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