| Issue |
179955
|
| Summary |
[llvm-mca] Warn the user or error if there is no accurate CPU model for selected -mcpu
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
ktkachov
|
Currently if a user passes an -mcpu=option to simulate a certain instruction sequence MCA will use whatever scheduling model is specified for that CPU internally in LLVM. Usually if there is no accurate scheduling model added when a new -mcpu option is upstreamed a sensible default is selected. This is fine for instruction scheduling optimisations, but I would argue is unhelpful for MCA users and can lead to confusion.
For example, in aarch64 many CPUs default to Cortex-A510 scheduling in the absence of dedicated models, but silently simulating a Cortex-A510 in MCA when an unsuspecting user selects a different -mcpu option is unhelpful
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
