glsl: move to compiler/

Signed-off-by: Emil Velikov <[email protected]>
Acked-by: Matt Turner <[email protected]>
Acked-by: Jose Fonseca <[email protected]>

1 files changed, 524 insertions, 0 deletions

diff --git a/src/compiler/glsl/opt_constant_propagation.cpp b/src/compiler/glsl/opt_constant_propagation.cpp
new file mode 100644
index 00000000000..416ba16a3c5
--- /dev/null
+++ b/src/compiler/glsl/opt_constant_propagation.cpp
@@ -0,0 +1,524 @@
+/*
+ * Copyright © 2010 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * constant of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, constant, 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 constantright 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 CONSTANTRIGHT 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.
+ */
+
+/**
+ * \file opt_constant_propagation.cpp
+ *
+ * Tracks assignments of constants to channels of variables, and
+ * usage of those constant channels with direct usage of the constants.
+ *
+ * This can lead to constant folding and algebraic optimizations in
+ * those later expressions, while causing no increase in instruction
+ * count (due to constants being generally free to load from a
+ * constant push buffer or as instruction immediate values) and
+ * possibly reducing register pressure.
+ */
+
+#include "ir.h"
+#include "ir_visitor.h"
+#include "ir_rvalue_visitor.h"
+#include "ir_basic_block.h"
+#include "ir_optimization.h"
+#include "compiler/glsl_types.h"
+#include "util/hash_table.h"
+
+namespace {
+
+class acp_entry : public exec_node
+{
+public:
+   acp_entry(ir_variable *var, unsigned write_mask, ir_constant *constant)
+   {
+      assert(var);
+      assert(constant);
+      this->var = var;
+      this->write_mask = write_mask;
+      this->constant = constant;
+      this->initial_values = write_mask;
+   }
+
+   acp_entry(const acp_entry *src)
+   {
+      this->var = src->var;
+      this->write_mask = src->write_mask;
+      this->constant = src->constant;
+      this->initial_values = src->initial_values;
+   }
+
+   ir_variable *var;
+   ir_constant *constant;
+   unsigned write_mask;
+
+   /** Mask of values initially available in the constant. */
+   unsigned initial_values;
+};
+
+
+class kill_entry : public exec_node
+{
+public:
+   kill_entry(ir_variable *var, unsigned write_mask)
+   {
+      assert(var);
+      this->var = var;
+      this->write_mask = write_mask;
+   }
+
+   ir_variable *var;
+   unsigned write_mask;
+};
+
+class ir_constant_propagation_visitor : public ir_rvalue_visitor {
+public:
+   ir_constant_propagation_visitor()
+   {
+      progress = false;
+      killed_all = false;
+      mem_ctx = ralloc_context(0);
+      this->acp = new(mem_ctx) exec_list;
+      this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
+                                            _mesa_key_pointer_equal);
+   }
+   ~ir_constant_propagation_visitor()
+   {
+      ralloc_free(mem_ctx);
+   }
+
+   virtual ir_visitor_status visit_enter(class ir_loop *);
+   virtual ir_visitor_status visit_enter(class ir_function_signature *);
+   virtual ir_visitor_status visit_enter(class ir_function *);
+   virtual ir_visitor_status visit_leave(class ir_assignment *);
+   virtual ir_visitor_status visit_enter(class ir_call *);
+   virtual ir_visitor_status visit_enter(class ir_if *);
+
+   void add_constant(ir_assignment *ir);
+   void constant_folding(ir_rvalue **rvalue);
+   void constant_propagation(ir_rvalue **rvalue);
+   void kill(ir_variable *ir, unsigned write_mask);
+   void handle_if_block(exec_list *instructions);
+   void handle_rvalue(ir_rvalue **rvalue);
+
+   /** List of acp_entry: The available constants to propagate */
+   exec_list *acp;
+
+   /**
+    * List of kill_entry: The masks of variables whose values were
+    * killed in this block.
+    */
+   hash_table *kills;
+
+   bool progress;
+
+   bool killed_all;
+
+   void *mem_ctx;
+};
+
+
+void
+ir_constant_propagation_visitor::constant_folding(ir_rvalue **rvalue) {
+
+   if (*rvalue == NULL || (*rvalue)->ir_type == ir_type_constant)
+      return;
+
+   /* Note that we visit rvalues one leaving.  So if an expression has a
+    * non-constant operand, no need to go looking down it to find if it's
+    * constant.  This cuts the time of this pass down drastically.
