Skip to content

[META][X86] Improve implementation and use of X86 scheduler models #31672

Closed
@RKSimon

Description

@RKSimon
Bugzilla Link 32325
Version trunk
OS Windows NT
Depends On #17741 #21274 #26792 #27166 #31673 #31834 #34896 #34935 #34936 #34937 #34938 #34939 #34958 #34960 #34991 #35132 #35296 #35304 #35327 #35461 #36128 #36176 #36236 #36243 #36245 #36407 #36479 #36658 #36855 #36949 #36950 #37202 #38536 #41327 #47454 #49694 #49704 #28947 #31564 #31778 #32148 #32204 #34510 #34894 #34897 #34954 #34956 #34987 #35023 #35024 #35061 #35062 #35358 #35432 #36074 #36129 #36130 #36131 #36140 #36156 #36175 #36229 #36235 #36238 #36242 #36246 #36256 #36272 #36278 #36279 #36299 #36305 #36441 #36634 #38813
Blocks #35898 #36073
CC @topperc,@filcab,@jryans,@LebedevRI,@MattPD,@rotateright,@ZviRackover

Extended Description

Meta ticket covering issues with the x86 scheduler models, which currently include:

X86ScheduleAtom.td
X86SchedSandyBridge.td
X86SchedHaswell.td
X86ScheduleSLM.td
X86ScheduleBtVer2.td

Including incorrect latency/throughput values and resources.

Also covers possible improvements to how we can use the schedulers to improve performance (e.g. use in the machine combiner for better instruction selection).

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions