|
From: Henrique de M. H. <hm...@hm...> - 2007-04-22 18:56:35
|
On Sat, 21 Apr 2007, Len Brown wrote: > On Saturday 21 April 2007 10:08, Henrique de Moraes Holschuh wrote: > > Use a bitfield to hold boolean module-wide flags, to conserve some memory. > > It is easy and it is clean, so we do it just for the heck of it even if it > > saves very little space. > > I think that is an under-statement. > This plus the patch before it actually increase text size > for a static size change of 0. > > -Len > > before: > text data bss dec hex filename > 24002 3286 728 28016 6d70 drivers/misc/thinkpad_acpi.o > > after > text data bss dec hex filename > 24082 3270 664 28016 6d70 drivers/misc/thinkpad_acpi.o Well, as module, same Debian gcc 4.1.2 pre-release I used before, and objdump -w -h on the .ko file: .bss: decreases from 0x22c to 0x20c .data: decreases from 0x7a4 to 0x5b4 .init.data: no change .text: decreases from 0x3658 to 0x3628 .init.text: increases from 0x174a to 0x1763 .rodata*: no change Same remark as before: maybe I am mistaken on how these things work when the linking is done at insmod time. I really did not even consider checking what happens in a static build, it is quite possible that size increases with these patches. One of the other patches really causes an unamusing size increase due to data structure alignment that I will address later (it is lower priority and would cause too much rework on patches I did not submit yet). It happened when some arrays were broken up to components and arrays of pointers to them. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh |
