[Mesa-dev] [PATCH 1/4] glsl: Merge opt_noop_swizzle, and opt_swizzle_swizzle

Thomas Helland thomashelland90 at gmail.com
Thu Apr 6 19:49:05 UTC 2017


The pass to optimize two swizzles in a row can be interpreted as
a noop swizzle optimization. Merge the two passes, and avoid having
the extra pass that runs a visitor pattern.
---
 src/compiler/Makefile.sources             |  1 -
 src/compiler/glsl/glsl_parser_extras.cpp  |  1 -
 src/compiler/glsl/ir_optimization.h       |  1 -
 src/compiler/glsl/opt_noop_swizzle.cpp    | 43 +++++++++++++-
 src/compiler/glsl/opt_swizzle_swizzle.cpp | 97 -------------------------------
 5 files changed, 42 insertions(+), 101 deletions(-)
 delete mode 100644 src/compiler/glsl/opt_swizzle_swizzle.cpp

diff --git a/src/compiler/Makefile.sources b/src/compiler/Makefile.sources
index 2455d4eb5a..8b4245612b 100644
--- a/src/compiler/Makefile.sources
+++ b/src/compiler/Makefile.sources
@@ -134,7 +134,6 @@ LIBGLSL_FILES = \
 	glsl/opt_rebalance_tree.cpp \
 	glsl/opt_redundant_jumps.cpp \
 	glsl/opt_structure_splitting.cpp \
-	glsl/opt_swizzle_swizzle.cpp \
 	glsl/opt_tree_grafting.cpp \
 	glsl/opt_vectorize.cpp \
 	glsl/program.h \
diff --git a/src/compiler/glsl/glsl_parser_extras.cpp b/src/compiler/glsl/glsl_parser_extras.cpp
index 48cbc01ba7..53fcb0e45a 100644
--- a/src/compiler/glsl/glsl_parser_extras.cpp
+++ b/src/compiler/glsl/glsl_parser_extras.cpp
@@ -2170,7 +2170,6 @@ do_common_optimization(exec_list *ir, bool linked,
        options->EmitNoCont, options->EmitNoLoops);
    OPT(do_vec_index_to_swizzle, ir);
    OPT(lower_vector_insert, ir, false);
-   OPT(do_swizzle_swizzle, ir);
    OPT(do_noop_swizzle, ir);
 
    OPT(optimize_split_arrays, ir, linked);
diff --git a/src/compiler/glsl/ir_optimization.h b/src/compiler/glsl/ir_optimization.h
index 67a7514c7d..65d3b3e0db 100644
--- a/src/compiler/glsl/ir_optimization.h
+++ b/src/compiler/glsl/ir_optimization.h
@@ -122,7 +122,6 @@ bool do_mat_op_to_vec(exec_list *instructions);
 bool do_minmax_prune(exec_list *instructions);
 bool do_noop_swizzle(exec_list *instructions);
 bool do_structure_splitting(exec_list *instructions);
-bool do_swizzle_swizzle(exec_list *instructions);
 bool do_vectorize(exec_list *instructions);
 bool do_tree_grafting(exec_list *instructions);
 bool do_vec_index_to_cond_assign(exec_list *instructions);
diff --git a/src/compiler/glsl/opt_noop_swizzle.cpp b/src/compiler/glsl/opt_noop_swizzle.cpp
index 41890ab2b1..082e0195fe 100644
--- a/src/compiler/glsl/opt_noop_swizzle.cpp
+++ b/src/compiler/glsl/opt_noop_swizzle.cpp
@@ -49,6 +49,39 @@ public:
 
