src/compiler/glsl/gl_nir_lower_atomics.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  *    Connor Abbott (cwabbott0@gmail.com)
  25  *
  26  */
  27
  28 #include "compiler/nir/nir.h"
  29 #include "gl_nir.h"
  30 #include "ir_uniform.h"
  31
  32 #include "main/config.h"
  33 #include "main/mtypes.h"
  34 #include <assert.h>
  35
  36 /*
  37  * replace atomic counter intrinsics that use a variable with intrinsics
  38  * that directly store the buffer index and byte offset
  39  */
  40
  41 static bool
  42 lower_instr(nir_intrinsic_instr *instr,
  43             const struct gl_shader_program *shader_program,
  44             nir_shader *shader, bool use_binding_as_idx)
  45 {
  46    nir_intrinsic_op op;
  47    switch (instr->intrinsic) {
  48    case nir_intrinsic_atomic_counter_read_var:
  49       op = nir_intrinsic_atomic_counter_read;
  50       break;
  51
  52    case nir_intrinsic_atomic_counter_inc_var:
  53       op = nir_intrinsic_atomic_counter_inc;
  54       break;
  55
  56    case nir_intrinsic_atomic_counter_dec_var:
  57       op = nir_intrinsic_atomic_counter_dec;
  58       break;
  59
  60    case nir_intrinsic_atomic_counter_add_var:
  61       op = nir_intrinsic_atomic_counter_add;
  62       break;
  63
  64    case nir_intrinsic_atomic_counter_min_var:
  65       op = nir_intrinsic_atomic_counter_min;
  66       break;
  67
  68    case nir_intrinsic_atomic_counter_max_var:
  69       op = nir_intrinsic_atomic_counter_max;
  70       break;
  71
  72    case nir_intrinsic_atomic_counter_and_var:
  73       op = nir_intrinsic_atomic_counter_and;
  74       break;
  75
  76    case nir_intrinsic_atomic_counter_or_var:
  77       op = nir_intrinsic_atomic_counter_or;
  78       break;
  79
  80    case nir_intrinsic_atomic_counter_xor_var:
  81       op = nir_intrinsic_atomic_counter_xor;
  82       break;
  83
  84    case nir_intrinsic_atomic_counter_exchange_var:
  85       op = nir_intrinsic_atomic_counter_exchange;
  86       break;
  87
  88    case nir_intrinsic_atomic_counter_comp_swap_var:
  89       op = nir_intrinsic_atomic_counter_comp_swap;
  90       break;
  91
  92    default:
  93       return false;
  94    }
  95
  96    if (instr->variables[0]->var->data.mode != nir_var_uniform &&
  97        instr->variables[0]->var->data.mode != nir_var_shader_storage &&
  98        instr->variables[0]->var->data.mode != nir_var_shared)
  99       return false; /* atomics passed as function arguments can't be lowered */
 100
 101    void *mem_ctx = ralloc_parent(instr);
 102    unsigned uniform_loc = instr->variables[0]->var->data.location;
 103
 104    unsigned idx = use_binding_as_idx ?
 105       instr->variables[0]->var->data.binding :
 106       shader_program->data->UniformStorage[uniform_loc].opaque[shader->info.stage].index;
 107
 108    nir_intrinsic_instr *new_instr = nir_intrinsic_instr_create(mem_ctx, op);
 109    nir_intrinsic_set_base(new_instr, idx);
 110
 111    nir_load_const_instr *offset_const =
 112       nir_load_const_instr_create(mem_ctx, 1, 32);
 113    offset_const->value.u32[0] = instr->variables[0]->var->data.offset;
 114
 115    nir_instr_insert_before(&instr->instr, &offset_const->instr);
 116
 117    nir_ssa_def *offset_def = &offset_const->def;
 118
 119    nir_deref *tail = &instr->variables[0]->deref;
 120    while (tail->child != NULL) {
 121       nir_deref_array *deref_array = nir_deref_as_array(tail->child);
 122       tail = tail->child;
 123
 124       unsigned child_array_elements = tail->child != NULL ?
