[MPlayer-dev-eng] [PATCH] Windows DLL support for OS X/Intel

[Valtteri Vuorikoski]

Diego Biurrun <diego at biurrun.de> writes:

> The OpenBSD hunk looks wrong/unrelated.

If you mean the stubs.s/loader_objfix.sh changes, it depends. As far
as I can see, my cpp-ified stubs.s and loader_objfix.sh result in
exactly the same thing, but without needing objcopy installed (the
required functionality is the same for Darwin and OpenBSD, that is,
adding _ to symbol names).

The loader_objfix.sh method can be preserved for OpenBSD, but that
would be sort of redundant.

> Skip the WRAPPER_OBJ variable and just add
>
>   # QTX emulation is not supported in Darwin yet
>   ifneq ($(TARGET_OS),Darwin)
>   LIB_OBJECTS+=wrapper.o
>   endif

I'll fix that.

> Other than that I think an updated patch will have a good chance of
> getting committed.

I'll clean up the mmap stuff that was discussed earlier first. It
might also be good to have a warning to the effect of "set
DYLD_BIND_AT_LAUNCH if you don't want to crash" if DYLD_BIND_AT_LAUNCH
isn't set, since all the workarounds I've managed to think for the
dyld issue of have been rather evil.

> You mean the stuff in FFmpeg?  Several hacks have been proposed, but
> none have been accepted so far...

Yes. I have only seen the #ifdef'ified version. Do you think
converting the current gas macros into cpp macros would be acceptable?
For example:

#define FDCT_ROW_SSE2_H2(p1,p2) \
        "movq      " #p1 "(%0), %%xmm2      \n\t"\
        "movq      " #p1 "+8(%0), %%xmm0    \n\t"\
        "movdqa    " #p1 "+32(%1), %%xmm3   \n\t"\
        "movdqa    " #p2 "+48(%1), %%xmm7   \n\t"

asm volatile(FDCT_ROW_SSE2_H2(4242,666) :: "r" (in), [...etc])



 -v