[Mesa-dev] [PATCH] glsl: Add an optimization pass to flatten simple nested if blocks.

[Eric Anholt]

Kenneth Graunke <kenneth at whitecape.org> writes:
> diff --git a/src/glsl/Makefile.sources b/src/glsl/Makefile.sources
> index c294aa4..b5282a6 100644
> --- a/src/glsl/Makefile.sources
> +++ b/src/glsl/Makefile.sources
> @@ -80,6 +80,7 @@ LIBGLSL_FILES = \
>  	$(GLSL_SRCDIR)/opt_dead_code.cpp \
>  	$(GLSL_SRCDIR)/opt_dead_code_local.cpp \
>  	$(GLSL_SRCDIR)/opt_dead_functions.cpp \
> +	$(GLSL_SRCDIR)/opt_flatten_nested_if_blocks.cpp \
>  	$(GLSL_SRCDIR)/opt_function_inlining.cpp \
>  	$(GLSL_SRCDIR)/opt_if_simplification.cpp \
>  	$(GLSL_SRCDIR)/opt_noop_swizzle.cpp \
> diff --git a/src/glsl/glsl_parser_extras.cpp b/src/glsl/glsl_parser_extras.cpp
> index 9740903..0992294 100644
> --- a/src/glsl/glsl_parser_extras.cpp
> +++ b/src/glsl/glsl_parser_extras.cpp
> @@ -1218,6 +1218,7 @@ do_common_optimization(exec_list *ir, bool linked,
>        progress = do_structure_splitting(ir) || progress;
>     }
>     progress = do_if_simplification(ir) || progress;
> +   progress = opt_flatten_nested_if_blocks(ir) || progress;
>     progress = do_copy_propagation(ir) || progress;
>     progress = do_copy_propagation_elements(ir) || progress;
>     if (linked)
> diff --git a/src/glsl/ir_optimization.h b/src/glsl/ir_optimization.h
> index 2454bbe..a8885d7 100644
> --- a/src/glsl/ir_optimization.h
> +++ b/src/glsl/ir_optimization.h
> @@ -82,6 +82,7 @@ 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);
>  bool do_if_simplification(exec_list *instructions);
> +bool opt_flatten_nested_if_blocks(exec_list *instructions);
>  bool do_discard_simplification(exec_list *instructions);
>  bool lower_if_to_cond_assign(exec_list *instructions, unsigned max_depth = 0);
>  bool do_mat_op_to_vec(exec_list *instructions);
> diff --git a/src/glsl/opt_flatten_nested_if_blocks.cpp b/src/glsl/opt_flatten_nested_if_blocks.cpp
> new file mode 100644
> index 0000000..c702102
> --- /dev/null
> +++ b/src/glsl/opt_flatten_nested_if_blocks.cpp
> @@ -0,0 +1,103 @@
> +/*
> + * 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_flatten_nested_if_blocks.cpp
> + *
> + * Flattens nested if blocks such as:
> + *
> + * if (x) {
> + *    if (y) {
> + *       ...
> + *    }
> + * }
> + *
> + * into a single if block with a combined condition:
> + *
> + * if (x && y) {
> + *    ...
> + * }
> + */
> +
> +#include "ir.h"
> +#include "ir_builder.h"
> +
> +using namespace ir_builder;
> +
> +namespace {
> +
> +class nested_if_flattener : public ir_hierarchical_visitor {
> +public:
> +   nested_if_flattener()
> +   {
> +      progress = false;
> +   }
> +
> +   ir_visitor_status visit_leave(ir_if *);
> +   ir_visitor_status visit_enter(ir_assignment *);
> +
> +   bool progress;
> +};
> +
> +} /* unnamed namespace */
> +
> +/* We only care about the top level "if" instructions, so don't
> + * descend into expressions.
> + */
> +ir_visitor_status
> +nested_if_flattener::visit_enter(ir_assignment *ir)
> +{
> +   (void) ir;
> +   return visit_continue_with_parent;
> +}
> +
> +bool
> +opt_flatten_nested_if_blocks(exec_list *instructions)
> +{
> +   nested_if_flattener v;
> +
> +   v.run(instructions);
> +   return v.progress;
> +}
> +
> +
> +ir_visitor_status
> +nested_if_flattener::visit_leave(ir_if *ir)
> +{
> +   /* Only handle a single ir_if within the then clause of an ir_if.  No extra
> +    * instructions, no else clauses, nothing.
> +    */
> +   if (ir->then_instructions.is_empty() || !ir->else_instructions.is_empty())
> +      return visit_continue;
> +
> +   ir_if *inner = ((ir_instruction *) ir->then_instructions.head)->as_if();
> +   if (!inner || !inner->next->is_tail_sentinel() ||
> +       !inner->else_instructions.is_empty())
> +      return visit_continue;
> +
> +   ir->condition = logic_and(ir->condition, inner->condition);
> +   inner->then_instructions.move_nodes_to(&ir->then_instructions);
> +

Also inner->remove() right now instead of leaving it up to the
empty-if-statement-remover?  Equivalent functionality, but may take one
less turn through the giant loop.

Other than that,
Reviewed-by: Eric Anholt <eric at anholt.net>

I'm glad to see this failry simple optimization made possible by us
having removed things-with-side-effects from expression trees.  We may
some day find a need to not do this optimization if the inner expression
tree is very large, but for now it looks like it's an obvious win on the
shaders we have on Linux.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20130404/e89f79f2/attachment.pgp>