Tiny Core Base > Alpha Releases

Tiny Core v4.0 Alpha 3

<< < (3/5) > >>

bmarkus:
Any idea how to avoid the 10-15% size increase of 4.0 built extensions compared to 3.x build? In most cases no need for GCC 4.61. Adding GCC from 3.x to 4.0 repo would help. BTW, 4.0 repo contains GCC 3.3.6 also.

curaga:
Why not ask the gcc folks on what changed with -Os from 4.4 to 4.6? If there's a reproducible test case, some small package etc, they should listen.

bmarkus:

--- Quote from: curaga on September 09, 2011, 03:52:02 AM ---Why not ask the gcc folks on what changed with -Os from 4.4 to 4.6? If there's a reproducible test case, some small package etc, they should listen.

--- End quote ---

For sure there are test cases (all extensions maintained by me :) ) Will make a try

Jason W:
Here is link noticing a change in the behavior of gcc 4.6, perhaps building gcc with "--enable-frame-pointer" can restore the smaller size binaries generated by the previous gcc.  May be worth a try, if I am reading that page right.  (EDIT:I don't see -fomit-frame-pointer along with -Os causing a larger binary size with the TC 3.x I have available to test with remotely, so may not be the cause.)

http://www.mail-archive.com/coreboot@coreboot.org/msg30467.html

robc:

--- Quote from: Jason W on September 09, 2011, 07:24:28 AM ---Here is link noticing a change in the behavior of gcc 4.6, perhaps building gcc with "--enable-frame-pointer" can restore the smaller size binaries generated by the previous gcc.  May be worth a try, if I am reading that page right.  (EDIT:I don't see -fomit-frame-pointer along with -Os causing a larger binary size with the TC 3.x I have available to test with remotely, so may not be the cause.)

http://www.mail-archive.com/coreboot@coreboot.org/msg30467.html

--- End quote ---
I ran some quick builds of ccache to look at the different optimizations.  The build tests failed without the -fno-omit-frame-pointer switch.  The builds finished and produced the binary but the binary failed at runtime with the testsuites.  I would recommend adding the --enable-frame-pointer into gcc to prevent any runtime problems.

Here is a table of optimizations ran:optimizationbinary size-Os74.9k-O175.6k-O272.2k-O389.0k-Ofast85.9kI find it interesting that the -O2 gives the smallest binary size, though it is over 33% larger then in 3.x.

Also, compiletc depends on linux-3.0.1_api-headers, is this correct? I thought the kernel was 3.0.3.

Navigation

[0] Message Index

[#] Next page

[*] Previous page