src/compiler/nir/nir_lower_system_values.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 "nir.h"
  29 #include "nir_builder.h"
  30
  31 static nir_ssa_def*
  32 build_local_group_size(nir_builder *b)
  33 {
  34    nir_const_value local_size;
  35    memset(&local_size, 0, sizeof(local_size));
  36    local_size.u32[0] = b->shader->info.cs.local_size[0];
  37    local_size.u32[1] = b->shader->info.cs.local_size[1];
  38    local_size.u32[2] = b->shader->info.cs.local_size[2];
  39    return nir_build_imm(b, 3, 32, local_size);
  40 }
  41
  42 static bool
  43 convert_block(nir_block *block, nir_builder *b)
  44 {
  45    bool progress = false;
  46
  47    nir_foreach_instr_safe(instr, block) {
  48       if (instr->type != nir_instr_type_intrinsic)
  49          continue;
  50
  51       nir_intrinsic_instr *load_deref = nir_instr_as_intrinsic(instr);
  52       if (load_deref->intrinsic != nir_intrinsic_load_deref)
  53          continue;
  54
  55       nir_deref_instr *deref = nir_src_as_deref(load_deref->src[0]);
  56       if (deref->mode != nir_var_system_value)
  57          continue;
  58
  59       if (deref->deref_type != nir_deref_type_var) {
  60          /* The only one system value that is an array and that is
  61           * gl_SampleMask which is always an array of one element.
  62           */
  63          assert(deref->deref_type == nir_deref_type_array);
  64          deref = nir_deref_instr_parent(deref);
  65          assert(deref->deref_type == nir_deref_type_var);
  66          assert(deref->var->data.location == SYSTEM_VALUE_SAMPLE_MASK_IN);
  67       }
  68       nir_variable *var = deref->var;
  69
  70       b->cursor = nir_after_instr(&load_deref->instr);
  71
  72       nir_ssa_def *sysval = NULL;
  73       switch (var->data.location) {
  74       case SYSTEM_VALUE_GLOBAL_INVOCATION_ID: {
  75          /* From the GLSL man page for gl_GlobalInvocationID:
  76           *
  77           *    "The value of gl_GlobalInvocationID is equal to
  78           *    gl_WorkGroupID * gl_WorkGroupSize + gl_LocalInvocationID"
  79           */
  80          nir_ssa_def *group_size = build_local_group_size(b);
  81          nir_ssa_def *group_id = nir_load_work_group_id(b);
  82          nir_ssa_def *local_id = nir_load_local_invocation_id(b);
  83
  84          sysval = nir_iadd(b, nir_imul(b, group_id, group_size), local_id);
  85          break;
  86       }
  87
  88       case SYSTEM_VALUE_LOCAL_INVOCATION_INDEX: {
  89          /* If lower_cs_local_index_from_id is true, then we derive the local
  90           * index from the local id.
  91           */
  92          if (!b->shader->options->lower_cs_local_index_from_id)
  93             break;
  94
  95          /* From the GLSL man page for gl_LocalInvocationIndex:
  96           *
  97           *    "The value of gl_LocalInvocationIndex is equal to
  98           *    gl_LocalInvocationID.z * gl_WorkGroupSize.x *
  99           *    gl_WorkGroupSize.y + gl_LocalInvocationID.y *
 100           *    gl_WorkGroupSize.x + gl_LocalInvocationID.x"
 101           */
 102          nir_ssa_def *local_id = nir_load_local_invocation_id(b);
 103
 104          nir_ssa_def *size_x =
 105             nir_imm_int(b, b->shader->info.cs.local_size[0]);
 106          nir_ssa_def *size_y =
 107             nir_imm_int(b, b->shader->info.cs.local_size[1]);
 108
 109          sysval = nir_imul(b, nir_channel(b, local_id, 2),
 110                               nir_imul(b, size_x, size_y));
 111          sysval = nir_iadd(b, sysval,
 112                               nir_imul(b, nir_channel(b, local_id, 1), size_x));
 113          sysval = nir_iadd(b, sysval, nir_channel(b, local_id, 0));
 114          break;
 115       }
 116
 117       case SYSTEM_VALUE_LOCAL_GROUP_SIZE: {
 118          sysval = build_local_group_size(b);
 119          break;
 120       }
 121
 122       case SYSTEM_VALUE_VERTEX_ID:
 123          if (b->shader->options->vertex_id_zero_based) {
 124             sysval = nir_iadd(b,
 125                               nir_load_vertex_id_zero_base(b),
 126                               nir_load_first_vertex(b));
 127          } else {
 128             sysval = nir_load_vertex_id(b);
 129          }
 130          break;
 131
 132       case SYSTEM_VALUE_BASE_VERTEX:
 133          /**
 134           * From the OpenGL 4.6 (11.1.3.9 Shader Inputs) specification:
 135           *
 136           * "gl_BaseVertex holds the integer value passed to the baseVertex
 137           * parameter to the command that resulted in the current shader
 138           * invocation. In the case where the command has no baseVertex
 139           * parameter, the value of gl_BaseVertex is zero."