 } /* unnamed namespace */
 
+static bool
+opt_swizzle_swizzle(ir_swizzle *ir)
+{
+   int mask2[4];
+
+   ir_swizzle *swiz2 = ir->val->as_swizzle();
+   if (!swiz2)
+      return visit_continue;
+
+   memset(&mask2, 0, sizeof(mask2));
+   if (swiz2->mask.num_components >= 1)
+      mask2[0] = swiz2->mask.x;
+   if (swiz2->mask.num_components >= 2)
+      mask2[1] = swiz2->mask.y;
+   if (swiz2->mask.num_components >= 3)
+      mask2[2] = swiz2->mask.z;
+   if (swiz2->mask.num_components >= 4)
+      mask2[3] = swiz2->mask.w;
+
+   if (ir->mask.num_components >= 1)
+      ir->mask.x = mask2[ir->mask.x];
+   if (ir->mask.num_components >= 2)
+      ir->mask.y = mask2[ir->mask.y];
+   if (ir->mask.num_components >= 3)
+      ir->mask.z = mask2[ir->mask.z];
+   if (ir->mask.num_components >= 4)
+      ir->mask.w = mask2[ir->mask.w];
+
+   ir->val = swiz2->val;
+
+   return true;
+}
+
 void
 ir_noop_swizzle_visitor::handle_rvalue(ir_rvalue **rvalue)
 {
@@ -56,7 +89,15 @@ ir_noop_swizzle_visitor::handle_rvalue(ir_rvalue **rvalue)
       return;
 
    ir_swizzle *swiz = (*rvalue)->as_swizzle();
-   if (!swiz || swiz->type != swiz->val->type)
+
+   if (!swiz)
+      return;
+
+   /* Check first if we can optimize two a sequence of two swizzles */
+   if (opt_swizzle_swizzle(swiz))
+      this->progress = true;
+
+   if (swiz->type != swiz->val->type)
       return;
 
    int elems = swiz->val->type->vector_elements;
diff --git a/src/compiler/glsl/opt_swizzle_swizzle.cpp b/src/compiler/glsl/opt_swizzle_swizzle.cpp
deleted file mode 100644
index 7285474b08..0000000000
--- a/src/compiler/glsl/opt_swizzle_swizzle.cpp
+++ /dev/null
@@ -1,97 +0,0 @@
-/*
- * Copyright © 2010 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_swizzle_swizzle.cpp
- *
- * Eliminates the second swizzle in a swizzle chain.
- */
-
-#include "ir.h"
-#include "ir_visitor.h"
-#include "ir_optimization.h"
-#include "compiler/glsl_types.h"
-
-namespace {
-
-class ir_swizzle_swizzle_visitor : public ir_hierarchical_visitor {
-public:
-   ir_swizzle_swizzle_visitor()
-   {
-      progress = false;
-   }
-
-   virtual ir_visitor_status visit_enter(ir_swizzle *);
-
-   bool progress;
-};
-
-} /* unnamed namespace */
-
-ir_visitor_status
-ir_swizzle_swizzle_visitor::visit_enter(ir_swizzle *ir)
-{
-   int mask2[4];
-
-   ir_swizzle *swiz2 = ir->val->as_swizzle();
-   if (!swiz2)
-      return visit_continue;
-
-   memset(&mask2, 0, sizeof(mask2));
-   if (swiz2->mask.num_components >= 1)
-      mask2[0] = swiz2->mask.x;
-   if (swiz2->mask.num_components >= 2)
-      mask2[1] = swiz2->mask.y;
-   if (swiz2->mask.num_components >= 3)
-      mask2[2] = swiz2->mask.z;
-   if (swiz2->mask.num_components >= 4)
-      mask2[3] = swiz2->mask.w;
-
-   if (ir->mask.num_components >= 1)
-      ir->mask.x = mask2[ir->mask.x];
-   if (ir->mask.num_components >= 2)
-      ir->mask.y = mask2[ir->mask.y];
-   if (ir->mask.num_components >= 3)
-      ir->mask.z = mask2[ir->mask.z];
-   if (ir->mask.num_components >= 4)
-      ir->mask.w = mask2[ir->mask.w];
-
-   ir->val = swiz2->val;
-
-   this->progress = true;
-
-   return visit_continue;
-}
-
-/**
- * Does a copy propagation pass on the code present in the instruction stream.
- */
-bool
-do_swizzle_swizzle(exec_list *instructions)
-{
-   ir_swizzle_swizzle_visitor v;
-
-   v.run(instructions);
-
-   return v.progress;
-}
-- 
2.11.1



More information about the mesa-dev mailing list