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 "ir.h"
  25 #include "ir_uniform.h"
  26 #include "linker.h"
  27 #include "program/hash_table.h"
  28 #include "main/macros.h"
  29
  30 namespace {
  31    /*
  32     * Atomic counter as seen by the program.
  33     */
  34    struct active_atomic_counter {
  35       unsigned id;
  36       ir_variable *var;
  37    };
  38
  39    /*
  40     * Atomic counter buffer referenced by the program.  There is a one
  41     * to one correspondence between these and the objects that can be
  42     * queried using glGetActiveAtomicCounterBufferiv().
  43     */
  44    struct active_atomic_buffer {
  45       active_atomic_buffer()
  46          : counters(0), num_counters(0), stage_references(), size(0)
  47       {}
  48
  49       ~active_atomic_buffer()
  50       {
  51          free(counters);
  52       }
  53
  54       void push_back(unsigned id, ir_variable *var)
  55       {
  56          counters = (active_atomic_counter *)
  57             realloc(counters, sizeof(active_atomic_counter) * (num_counters + 1));
  58
  59          counters[num_counters].id = id;
  60          counters[num_counters].var = var;
  61          num_counters++;
  62       }
  63
  64       active_atomic_counter *counters;
  65       unsigned num_counters;
  66       unsigned stage_references[MESA_SHADER_TYPES];
  67       unsigned size;
  68    };
  69
  70    int
  71    cmp_actives(const void *a, const void *b)
  72    {
  73       const active_atomic_counter *const first = (active_atomic_counter *) a;
  74       const active_atomic_counter *const second = (active_atomic_counter *) b;
  75
  76       return int(first->var->atomic.offset) - int(second->var->atomic.offset);
  77    }
  78
  79    bool
  80    check_atomic_counters_overlap(const ir_variable *x, const ir_variable *y)
  81    {
  82       return ((x->atomic.offset >= y->atomic.offset &&
  83                x->atomic.offset < y->atomic.offset + y->type->atomic_size()) ||
  84               (y->atomic.offset >= x->atomic.offset &&
  85                y->atomic.offset < x->atomic.offset + x->type->atomic_size()));
  86    }
  87
  88    active_atomic_buffer *
  89    find_active_atomic_counters(struct gl_context *ctx,
  90                                struct gl_shader_program *prog,
  91                                unsigned *num_buffers)
  92    {
  93       active_atomic_buffer *const buffers =
  94          new active_atomic_buffer[ctx->Const.MaxAtomicBufferBindings];
  95
  96       *num_buffers = 0;
  97
  98       for (unsigned i = 0; i < MESA_SHADER_TYPES; ++i) {
  99          struct gl_shader *sh = prog->_LinkedShaders[i];
 100          if (sh == NULL)
 101             continue;
 102
 103          foreach_list(node, sh->ir) {
 104             ir_variable *var = ((ir_instruction *)node)->as_variable();
 105
 106             if (var && var->type->contains_atomic()) {
 107                unsigned id;
 108                bool found = prog->UniformHash->get(id, var->name);
 109                assert(found);
 110                active_atomic_buffer *buf = &buffers[var->binding];
 111
 112                /* If this is the first time the buffer is used, increment
 113                 * the counter of buffers used.
 114                 */
 115                if (buf->size == 0)
 116                   (*num_buffers)++;
 117
 118                buf->push_back(id, var);
 119
 120                buf->stage_references[i]++;
 121                buf->size = MAX2(buf->size, var->atomic.offset +
 122                                 var->type->atomic_size());
 123             }
 124          }
 125       }
 126
 127       for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
 128          if (buffers[i].size == 0)
 129             continue;
 130
 131          qsort(buffers[i].counters, buffers[i].num_counters,
 132                sizeof(active_atomic_counter),
 133                cmp_actives);
 134
 135          for (unsigned j = 1; j < buffers[i].num_counters; j++) {
 136             /* If an overlapping counter found, it must be a reference to the
 137              * same counter from a different shader stage.
 138              */
 139             if (check_atomic_counters_overlap(buffers[i].counters[j-1].var,
 140                                               buffers[i].counters[j].var)
 141                 && strcmp(buffers[i].counters[j-1].var->name,
 142                           buffers[i].counters[j].var->name) != 0) {
 143                linker_error(prog, "Atomic counter %s declared at offset %d "
 144                             "which is already in use.",
 145                             buffers[i].counters[j].var->name,
 146                             buffers[i].counters[j].var->atomic.offset);
 147             }
 148          }
 149       }
 150       return buffers;
 151    }
 152 }
 153
 154 void
 155 link_assign_atomic_counter_resources(struct gl_context *ctx,
 156                                      struct gl_shader_program *prog)
 157 {
 158    unsigned num_buffers;
 159    active_atomic_buffer *abs =
 160       find_active_atomic_counters(ctx, prog, &num_buffers);
 161
 162    prog->AtomicBuffers = rzalloc_array(prog, gl_active_atomic_buffer,
 163                                        num_buffers);
 164    prog->NumAtomicBuffers = num_buffers;
 165
 166    unsigned i = 0;
 167    for (unsigned binding = 0;
 168         binding < ctx->Const.MaxAtomicBufferBindings;
 169         binding++) {
 170
 171       /* If the binding was not used, skip.
