src/compiler/nir/nir_gather_info.c

   1 /*
   2  * Copyright © 2015 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 "main/menums.h"
  26
  27 static void
  28 set_io_mask(nir_shader *shader, nir_variable *var, int offset, int len,
  29             bool is_output_read)
  30 {
  31    for (int i = 0; i < len; i++) {
  32       assert(var->data.location != -1);
  33
  34       int idx = var->data.location + offset + i;
  35       bool is_patch_generic = var->data.patch &&
  36                               idx != VARYING_SLOT_TESS_LEVEL_INNER &&
  37                               idx != VARYING_SLOT_TESS_LEVEL_OUTER &&
  38                               idx != VARYING_SLOT_BOUNDING_BOX0 &&
  39                               idx != VARYING_SLOT_BOUNDING_BOX1;
  40       uint64_t bitfield;
  41
  42       if (is_patch_generic) {
  43          assert(idx >= VARYING_SLOT_PATCH0 && idx < VARYING_SLOT_TESS_MAX);
  44          bitfield = BITFIELD64_BIT(idx - VARYING_SLOT_PATCH0);
  45       }
  46       else {
  47          assert(idx < VARYING_SLOT_MAX);
  48          bitfield = BITFIELD64_BIT(idx);
  49       }
  50
  51       if (var->data.mode == nir_var_shader_in) {
  52          if (is_patch_generic)
  53             shader->info.patch_inputs_read |= bitfield;
  54          else
  55             shader->info.inputs_read |= bitfield;
  56
  57          /* double inputs read is only for vertex inputs */
  58          if (shader->info.stage == MESA_SHADER_VERTEX &&
  59              glsl_type_is_dual_slot(glsl_without_array(var->type)))
  60             shader->info.vs.double_inputs_read |= bitfield;
  61
  62          if (shader->info.stage == MESA_SHADER_FRAGMENT) {
  63             shader->info.fs.uses_sample_qualifier |= var->data.sample;
  64          }
  65       } else {
  66          assert(var->data.mode == nir_var_shader_out);
  67          if (is_output_read) {
  68             if (is_patch_generic) {
  69                shader->info.patch_outputs_read |= bitfield;
  70             } else {
  71                shader->info.outputs_read |= bitfield;
  72             }
  73          } else {
  74             if (is_patch_generic) {
  75                shader->info.patch_outputs_written |= bitfield;
  76             } else if (!var->data.read_only) {
  77                shader->info.outputs_written |= bitfield;
  78             }
  79          }
  80
  81
  82          if (var->data.fb_fetch_output)
  83             shader->info.outputs_read |= bitfield;
  84       }
  85    }
  86 }
  87
  88 /**
  89  * Mark an entire variable as used.  Caller must ensure that the variable
  90  * represents a shader input or output.
  91  */
  92 static void
  93 mark_whole_variable(nir_shader *shader, nir_variable *var, bool is_output_read)
  94 {
  95    const struct glsl_type *type = var->type;
  96    bool is_vertex_input = false;
  97
  98    if (nir_is_per_vertex_io(var, shader->info.stage)) {
  99       assert(glsl_type_is_array(type));
 100       type = glsl_get_array_element(type);
 101    }
 102
 103    if (!shader->options->vs_inputs_dual_locations &&
 104        shader->info.stage == MESA_SHADER_VERTEX &&
 105        var->data.mode == nir_var_shader_in)
 106       is_vertex_input = true;
 107
 108    const unsigned slots =
 109       var->data.compact ? DIV_ROUND_UP(glsl_get_length(type), 4)
 110                         : glsl_count_attribute_slots(type, is_vertex_input);
 111
 112    set_io_mask(shader, var, 0, slots, is_output_read);
 113 }
 114
 115 static unsigned
 116 get_io_offset(nir_deref_var *deref, bool is_vertex_input)
 117 {
 118    unsigned offset = 0;
 119
 120    nir_deref *tail = &deref->deref;
 121    while (tail->child != NULL) {
 122       tail = tail->child;
 123
 124       if (tail->deref_type == nir_deref_type_array) {
 125          nir_deref_array *deref_array = nir_deref_as_array(tail);
 126
 127          if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
 128             return -1;
 129          }
 130
 131          offset += glsl_count_attribute_slots(tail->type, is_vertex_input) *
 132             deref_array->base_offset;
 133       }
 134       /* TODO: we can get the offset for structs here see nir_lower_io() */
 135    }
 136
 137    return offset;
 138 }
 139
 140 /**
 141  * Try to mark a portion of the given varying as used.  Caller must ensure
 142  * that the variable represents a shader input or output.
 143  *
 144  * If the index can't be interpreted as a constant, or some other problem
 145  * occurs, then nothing will be marked and false will be returned.