+    */
+   ir_expression *expr = (*rvalue)->as_expression();
+   if (expr) {
+      for (unsigned int i = 0; i < expr->get_num_operands(); i++) {
+	 if (!expr->operands[i]->as_constant())
+	    return;
+      }
+   }
+
+   /* Ditto for swizzles. */
+   ir_swizzle *swiz = (*rvalue)->as_swizzle();
+   if (swiz && !swiz->val->as_constant())
+      return;
+
+   ir_constant *constant = (*rvalue)->constant_expression_value();
+   if (constant) {
+      *rvalue = constant;
+      this->progress = true;
+   }
+}
+
+void
+ir_constant_propagation_visitor::constant_propagation(ir_rvalue **rvalue) {
+
+   if (this->in_assignee || !*rvalue)
+      return;
+
+   const glsl_type *type = (*rvalue)->type;
+   if (!type->is_scalar() && !type->is_vector())
+      return;
+
+   ir_swizzle *swiz = NULL;
+   ir_dereference_variable *deref = (*rvalue)->as_dereference_variable();
+   if (!deref) {
+      swiz = (*rvalue)->as_swizzle();
+      if (!swiz)
+	 return;
+
+      deref = swiz->val->as_dereference_variable();
+      if (!deref)
+	 return;
+   }
+
+   ir_constant_data data;
+   memset(&data, 0, sizeof(data));
+
+   for (unsigned int i = 0; i < type->components(); i++) {
+      int channel;
+      acp_entry *found = NULL;
+
+      if (swiz) {
+	 switch (i) {
+	 case 0: channel = swiz->mask.x; break;
+	 case 1: channel = swiz->mask.y; break;
+	 case 2: channel = swiz->mask.z; break;
+	 case 3: channel = swiz->mask.w; break;
+	 default: assert(!"shouldn't be reached"); channel = 0; break;
+	 }
+      } else {
+	 channel = i;
+      }
+
+      foreach_in_list(acp_entry, entry, this->acp) {
+	 if (entry->var == deref->var && entry->write_mask & (1 << channel)) {
+	    found = entry;
+	    break;
+	 }
+      }
+
+      if (!found)
+	 return;
+
+      int rhs_channel = 0;
+      for (int j = 0; j < 4; j++) {
+	 if (j == channel)
+	    break;
+	 if (found->initial_values & (1 << j))
+	    rhs_channel++;
+      }
+
+      switch (type->base_type) {
+      case GLSL_TYPE_FLOAT:
+	 data.f[i] = found->constant->value.f[rhs_channel];
+	 break;
+      case GLSL_TYPE_DOUBLE:
+	 data.d[i] = found->constant->value.d[rhs_channel];
+	 break;
+      case GLSL_TYPE_INT:
+	 data.i[i] = found->constant->value.i[rhs_channel];
+	 break;
+      case GLSL_TYPE_UINT:
+	 data.u[i] = found->constant->value.u[rhs_channel];
+	 break;
+      case GLSL_TYPE_BOOL:
+	 data.b[i] = found->constant->value.b[rhs_channel];
+	 break;
+      default:
+	 assert(!"not reached");
+	 break;
+      }
+   }
+
+   *rvalue = new(ralloc_parent(deref)) ir_constant(type, &data);
+   this->progress = true;
+}
+
+void
+ir_constant_propagation_visitor::handle_rvalue(ir_rvalue **rvalue)
+{
+   constant_propagation(rvalue);
+   constant_folding(rvalue);
+}
+
+ir_visitor_status
+ir_constant_propagation_visitor::visit_enter(ir_function_signature *ir)
+{
+   /* Treat entry into a function signature as a completely separate
+    * block.  Any instructions at global scope will be shuffled into
+    * main() at link time, so they're irrelevant to us.
