This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Boolize darwin's target files, and clean up and fix compat testcases
On Apr 12, 2004, at 12:22 PM, Andrew Pinski wrote:
On Apr 12, 2004, at 15:03, Dale Johannesen wrote:
On Apr 12, 2004, at 6:29 AM, Andrew Pinski wrote:
To help out for fixing part of the darwin back-end so that it will
only find out if the
function needs a stub only when outputting the symbol, I made two
machopic_define_data_name and machopic_define_function_name which
right now just do the
same as machopic_define_name just to do.
So you expect these to diverge in the future, right?
Yes, I am changing when the back-end figures out if the symbol needs
to be a stub or not
to fix a regression which I introduced after fixing a build regression
which was introduced
by one of Roger Sayle's combine patch. Here is the plan I had in mind.
When a function is declared (meaning a function body is there),
That's defined, not declared, in the standard terminology. (I wouldn't
care except that
declared means something else.)
the function gets added to a
declared list. When the function is being outputted the branches to a
symbol is checked to see
if it is in the declared list and if it is, then just output the
symbol name. Otherwise, output
the name of the stub for the symbol and then add it to the stub list
if it is not already there.
I don't know about the bug, but in general this isn't a very useful
thing to do. If the compiler
just inserts stubs for all calls the linker will get rid of all that
the compiler could have.
Furthermore when it was tried before a couple of years ago it broke
things, although I forget how.
Right now eon in SPEC 2K cannot compile because the patch which I
added to fix the build regression
and I was hoping to get some numbers so that I can compare the
tree-ssa with the mainline soon
for the FSF's gcc.