]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commit
Give up on pushing CC_OPTIMIZE_FOR_SIZE
authorLinus Torvalds <torvalds@linux-foundation.org>
Sun, 22 May 2011 21:30:36 +0000 (14:30 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sun, 22 May 2011 21:30:36 +0000 (14:30 -0700)
commit281dc5c5ec0fb299514567cbc358562649c1af95
tree2ecb019795e99147ce8d90b5e97c2bc3051f7855
parent71a8638480eb8fb6cfabe2ee9ca3fbc6e3453a14
Give up on pushing CC_OPTIMIZE_FOR_SIZE

I still happen to believe that I$ miss costs are a major thing, but
sadly, -Os doesn't seem to be the solution.  With or without it, gcc
will miss some obvious code size improvements, and with it enabled gcc
will sometimes make choices that aren't good even with high I$ miss
ratios.

For example, with -Os, gcc on x86 will turn a 20-byte constant memcpy
into a "rep movsl".  While I sincerely hope that x86 CPU's will some day
do a good job at that, they certainly don't do it yet, and the cost is
higher than a L1 I$ miss would be.

Some day I hope we can re-enable this.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
init/Kconfig