glsl: Linker support for ARB_shader_atomic_counters.

v2: Add comments on the purpose of the auxiliary data structures.
    Check for atomic counter overlaps.  Use the contains_atomic()
    convenience method.  Add static assert with the number of expected
    shader stages.
v3: Don't resize atomic arrays.
v4: Add comment on the reason why we don't resize atomic counter
    arrays.  Use 'strcmp(...) == 0' instead of '!strcmp(...)'.
v5 (idr):  Don't use STL in the linker.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>

src/glsl/Makefile.sources

@@ -47,6 +47,7 @@ LIBGLSL_FILES = \
 	$(GLSL_SRCDIR)/ir_validate.cpp \
 	$(GLSL_SRCDIR)/ir_variable_refcount.cpp \
 	$(GLSL_SRCDIR)/linker.cpp \
+	$(GLSL_SRCDIR)/link_atomics.cpp \
 	$(GLSL_SRCDIR)/link_functions.cpp \
 	$(GLSL_SRCDIR)/link_interface_blocks.cpp \
 	$(GLSL_SRCDIR)/link_uniforms.cpp \

src/glsl/link_atomics.cpp

@@ -0,0 +1,277 @@
+/*
+ * Copyright © 2013 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include "ir.h"
+#include "ir_uniform.h"
+#include "linker.h"
+#include "program/hash_table.h"
+#include "main/macros.h"
+
+namespace {
+   /*
+    * Atomic counter as seen by the program.
+    */
+   struct active_atomic_counter {
+      unsigned id;
+      ir_variable *var;
+   };
+
+   /*
+    * Atomic counter buffer referenced by the program.  There is a one
+    * to one correspondence between these and the objects that can be
+    * queried using glGetActiveAtomicCounterBufferiv().
+    */
+   struct active_atomic_buffer {
+      active_atomic_buffer()
+         : counters(0), num_counters(0), stage_references(), size(0)
+      {}
+
+      ~active_atomic_buffer()
+      {
+         free(counters);
+      }
+
+      void push_back(unsigned id, ir_variable *var)
+      {
+         counters = (active_atomic_counter *)
+            realloc(counters, sizeof(active_atomic_counter) * (num_counters + 1));
+
+         counters[num_counters].id = id;
+         counters[num_counters].var = var;
+         num_counters++;
+      }
+
+      active_atomic_counter *counters;
+      unsigned num_counters;
+      unsigned stage_references[MESA_SHADER_TYPES];
+      unsigned size;
+   };
+
+   int
+   cmp_actives(const void *a, const void *b)
+   {
+      const active_atomic_counter *const first = (active_atomic_counter *) a;
+      const active_atomic_counter *const second = (active_atomic_counter *) b;
+
+      return int(first->var->atomic.offset) - int(second->var->atomic.offset);
+   }
+
+   bool
+   check_atomic_counters_overlap(const ir_variable *x, const ir_variable *y)
+   {
+      return ((x->atomic.offset >= y->atomic.offset &&
+               x->atomic.offset < y->atomic.offset + y->type->atomic_size()) ||
+              (y->atomic.offset >= x->atomic.offset &&
+               y->atomic.offset < x->atomic.offset + x->type->atomic_size()));
+   }
+
+   active_atomic_buffer *
+   find_active_atomic_counters(struct gl_context *ctx,
+                               struct gl_shader_program *prog,
+                               unsigned *num_buffers)
+   {
+      active_atomic_buffer *const buffers =
+         new active_atomic_buffer[ctx->Const.MaxAtomicBufferBindings];
+
+      *num_buffers = 0;
+
+      for (unsigned i = 0; i < MESA_SHADER_TYPES; ++i) {
+         struct gl_shader *sh = prog->_LinkedShaders[i];
+         if (sh == NULL)
+            continue;
+
+         foreach_list(node, sh->ir) {
+            ir_variable *var = ((ir_instruction *)node)->as_variable();
+
+            if (var && var->type->contains_atomic()) {
+               unsigned id;
+               bool found = prog->UniformHash->get(id, var->name);
+               assert(found);
+               active_atomic_buffer *buf = &buffers[var->binding];
+
+               /* If this is the first time the buffer is used, increment
+                * the counter of buffers used.
+                */
+               if (buf->size == 0)
+                  (*num_buffers)++;
+
+               buf->push_back(id, var);
+
+               buf->stage_references[i]++;
+               buf->size = MAX2(buf->size, var->atomic.offset +
+                                var->type->atomic_size());
+            }
+         }
+      }
+
+      for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
+         if (buffers[i].size == 0)
+            continue;
+
+         qsort(buffers[i].counters, buffers[i].num_counters,
+               sizeof(active_atomic_counter),
+               cmp_actives);
+
+         for (unsigned j = 1; j < buffers[i].num_counters; j++) {
+            /* If an overlapping counter found, it must be a reference to the
+             * same counter from a different shader stage.
+             */
+            if (check_atomic_counters_overlap(buffers[i].counters[j-1].var,
+                                              buffers[i].counters[j].var)
