);
 108
 109    return progress;
 110 }
 111
 112 static bool
 113 remove_dead_write_vars_local(void *mem_ctx, nir_block *block)
 114 {
 115    bool progress = false;
 116
 117    struct util_dynarray unused_writes;
 118    util_dynarray_init(&unused_writes, mem_ctx);
 119
 120    nir_foreach_instr_safe(instr, block) {
 121       if (instr->type == nir_instr_type_call) {
 122          clear_unused_for_modes(&unused_writes, nir_var_shader_out |
 123                                                 nir_var_global |
 124                                                 nir_var_local |
 125                                                 nir_var_shader_storage |
 126                                                 nir_var_shared);
 127          continue;
 128       }
 129
 130       if (instr->type != nir_instr_type_intrinsic)
 131          continue;
 132
 133       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 134       switch (intrin->intrinsic) {
 135       case nir_intrinsic_barrier:
 136       case nir_intrinsic_memory_barrier: {
 137          clear_unused_for_modes(&unused_writes, nir_var_shader_out |
 138                                                 nir_var_shader_storage |
 139                                                 nir_var_shared);
 140          break;
 141       }
 142
 143       case nir_intrinsic_emit_vertex:
 144       case nir_intrinsic_emit_vertex_with_counter: {
 145          clear_unused_for_modes(&unused_writes, nir_var_shader_out);
 146          break;
 147       }
 148
 149       case nir_intrinsic_load_deref: {
 150          nir_deref_instr *src = nir_src_as_deref(intrin->src[0]);
 151          clear_unused_for_read(&unused_writes, src);
 152          break;
 153       }
 154
 155       case nir_intrinsic_store_deref: {
 156          nir_deref_instr *dst = nir_src_as_deref(intrin->src[0]);
 157          nir_component_mask_t mask = nir_intrinsic_write_mask(intrin);
 158          progress |= update_unused_writes(&unused_writes, intrin, dst, mask);
 159          break;
 160       }
 161
 162       case nir_intrinsic_copy_deref: {
 163          nir_deref_instr *src = nir_src_as_deref(intrin->src[1]);
 164          nir_deref_instr *dst = nir_src_as_deref(intrin->src[0]);
 165
 166          /* Self-copy is removed. */
 167          if (nir_compare_derefs(src, dst) & nir_derefs_equal_bit) {
 168             nir_instr_remove(instr);
 169             progress = true;
 170             break;
 171          }
 172
 173          clear_unused_for_read(&unused_writes, src);
 174          nir_component_mask_t mask = (1 << glsl_get_vector_elements(dst->type)) - 1;
 175          progress |= update_unused_writes(&unused_writes, intrin, dst, mask);
 176          break;
 177       }
 178
 179       default:
 180          break;
 181       }
 182    }
 183
 184    /* All unused writes at the end of the block are kept, since we can't be
 185     * sure they'll be overwritten or not with local analysis only.
 186     */
 187
 188    return progress;
 189 }
 190
 191 static bool
 192 remove_dead_write_vars_impl(void *mem_ctx, nir_function_impl *impl)
 193 {
 194    bool progress = false;
 195
 196    nir_metadata_require(impl, nir_metadata_block_index);
 197
 198    nir_foreach_block(block, impl)
 199       progress |= remove_dead_write_vars_local(mem_ctx, block);
 200
 201    if (progress) {
 202       nir_metadata_preserve(impl, nir_metadata_block_index |
 203                                   nir_metadata_dominance);
 204    }
 205
 206    return progress;
 207 }
 208
 209 bool
 210 nir_opt_dead_write_vars(nir_shader *shader)
 211 {
 212    void *mem_ctx = ralloc_context(NULL);
 213    bool progress = false;
 214
 215    nir_foreach_function(function, shader) {
 216       if (!function->impl)
 217          continue;
 218       progress |= remove_dead_write_vars_impl(mem_ctx, function->impl);
 219    }
 220
 221    ralloc_free(mem_ctx);
 222    return progress;
 223 }