On Monday 19 October 2020 10:09:19 CEST David Laight wrote:
> From: Joe Perches
> > Sent: 17 October 2020 01:12
> >
> > On Fri, 2020-10-16 at 23:33 +0100, Colin King wrote:
> > > From: Colin Ian King <colin.king@xxxxxxxxxxxxx>
> > >
> > > Don't populate const array filter_ies on the stack but instead
> > > make it static. Makes the object code smaller by 261 bytes.
> > >
> > > Before:
> > > text data bss dec hex filename
> > > 21674 3166 448 25288 62c8 drivers/staging/wfx/sta.o
> > >
> > > After:
> > > text data bss dec hex filename
> > > 21349 3230 448 25027 61c3 drivers/staging/wfx/sta.o
> >
> > Thanks.
> >
> > It's odd to me it's so large a change as it's only
> > 24 bytes of initialization. (3 entries, each 8 bytes)
>
> Perhaps the 'stack protector' crap?
>
> Interestingly, loading the data from the 'readonly' section
> is probably a data cache miss.
> Which might end up being slower than the extra code to
> update the on-stack data.
> The extra code might get prefetched...
I had never realized the difference between "const" and "static const" in
this case.
With my gcc fro arm, the output of "objdump -h sta.o" gives:
Before:
0 .text 000019fc 00000000 00000000 00000034 2**2
CONTENTS, ALLOC, LOAD, READONLY, CODE
[...]
7 .rodata 00000015 00000000 00000000 00001e78 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
After:
0 .text 00001974 00000000 00000000 00000034 2**2
CONTENTS, ALLOC, LOAD, READONLY, CODE
[...]
7 .rodata 0000002d 00000000 00000000 00001dd4 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
The difference of .rodata is exactly what is expected (24 bytes) and we
save 115 bytes of code.
Reviewed-by: Jérôme Pouiller <jerome.pouiller@xxxxxxxxxx>
--
Jérôme Pouiller
