This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] PR 32475 stack setup code being deleted.
On 7/3/07, Paolo Bonzini <email@example.com> wrote:
Seongbae Park wrote:
> Attached is the patch to fix 32475
> where the stack pointer becomes dead after epilogue generation,
> due to the stack pointer restoration codein the epilogue.
> The patch adds a df_ref of use of the stack pointer
> for every write to the stack pointer - which essentially keeps
> the stack pointer alive always.
So, why not munging the live bitmaps directly, so that when the frame
pointer is live, the stack pointer is too?
I'd like to avoid munging the live bitmaps for various reasons:
1) munging the live vector doesn't give us the correct reg chain.
2) munging is more fragile - scanning is the lower layer than live analysis,
and this fix is very small in the size, with effectively the same outcome.
#pragma ident "Seongbae Park, compiler, http://seongbae.blogspot.com"