 125          glsl_get_aoa_size(tail->type) : 1;
 126
 127       offset_const->value.u32[0] += deref_array->base_offset *
 128          child_array_elements * ATOMIC_COUNTER_SIZE;
 129
 130       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 131          nir_load_const_instr *atomic_counter_size =
 132             nir_load_const_instr_create(mem_ctx, 1, 32);
 133          atomic_counter_size->value.u32[0] = child_array_elements * ATOMIC_COUNTER_SIZE;
 134          nir_instr_insert_before(&instr->instr, &atomic_counter_size->instr);
 135
 136          nir_alu_instr *mul = nir_alu_instr_create(mem_ctx, nir_op_imul);
 137          nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, 32, NULL);
 138          mul->dest.write_mask = 0x1;
 139          nir_src_copy(&mul->src[0].src, &deref_array->indirect, mul);
 140          mul->src[1].src.is_ssa = true;
 141          mul->src[1].src.ssa = &atomic_counter_size->def;
 142          nir_instr_insert_before(&instr->instr, &mul->instr);
 143
 144          nir_alu_instr *add = nir_alu_instr_create(mem_ctx, nir_op_iadd);
 145          nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, 32, NULL);
 146          add->dest.write_mask = 0x1;
 147          add->src[0].src.is_ssa = true;
 148          add->src[0].src.ssa = &mul->dest.dest.ssa;
 149          add->src[1].src.is_ssa = true;
 150          add->src[1].src.ssa = offset_def;
 151          nir_instr_insert_before(&instr->instr, &add->instr);
 152
 153          offset_def = &add->dest.dest.ssa;
 154       }
 155    }
 156
 157    new_instr->src[0].is_ssa = true;
 158    new_instr->src[0].ssa = offset_def;
 159
 160    /* Copy the other sources, if any, from the original instruction to the new
 161     * instruction.
 162     */
 163    for (unsigned i = 0; i < nir_intrinsic_infos[instr->intrinsic].num_srcs; i++)
 164       nir_src_copy(&new_instr->src[i + 1], &instr->src[i], new_instr);
 165
 166    if (instr->dest.is_ssa) {
 167       nir_ssa_dest_init(&new_instr->instr, &new_instr->dest,
 168                         instr->dest.ssa.num_components, 32, NULL);
 169       nir_ssa_def_rewrite_uses(&instr->dest.ssa,
 170                                nir_src_for_ssa(&new_instr->dest.ssa));
 171    } else {
 172       nir_dest_copy(&new_instr->dest, &instr->dest, mem_ctx);
 173    }
 174
 175    nir_instr_insert_before(&instr->instr, &new_instr->instr);
 176    nir_instr_remove(&instr->instr);
 177
 178    return true;
 179 }
 180
 181 bool
 182 gl_nir_lower_atomics(nir_shader *shader,
 183                      const struct gl_shader_program *shader_program,
 184                      bool use_binding_as_idx)
 185 {
 186    bool progress = false;
 187
 188    nir_assert_lowered_derefs(shader, nir_lower_atomic_counter_derefs);
 189
 190    nir_foreach_function(function, shader) {
 191       if (!function->impl)
 192          continue;
 193
 194       bool impl_progress = false;
 195
 196       nir_foreach_block(block, function->impl) {
 197          nir_foreach_instr_safe(instr, block) {
 198             if (instr->type != nir_instr_type_intrinsic)
 199                continue;
 200
 201             impl_progress |= lower_instr(nir_instr_as_intrinsic(instr),
 202                                          shader_program, shader,
 203                                          use_binding_as_idx);
 204          }
 205       }
 206
 207       if (impl_progress) {
 208          nir_metadata_preserve(function->impl, nir_metadata_block_index |
 209                                                nir_metadata_dominance);
 210          progress = true;
 211       }
 212    }
 213
 214    return progress;
 215 }