[LLVM][ARM] Wrong profiling information in llvm-mca for cortex-m52

[shazha01]

llvm-mca already supports Cortex-M52 target but provides Cortex-M55 information for Cortex-M52 when using llvm-mca-18/-19/-21 to profile assembly code, such as "Instruction Info", unit "Resources"...
llvm-mca -mtriple=arm-none-eabi -mcpu=cortex-m52 <func>.s

And the confusing code can be found in llvm/lib/Target/ARM/ARMProcessors.td
def : ProcessorModel<"cortex-m52", CortexM55Model,

I would like to know if it is a typo? Or will Cortex-M52 be completely supported in the next version?

Thanks!

[llvmbot]

@llvm/issue-subscribers-backend-arm

Author: None (shazha01)

llvm-mca already supports Cortex-M52 target but provides Cortex-M55 information for Cortex-M52 when using llvm-mca-18/-19/-21 to profile assembly code, such as "Instruction Info", unit "Resources"... `llvm-mca -mtriple=arm-none-eabi -mcpu=cortex-m52 <func>.s`

And the confusing code can be found in llvm/lib/Target/ARM/ARMProcessors.td
def : ProcessorModel<"cortex-m52", CortexM55Model,

I would like to know if it is a typo? Or will Cortex-M52 be completely supported in the next version?

Thanks!

[davemgreen]

@AlbertHuang-CPU FYI.

The way LLVM / llvm-mca works is by having scheduling models for various cores. LLVM-MCA re-uses the model that we use for scheduling during codegen to calculate latency and resource usage. If we don't have a scheduling model for a particular core then it is common to reuse the next-closest as an approximation.

If anyone contributes an cortex-m52 scheduling model then we can use it for both instruction scheduling and llvm-mca. The M55 model is defined in https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/ARM/ARMScheduleM55.td.

[AlbertHuang-CPU]

Thanks @davemgreen for the information. Yes, we're aware of it and currently it's fine using M55 model for M52. We don't have a plan at the moment to create a model for M52 separately, but highly appriciate if anyone will contribute it. And will let @shazha01 know if any progress.