 146  */
 147 static bool
 148 try_mask_partial_io(nir_shader *shader, nir_deref_var *deref, bool is_output_read)
 149 {
 150    nir_variable *var = deref->var;
 151    const struct glsl_type *type = var->type;
 152
 153    if (nir_is_per_vertex_io(var, shader->info.stage)) {
 154       assert(glsl_type_is_array(type));
 155       type = glsl_get_array_element(type);
 156    }
 157
 158    /* The code below only handles:
 159     *
 160     * - Indexing into matrices
 161     * - Indexing into arrays of (arrays, matrices, vectors, or scalars)
 162     *
 163     * For now, we just give up if we see varying structs and arrays of structs
 164     * here marking the entire variable as used.
 165     */
 166    if (!(glsl_type_is_matrix(type) ||
 167          (glsl_type_is_array(type) && !var->data.compact &&
 168           (glsl_type_is_numeric(glsl_without_array(type)) ||
 169            glsl_type_is_boolean(glsl_without_array(type)))))) {
 170
 171       /* If we don't know how to handle this case, give up and let the
 172        * caller mark the whole variable as used.
 173        */
 174       return false;
 175    }
 176
 177    bool is_vertex_input = false;
 178    if (!shader->options->vs_inputs_dual_locations &&
 179        shader->info.stage == MESA_SHADER_VERTEX &&
 180        var->data.mode == nir_var_shader_in)
 181       is_vertex_input = true;
 182
 183    unsigned offset = get_io_offset(deref, is_vertex_input);
 184    if (offset == -1)
 185       return false;
 186
 187    unsigned num_elems;
 188    unsigned elem_width = 1;
 189    unsigned mat_cols = 1;
 190    if (glsl_type_is_array(type)) {
 191       num_elems = glsl_get_aoa_size(type);
 192       if (glsl_type_is_matrix(glsl_without_array(type)))
 193          mat_cols = glsl_get_matrix_columns(glsl_without_array(type));
 194    } else {
 195       num_elems = glsl_get_matrix_columns(type);
 196    }
 197
 198    /* double element width for double types that takes two slots */
 199    if (!is_vertex_input &&
 200        glsl_type_is_dual_slot(glsl_without_array(type))) {
 201       elem_width *= 2;
 202    }
 203
 204    if (offset >= num_elems * elem_width * mat_cols) {
 205       /* Constant index outside the bounds of the matrix/array.  This could
 206        * arise as a result of constant folding of a legal GLSL program.
 207        *
 208        * Even though the spec says that indexing outside the bounds of a
 209        * matrix/array results in undefined behaviour, we don't want to pass
 210        * out-of-range values to set_io_mask() (since this could result in
 211        * slots that don't exist being marked as used), so just let the caller
 212        * mark the whole variable as used.
 213        */
 214       return false;
 215    }
 216
 217    set_io_mask(shader, var, offset, elem_width, is_output_read);
 218    return true;
 219 }
 220
 221 static void
 222 gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader)
 223 {
 224    switch (instr->intrinsic) {
 225    case nir_intrinsic_discard:
 226    case nir_intrinsic_discard_if:
 227       assert(shader->info.stage == MESA_SHADER_FRAGMENT);
 228       shader->info.fs.uses_discard = true;
 229       break;
 230
 231    case nir_intrinsic_interp_var_at_centroid:
 232    case nir_intrinsic_interp_var_at_sample:
 233    case nir_intrinsic_interp_var_at_offset:
 234    case nir_intrinsic_load_var:
 235    case nir_intrinsic_store_var: {
 236       nir_variable *var = instr->variables[0]->var;
 237
 238       if (var->data.mode == nir_var_shader_in ||
 239           var->data.mode == nir_var_shader_out) {
 240          bool is_output_read = false;
 241          if (var->data.mode == nir_var_shader_out &&
 242              instr->intrinsic == nir_intrinsic_load_var)
 243             is_output_read = true;
 244
 245          if (!try_mask_partial_io(shader, instr->variables[0], is_output_read))
 246             mark_whole_variable(shader, var, is_output_read);
 247
 248          /* We need to track which input_reads bits correspond to a
 249           * dvec3/dvec4 input attribute */
 250          if (shader->info.stage == MESA_SHADER_VERTEX &&
 251              var->data.mode == nir_var_shader_in &&
 252              glsl_type_is_dual_slot(glsl_without_array(var->type))) {
 253             for (unsigned i = 0; i < glsl_count_attribute_slots(var->type, false); i++) {
 254                int idx = var->data.location + i;
 255                shader->info.vs.double_inputs |= BITFIELD64_BIT(idx);
