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