+                && strcmp(buffers[i].counters[j-1].var->name,
+                          buffers[i].counters[j].var->name) != 0) {
+               linker_error(prog, "Atomic counter %s declared at offset %d "
+                            "which is already in use.",
+                            buffers[i].counters[j].var->name,
+                            buffers[i].counters[j].var->atomic.offset);
+            }
+         }
+      }
+      return buffers;
+   }
+}
+
+void
+link_assign_atomic_counter_resources(struct gl_context *ctx,
+                                     struct gl_shader_program *prog)
+{
+   unsigned num_buffers;
+   active_atomic_buffer *abs =
+      find_active_atomic_counters(ctx, prog, &num_buffers);
+
+   prog->AtomicBuffers = rzalloc_array(prog, gl_active_atomic_buffer,
+                                       num_buffers);
+   prog->NumAtomicBuffers = num_buffers;
+
+   unsigned i = 0;
+   for (unsigned binding = 0;
+        binding < ctx->Const.MaxAtomicBufferBindings;
+        binding++) {
+
+      /* If the binding was not used, skip.
+       */
+      if (abs[binding].size == 0)
+         continue;
+
+      active_atomic_buffer &ab = abs[binding];
+      gl_active_atomic_buffer &mab = prog->AtomicBuffers[i];
+
+      /* Assign buffer-specific fields. */
+      mab.Binding = binding;
+      mab.MinimumSize = ab.size;
+      mab.Uniforms = rzalloc_array(prog->AtomicBuffers, GLuint,
+                                   ab.num_counters);
+      mab.NumUniforms = ab.num_counters;
+
+      /* Assign counter-specific fields. */
+      for (unsigned j = 0; j < ab.num_counters; j++) {
+         ir_variable *const var = ab.counters[j].var;
+         const unsigned id = ab.counters[j].id;
+         gl_uniform_storage *const storage = &prog->UniformStorage[id];
+
+         mab.Uniforms[j] = id;
+         var->atomic.buffer_index = i;
+         storage->atomic_buffer_index = i;
+         storage->offset = var->atomic.offset;
+         storage->array_stride = (var->type->is_array() ?
+                                  var->type->element_type()->atomic_size() : 0);
+      }
+
+      /* Assign stage-specific fields. */
+      for (unsigned j = 0; j < MESA_SHADER_TYPES; ++j)
+         mab.StageReferences[j] =
+            (ab.stage_references[j] ? GL_TRUE : GL_FALSE);
+
+      i++;
+   }
+
+   delete [] abs;
+   assert(i == num_buffers);
+}
+
+void
+link_check_atomic_counter_resources(struct gl_context *ctx,
+                                    struct gl_shader_program *prog)
+{
+   STATIC_ASSERT(MESA_SHADER_TYPES == 3);
+   static const char *shader_names[MESA_SHADER_TYPES] = {
+      "vertex", "geometry", "fragment"
+   };
+   const unsigned max_atomic_counters[MESA_SHADER_TYPES] = {
+      ctx->Const.VertexProgram.MaxAtomicCounters,
+      ctx->Const.GeometryProgram.MaxAtomicCounters,
+      ctx->Const.FragmentProgram.MaxAtomicCounters
+   };
+   const unsigned max_atomic_buffers[MESA_SHADER_TYPES] = {
+      ctx->Const.VertexProgram.MaxAtomicBuffers,
+      ctx->Const.GeometryProgram.MaxAtomicBuffers,
+      ctx->Const.FragmentProgram.MaxAtomicBuffers
+   };
+   unsigned num_buffers;
+   active_atomic_buffer *const abs =
+      find_active_atomic_counters(ctx, prog, &num_buffers);
+   unsigned atomic_counters[MESA_SHADER_TYPES] = {};
+   unsigned atomic_buffers[MESA_SHADER_TYPES] = {};
+   unsigned total_atomic_counters = 0;
+   unsigned total_atomic_buffers = 0;
+
+   /* Sum the required resources.  Note that this counts buffers and
+    * counters referenced by several shader stages multiple times
+    * against the combined limit -- That's the behavior the spec
+    * requires.
+    */
+   for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
+      if (abs[i].size == 0)
+         continue;
+
+      for (unsigned j = 0; j < MESA_SHADER_TYPES; ++j) {
+         const unsigned n = abs[i].stage_references[j];
+
+         if (n) {
+            atomic_counters[j] += n;
+            total_atomic_counters += n;
+            atomic_buffers[j]++;
+            total_atomic_buffers++;
+         }
+      }
+   }
+
+   /* Check that they are within the supported limits. */
+   for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
+      if (atomic_counters[i] > max_atomic_counters[i])
+         linker_error(prog, "Too many %s shader atomic counters",
+                      shader_names[i]);
+
+      if (atomic_buffers[i] > max_atomic_buffers[i])
+         linker_error(prog, "Too many %s shader atomic counter buffers",
+                      shader_names[i]);
+   }
+
+   if (total_atomic_counters > ctx->Const.MaxCombinedAtomicCounters)
+      linker_error(prog, "Too many combined atomic counters");
+
+   if (total_atomic_buffers > ctx->Const.MaxCombinedAtomicBuffers)
+      linker_error(prog, "Too many combined atomic buffers");
+
+   delete [] abs;
+}

