Commit Graph

2 Commits

Author SHA1 Message Date
Andrew Boie f6dd6d5fd8 x86: atomic: don't use '__asm__ goto'
Not supported in LLVM/clang.  As it turns out,
this new implementation is 4 bytes shorter than its
predecessor.

Change-Id: I1f63b7a245dafcfc5a6dadc293875f00d02b997c
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
2016-02-05 20:24:47 -05:00
Andrew Boie 6bf328b6ea x86: use GCC inline assembly for atomic operations
This doesn't make any assumptions on calling conventions or
the structure of the stack, and should thus be portable to
compilers that implement different C calling conventions.

In order for the rewritten functions to take up the same code
size as the pure-asm counterparts, -fomit-frame-pointer has
been specified for each of them, otherwise an extra 4 bytes
is used for every function.

The generated assembly code by these new functions has been
verified in GDB to be the same as the old ones, except a few
trivial things like particular registers used.

Change-Id: I9a896cbfc3e7f4c2497d749140729d28b32f1c9d
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
2016-02-05 20:24:42 -05:00