Instruction scheduling: Difference between revisions

Revision as of 18:55, 26 June 2019

Latest revision as of 23:01, 5 July 2025

(21 intermediate revisions by 9 users not shown)

Line 1:

{{Short description|Compiler optimization technique}}

{{Refimprove|date=April 2018}}

In [[computer science]], '''instruction scheduling''' is a [[compiler optimization]] used to improve [[instruction-level parallelism]], which improves performance on machines with [[instruction pipeline]]s. Put more simply, it tries to do the following without changing the meaning of the code:

Line 27 ⟶ 28:

* The processor resources used by the already scheduled instructions are recorded. If a candidate uses a resource that is occupied, its priority will drop.

* If a candidate is scheduled closer to its predecessors than the associated latency, its priority will drop.

* If a candidate lies on the critical path of the graph, its priority will rise. This heuristic provides some form of look-ahead in an otherwise local decision process.

* If choosing a candidate will create many new sources, its priority will rise. This heuristic tends to generate more freedom for the scheduler.

== The phasePhase order of instruction scheduling ==

Instruction scheduling may be done either before or after [[register allocation]] or both before and after it. The advantage of doing it before register allocation is that this results in maximum parallelism. The disadvantage of doing it before register allocation is that this can result in the register allocator needing to use a number of registers exceeding those available. This will cause spill/fill code to be introduced, which will reduce the performance of the section of code in question.

If the architecture being scheduled has instruction sequences that have potentially illegal combinations (due to a lack of instruction interlocks), the instructions must be scheduled after register allocation. This second scheduling pass will also improve the placement of the spill/fill code.

If scheduling is only done after register allocation, then there will be false dependencies introduced by the register allocation that will limit the amount of instruction motion possible by the scheduler.

== Types of instruction scheduling ==

There are several types of instruction scheduling:

#''Local'' (''basic block'') ''scheduling'': instructions can't move across basic block boundaries.

#''Global scheduling'': instructions can move across basic block boundaries.

#''Modulo scheduling'': an algorithm for generating [[software pipelining]], which is a way of increasing instruction level parallelism by interleaving different iterations of an inner [[Control_flowControl flow#Loops|loop]].

#''[[Trace scheduling]]'': the first practical approach for global scheduling, trace scheduling tries to optimize the control flow path that is executed most often.

#''Superblock scheduling'': a simplified form of trace scheduling which does not attempt to merge control flow paths at trace "side entrances". Instead, code can be implemented by more than one schedule, vastly simplifying the code generator.

== Compiler examples ==

The [[GNU Compiler Collection]] is one compiler known to perform instruction scheduling, using the {{code|-march}} (both instruction set and scheduling) or {{code|-mtune}} (only scheduling) flags. It uses descriptions of instruction latencies and what instructions can be run in parallel (or equivalently, which "port" each use) for each microarchitecture to perform the task. This feature is available to almost all architectures that GCC supports.<ref>{{cite web |title=x86 Options |url=https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html |website=Using the GNU Compiler Collection (GCC)}}</ref>

Until version 12.0.0, the instruction scheduling in [[LLVM]]/Clang could only accept a {{code|-march}} (called {{code|target-cpu}} in LLVM parlance) switch for both instruction set and scheduling. Version 12 adds support for {{code|-mtune}} ({{code|tune-cpu}}) for x86 only.<ref>{{cite web |title=⚙ D85384 [X86] Add basic support for -mtune command line option in clang |url=http://reviews.llvm.org/D85384 |website=reviews.llvm.org}}</ref>

Sources of information on latency and port usage include:

* GCC and LLVM;

* [[Agner Fog]], who compiles extensive data for the [[x86 architecture]];<ref name="optimize">{{cite web |title=Software optimization resources. C++ and assembly. Windows, Linux, BSD, Mac OS X |url=https://www.agner.org/optimize/ |website=Agner Fog}}</ref>

* InstLatx64, which uses [[AIDA64]] to collect data on x86 CPUs.<ref>{{cite web |title=x86, x64 Instruction Latency, Memory Latency and CPUID dumps |url=http://instlatx64.atw.hu/ |website=instlatx64.atw.hu}} See also the "Comments" link on the page.</ref>

* uops.info, which provides latency, throughput, and port usage information for x86 microarchitectures.<ref>[https://uops.info/ uops.info]</ref><ref>{{Cite conference |title=uops.info: Characterizing Latency, Throughput, and Port Usage of Instructions on Intel Microarchitectures |last1=Abel |first1=Andreas |last2=Reineke |first2=Jan |date=2019 |conference=[[International Conference on Architectural Support for Programming Languages and Operating Systems]], Providence, RI, USA, April 13–17, 2019 |arxiv=1810.04610 |doi=10.1145/3297858.3304062 |isbn=978-1-4503-6240-5 |pages=673-686 |book-title=ASPLOS '19: Proceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems |publication-date=April 2019 |publication-place=New York, NY, USA |publisher=[[Association for Computing Machinery|ACM]] }}</ref>

LLVM's {{code|llvm-exegesis}} should be usable on all machines, especially to gather information on non-x86 ones.<ref>{{cite web |title=llvm-exegesis - LLVM Machine Instruction Benchmark |url=https://llvm.org/docs/CommandGuide/llvm-exegesis.html |website=LLVM 12 Documentation}}</ref>

==See also ==

* [[Branch predication]]

{{Portal|Computer science}}

* [[Code generation (compiler)|Code generation]]

* [[Instruction unit]]

* [[BranchOut-of-order predicationexecution]]

==References==

Line 58 ⟶ 72:

==Further reading==

* {{cite journal |author-first=Joseph A. |author-last=Fisher |author-link=Joseph A. Fisher |title=Trace Scheduling: A Technique for Global Microcode Compaction |journal=IEEE Transactions on Computers |volume=30 |number=7 |pages=478–490 |date=1981 |urldoi=http:10.1109//ieeexploreTC.ieee1981.org/xpl/login.jsp?tp1675827 |s2cid=&arnumber=16758271650655 }} (''[[Trace scheduling]]'')

* {{cite journal |author-first1=Alexandru |author-last1=Nicolau |author-first2=Joseph A. |author-last2=Fisher |author-link2=Joseph A. Fisher |title=Measuring the Parallelism Available for Very Long Instruction Word Architectures |journal=IEEE Transactions on Computers |volume=33 |number=11 |date=1984}} (''Percolation scheduling'')

* {{cite journal |author-first1=David |author-last1=Bernstein |author-first2=Michael |author-last2=Rodeh |title=Global Instruction Scheduling for Superscalar Machines |journal=Proceedings of the ACM, SIGPLAN '91 Conference on Programming Language Design and Implementation |date=June 1991|url=http://pages.cs.wisc.edu/~fischer/cs701.f06/berstein_rodeh.pdf}} (''Global scheduling'')

* {{cite web |last1=Cordes |first1=Peter |title=assembly - Instruction reordering in x86 / x64 asm - performance optimisation with latest CPUs |url=https://stackoverflow.com/a/45970664 |website=Stack Overflow}}

{{Compiler optimizations}}