src/glsl/linker.cpp

@@ -651,6 +651,14 @@ cross_validate_globals(struct gl_shader_program *prog,
                existing->explicit_binding = true;
             }
 
+            if (var->type->contains_atomic() &&
+                var->atomic.offset != existing->atomic.offset) {
+               linker_error(prog, "offset specifications for %s "
+                            "`%s' have differing values\n",
+                            mode_string(var), var->name);
+               return;
+            }
+
 	    /* Validate layout qualifiers for gl_FragDepth.
 	     *
 	     * From the AMD/ARB_conservative_depth specs:
@@ -1485,8 +1493,12 @@ update_array_sizes(struct gl_shader_program *prog)
 	 /* GL_ARB_uniform_buffer_object says that std140 uniforms
 	  * will not be eliminated.  Since we always do std140, just
 	  * don't resize arrays in UBOs.
+          *
+          * Atomic counters are supposed to get deterministic
+          * locations assigned based on the declaration ordering and
+          * sizes, array compaction would mess that up.
 	  */
-	 if (var->is_in_uniform_block())
+	 if (var->is_in_uniform_block() || var->type->contains_atomic())
 	    continue;
 
 	 unsigned int size = var->max_array_access;
@@ -1991,6 +2003,10 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
       prog->UniformBlockStageIndex[i] = NULL;
    }
 
+   ralloc_free(prog->AtomicBuffers);
+   prog->AtomicBuffers = NULL;
+   prog->NumAtomicBuffers = 0;
+
    /* Separate the shaders into groups based on their type.
     */
    struct gl_shader **vert_shader_list;
@@ -2342,9 +2358,12 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
 
    update_array_sizes(prog);
    link_assign_uniform_locations(prog);
+   link_assign_atomic_counter_resources(ctx, prog);
    store_fragdepth_layout(prog);
 
    check_resources(ctx, prog);
+   link_check_atomic_counter_resources(ctx, prog);
+
    if (!prog->LinkStatus)
       goto done;
 

src/glsl/linker.h

@@ -69,6 +69,14 @@ validate_interstage_interface_blocks(struct gl_shader_program *prog,
                                      const gl_shader *producer,
                                      const gl_shader *consumer);
 
+extern void
+link_assign_atomic_counter_resources(struct gl_context *ctx,
+                                     struct gl_shader_program *prog);
+
+extern void
+link_check_atomic_counter_resources(struct gl_context *ctx,
+                                    struct gl_shader_program *prog);
+
 /**
  * Class for processing all of the leaf fields of a variable that corresponds
  * to a program resource.