glsl: Add lowering pass for ir_triop_vector_insert

This will eventually replace do_vec_index_to_cond_assign.  This lowering
pass is called in all the places where do_vec_index_to_cond_assign or
do_vec_index_to_swizzle is called.

v2: Use WRITEMASK_* instead of integer literals.  Use a more concise
method of generating broadcast_index.  Both suggested by Eric.

v3: Use a series of scalar compares instead of a single vector compare.
Suggested by Eric and Ken.  It still uses 'if (cond) v.x = y;' instead
of conditional assignments because ir_builder doesn't do conditional
assignments, and I'd rather keep the code simple.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

src/glsl/Makefile.sources

@@ -69,6 +69,7 @@ LIBGLSL_FILES = \
 	$(GLSL_SRCDIR)/lower_vec_index_to_cond_assign.cpp \
 	$(GLSL_SRCDIR)/lower_vec_index_to_swizzle.cpp \
 	$(GLSL_SRCDIR)/lower_vector.cpp \
+	$(GLSL_SRCDIR)/lower_vector_insert.cpp \
 	$(GLSL_SRCDIR)/lower_output_reads.cpp \
 	$(GLSL_SRCDIR)/lower_ubo_reference.cpp \
 	$(GLSL_SRCDIR)/opt_algebraic.cpp \

src/glsl/glsl_parser_extras.cpp

@@ -1243,6 +1243,7 @@ do_common_optimization(exec_list *ir, bool linked,
    progress = do_algebraic(ir) || progress;
    progress = do_lower_jumps(ir) || progress;
    progress = do_vec_index_to_swizzle(ir) || progress;
+   progress = lower_vector_insert(ir, false) || progress;
    progress = do_swizzle_swizzle(ir) || progress;
    progress = do_noop_swizzle(ir) || progress;
 

src/glsl/ir_optimization.h

@@ -109,6 +109,7 @@ void lower_ubo_reference(struct gl_shader *shader, exec_list *instructions);
 void lower_packed_varyings(void *mem_ctx, unsigned location_base,
                            unsigned locations_used, ir_variable_mode mode,
                            gl_shader *shader);
+bool lower_vector_insert(exec_list *instructions, bool lower_nonconstant_index);
 bool optimize_redundant_jumps(exec_list *instructions);
 bool optimize_split_arrays(exec_list *instructions, bool linked);
 

src/glsl/lower_vector_insert.cpp

@@ -0,0 +1,139 @@
+/*
+ * 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_builder.h"
+#include "ir_rvalue_visitor.h"
+#include "ir_optimization.h"
+
+using namespace ir_builder;
+
+class vector_insert_visitor : public ir_rvalue_visitor {
+public:
+   vector_insert_visitor(bool lower_nonconstant_index)
+      : progress(false), lower_nonconstant_index(lower_nonconstant_index)
+   {
+      factory.instructions = &factory_instructions;
+   }
+
+   virtual ~vector_insert_visitor()
+   {
+      assert(factory_instructions.is_empty());
+   }
+
+   virtual void handle_rvalue(ir_rvalue **rv);
+
+   ir_factory factory;
+   exec_list factory_instructions;
+   bool progress;
+   bool lower_nonconstant_index;
+};
+
+
+void
+vector_insert_visitor::handle_rvalue(ir_rvalue **rv)
+{
+   if (*rv == NULL || (*rv)->ir_type != ir_type_expression)
+      return;
+
+   ir_expression *const expr = (ir_expression *) *rv;
+
+   if (likely(expr->operation != ir_triop_vector_insert))
+      return;
+
+   factory.mem_ctx = ralloc_parent(expr);
+
+   ir_constant *const idx = expr->operands[2]->constant_expression_value();
+   if (idx != NULL) {
+      /* Replace (vector_insert (vec) (scalar) (index)) with a dereference of
+       * a new temporary.  The new temporary gets assigned as
+       *
+       *     t = vec
+       *     t.mask = scalar
+       *
+       * where mask is the component selected by index.
+       */
+      ir_variable *const temp =
+         factory.make_temp(expr->operands[0]->type, "vec_tmp");
+
+      const int mask = 1 << idx->value.i[0];
+
+      factory.emit(assign(temp, expr->operands[0]));
+      factory.emit(assign(temp, expr->operands[1], mask));
+
+      this->progress = true;
+      *rv = new(factory.mem_ctx) ir_dereference_variable(temp);
+   } else if (this->lower_nonconstant_index) {
+      /* Replace (vector_insert (vec) (scalar) (index)) with a dereference of
+       * a new temporary.  The new temporary gets assigned as
+       *
+       *     t = vec
+       *     if (index == 0)
+       *         t.x = scalar
+       *     if (index == 1)
+       *         t.y = scalar
+       *     if (index == 2)
+       *         t.z = scalar
+       *     if (index == 3)
+       *         t.w = scalar
+       */
+      ir_variable *const temp =
+         factory.make_temp(expr->operands[0]->type, "vec_tmp");
+
+      ir_variable *const src_temp =
+         factory.make_temp(expr->operands[1]->type, "src_temp");
+
+      factory.emit(assign(temp, expr->operands[0]));
+      factory.emit(assign(src_temp, expr->operands[1]));
+
+      for (unsigned i = 0; i < expr->type->vector_elements; i++) {
+         ir_constant *const cmp_index =
+            new(factory.mem_ctx) ir_constant(int(i));
+
+         ir_variable *const cmp_result =
+            factory.make_temp(glsl_type::bool_type, "index_condition");
+
+         factory.emit(assign(cmp_result,
+                             equal(expr->operands[2]->clone(factory.mem_ctx,
+                                                            NULL),
+                                   cmp_index)));
+
+         factory.emit(if_tree(cmp_result,
+                              assign(temp, src_temp, WRITEMASK_X << i)));
+      }
+
+      this->progress = true;
+      *rv = new(factory.mem_ctx) ir_dereference_variable(temp);
+   }
+
+   base_ir->insert_before(factory.instructions);
+}
+
+bool
+lower_vector_insert(exec_list *instructions, bool lower_nonconstant_index)
+{
+   vector_insert_visitor v(lower_nonconstant_index);
+
+   visit_list_elements(&v, instructions);
+
+   return v.progress;
+}

src/mesa/drivers/dri/i965/brw_shader.cpp

@@ -174,6 +174,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
       do_lower_texture_projection(shader->ir);
       brw_lower_texture_gradients(intel, shader->ir);
       do_vec_index_to_cond_assign(shader->ir);
+      lower_vector_insert(shader->ir, true);
       brw_do_cubemap_normalize(shader->ir);
       lower_noise(shader->ir);
       lower_quadop_vector(shader->ir, false);

src/mesa/program/ir_to_mesa.cpp

@@ -3055,6 +3055,7 @@ _mesa_ir_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
 	     || progress;
 
 	 progress = do_vec_index_to_cond_assign(ir) || progress;
+         progress = lower_vector_insert(ir, true) || progress;
       } while (progress);
 
       validate_ir_tree(ir);

src/mesa/state_tracker/st_glsl_to_tgsi.cpp

@@ -5243,6 +5243,7 @@ st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
 
       lower_ubo_reference(prog->_LinkedShaders[i], ir);
       do_vec_index_to_cond_assign(ir);
+      lower_vector_insert(ir, true);
       lower_quadop_vector(ir, false);
       lower_noise(ir);
       if (options->MaxIfDepth == 0) {