https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69874
Markus Trippelsdorf <trippels at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |hjl.tools at gmail dot com
--- Comment #5 from Markus Trippelsdorf <trippels at gcc dot gnu.org> ---
Ah the misalignment will be fixed for 4.9.4, PR58066.
H.J.:
sysdeps/x86_64/dl-trampoline.S says:
23 #ifndef DL_STACK_ALIGNMENT
24 /* Due to GCC bug:
25
26 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58066
27
28 __tls_get_addr may be called with 8-byte stack alignment. Although
29 this bug has been fixed in GCC 4.9.4, 5.3 and 6, we can't assume
30 that stack will be always aligned at 16 bytes. We use unaligned
31 16-byte move to load and store SSE registers, which has no penalty
32 on modern processors if stack is 16-byte aligned. */
33 # define DL_STACK_ALIGNMENT 8
34 #endif
35
36 #ifndef DL_RUNIME_UNALIGNED_VEC_SIZE
37 /* The maximum size of unaligned vector load and store. */
38 # define DL_RUNIME_UNALIGNED_VEC_SIZE 16
39 #endif
40
41 /* True if _dl_runtime_resolve should align stack to VEC_SIZE bytes. */
42 #define DL_RUNIME_RESOLVE_REALIGN_STACK \
43 (VEC_SIZE > DL_STACK_ALIGNMENT \
44 && VEC_SIZE > DL_RUNIME_UNALIGNED_VEC_SIZE)
this apparently doesn't work for _dl_runtime_resolve_sse.
