src/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 id;
  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 id, ir_variable *var)
  56       {
  57          counters = (active_atomic_counter *)
  58             realloc(counters, sizeof(active_atomic_counter) * (num_counters + 1));
  59
  60          counters[num_counters].id = id;
  61          counters[num_counters].var = var;
  62          num_counters++;
  63       }
  64
  65       active_atomic_counter *counters;
  66       unsigned num_counters;
  67       unsigned stage_references[MESA_SHADER_TYPES];
  68       unsigned size;
  69    };
  70
  71    int
  72    cmp_actives(const void *a, const void *b)
  73    {
  74       const active_atomic_counter *const first = (active_atomic_counter *) a;
  75       const active_atomic_counter *const second = (active_atomic_counter *) b;
  76
  77       return int(first->var->data.atomic.offset) - int(second->var->data.atomic.offset);
  78    }
  79
  80    bool
  81    check_atomic_counters_overlap(const ir_variable *x, const ir_variable *y)
  82    {
  83       return ((x->data.atomic.offset >= y->data.atomic.offset &&
  84                x->data.atomic.offset < y->data.atomic.offset + y->type->atomic_size()) ||
  85               (y->data.atomic.offset >= x->data.atomic.offset &&
  86                y->data.atomic.offset < x->data.atomic.offset + x->type->atomic_size()));
  87    }
  88
  89    active_atomic_buffer *
  90    find_active_atomic_counters(struct gl_context *ctx,
  91                                struct gl_shader_program *prog,
  92                                unsigned *num_buffers)
  93    {
  94       active_atomic_buffer *const buffers =
  95          new active_atomic_buffer[ctx->Const.MaxAtomicBufferBindings];
  96
  97       *num_buffers = 0;
  98
  99       for (unsigned i = 0; i < MESA_SHADER_TYPES; ++i) {
 100          struct gl_shader *sh = prog->_LinkedShaders[i];
 101          if (sh == NULL)
 102             continue;
 103
 104          foreach_list(node, sh->ir) {
 105             ir_variable *var = ((ir_instruction *)node)->as_variable();
 106
 107             if (var && var->type->contains_atomic()) {
 108                unsigned id;
 109                bool found = prog->UniformHash->get(id, var->name);
 110                assert(found);
 111                active_atomic_buffer *buf = &buffers[var->data.binding];
 112
 113                /* If this is the first time the buffer is used, increment
 114                 * the counter of buffers used.
 115                 */
 116                if (buf->size == 0)
 117                   (*num_buffers)++;
 118
 119                buf->push_back(id, var);
 120
 121                buf->stage_references[i]++;
 122                buf->size = MAX2(buf->size, var->data.atomic.offset +
 123                                 var->type->atomic_size());
 124             }
 125          }
 126       }
 127
 128       for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
 129          if (buffers[i].size == 0)
 130             continue;
 131
 132          qsort(buffers[i].counters, buffers[i].num_counters,
 133                sizeof(active_atomic_counter),
 134                cmp_actives);
 135
 136          for (unsigned j = 1; j < buffers[i].num_counters; j++) {
 137             /* If an overlapping counter found, it must be a reference to the
 138              * same counter from a different shader stage.
 139              */
 140             if (check_atomic_counters_overlap(buffers[i].counters[j-1].var,
 141                                               buffers[i].counters[j].var)
 142                 && strcmp(buffers[i].counters[j-1].var->name,
 143                           buffers[i].counters[j].var->name) != 0) {
 144                linker_error(prog, "Atomic counter %s declared at offset %d "
 145                             "which is already in use.",
 146                             buffers[i].counters[j].var->name,
 147                             buffers[i].counters[j].var->data.atomic.offset);
 148             }
 149          }
 150       }
 151       return buffers;
 152    }
 153 }
 154
 155 void
 156 link_assign_atomic_counter_resources(struct gl_context *ctx,
 157                                      struct gl_shader_program *prog)
 158 {
 159    unsigned num_buffers;
 160    active_atomic_buffer *abs =
 161       find_active_atomic_counters(ctx, prog, &num_buffers);
 162
 163    prog->AtomicBuffers = rzalloc_array(prog, gl_active_atomic_buffer,
 164                                        num_buffers);
 165    prog->NumAtomicBuffers = num_buffers;
 166
 167    unsigned i = 0;
 168    for (unsigned binding = 0;
 169         binding < ctx->Const.MaxAtomicBufferBindings;
 170         binding++) {
 171
 172       /* If the binding was not used, skip.
 173        */
 174       if (abs[binding].size == 0)
 175          continue;
 176
 177       active_atomic_buffer &ab = abs[binding];
 178       gl_active_atomic_buffer &mab = prog->AtomicBuffers[i];
 179
 180       /* Assign buffer-specific fields. */
 181       mab.Binding = binding;
 182       mab.MinimumSize = ab.size;
 183       mab.Uniforms = rzalloc_array(prog->AtomicBuffers, GLuint,
 184                                    ab.num_counters);
 185       mab.NumUniforms = ab.num_counters;
 186
 187       /* Assign counter-specific fields. */
 188       for (unsigned j = 0; j < ab.num_counters; j++) {
 189          ir_variable *const var = ab.counters[j].var;
 190          const unsigned id = ab.counters[j].id;
 191          gl_uniform_storage *const storage = &prog->UniformStorage[id];
 192
 193          mab.Uniforms[j] = id;
 194          var->data.atomic.buffer_index = i;
 195          storage->atomic_buffer_index = i;
 196          storage->offset = var->data.atomic.offset;
 197          storage->array_stride = (var->type->is_array() ?
 198                                   var->type->element_type()->atomic_size() : 0);
 199       }
 200
 201       /* Assign stage-specific fields. */
 202       for (unsigned j = 0; j < MESA_SHADER_TYPES; ++j)
 203          mab.StageReferences[j] =
 204             (ab.stage_references[j] ? GL_TRUE : GL_FALSE);
 205
 206       i++;
 207    }
 208
 209    delete [] abs;
 210    assert(i == num_buffers);
 211 }
 212
 213 void
 214 link_check_atomic_counter_resources(struct gl_context *ctx,
 215                                     struct gl_shader_program *prog)
 216 {
 217    STATIC_ASSERT(MESA_SHADER_TYPES == 3);
 218    const unsigned max_atomic_counters[MESA_SHADER_TYPES] = {
 219       ctx->Const.VertexProgram.MaxAtomicCounters,
 220       ctx->Const.GeometryProgram.MaxAtomicCounters,
 221       ctx->Const.FragmentProgram.MaxAtomicCounters
 222    };
 223    const unsigned max_atomic_buffers[MESA_SHADER_TYPES] = {
 224       ctx->Const.VertexProgram.MaxAtomicBuffers,
 225       ctx->Const.GeometryProgram.MaxAtomicBuffers,
 226       ctx->Const.FragmentProgram.MaxAtomicBuffers
 227    };
 228    unsigned num_buffers;
 229    active_atomic_buffer *const abs =
 230       find_active_atomic_counters(ctx, prog, &num_buffers);
 231    unsigned atomic_counters[MESA_SHADER_TYPES] = {};
 232    unsigned atomic_buffers[MESA_SHADER_TYPES] = {};
 233    unsigned total_atomic_counters = 0;
 234    unsigned total_atomic_buffers = 0;
 235
 236    /* Sum the required resources.  Note that this counts buffers and
 237     * counters referenced by several shader stages multiple times
 238     * against the combined limit -- That's the behavior the spec
 239     * requires.
 240     */
 241    for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
 242       if (abs[i].size == 0)
 243          continue;
 244
 245       for (unsigned j = 0; j < MESA_SHADER_TYPES; ++j) {
 246          const unsigned n = abs[i].stage_references[j];
 247
 248          if (n) {
 249             atomic_counters[j] += n;
 250             total_atomic_counters += n;
 251             atomic_buffers[j]++;
 252             total_atomic_buffers++;
 253          }
 254       }
 255    }
 256
 257    /* Check that they are within the supported limits. */
 258    for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
 259       if (atomic_counters[i] > max_atomic_counters[i])
 260          linker_error(prog, "Too many %s shader atomic counters",
 261                       _mesa_shader_type_to_string(i));
 262
 263       if (atomic_buffers[i] > max_atomic_buffers[i])
 264          linker_error(prog, "Too many %s shader atomic counter buffers",
 265                       _mesa_shader_type_to_string(i));
 266    }
 267
 268    if (total_atomic_counters > ctx->Const.MaxCombinedAtomicCounters)
 269       linker_error(prog, "Too many combined atomic counters");
 270
 271    if (total_atomic_buffers > ctx->Const.MaxCombinedAtomicBuffers)
 272       linker_error(prog, "Too many combined atomic buffers");
 273
 274    delete [] abs;
 275 }