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          if (shader->info.stage == MESA_SHADER_FRAGMENT) {
  58             shader->info.fs.uses_sample_qualifier |= var->data.sample;
  59          }
  60       } else {
  61          assert(var->data.mode == nir_var_shader_out);
  62          if (is_output_read) {
  63             if (is_patch_generic) {
  64                shader->info.patch_outputs_read |= bitfield;
  65             } else {
  66                shader->info.outputs_read |= bitfield;
  67             }
  68          } else {
  69             if (is_patch_generic) {
  70                shader->info.patch_outputs_written |= bitfield;
  71             } else if (!var->data.read_only) {
  72                shader->info.outputs_written |= bitfield;
  73             }
  74          }
  75
  76
  77          if (var->data.fb_fetch_output)
  78             shader->info.outputs_read |= bitfield;
  79       }
  80    }
  81 }
  82
  83 /**
  84  * Mark an entire variable as used.  Caller must ensure that the variable
  85  * represents a shader input or output.
  86  */
  87 static void
  88 mark_whole_variable(nir_shader *shader, nir_variable *var, bool is_output_read)
  89 {
  90    const struct glsl_type *type = var->type;
  91
  92    if (nir_is_per_vertex_io(var, shader->info.stage)) {
  93       assert(glsl_type_is_array(type));
  94       type = glsl_get_array_element(type);
  95    }
  96
  97    const unsigned slots =
  98       var->data.compact ? DIV_ROUND_UP(glsl_get_length(type), 4)
  99                         : glsl_count_attribute_slots(type, false);
 100
 101    set_io_mask(shader, var, 0, slots, is_output_read);
 102 }
 103
 104 static unsigned
 105 get_io_offset(nir_deref_instr *deref, bool is_vertex_input)
 106 {
 107    unsigned offset = 0;
 108
 109    for (nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) {
 110       if (d->deref_type == nir_deref_type_array) {
 111          nir_const_value *const_index = nir_src_as_const_value(d->arr.index);
 112
 113          if (!const_index)
 114             return -1;
 115
 116          offset += glsl_count_attribute_slots(d->type, is_vertex_input) *
 117             const_index->u32[0];
 118       }
 119       /* TODO: we can get the offset for structs here see nir_lower_io() */
 120    }
 121
 122    return offset;
 123 }
 124
 125 /**
 126  * Try to mark a portion of the given varying as used.  Caller must ensure
 127  * that the variable represents a shader input or output.
 128  *
 129  * If the index can't be interpreted as a constant, or some other problem
 130  * occurs, then nothing will be marked and false will be returned.
 131  */
 132 static bool
 133 try_mask_partial_io(nir_shader *shader, nir_variable *var,
 134                     nir_deref_instr *deref, bool is_output_read)
 135 {
 136    const struct glsl_type *type = var->type;
 137
 138    if (nir_is_per_vertex_io(var, shader->info.stage)) {
 139       assert(glsl_type_is_array(type));
 140       type = glsl_get_array_element(type);
 141    }
 142
 143    /* The code below only handles:
 144     *
 145     * - Indexing into matrices
 146     * - Indexing into arrays of (arrays, matrices, vectors, or scalars)
 147     *
 148     * For now, we just give up if we see varying structs and arrays of structs
 149     * here marking the entire variable as used.
 150     */
 151    if (!(glsl_type_is_matrix(type) ||
 152          (glsl_type_is_array(type) && !var->data.compact &&
 153           (glsl_type_is_numeric(glsl_without_array(type)) ||
 154            glsl_type_is_boolean(glsl_without_array(type)))))) {
 155
 156       /* If we don't know how to handle this case, give up and let the
 157        * caller mark the whole variable as used.
 158        */
 159       return false;
 160    }
 161
 162    unsigned offset = get_io_offset(deref, false);
 163    if (offset == -1)
 164       return false;
 165
 166    unsigned num_elems;
 167    unsigned elem_width = 1;
 168    unsigned mat_cols = 1;
 169    if (glsl_type_is_array(type)) {
 170       num_elems = glsl_get_aoa_size(type);
 171       if (glsl_type_is_matrix(glsl_without_array(type)))
 172          mat_cols = glsl_get_matrix_columns(glsl_without_array(type));
 173    } else {
 174       num_elems = glsl_get_matrix_columns(type);
 175    }
 176
 177    /* double element width for double types that takes two slots */
 178    if (glsl_type_is_dual_slot(glsl_without_array(type)))
 179       elem_width *= 2;
 180
 181    if (offset >= num_elems * elem_width * mat_cols) {
 182       /* Constant index outside the bounds of the matrix/array.  This could
 183        * arise as a result of constant folding of a legal GLSL program.