+    */
+   exec_list *orig_acp = this->acp;
+   hash_table *orig_kills = this->kills;
+   bool orig_killed_all = this->killed_all;
+
+   this->acp = new(mem_ctx) exec_list;
+   this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
+                                         _mesa_key_pointer_equal);
+   this->killed_all = false;
+
+   visit_list_elements(this, &ir->body);
+
+   this->kills = orig_kills;
+   this->acp = orig_acp;
+   this->killed_all = orig_killed_all;
+
+   return visit_continue_with_parent;
+}
+
+ir_visitor_status
+ir_constant_propagation_visitor::visit_leave(ir_assignment *ir)
+{
+  constant_folding(&ir->rhs);
+
+   if (this->in_assignee)
+      return visit_continue;
+
+   unsigned kill_mask = ir->write_mask;
+   if (ir->lhs->as_dereference_array()) {
+      /* The LHS of the assignment uses an array indexing operator (e.g. v[i]
+       * = ...;).  Since we only try to constant propagate vectors and
+       * scalars, this means that either (a) array indexing is being used to
+       * select a vector component, or (b) the variable in question is neither
+       * a scalar or a vector, so we don't care about it.  In the former case,
+       * we want to kill the whole vector, since in general we can't predict
+       * which vector component will be selected by array indexing.  In the
+       * latter case, it doesn't matter what we do, so go ahead and kill the
+       * whole variable anyway.
+       *
+       * Note that if the array index is constant (e.g. v[2] = ...;), we could
+       * in principle be smarter, but we don't need to, because a future
+       * optimization pass will convert it to a simple assignment with the
+       * correct mask.
+       */
+      kill_mask = ~0;
+   }
+   kill(ir->lhs->variable_referenced(), kill_mask);
+
+   add_constant(ir);
+
+   return visit_continue;
+}
+
+ir_visitor_status
+ir_constant_propagation_visitor::visit_enter(ir_function *ir)
+{
+   (void) ir;
+   return visit_continue;
+}
+
+ir_visitor_status
+ir_constant_propagation_visitor::visit_enter(ir_call *ir)
+{
+   /* Do constant propagation on call parameters, but skip any out params */
+   foreach_two_lists(formal_node, &ir->callee->parameters,
+                     actual_node, &ir->actual_parameters) {
+      ir_variable *sig_param = (ir_variable *) formal_node;
+      ir_rvalue *param = (ir_rvalue *) actual_node;
+      if (sig_param->data.mode != ir_var_function_out
+          && sig_param->data.mode != ir_var_function_inout) {
+	 ir_rvalue *new_param = param;
+	 handle_rvalue(&new_param);
+         if (new_param != param)
+	    param->replace_with(new_param);
+	 else
+	    param->accept(this);
+      }
+   }
+
+   /* Since we're unlinked, we don't (necssarily) know the side effects of
+    * this call.  So kill all copies.
+    */
+   acp->make_empty();
+   this->killed_all = true;
+
+   return visit_continue_with_parent;
+}
+
+void
+ir_constant_propagation_visitor::handle_if_block(exec_list *instructions)
+{
+   exec_list *orig_acp = this->acp;
+   hash_table *orig_kills = this->kills;
+   bool orig_killed_all = this->killed_all;
+
+   this->acp = new(mem_ctx) exec_list;
+   this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
+                                         _mesa_key_pointer_equal);
+   this->killed_all = false;
+
+   /* Populate the initial acp with a constant of the original */
+   foreach_in_list(acp_entry, a, orig_acp) {
+      this->acp->push_tail(new(this->mem_ctx) acp_entry(a));
+   }
+
+   visit_list_elements(this, instructions);
+
+   if (this->killed_all) {
+      orig_acp->make_empty();
+   }
+
+   hash_table *new_kills = this->kills;
+   this->kills = orig_kills;
+   this->acp = orig_acp;
+   this->killed_all = this->killed_all || orig_killed_all;
+
+   hash_entry *htk;
+   hash_table_foreach(new_kills, htk) {
+      kill_entry *k = (kill_entry *) htk->data;
+      kill(k->var, k->write_mask);
+   }
+}
+
+ir_visitor_status
+ir_constant_propagation_visitor::visit_enter(ir_if *ir)
+{
+   ir->condition->accept(this);
+   handle_rvalue(&ir->condition);
+
+   handle_if_block(&ir->then_instructions);
+   handle_if_block(&ir->else_instructions);
+
+   /* handle_if_block() already descended into the children. */
+   return visit_continue_with_parent;
+}
+
+ir_visitor_status
+ir_constant_propagation_visitor::visit_enter(ir_loop *ir)
+{
+   exec_list *orig_acp = this->acp;
+   hash_table *orig_kills = this->kills;
+   bool orig_killed_all = this->killed_all;
+
+   /* FINISHME: For now, the initial acp for loops is totally empty.
