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_instr *deref, bool is_vertex_input)
 117 {
 118    unsigned offset = 0;
 119
 120    for (nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) {
 121       if (d->deref_type == nir_deref_type_array) {
 122          nir_const_value *const_index = nir_src_as_const_value(d->arr.index);
 123
 124          if (!const_index)
 125             return -1;
 126
 127          offset += glsl_count_attribute_slots(d->type, is_vertex_input) *
 128             const_index->u32[0];
 129       }
 130       /* TODO: we can get the offset for structs here see nir_lower_io() */
 131    }
 132
 133    return offset;
 134 }
 135
 136 /**
 137  * Try to mark a portion of the given varying as used.  Caller must ensure
 138  * that the variable represents a shader input or output.
 139  *
 140  * If the index can't be interpreted as a constant, or some other problem
 141  * occurs, then nothing will be marked and false will be returned.
 142  */
 143 static bool
 144 try_mask_partial_io(nir_shader *shader, nir_variable *var,
 145                     nir_deref_instr *deref, bool is_output_read)
 146 {
 147    const struct glsl_type *type = var->type;
 148
 149    if (nir_is_per_vertex_io(var, shader->info.stage)) {
 150       assert(glsl_type_is_array(type));
 151       type = glsl_get_array_element(type);
 152    }
 153
 154    /* The code below only handles:
 155     *
 156     * - Indexing into matrices
 157     * - Indexing into arrays of (arrays, matrices, vectors, or scalars)
 158     *
 159     * For now, we just give up if we see varying structs and arrays of structs
 160     * here marking the entire variable as used.
 161     */
 162    if (!(glsl_type_is_matrix(type) ||
 163          (glsl_type_is_array(type) && !var->data.compact &&
 164           (glsl_type_is_numeric(glsl_without_array(type)) ||
 165            glsl_type_is_boolean(glsl_without_array(type)))))) {
 166
 167       /* If we don't know how to handle this case, give up and let the
 168        * caller mark the whole variable as used.
 169        */
 170       return false;
 171    }
 172
 173    bool is_vertex_input = false;
 174    if (!shader->options->vs_inputs_dual_locations &&
 175        shader->info.stage == MESA_SHADER_VERTEX &&
 176        var->data.mode == nir_var_shader_in)
 177       is_vertex_input = true;
 178
 179    unsigned offset = get_io_offset(deref, is_vertex_input);
 180    if (offset == -1)
 181       return false;
 182
 183    unsigned num_elems;
 184    unsigned elem_width = 1;
 185    unsigned mat_cols = 1;
 186    if (glsl_type_is_array(type)) {
 187       num_elems = glsl_get_aoa_size(type);
 188       if (glsl_type_is_matrix(glsl_without_array(type)))
 189          mat_cols = glsl_get_matrix_columns(glsl_without_array(type));
 190    } else {
 191       num_elems = glsl_get_matrix_columns(type);
 192    }
 193
 194    /* double element width for double types that takes two slots */
 195    if (!is_vertex_input &&
 196        glsl_type_is_dual_slot(glsl_without_array(type))) {
 197       elem_width *= 2;
 198    }
 199
 200    if (offset >= num_elems * elem_width * mat_cols) {
 201       /* Constant index outside the bounds of the matrix/array.  This could
 202        * arise as a result of constant folding of a legal GLSL program.
 203        *
 204        * Even though the spec says that indexing outside the bounds of a
 205        * matrix/array results in undefined behaviour, we don't want to pass
 206        * out-of-range values to set_io_mask() (since this could result in
 207        * slots that don't exist being marked as used), so just let the caller
 208        * mark the whole variable as used.
 209        */
 210       return false;
 211    }
 212
 213    set_io_mask(shader, var, offset, elem_width, is_output_read);
 214    return true;
 215 }
 216
 217 static void
 218 gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader,
 219                       void *dead_ctx)
 220 {
 221    switch (instr->intrinsic) {
 222    case nir_intrinsic_discard:
 223    case nir_intrinsic_discard_if:
 224       assert(shader->info.stage == MESA_SHADER_FRAGMENT);
 225       shader->info.fs.uses_discard = true;
 226       break;
 227
 228    case nir_intrinsic_interp_deref_at_centroid:
 229    case nir_intrinsic_interp_deref_at_sample:
 230    case nir_intrinsic_interp_deref_at_offset:
 231    case nir_intrinsic_load_deref:
 232    case nir_intrinsic_store_deref:{
 233       nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
 234       nir_variable *var = nir_deref_instr_get_variable(deref);
 235
 236       if (var->data.mode == nir_var_shader_in ||
 237           var->data.mode == nir_var_shader_out) {
 238          bool is_output_read = false;
 239          if (var->data.mode == nir_var_shader_out &&
 240              instr->intrinsic == nir_intrinsic_load_deref)
 241             is_output_read = true;
 242
 243          if (!try_mask_partial_io(shader, var, deref, is_output_read))
 244             mark_whole_variable(shader, var, is_output_read);
 245
 246          /* We need to track which input_reads bits correspond to a