 184        *
 185        * Even though the spec says that indexing outside the bounds of a
 186        * matrix/array results in undefined behaviour, we don't want to pass
 187        * out-of-range values to set_io_mask() (since this could result in
 188        * slots that don't exist being marked as used), so just let the caller
 189        * mark the whole variable as used.
 190        */
 191       return false;
 192    }
 193
 194    set_io_mask(shader, var, offset, elem_width, is_output_read);
 195    return true;
 196 }
 197
 198 static void
 199 gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader,
 200                       void *dead_ctx)
 201 {
 202    switch (instr->intrinsic) {
 203    case nir_intrinsic_discard:
 204    case nir_intrinsic_discard_if:
 205       assert(shader->info.stage == MESA_SHADER_FRAGMENT);
 206       shader->info.fs.uses_discard = true;
 207       break;
 208
 209    case nir_intrinsic_interp_deref_at_centroid:
 210    case nir_intrinsic_interp_deref_at_sample:
 211    case nir_intrinsic_interp_deref_at_offset:
 212    case nir_intrinsic_load_deref:
 213    case nir_intrinsic_store_deref:{
 214       nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
 215       nir_variable *var = nir_deref_instr_get_variable(deref);
 216
 217       if (var->data.mode == nir_var_shader_in ||
 218           var->data.mode == nir_var_shader_out) {
 219          bool is_output_read = false;
 220          if (var->data.mode == nir_var_shader_out &&
 221              instr->intrinsic == nir_intrinsic_load_deref)
 222             is_output_read = true;
 223
 224          if (!try_mask_partial_io(shader, var, deref, is_output_read))
 225             mark_whole_variable(shader, var, is_output_read);
 226
 227          /* We need to track which input_reads bits correspond to a
 228           * dvec3/dvec4 input attribute */
 229          if (shader->info.stage == MESA_SHADER_VERTEX &&
 230              var->data.mode == nir_var_shader_in &&
 231              glsl_type_is_dual_slot(glsl_without_array(var->type))) {
 232             for (unsigned i = 0; i < glsl_count_attribute_slots(var->type, false); i++) {
 233                int idx = var->data.location + i;
 234                shader->info.vs.double_inputs |= BITFIELD64_BIT(idx);
 235             }
 236          }
 237       }
 238       break;
 239    }
 240
 241    case nir_intrinsic_load_draw_id:
 242    case nir_intrinsic_load_frag_coord:
 243    case nir_intrinsic_load_front_face:
 244    case nir_intrinsic_load_vertex_id:
 245    case nir_intrinsic_load_vertex_id_zero_base:
 246    case nir_intrinsic_load_base_vertex:
 247    case nir_intrinsic_load_first_vertex:
 248    case nir_intrinsic_load_is_indexed_draw:
 249    case nir_intrinsic_load_base_instance:
 250    case nir_intrinsic_load_instance_id:
 251    case nir_intrinsic_load_sample_id:
 252    case nir_intrinsic_load_sample_pos:
 253    case nir_intrinsic_load_sample_mask_in:
 254    case nir_intrinsic_load_primitive_id:
 255    case nir_intrinsic_load_invocation_id:
 256    case nir_intrinsic_load_local_invocation_id:
 257    case nir_intrinsic_load_local_invocation_index:
 258    case nir_intrinsic_load_work_group_id:
 259    case nir_intrinsic_load_num_work_groups:
 260    case nir_intrinsic_load_tess_coord:
 261    case nir_intrinsic_load_tess_level_outer:
 262    case nir_intrinsic_load_tess_level_inner:
 263    case nir_intrinsic_load_patch_vertices_in:
 264       shader->info.system_values_read |=
 265          (1ull << nir_system_value_from_intrinsic(instr->intrinsic));
 266       break;
 267
 268    case nir_intrinsic_end_primitive:
 269    case nir_intrinsic_end_primitive_with_counter:
 270       assert(shader->info.stage == MESA_SHADER_GEOMETRY);
 271       shader->info.gs.uses_end_primitive = 1;
 272       /* fall through */
 273
 274    case nir_intrinsic_emit_vertex:
 275       if (nir_intrinsic_stream_id(instr) > 0)
 276          shader->info.gs.uses_streams = true;
 277
 278       break;
