[PATCH/RFA] SH TLS support
Joern Rennecke
joern.rennecke@superh.com
Fri Feb 21 12:49:00 GMT 2003
kaz Kojima wrote:
>
> Joern Rennecke <joern.rennecke@superh.com> wrote:
> > I don't see why we need new relocation numbers. Can't we just extend
> > the scope of the existing BFD_RELOC_SH_COUNT and BFD_RELOC_SH_USES
> That's a very good idea for GD and LD cases. It seems that Initial
> Exec (IE) case needs new relocations yet. If non-fixed sequence is
> used, IE code will be like:
>
> mov.l L1, r0
> stc gbr, r1
> mov.l @(r0,r12), r0
> add r1, r0
> ...
> (constant pool)
> L1: .long x@GOTTPOFF
>
> and all 4 instructions should be marked as TLS code.
>
> BTW, I took a simple linker statistics in cross build of glibc.
>
> (type) (in) (out)
> GD 8793 8793
> LD 72 72
> IE 1347 125
> LE 34 1256
>
> I could be wrong about it and it would be not typical, but it seems
> high percentage of TLS relocations are non relaxable GD and almost
> transitions occur as IE->LE. If it is general tendency, the loss of
> fixed sequences would be small and permissible for SH-3/4 case.
What does LE stand for? How does such an IE->LE transition look like?
--
--------------------------
SuperH (UK) Ltd.
2410 Aztec West / Almondsbury / BRISTOL / BS32 4QX
T:+44 1454 465658
More information about the Gcc-patches
mailing list