Hi Sughosh,
On Sun, 28 Jul 2024 at 12:07, Sughosh Ganu sughosh.ganu@linaro.org wrote:
On Thu, 25 Jul 2024 at 18:24, Sughosh Ganu sughosh.ganu@linaro.org wrote:
On Thu, 25 Jul 2024 at 02:24, Tom Rini trini@konsulko.com wrote:
On Wed, Jul 24, 2024 at 11:31:48AM +0530, Sughosh Ganu wrote:
From: Simon Glass sjg@chromium.org
Use this new data structure in the utility function.
Signed-off-by: Simon Glass sjg@chromium.org Signed-off-by: Sughosh Ganu sughosh.ganu@linaro.org
lib/strto.c | 35 +++++++++++++++++++---------------- 1 file changed, 19 insertions(+), 16 deletions(-)
This is rather big growth when we didn't already have realloc: 05: lib: Convert str_to_list() to use alist aarch64: (for 1/1 boards) all +1765.0 rodata +37.0 text +1728.0 xilinx_versal_mini_emmc0: all +1765 rodata +37 text +1728 u-boot: add: 7/0, grow: 1/0 bytes: 1728/0 (1728) function old new delta realloc - 1120 +1120 alist_ensure_ptr - 140 +140 alist_expand_to - 136 +136 alist_init - 108 +108 alist_uninit_move_ptr - 76 +76 alist_add_ptr - 72 +72 alist_uninit - 48 +48 str_to_list 204 232 +28
I am working on an implementation of lmb maps using lists. The list nodes are then allocated with calloc, which I believe is included in most of the board images. We can then compare the size impact with the two implementations (alist vs list). Thanks.
I have worked on implementing the LMB maps using simple lists, and we do get a good size benefit with the list based implementation. I have used the script that was shared by Tom some time back to get the size impact [1]. The readings are with 1) alists with realloc 2) alist with malloc and 3) using linked list. These are on a RPI4 build, which has LMB enabled. I have checked the impact on a xilinx_versal_mini config as well, which does not have LMB enabled, and the size impact result is best with lists. The list based changes can be checked on my github [2].
Thanks for doing that. It is quite hard to read with the LTO enabled. You can try with the -L flag and should get a cleaner result. Also see my comment about how you are using alist directly instead of going through the API...that might affect things.
Anyway (once you are using the API), my assumption from my previous refactor attempt was that lmb would be easier to deal with with simple arrays, which I why I wrote alist. Is that true, or not?
-sughosh
[1] - https://gist.github.com/sughoshg/d4f9bda8d8a33f715dab892738d192ba [2] - https://github.com/sughoshg/u-boot/tree/lmb_only_linked_lists_v3
Regards, Simon