On 4/23/25 12:53, Philippe Mathieu-Daudé wrote:
On 23/4/25 21:33, Pierrick Bouvier wrote:
On 4/23/25 12:12, Richard Henderson wrote:
@Richard:
Is it a concern regarding code maintenance, or potential impact
on .data?
I was thinking of impact on .data, especially with so many.
du qemu-system-aarch64 optimized and stripped (in kB):
31880 upstream
31896 upstream + this series
FYI same tag as a branch with InterfaceInfo constified:
https://gitlab.com/philmd/qemu/-/tree/single-binary-hw-arm-rfc-v4-const
Maybe I missed something, but machine interfaces are not const on this
branch.
So we have +16kB which is a size increase of +0.0005%.
Even if we project something similar on other architectures (let's say
x10), the final impact on binary size should be < 0.005%.
Maybe it's a reasonable impact considering the trade off on coherency
and readability through the codebase?
Else, in case we make this array const, can we expect the linker to
deduplicate it? I'm not familiar with how final .data section is assembled.
