[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [RFC PATCH 2/2] gitlab-ci: Add a job building TCI with Clang

From: Alex Bennée
Subject: Re: [RFC PATCH 2/2] gitlab-ci: Add a job building TCI with Clang
Date: Tue, 26 Jan 2021 17:08:30 +0000
User-agent: mu4e 1.5.7; emacs 28.0.50

Stefan Weil <sw@weilnetz.de> writes:

> Am 23.01.21 um 09:59 schrieb Wataru Ashihara:
>> Actually I use TCI also on macOS. Like the use case quoted by Philippe,
>> there're even other reasons to use TCI:
>> 1. Learning TCG ops.
>> 2. Debugging QEMU with gdb. e.g. diagnose codegen or stepping into
>>     helper functions from tci.c:tcg_qemu_tb_exec().
>> 3. Guest instruction tracing. TCI is faster than TCG or KVM when tracing
>>     the guest ops [1]. I guess qira is using TCI for this reason [2].
>> [1]: https://twitter.com/wata_ash/status/1352899988032942080
>> [2]: https://github.com/geohot/qira/blob/v1.3/tracers/qemu_build.sh#L55
> Yes, TCI can help a lot for debugging, especially also when porting TCG 
> to a new host architecture.
> If we had binaries which can switch from native to interpreted TCG, it 
> could also be a reference implementation used for unit tests, comparing 
> the results for each TCG opcode.
> Using TCI with profiling like gprof is useful to count the frequency of 
> the different TCG opcodes in practical scenarios and can be used to 
> detect bottlenecks (and less frequent or unused opcodes) for native TCG, 
> too.

FWIW I had a bunch of JIT profiling changes that exposed the TCG op
counts via the JIT profiler.  I think I even enabled the op counting by
default because it was fairly lightweight to add.

IOW I think more introspection can be brought into the core TCG code
rather than relying on TCI to achieve it.

Alex Bennée

reply via email to

[Prev in Thread] Current Thread [Next in Thread]