Re: [PATCH v2 12/13] gpu: nova-core: add PIO support for loading firmware images

[John Hubbard <jhubbard@nvidia.com>]

On 12/2/25 6:14 PM, Joel Fernandes wrote:
>> On Dec 2, 2025, at 7:46 PM, John Hubbard <jhubbard@nvidia.com> wrote:
...
>>> suspect that in both cases, the compiler cannot tell that each chunk is always 4 bytes and
>>> try_into() will never panic.  In my case, word[3] always exists and will never panic either.  So I'm
>>> guess that the compiler will still emit code to check for panic.  I don't know.
>>>
>>
>> Good question. Maybe someone with Rust experience can enlighten us on
>> that one.
> 
> If the optimizer could not eliminate the dead code, IMO that is a bug, pretty much like how we rely on dead code elimination optimization for build_assert.
> 
> It is worth compiling and checking but I am almost certain the dead code for an unreachable case (in this case, panicking, will be eliminated).
> 

Well yes, but the question remains: is the compiler able to even identify
this as dead code? Because it depends on the early -EINVAL return, in
order to be correct. Is rustc able to figure out that kind of situation?

thanks,
-- 
John Hubbard