diff options
Diffstat (limited to 'source/blender/functions/FN_multi_function_procedure_optimization.hh')
-rw-r--r-- | source/blender/functions/FN_multi_function_procedure_optimization.hh | 61 |
1 files changed, 61 insertions, 0 deletions
diff --git a/source/blender/functions/FN_multi_function_procedure_optimization.hh b/source/blender/functions/FN_multi_function_procedure_optimization.hh new file mode 100644 index 00000000000..e5ffc12b241 --- /dev/null +++ b/source/blender/functions/FN_multi_function_procedure_optimization.hh @@ -0,0 +1,61 @@ +/* + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version 2 + * of the License, or (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software Foundation, + * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. + */ + +#pragma once + +/** \file + * \ingroup fn + * + * A #MFProcedure optimization pass takes an existing procedure and changes it in a way that + * improves its performance when executed. + * + * Oftentimes it would also be possible to implement a specific optimization directly during + * construction of the initial #MFProcedure. There is a trade-off between doing that or just + * building a "simple" procedure and then optimizing it uses separate optimization passes. + * - Doing optimizations directly during construction is typically faster than doing it as a + * separate pass. However, it would be much harder to turn the optimization off when it is not + * necessary, making the construction potentially slower in those cases. + * - Doing optimizations directly would also make code more complex, because it mixes the logic + * that generates the procedure from some other data with optimization decisions. + * - Having a separate pass allows us to use it in different places when necessary. + * - Having a separate pass allows us to enable and disable it easily to better understand its + * impact on performance. + */ + +#include "FN_multi_function_procedure.hh" + +namespace blender::fn::procedure_optimization { + +/** + * When generating a procedure, destruct instructions (#MFDestructInstruction) have to be inserted + * for all variables that are not outputs. Often the simplest approach is to add these instructions + * at the very end. However, when the procedure is executed this is not optimal, because many more + * variables are initialized at the same time than necessary. This inhibits the reuse of memory + * buffers which decreases performance and increases memory use. + * + * This optimization pass moves destruct instructions up in the procedure. The goal is to destruct + * each variable right after its last use. + * + * For simplicity, and because this is the most common use case, this optimization currently only + * works on a single chain of instructions. Destruct instructions are not moved across branches. + * + * \param procedure The procedure that should be optimized. + * \param block_end_instr The instruction that points to the last instruction within a linear chain + * of instructions. The algorithm moves instructions backward starting at this instruction. + */ +void move_destructs_up(MFProcedure &procedure, MFInstruction &block_end_instr); + +} // namespace blender::fn::procedure_optimization |