src/compiler/glsl/link_atomics.cpp

   1 /*
   2  * Copyright © 2013 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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "glsl_parser_extras.h"
  25 #include "ir.h"
  26 #include "ir_uniform.h"
  27 #include "linker.h"
  28 #include "program/hash_table.h"
  29 #include "main/macros.h"
  30
  31 namespace {
  32    /*
  33     * Atomic counter as seen by the program.
  34     */
  35    struct active_atomic_counter {
  36       unsigned uniform_loc;
  37       ir_variable *var;
  38    };
  39
  40    /*
  41     * Atomic counter buffer referenced by the program.  There is a one
  42     * to one correspondence between these and the objects that can be
  43     * queried using glGetActiveAtomicCounterBufferiv().
  44     */
  45    struct active_atomic_buffer {
  46       active_atomic_buffer()
  47          : counters(0), num_counters(0), stage_references(), size(0)
  48       {}
  49
  50       ~active_atomic_buffer()
  51       {
  52          free(counters);
  53       }
  54
  55       void push_back(unsigned uniform_loc, ir_variable *var)
  56       {
  57          active_atomic_counter *new_counters;
  58
  59          new_counters = (active_atomic_counter *)
  60             realloc(counters, sizeof(active_atomic_counter) *
  61                     (num_counters + 1));
  62
  63          if (new_counters == NULL) {
  64             _mesa_error_no_memory(__func__);
  65             return;
  66          }
  67
  68          counters = new_counters;
  69          counters[num_counters].uniform_loc = uniform_loc;
  70          counters[num_counters].var = var;
  71          num_counters++;
  72       }
  73
  74       active_atomic_counter *counters;
  75       unsigned num_counters;
  76       unsigned stage_references[MESA_SHADER_STAGES];
  77       unsigned size;
  78    };
  79
  80    int
  81    cmp_actives(const void *a, const void *b)
  82    {
  83       const active_atomic_counter *const first = (active_atomic_counter *) a;
  84       const active_atomic_counter *const second = (active_atomic_counter *) b;
  85
  86       return int(first->var->data.offset) - int(second->var->data.offset);
  87    }
  88
  89    bool
  90    check_atomic_counters_overlap(const ir_variable *x, const ir_variable *y)
  91    {
  92       return ((x->data.offset >= y->data.offset &&
  93                x->data.offset < y->data.offset + y->type->atomic_size()) ||
  94               (y->data.offset >= x->data.offset &&
  95                y->data.offset < x->data.offset + x->type->atomic_size()));
  96    }
  97
  98    void
  99    process_atomic_variable(const glsl_type *t, struct gl_shader_program *prog,
 100                            unsigned *uniform_loc, ir_variable *var,
 101                            active_atomic_buffer *const buffers,
 102                            unsigned *num_buffers, int *offset,
 103                            const unsigned shader_stage)
 104    {
 105       /* FIXME: Arrays of arrays get counted separately. For example:
 106        * x1[3][3][2] = 9 uniforms, 18 atomic counters
 107        * x2[3][2]    = 3 uniforms, 6 atomic counters
 108        * x3[2]       = 1 uniform, 2 atomic counters
 109        *
 110        * However this code marks all the counters as active even when they
 111        * might not be used.
 112        */
 113       if (t->is_array() && t->fields.array->is_array()) {
 114          for (unsigned i = 0; i < t->length; i++) {
 115             process_atomic_variable(t->fields.array, prog, uniform_loc,
 116                                     var, buffers, num_buffers, offset,
 117                                     shader_stage);
 118          }
 119       } else {
 120          active_atomic_buffer *buf = &buffers[var->data.binding];
 121          gl_uniform_storage *const storage =
 122             &prog->UniformStorage[*uniform_loc];
 123
 124          /* If this is the first time the buffer is used, increment
 125           * the counter of buffers used.
 126           */
 127          if (buf->size == 0)
 128             (*num_buffers)++;
 129
 130          buf->push_back(*uniform_loc, var);
 131
 132          /* When checking for atomic counters we should count every member in
 133           * an array as an atomic counter reference.