 279
 280    default:
 281       break;
 282    }
 283 }
 284
 285 static void
 286 gather_tex_info(nir_tex_instr *instr, nir_shader *shader)
 287 {
 288    switch (instr->op) {
 289    case nir_texop_tg4:
 290       shader->info.uses_texture_gather = true;
 291       break;
 292    case nir_texop_txf:
 293    case nir_texop_txf_ms:
 294    case nir_texop_txf_ms_mcs:
 295       shader->info.textures_used_by_txf |=
 296          ((1 << MAX2(instr->texture_array_size, 1)) - 1) <<
 297          instr->texture_index;
 298       break;
 299    default:
 300       break;
 301    }
 302 }
 303
 304 static void
 305 gather_alu_info(nir_alu_instr *instr, nir_shader *shader)
 306 {
 307    switch (instr->op) {
 308    case nir_op_fddx:
 309    case nir_op_fddy:
 310       shader->info.uses_fddx_fddy = true;
 311       break;
 312    default:
 313       break;
 314    }
 315 }
 316
 317 static void
 318 gather_info_block(nir_block *block, nir_shader *shader, void *dead_ctx)
 319 {
 320    nir_foreach_instr(instr, block) {
 321       switch (instr->type) {
 322       case nir_instr_type_alu:
 323          gather_alu_info(nir_instr_as_alu(instr), shader);
 324          break;
 325       case nir_instr_type_intrinsic:
 326          gather_intrinsic_info(nir_instr_as_intrinsic(instr), shader, dead_ctx);
 327          break;
 328       case nir_instr_type_tex:
 329          gather_tex_info(nir_instr_as_tex(instr), shader);
 330          break;
 331       case nir_instr_type_call:
 332          assert(!"nir_shader_gather_info only works if functions are inlined");
 333          break;
 334       default:
 335          break;
 336       }
 337    }
 338 }
 339
 340 static unsigned
 341 glsl_type_get_sampler_count(const struct glsl_type *type)
 342 {
 343    if (glsl_type_is_array(type)) {
 344       return (glsl_get_aoa_size(type) *
 345               glsl_type_get_sampler_count(glsl_without_array(type)));
 346    }
 347
 348    if (glsl_type_is_struct(type)) {
 349       unsigned count = 0;
 350       for (int i = 0; i < glsl_get_length(type); i++)
 351          count += glsl_type_get_sampler_count(glsl_get_struct_field(type, i));
 352       return count;
 353    }
 354
 355    if (glsl_type_is_sampler(type))
 356       return 1;
 357
 358    return 0;
 359 }
 360
 361 static unsigned
 362 glsl_type_get_image_count(const struct glsl_type *type)
 363 {
 364    if (glsl_type_is_array(type)) {
 365       return (glsl_get_aoa_size(type) *
 366               glsl_type_get_image_count(glsl_without_array(type)));
 367    }
 368
 369    if (glsl_type_is_struct(type)) {
 370       unsigned count = 0;
 371       for (int i = 0; i < glsl_get_length(type); i++)
 372          count += glsl_type_get_image_count(glsl_get_struct_field(type, i));
 373       return count;
 374    }
 375
 376    if (glsl_type_is_image(type))
 377       return 1;
 378
 379    return 0;
 380 }
 381
 382 void
 383 nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint)
 384 {
 385    shader->info.num_textures = 0;
 386    shader->info.num_images = 0;
 387    nir_foreach_variable(var, &shader->uniforms) {
 388       shader->info.num_textures += glsl_type_get_sampler_count(var->type);
 389       shader->info.num_images += glsl_type_get_image_count(var->type);
 390    }
 391
 392    shader->info.inputs_read = 0;
 393    shader->info.outputs_written = 0;
 394    shader->info.outputs_read = 0;
 395    shader->info.patch_outputs_read = 0;
 396    shader->info.patch_inputs_read = 0;
 397    shader->info.patch_outputs_written = 0;
 398    shader->info.system_values_read = 0;
 399    if (shader->info.stage == MESA_SHADER_VERTEX) {
 400       shader->info.vs.double_inputs = 0;
 401    }
 402    if (shader->info.stage == MESA_SHADER_FRAGMENT) {
 403       shader->info.fs.uses_sample_qualifier = false;
 404    }
 405
 406    void *dead_ctx = ralloc_context(NULL);
 407    nir_foreach_block(block, entrypoint) {
 408       gather_info_block(block, shader, dead_ctx);
 409    }
 410    ralloc_free(dead_ctx);
 411 }