Re: [PATCH v2] sh: Remove IO memcpy and memset from sh code

[John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>]

Hi Geert,

On Thu, 2025-01-30 at 10:13 +0100, Geert Uytterhoeven wrote:
> Hi Adrian,
> 
> On Thu, 30 Jan 2025 at 09:44, John Paul Adrian Glaubitz
> <glaubitz@physik.fu-berlin.de> wrote:
> > On Tue, 2025-01-28 at 11:13 +0100, Julian Vetter wrote:
> > > Remove IO memcpy and memset from sh specific code and fall back to the
> > > new implementations from lib/iomem_copy.c. They use word accesses if the
> > > buffers are aligned and only fall back to byte accesses for potentially
> > > unaligned parts of a buffer.
> > > 
> > > Signed-off-by: Julian Vetter <julian@outer-limits.org>
> > > ---
> > > Changes for V2:
> > > - Removed also SH4 specific memcpy_fromio code
> 
> > I'm not sure that I understand the motivation to remove hand-optimized sh4 assembler
> > code for memset and drop it in favor of potentially slower generic C code. What is
> > the reasoning behind this?
> 
> See Arnd's feedback on v1
> https://lore.kernel.org/all/ffe019a1-11b4-4ad7-bbe2-8ef3e01ffeb0@app.fastmail.com

I read Arnd's feedback but I don't really know whether GCC produces better code than
this hand-written assembly. Is there any compelling argument?

I'm just worried we would slow down something as fundamental as memset().

> > Also, it seems that this patch would make your other patch
> > 
> >         "sh: Remove memset_io from sh specific code"
> > 
> > obsolete.
> 
> Yeah, that should have been mentioned under the ---.

Thought so.

Adrian

-- 
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913