glsl: Add a pass to flip matrix/vector multiplies to use dot products.

This pass flips (matrix * vector) operations to (vector *
matrixTranspose) for certain built-in matrices (currently
gl_ModelViewProjectionMatrix and gl_TextureMatrix).

This is equivalent, but results in dot products rather than multiplies
and adds.  On some hardware, this is more efficient.

This pass is conditionalized on ctx->mvp_with_dp4, the flag drivers set
to indicate they prefer dot products.

Improves performance in Lightsmark by 1.01131% +/- 0.162069% (n = 10)
on a Haswell GT2 system.  Passes Piglit on Ivybridge.

v2: Use struct gl_shader_compiler_options instead of plumbing through
    another boolean flag for this purpose.

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

src/glsl/Makefile.sources

@@ -82,6 +82,7 @@ LIBGLSL_FILES = \
 	$(GLSL_SRCDIR)/opt_dead_code_local.cpp \
 	$(GLSL_SRCDIR)/opt_dead_functions.cpp \
 	$(GLSL_SRCDIR)/opt_flatten_nested_if_blocks.cpp \
+	$(GLSL_SRCDIR)/opt_flip_matrices.cpp \
 	$(GLSL_SRCDIR)/opt_function_inlining.cpp \
 	$(GLSL_SRCDIR)/opt_if_simplification.cpp \
 	$(GLSL_SRCDIR)/opt_noop_swizzle.cpp \

src/glsl/glsl_parser_extras.cpp

@@ -1224,6 +1224,10 @@ do_common_optimization(exec_list *ir, bool linked,
    progress = opt_flatten_nested_if_blocks(ir) || progress;
    progress = do_copy_propagation(ir) || progress;
    progress = do_copy_propagation_elements(ir) || progress;
+
+   if (options->PreferDP4 && !linked)
+      progress = opt_flip_matrices(ir) || progress;
+
    if (linked)
       progress = do_dead_code(ir, uniform_locations_assigned) || progress;
    else

src/glsl/ir_optimization.h

@@ -80,6 +80,7 @@ bool do_dead_code(exec_list *instructions, bool uniform_locations_assigned);
 bool do_dead_code_local(exec_list *instructions);
 bool do_dead_code_unlinked(exec_list *instructions);
 bool do_dead_functions(exec_list *instructions);
+bool opt_flip_matrices(exec_list *instructions);
 bool do_function_inlining(exec_list *instructions);
 bool do_lower_jumps(exec_list *instructions, bool pull_out_jumps = true, bool lower_sub_return = true, bool lower_main_return = false, bool lower_continue = false, bool lower_break = false);
 bool do_lower_texture_projection(exec_list *instructions);

src/glsl/opt_flip_matrices.cpp

@@ -0,0 +1,122 @@
+/*
+ * 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.
+ */
+
+/**
+ * \file opt_flip_matrices.cpp
+ *
+ * Convert (matrix * vector) operations to (vector * matrixTranspose),
+ * which can be done using dot products rather than multiplies and adds.
+ * On some hardware, this is more efficient.
+ *
+ * This currently only does the conversion for built-in matrices which
+ * already have transposed equivalents.  Namely, gl_ModelViewProjectionMatrix
+ * and gl_TextureMatrix.
+ */
+#include "ir.h"
+#include "ir_optimization.h"
+#include "main/macros.h"
+
+namespace {
+class matrix_flipper : public ir_hierarchical_visitor {
+public:
+   matrix_flipper(exec_list *instructions)
+   {
+      progress = false;
+      mvp_transpose = NULL;
+      texmat_transpose = NULL;
+
+      foreach_list(n, instructions) {
+         ir_instruction *ir = (ir_instruction *) n;
+         ir_variable *var = ir->as_variable();
+         if (!var)
+            continue;
+         if (strcmp(var->name, "gl_ModelViewProjectionMatrixTranspose") == 0)
+            mvp_transpose = var;
+         if (strcmp(var->name, "gl_TextureMatrixTranspose") == 0)
+            texmat_transpose = var;
+      }
+   }
+
+   ir_visitor_status visit_enter(ir_expression *ir);
+
+   bool progress;
+
+private:
+   ir_variable *mvp_transpose;
+   ir_variable *texmat_transpose;
+};
+}
+
+ir_visitor_status
+matrix_flipper::visit_enter(ir_expression *ir)
+{
+   if (ir->operation != ir_binop_mul ||
+       !ir->operands[0]->type->is_matrix() ||
+       !ir->operands[1]->type->is_vector())
+      return visit_continue;
+
+   ir_variable *mat_var = ir->operands[0]->variable_referenced();
+   if (!mat_var)
+      return visit_continue;
+
+   if (mvp_transpose &&
+       strcmp(mat_var->name, "gl_ModelViewProjectionMatrix") == 0) {
+      ir_dereference_variable *deref = ir->operands[0]->as_dereference_variable();
+      assert(deref && deref->var == mat_var);
+
+      void *mem_ctx = ralloc_parent(ir);
+
+      ir->operands[0] = ir->operands[1];
+      ir->operands[1] = new(mem_ctx) ir_dereference_variable(mvp_transpose);
+
+      progress = true;
+   } else if (texmat_transpose &&
+              strcmp(mat_var->name, "gl_TextureMatrix") == 0) {
+      ir_dereference_array *array_ref = ir->operands[0]->as_dereference_array();
+      assert(array_ref != NULL);
+      ir_dereference_variable *var_ref = array_ref->array->as_dereference_variable();
+      assert(var_ref && var_ref->var == mat_var);
+
+      ir->operands[0] = ir->operands[1];
+      ir->operands[1] = array_ref;
+
+      var_ref->var = texmat_transpose;
+
+      texmat_transpose->max_array_access =
+         MAX2(texmat_transpose->max_array_access, mat_var->max_array_access);
+
+      progress = true;
+   }
+
+   return visit_continue;
+}
+
+bool
+opt_flip_matrices(struct exec_list *instructions)
+{
+   matrix_flipper v(instructions);
+
+   visit_list_elements(&v, instructions);
+
+   return v.progress;
+}