 247           * dvec3/dvec4 input attribute */
 248          if (shader->info.stage == MESA_SHADER_VERTEX &&
 249              var->data.mode == nir_var_shader_in &&
 250              glsl_type_is_dual_slot(glsl_without_array(var->type))) {
 251             for (unsigned i = 0; i < glsl_count_attribute_slots(var->type, false); i++) {
 252                int idx = var->data.location + i;
 253                shader->info.vs.double_inputs |= BITFIELD64_BIT(idx);
 254             }
 255          }
 256       }
 257       break;
 258    }
 259
 260    case nir_intrinsic_load_draw_id:
 261    case nir_intrinsic_load_frag_coord:
 262    case nir_intrinsic_load_front_face:
 263    case nir_intrinsic_load_vertex_id:
 264    case nir_intrinsic_load_vertex_id_zero_base:
 265    case nir_intrinsic_load_base_vertex:
 266    case nir_intrinsic_load_first_vertex:
 267    case nir_intrinsic_load_is_indexed_draw:
 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
 292    case nir_intrinsic_emit_vertex:
 293       if (nir_intrinsic_stream_id(instr) > 0)
 294          shader->info.gs.uses_streams = true;
 295
 296       break;
 297
 298    default:
 299       break;
 300    }
 301 }
 302
 303 static void
 304 gather_tex_info(nir_tex_instr *instr, nir_shader *shader)
 305 {
 306    switch (instr->op) {
 307    case nir_texop_tg4:
 308       shader->info.uses_texture_gather = true;
 309       break;
 310    case nir_texop_txf:
 311    case nir_texop_txf_ms:
 312    case nir_texop_txf_ms_mcs:
 313       shader->info.textures_used_by_txf |=
 314          ((1 << MAX2(instr->texture_array_size, 1)) - 1) <<
 315          instr->texture_index;
 316       break;
 317    default:
 318       break;
 319    }
 320 }
 321
 322 static void
 323 gather_alu_info(nir_alu_instr *instr, nir_shader *shader)
 324 {
 325    switch (instr->op) {
 326    case nir_op_fddx:
 327    case nir_op_fddy:
 328       shader->info.uses_fddx_fddy = true;
 329       break;
 330    default:
 331       break;
 332    }
 333 }
 334
 335 static void
 336 gather_info_block(nir_block *block, nir_shader *shader, void *dead_ctx)
 337 {
 338    nir_foreach_instr(instr, block) {
 339       switch (instr->type) {
 340       case nir_instr_type_alu:
 341          gather_alu_info(nir_instr_as_alu(instr), shader);
 342          break;
 343       case nir_instr_type_intrinsic:
 344          gather_intrinsic_info(nir_instr_as_intrinsic(instr), shader, dead_ctx);
 345          break;
 346       case nir_instr_type_tex:
 347          gather_tex_info(nir_instr_as_tex(instr), shader);
 348          break;
 349       case nir_instr_type_call:
 350          assert(!"nir_shader_gather_info only works if functions are inlined");
 351          break;
 352       default:
 353          break;
 354       }
 355    }
 356 }
 357
 358 static unsigned
 359 glsl_type_get_sampler_count(const struct glsl_type *type)
 360 {
 361    if (glsl_type_is_array(type)) {
 362       return (glsl_get_aoa_size(type) *
 363               glsl_type_get_sampler_count(glsl_without_array(type)));
 364    }
 365
 366    if (glsl_type_is_struct(type)) {
 367       unsigned count = 0;
 368       for (int i = 0; i < glsl_get_length(type); i++)
 369          count += glsl_type_get_sampler_count(glsl_get_struct_field(type, i));
 370       return count;
 371    }
 372
 373    if (glsl_type_is_sampler(type))
 374       return 1;
 375
 376    return 0;
 377 }
 378
 379 static unsigned
 380 glsl_type_get_image_count(const struct glsl_type *type)
 381 {
 382    if (glsl_type_is_array(type)) {
 383       return (glsl_get_aoa_size(type) *
 384               glsl_type_get_image_count(glsl_without_array(type)));
 385    }
 386
 387    if (glsl_type_is_struct(type)) {
 388       unsigned count = 0;
 389       for (int i = 0; i < glsl_get_length(type); i++)
 390          count += glsl_type_get_image_count(glsl_get_struct_field(type, i));
 391       return count;
 392    }
 393
 394    if (glsl_type_is_image(type))
 395       return 1;
 396
 397    return 0;
 398 }
 399
 400 void
 401 nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint)
 402 {
 403    shader->info.num_textures = 0;
 404    shader->info.num_images = 0;
 405    nir_foreach_variable(var, &shader->uniforms) {
 406       shader->info.num_textures += glsl_type_get_sampler_count(var->type);
 407       shader->info.num_images += glsl_type_get_image_count(var->type);
 408    }
 409
 410    shader->info.inputs_read = 0;
 411    shader->info.outputs_written = 0;
 412    shader->info.outputs_read = 0;
 413    shader->info.patch_outputs_read = 0;
 414    shader->info.patch_inputs_read = 0;
 415    shader->info.patch_outputs_written = 0;
 416    shader->info.system_values_read = 0;
 417    if (shader->info.stage == MESA_SHADER_VERTEX) {
 418       shader->info.vs.double_inputs = 0;
 419       shader->info.vs.double_inputs_read = 0;
 420    }
 421    if (shader->info.stage == MESA_SHADER_FRAGMENT) {
 422       shader->info.fs.uses_sample_qualifier = false;
 423    }
 424
 425    void *dead_ctx = ralloc_context(NULL);
 426    nir_foreach_block(block, entrypoint) {
 427       gather_info_block(block, shader, dead_ctx);
 428    }
 429    ralloc_free(dead_ctx);
 430 }