Re: [Patch v5 00/19] Support SIMD/eGPRs/SSP registers sampling for perf

[Mi, Dapeng]

On 12/16/2025 12:42 PM, Ravi Bangoria wrote:
> Hi Dapeng,
>
>> While the hardware solution remains preferable due to its lower
>> overhead and higher accuracy, this software approach provides a
>> viable alternative.
> Lower accuracy in the software approach is due to the delay in an NMI
> delivery which will make the SIMD data misaligned a bit? Something like:
>
>    insn1
>    insn2  -> Overflow. RIP, GPRs captured by PEBS and NMI triggered
>    insn3
>    insn4
>    insn5  -> NMI delivered here, so SIMD regs are captured here?
>    insn6
>
> Am I interpreting it correctly?

Yes, there is always a delay with software-based (specifically PMI-based)
sampling. Hardware-based sampling like PEBS is preferable when available.


>
> Thanks,
> Ravi