Re: tinyconfig x86-32 vmlinux sizes by gcc compiler version

[Joe Perches <joe@perches.com>]

On Thu, 2014-12-04 at 14:25 -0800, josh@joshtriplett.org wrote:
> On Thu, Dec 04, 2014 at 01:32:59PM -0800, Joe Perches wrote:
> > Just fyi.
> > 
> > At least for x86-32, it seems later versions of gcc
> > are producing smaller images.
> > 
> > $ size vmlinux.*
> >    text	   data	    bss	    dec	    hex	filename
> >  657725	 118496	1189040	1965261	 1dfccd	vmlinux.4.4
> >  633563	 118528	1189448	1941539	 1da023	vmlinux.4.6
> >  633277	 118496	1189592	1941365	 1d9f75	vmlinux.4.7
> >  632299	 121120	1192784	1946203	 1db25b	vmlinux.4.9
> 
> I would certainly hope that GCC's -Os gets better over time.  However, I
> find the increase in data/bss and thus overall size in 4.9 concerning.
> Any idea what that comes from?

No.

It's all over the place.

For instance:

$ objdump -t vmlinux.4.7 | awk '{if (NF>=2) {print $(NF) "\t" $(NF-1)}}' | sort > 4.7
$ objdump -t vmlinux.4.9 | awk '{if (NF>=2) {print $(NF) "\t" $(NF-1)}}' | sort > 4.9

$ diff -u0 -rN 4.7 4.9 | wc -l
6944

$ diff -u0 -rN 4.7 4.9 | head -20
--- 4.7 2014-12-04 14:34:41.455533301 -0800
+++ 4.9 2014-12-04 14:34:16.703531656 -0800
@@ -5 +5 @@
-access_process_vm      00000040
+access_process_vm      0000003e
@@ -9,3 +9,2 @@
-account_event_cpu.isra.57      0000001f
-account_event.part.60  00000065
-account_group_user_time        0000001f
+account_event_cpu.isra.56      0000001f
+account_event.part.61  00000069
@@ -14,2 +13 @@
-account_kernel_stack   00000029
-account_page_dirtied   00000052
+account_page_dirtied   0000005f
@@ -20,2 +18,2 @@
-account_system_time    000000e3
-account_user_time      00000047
+account_system_time    0000010d
+account_user_time      0000005d