 172        */
 173       if (abs[binding].size == 0)
 174          continue;
 175
 176       active_atomic_buffer &ab = abs[binding];
 177       gl_active_atomic_buffer &mab = prog->AtomicBuffers[i];
 178
 179       /* Assign buffer-specific fields. */
 180       mab.Binding = binding;
 181       mab.MinimumSize = ab.size;
 182       mab.Uniforms = rzalloc_array(prog->AtomicBuffers, GLuint,
 183                                    ab.num_counters);
 184       mab.NumUniforms = ab.num_counters;
 185
 186       /* Assign counter-specific fields. */
 187       for (unsigned j = 0; j < ab.num_counters; j++) {
 188          ir_variable *const var = ab.counters[j].var;
 189          const unsigned id = ab.counters[j].id;
 190          gl_uniform_storage *const storage = &prog->UniformStorage[id];
 191
 192          mab.Uniforms[j] = id;
 193          var->atomic.buffer_index = i;
 194          storage->atomic_buffer_index = i;
 195          storage->offset = var->atomic.offset;
 196          storage->array_stride = (var->type->is_array() ?
 197                                   var->type->element_type()->atomic_size() : 0);
 198       }
 199
 200       /* Assign stage-specific fields. */
 201       for (unsigned j = 0; j < MESA_SHADER_TYPES; ++j)
 202          mab.StageReferences[j] =
 203             (ab.stage_references[j] ? GL_TRUE : GL_FALSE);
 204
 205       i++;
 206    }
 207
 208    delete [] abs;
 209    assert(i == num_buffers);
 210 }
 211
 212 void
 213 link_check_atomic_counter_resources(struct gl_context *ctx,
 214                                     struct gl_shader_program *prog)
 215 {
 216    STATIC_ASSERT(MESA_SHADER_TYPES == 3);
 217    static const char *shader_names[MESA_SHADER_TYPES] = {
 218       "vertex", "geometry", "fragment"
 219    };
 220    const unsigned max_atomic_counters[MESA_SHADER_TYPES] = {
 221       ctx->Const.VertexProgram.MaxAtomicCounters,
 222       ctx->Const.GeometryProgram.MaxAtomicCounters,
 223       ctx->Const.FragmentProgram.MaxAtomicCounters
 224    };
 225    const unsigned max_atomic_buffers[MESA_SHADER_TYPES] = {
 226       ctx->Const.VertexProgram.MaxAtomicBuffers,
 227       ctx->Const.GeometryProgram.MaxAtomicBuffers,
 228       ctx->Const.FragmentProgram.MaxAtomicBuffers
 229    };
 230    unsigned num_buffers;
 231    active_atomic_buffer *const abs =
 232       find_active_atomic_counters(ctx, prog, &num_buffers);
 233    unsigned atomic_counters[MESA_SHADER_TYPES] = {};
 234    unsigned atomic_buffers[MESA_SHADER_TYPES] = {};
 235    unsigned total_atomic_counters = 0;
 236    unsigned total_atomic_buffers = 0;
 237
 238    /* Sum the required resources.  Note that this counts buffers and
 239     * counters referenced by several shader stages multiple times
 240     * against the combined limit -- That's the behavior the spec
 241     * requires.
 242     */
 243    for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
 244       if (abs[i].size == 0)
 245          continue;
 246
 247       for (unsigned j = 0; j < MESA_SHADER_TYPES; ++j) {
 248          const unsigned n = abs[i].stage_references[j];
 249
 250          if (n) {
 251             atomic_counters[j] += n;
 252             total_atomic_counters += n;
 253             atomic_buffers[j]++;
 254             total_atomic_buffers++;
 255          }
 256       }
 257    }
 258
 259    /* Check that they are within the supported limits. */
 260    for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
 261       if (atomic_counters[i] > max_atomic_counters[i])
 262          linker_error(prog, "Too many %s shader atomic counters",
 263                       shader_names[i]);
 264
 265       if (atomic_buffers[i] > max_atomic_buffers[i])
 266          linker_error(prog, "Too many %s shader atomic counter buffers",
 267                       shader_names[i]);
 268    }
 269
 270    if (total_atomic_counters > ctx->Const.MaxCombinedAtomicCounters)
 271       linker_error(prog, "Too many combined atomic counters");
 272
 273    if (total_atomic_buffers > ctx->Const.MaxCombinedAtomicBuffers)
 274       linker_error(prog, "Too many combined atomic buffers");
 275
 276    delete [] abs;
 277 }