 256             }
 257          }
 258       }
 259       break;
 260    }
 261
 262    case nir_intrinsic_load_draw_id:
 263    case nir_intrinsic_load_frag_coord:
 264    case nir_intrinsic_load_front_face:
 265    case nir_intrinsic_load_vertex_id:
 266    case nir_intrinsic_load_vertex_id_zero_base:
 267    case nir_intrinsic_load_base_vertex:
 268    case nir_intrinsic_load_base_instance:
 269    case nir_intrinsic_load_instance_id:
 270    case nir_intrinsic_load_sample_id:
 271    case nir_intrinsic_load_sample_pos:
 272    case nir_intrinsic_load_sample_mask_in:
 273    case nir_intrinsic_load_primitive_id:
 274    case nir_intrinsic_load_invocation_id:
 275    case nir_intrinsic_load_local_invocation_id:
 276    case nir_intrinsic_load_local_invocation_index:
 277    case nir_intrinsic_load_work_group_id:
 278    case nir_intrinsic_load_num_work_groups:
 279    case nir_intrinsic_load_tess_coord:
 280    case nir_intrinsic_load_tess_level_outer:
 281    case nir_intrinsic_load_tess_level_inner:
 282    case nir_intrinsic_load_patch_vertices_in:
 283       shader->info.system_values_read |=
 284          (1ull << nir_system_value_from_intrinsic(instr->intrinsic));
 285       break;
 286
 287    case nir_intrinsic_end_primitive:
 288    case nir_intrinsic_end_primitive_with_counter:
 289       assert(shader->info.stage == MESA_SHADER_GEOMETRY);
 290       shader->info.gs.uses_end_primitive = 1;
 291       break;
 292
 293    default:
 294       break;
 295    }
 296 }
 297
 298 static void
 299 gather_tex_info(nir_tex_instr *instr, nir_shader *shader)
 300 {
 301    switch (instr->op) {
 302    case nir_texop_tg4:
 303       shader->info.uses_texture_gather = true;
 304       break;
 305    case nir_texop_txf:
 306    case nir_texop_txf_ms:
 307    case nir_texop_txf_ms_mcs:
 308       shader->info.textures_used_by_txf |=
 309          ((1 << MAX2(instr->texture_array_size, 1)) - 1) <<
 310          instr->texture_index;
 311       break;
 312    default:
 313       break;
 314    }
 315 }
 316
 317 static void
 318 gather_alu_info(nir_alu_instr *instr, nir_shader *shader)
 319 {
 320    switch (instr->op) {
 321    case nir_op_fddx:
 322    case nir_op_fddy:
 323       shader->info.uses_fddx_fddy = true;
 324       break;
 325    default:
 326       break;
 327    }
 328 }
 329
 330 static void
 331 gather_info_block(nir_block *block, nir_shader *shader)
 332 {
 333    nir_foreach_instr(instr, block) {
 334       switch (instr->type) {
 335       case nir_instr_type_alu:
 336          gather_alu_info(nir_instr_as_alu(instr), shader);
 337          break;
 338       case nir_instr_type_intrinsic:
 339          gather_intrinsic_info(nir_instr_as_intrinsic(instr), shader);
 340          break;
 341       case nir_instr_type_tex:
 342          gather_tex_info(nir_instr_as_tex(instr), shader);
 343          break;
 344       case nir_instr_type_call:
 345          assert(!"nir_shader_gather_info only works if functions are inlined");
 346          break;
 347       default:
 348          break;
 349       }
 350    }
 351 }
 352
 353 void
 354 nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint)
 355 {
 356    shader->info.num_textures = 0;
 357    shader->info.num_images = 0;
 358    nir_foreach_variable(var, &shader->uniforms) {
 359       const struct glsl_type *type = var->type;
 360       unsigned count = 1;
 361       if (glsl_type_is_array(type)) {
 362          count = glsl_get_aoa_size(type);
 363          type = glsl_without_array(type);
 364       }
 365
 366       if (glsl_type_is_image(type)) {
 367          shader->info.num_images += count;
 368       } else if (glsl_type_is_sampler(type)) {
 369          shader->info.num_textures += count;
 370       }
 371    }
 372
 373    shader->info.inputs_read = 0;
 374    shader->info.outputs_written = 0;
 375    shader->info.outputs_read = 0;
 376    shader->info.patch_outputs_read = 0;
 377    shader->info.patch_inputs_read = 0;
 378    shader->info.patch_outputs_written = 0;
 379    shader->info.system_values_read = 0;
 380    if (shader->info.stage == MESA_SHADER_VERTEX) {
 381       shader->info.vs.double_inputs = 0;
 382       shader->info.vs.double_inputs_read = 0;
 383    }
 384    if (shader->info.stage == MESA_SHADER_FRAGMENT) {
 385       shader->info.fs.uses_sample_qualifier = false;
 386    }
 387    nir_foreach_block(block, entrypoint) {
 388       gather_info_block(block, shader);
 389    }
 390 }