 140           */
 141          if (b->shader->options->lower_base_vertex)
 142             sysval = nir_iand(b,
 143                               nir_load_is_indexed_draw(b),
 144                               nir_load_first_vertex(b));
 145          break;
 146
 147       case SYSTEM_VALUE_HELPER_INVOCATION:
 148          if (b->shader->options->lower_helper_invocation) {
 149             nir_ssa_def *tmp;
 150
 151             tmp = nir_ishl(b,
 152                            nir_imm_int(b, 1),
 153                            nir_load_sample_id_no_per_sample(b));
 154
 155             tmp = nir_iand(b,
 156                            nir_load_sample_mask_in(b),
 157                            tmp);
 158
 159             sysval = nir_inot(b, nir_i2b(b, tmp));
 160          }
 161
 162          break;
 163
 164       case SYSTEM_VALUE_INSTANCE_INDEX:
 165          sysval = nir_iadd(b,
 166                            nir_load_instance_id(b),
 167                            nir_load_base_instance(b));
 168          break;
 169
 170       case SYSTEM_VALUE_SUBGROUP_EQ_MASK:
 171       case SYSTEM_VALUE_SUBGROUP_GE_MASK:
 172       case SYSTEM_VALUE_SUBGROUP_GT_MASK:
 173       case SYSTEM_VALUE_SUBGROUP_LE_MASK:
 174       case SYSTEM_VALUE_SUBGROUP_LT_MASK: {
 175          nir_intrinsic_op op =
 176             nir_intrinsic_from_system_value(var->data.location);
 177          nir_intrinsic_instr *load = nir_intrinsic_instr_create(b->shader, op);
 178          nir_ssa_dest_init_for_type(&load->instr, &load->dest,
 179                                     var->type, NULL);
 180          load->num_components = load->dest.ssa.num_components;
 181          nir_builder_instr_insert(b, &load->instr);
 182          sysval = &load->dest.ssa;
 183          break;
 184       }
 185
 186       case SYSTEM_VALUE_DEVICE_INDEX:
 187          if (b->shader->options->lower_device_index_to_zero)
 188             sysval = nir_imm_int(b, 0);
 189          break;
 190
 191       case SYSTEM_VALUE_GLOBAL_GROUP_SIZE: {
 192          nir_ssa_def *group_size = build_local_group_size(b);
 193          nir_ssa_def *num_work_groups = nir_load_num_work_groups(b);
 194          sysval = nir_imul(b, group_size, num_work_groups);
 195          break;
 196       }
 197
 198       default:
 199          break;
 200       }
 201
 202       if (sysval == NULL) {
 203          nir_intrinsic_op sysval_op =
 204             nir_intrinsic_from_system_value(var->data.location);
 205          sysval = nir_load_system_value(b, sysval_op, 0);
 206          sysval->bit_size = load_deref->dest.ssa.bit_size;
 207       }
 208
 209       nir_ssa_def_rewrite_uses(&load_deref->dest.ssa, nir_src_for_ssa(sysval));
 210       nir_instr_remove(&load_deref->instr);
 211
 212       progress = true;
 213    }
 214
 215    return progress;
 216 }
 217
 218 static bool
 219 convert_impl(nir_function_impl *impl)
 220 {
 221    bool progress = false;
 222    nir_builder builder;
 223    nir_builder_init(&builder, impl);
 224
 225    nir_foreach_block(block, impl) {
 226       progress |= convert_block(block, &builder);
 227    }
 228
 229    nir_metadata_preserve(impl, nir_metadata_block_index |
 230                                nir_metadata_dominance);
 231    return progress;
 232 }
 233
 234 bool
 235 nir_lower_system_values(nir_shader *shader)
 236 {
 237    bool progress = false;
 238
 239    nir_foreach_function(function, shader) {
 240       if (function->impl)
 241          progress = convert_impl(function->impl) || progress;
 242    }
 243
 244    /* We're going to delete the variables so we need to clean up all those
 245     * derefs we left lying around.
 246     */
 247    nir_remove_dead_derefs(shader);
 248
 249    exec_list_make_empty(&shader->system_values);
 250
 251    return progress;
 252 }