[FFmpeg-devel] [PATCH 1/2] lavu/bswap: remove some inline assembler

[James Almer]

On 6/11/2024 12:57 PM, Michael Niedermayer wrote:
> On Tue, Jun 11, 2024 at 12:38:37PM -0300, James Almer wrote:
>> On 6/11/2024 10:15 AM, Michael Niedermayer wrote:
>>> On Fri, Jun 07, 2024 at 09:19:46PM +0300, Rémi Denis-Courmont wrote:
>>>> C code or compiler built-ins are preferable over inline assembler for
>>>> byte-swaps as it allows for better optimisations (e.g. instruction
>>>> scheduling) which would otherwise be impossible.
>>>>
>>>> As with f64c2e710fa1a7b59753224e717f57c48462076f for x86 and Arm,
>>>> this removes the inline assembler on GCC (and Clang) since we now
>>>> require recent enough compiler versions (this indeed seems to work on
>>>> AArch64).
>>>> ---
>>>>    libavutil/aarch64/bswap.h | 56 ---------------------------------------
>>>>    libavutil/avr32/bswap.h   | 44 ------------------------------
>>>>    libavutil/bswap.h         |  8 +-----
>>>>    libavutil/sh4/bswap.h     | 48 ---------------------------------
>>>
>>> As you are writing that this preferrable for better optimisations
>>> Please provide benchmarks (for sh4, avr32)
>>
>> This is a ridiculous request, considering nobody has such hardware at all.
> 
> Then I think its a ridiculous claim that this optimizes the code
> 
> I mean, at some point there was hardware and these optimisations did improve
> speed.
> 
> This patch is not removing the code because its a rare (or dead) platform, it removes
> it with the claim that this would "allows for better optimisations"
> Iam sorry but i do not see why asking for the claim in the commit message
> to be backed up with facts being ridiculous
> The claim in the commit message may be ridiculous

Compilers have come a long way since 20 years ago when this code was added.
See https://godbolt.org/z/jPose4rj3, where new GCC generates the same 
code for sh4. And no inline assembly means instruction scheduling will 
take these functions into account.

> 
> thx
> 
> [...]
> 
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".