+    * We could go through once, then go through again with the acp
+    * cloned minus the killed entries after the first run through.
+    */
+   this->acp = new(mem_ctx) exec_list;
+   this->kills = _mesa_hash_table_create(mem_ctx, _mesa_hash_pointer,
+                                         _mesa_key_pointer_equal);
+   this->killed_all = false;
+
+   visit_list_elements(this, &ir->body_instructions);
+
+   if (this->killed_all) {
+      orig_acp->make_empty();
+   }
+
+   hash_table *new_kills = this->kills;
+   this->kills = orig_kills;
+   this->acp = orig_acp;
+   this->killed_all = this->killed_all || orig_killed_all;
+
+   hash_entry *htk;
+   hash_table_foreach(new_kills, htk) {
+      kill_entry *k = (kill_entry *) htk->data;
+      kill(k->var, k->write_mask);
+   }
+
+   /* already descended into the children. */
+   return visit_continue_with_parent;
+}
+
+void
+ir_constant_propagation_visitor::kill(ir_variable *var, unsigned write_mask)
+{
+   assert(var != NULL);
+
+   /* We don't track non-vectors. */
+   if (!var->type->is_vector() && !var->type->is_scalar())
+      return;
+
+   /* Remove any entries currently in the ACP for this kill. */
+   foreach_in_list_safe(acp_entry, entry, this->acp) {
+      if (entry->var == var) {
+	 entry->write_mask &= ~write_mask;
+	 if (entry->write_mask == 0)
+	    entry->remove();
+      }
+   }
+
+   /* Add this writemask of the variable to the list of killed
+    * variables in this block.
+    */
+   hash_entry *kill_hash_entry = _mesa_hash_table_search(this->kills, var);
+   if (kill_hash_entry) {
+      kill_entry *entry = (kill_entry *) kill_hash_entry->data;
+      entry->write_mask |= write_mask;
+      return;
+   }
+   /* Not already in the list.  Make new entry. */
+   _mesa_hash_table_insert(this->kills, var,
+                           new(this->mem_ctx) kill_entry(var, write_mask));
+}
+
+/**
+ * Adds an entry to the available constant list if it's a plain assignment
+ * of a variable to a variable.
+ */
+void
+ir_constant_propagation_visitor::add_constant(ir_assignment *ir)
+{
+   acp_entry *entry;
+
+   if (ir->condition)
+      return;
+
+   if (!ir->write_mask)
+      return;
+
+   ir_dereference_variable *deref = ir->lhs->as_dereference_variable();
+   ir_constant *constant = ir->rhs->as_constant();
+
+   if (!deref || !constant)
+      return;
+
+   /* Only do constant propagation on vectors.  Constant matrices,
+    * arrays, or structures would require more work elsewhere.
+    */
+   if (!deref->var->type->is_vector() && !deref->var->type->is_scalar())
+      return;
+
+   /* We can't do copy propagation on buffer variables, since the underlying
+    * memory storage is shared across multiple threads we can't be sure that
+    * the variable value isn't modified between this assignment and the next
+    * instruction where its value is read.
+    */
+   if (deref->var->data.mode == ir_var_shader_storage ||
+       deref->var->data.mode == ir_var_shader_shared)
+      return;
+
+   entry = new(this->mem_ctx) acp_entry(deref->var, ir->write_mask, constant);
+   this->acp->push_tail(entry);
+}
+
+} /* unnamed namespace */
+
+/**
+ * Does a constant propagation pass on the code present in the instruction stream.
+ */
+bool
+do_constant_propagation(exec_list *instructions)
+{
+   ir_constant_propagation_visitor v;
+
+   visit_list_elements(&v, instructions);
+
+   return v.progress;
+}