 134           */
 135          if (t->is_array())
 136             buf->stage_references[shader_stage] += t->length;
 137          else
 138             buf->stage_references[shader_stage]++;
 139          buf->size = MAX2(buf->size, *offset + t->atomic_size());
 140
 141          storage->offset = *offset;
 142          *offset += t->atomic_size();
 143
 144          (*uniform_loc)++;
 145       }
 146    }
 147
 148    active_atomic_buffer *
 149    find_active_atomic_counters(struct gl_context *ctx,
 150                                struct gl_shader_program *prog,
 151                                unsigned *num_buffers)
 152    {
 153       active_atomic_buffer *const buffers =
 154          new active_atomic_buffer[ctx->Const.MaxAtomicBufferBindings];
 155
 156       *num_buffers = 0;
 157
 158       for (unsigned i = 0; i < MESA_SHADER_STAGES; ++i) {
 159          struct gl_linked_shader *sh = prog->_LinkedShaders[i];
 160          if (sh == NULL)
 161             continue;
 162
 163          foreach_in_list(ir_instruction, node, sh->ir) {
 164             ir_variable *var = node->as_variable();
 165
 166             if (var && var->type->contains_atomic()) {
 167                int offset = var->data.offset;
 168                unsigned uniform_loc = var->data.location;
 169                process_atomic_variable(var->type, prog, &uniform_loc,
 170                                        var, buffers, num_buffers, &offset, i);
 171             }
 172          }
 173       }
 174
 175       for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
 176          if (buffers[i].size == 0)
 177             continue;
 178
 179          qsort(buffers[i].counters, buffers[i].num_counters,
 180                sizeof(active_atomic_counter),
 181                cmp_actives);
 182
 183          for (unsigned j = 1; j < buffers[i].num_counters; j++) {
 184             /* If an overlapping counter found, it must be a reference to the
 185              * same counter from a different shader stage.
 186              */
 187             if (check_atomic_counters_overlap(buffers[i].counters[j-1].var,
 188                                               buffers[i].counters[j].var)
 189                 && strcmp(buffers[i].counters[j-1].var->name,
 190                           buffers[i].counters[j].var->name) != 0) {
 191                linker_error(prog, "Atomic counter %s declared at offset %d "
 192                             "which is already in use.",
 193                             buffers[i].counters[j].var->name,
 194                             buffers[i].counters[j].var->data.offset);
 195             }
 196          }
 197       }
 198       return buffers;
 199    }
 200 }
 201
 202 void
 203 link_assign_atomic_counter_resources(struct gl_context *ctx,
 204                                      struct gl_shader_program *prog)
 205 {
 206    unsigned num_buffers;
 207    unsigned num_atomic_buffers[MESA_SHADER_STAGES] = {};
 208    active_atomic_buffer *abs =
 209       find_active_atomic_counters(ctx, prog, &num_buffers);
 210
 211    prog->AtomicBuffers = rzalloc_array(prog, gl_active_atomic_buffer,
 212                                        num_buffers);
 213    prog->NumAtomicBuffers = num_buffers;
 214
 215    unsigned i = 0;
 216    for (unsigned binding = 0;
 217         binding < ctx->Const.MaxAtomicBufferBindings;
 218         binding++) {
 219
 220       /* If the binding was not used, skip.
 221        */
 222       if (abs[binding].size == 0)
 223          continue;
 224
 225       active_atomic_buffer &ab = abs[binding];
 226       gl_active_atomic_buffer &mab = prog->AtomicBuffers[i];
 227
 228       /* Assign buffer-specific fields. */
 229       mab.Binding = binding;
 230       mab.MinimumSize = ab.size;
 231       mab.Uniforms = rzalloc_array(prog->AtomicBuffers, GLuint,
 232                                    ab.num_counters);
 233       mab.NumUniforms = ab.num_counters;
 234
 235       /* Assign counter-specific fields. */
 236       for (unsigned j = 0; j < ab.num_counters; j++) {
 237          ir_variable *const var = ab.counters[j].var;
 238          gl_uniform_storage *const storage =
 239             &prog->UniformStorage[ab.counters[j].uniform_loc];
 240
 241          mab.Uniforms[j] = ab.counters[j].uniform_loc;
 242          if (!var->data.explicit_binding)
 243             var->data.binding = i;
 244
 245          storage->atomic_buffer_index = i;
 246          storage->offset = var->data.offset;
 247          storage->array_stride = (var->type->is_array() ?
