1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
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
|