 248                                   var->type->without_array()->atomic_size() : 0);
 249          if (!var->type->is_matrix())
 250             storage->matrix_stride = 0;
 251       }
 252
 253       /* Assign stage-specific fields. */
 254       for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j) {
 255          if (ab.stage_references[j]) {
 256             mab.StageReferences[j] = GL_TRUE;
 257             num_atomic_buffers[j]++;
 258          } else {
 259             mab.StageReferences[j] = GL_FALSE;
 260          }
 261       }
 262
 263       i++;
 264    }
 265
 266    /* Store a list pointers to atomic buffers per stage and store the index
 267     * to the intra-stage buffer list in uniform storage.
 268     */
 269    for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j) {
 270       if (prog->_LinkedShaders[j] && num_atomic_buffers[j] > 0) {
 271          prog->_LinkedShaders[j]->NumAtomicBuffers = num_atomic_buffers[j];
 272          prog->_LinkedShaders[j]->AtomicBuffers =
 273             rzalloc_array(prog, gl_active_atomic_buffer *,
 274                           num_atomic_buffers[j]);
 275
 276          unsigned intra_stage_idx = 0;
 277          for (unsigned i = 0; i < num_buffers; i++) {
 278             struct gl_active_atomic_buffer *atomic_buffer =
 279                &prog->AtomicBuffers[i];
 280             if (atomic_buffer->StageReferences[j]) {
 281                prog->_LinkedShaders[j]->AtomicBuffers[intra_stage_idx] =
 282                   atomic_buffer;
 283
 284                for (unsigned u = 0; u < atomic_buffer->NumUniforms; u++) {
 285                   prog->UniformStorage[atomic_buffer->Uniforms[u]].opaque[j].index =
 286                      intra_stage_idx;
 287                   prog->UniformStorage[atomic_buffer->Uniforms[u]].opaque[j].active =
 288                      true;
 289                }
 290
 291                intra_stage_idx++;
 292             }
 293          }
 294       }
 295    }
 296
 297    delete [] abs;
 298    assert(i == num_buffers);
 299 }
 300
 301 void
 302 link_check_atomic_counter_resources(struct gl_context *ctx,
 303                                     struct gl_shader_program *prog)
 304 {
 305    unsigned num_buffers;
 306    active_atomic_buffer *const abs =
 307       find_active_atomic_counters(ctx, prog, &num_buffers);
 308    unsigned atomic_counters[MESA_SHADER_STAGES] = {};
 309    unsigned atomic_buffers[MESA_SHADER_STAGES] = {};
 310    unsigned total_atomic_counters = 0;
 311    unsigned total_atomic_buffers = 0;
 312
 313    /* Sum the required resources.  Note that this counts buffers and
 314     * counters referenced by several shader stages multiple times
 315     * against the combined limit -- That's the behavior the spec
 316     * requires.
 317     */
 318    for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
 319       if (abs[i].size == 0)
 320          continue;
 321
 322       for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j) {
 323          const unsigned n = abs[i].stage_references[j];
 324
 325          if (n) {
 326             atomic_counters[j] += n;
 327             total_atomic_counters += n;
 328             atomic_buffers[j]++;
 329             total_atomic_buffers++;
 330          }
 331       }
 332    }
 333
 334    /* Check that they are within the supported limits. */
 335    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
 336       if (atomic_counters[i] > ctx->Const.Program[i].MaxAtomicCounters)
 337          linker_error(prog, "Too many %s shader atomic counters",
 338                       _mesa_shader_stage_to_string(i));
 339
 340       if (atomic_buffers[i] > ctx->Const.Program[i].MaxAtomicBuffers)
 341          linker_error(prog, "Too many %s shader atomic counter buffers",
 342                       _mesa_shader_stage_to_string(i));
 343    }
 344
 345    if (total_atomic_counters > ctx->Const.MaxCombinedAtomicCounters)
 346       linker_error(prog, "Too many combined atomic counters");
 347
 348    if (total_atomic_buffers > ctx->Const.MaxCombinedAtomicBuffers)
 349       linker_error(prog, "Too many combined atomic buffers");
 350
 351    delete [] abs;
 352 }