In order to support to the following instruction described by ABI,
dynasm needed to be updated:

"""
The following code might appear in a PIC code setup sequence to compute
the distance from a function entry point to the TOC base:
addis 2,12,.TOC.-func@ha
addi 2,2,.TOC.-func@l
"""

Power Architecture 64-Bit ELF V2 ABI Specification, version 1.0, page 99
Source: http://openpowerfoundation.org/technical/technical-resources/technical-specifications/

The TOC register was not set correctly when branching with other
registers as the PIC code setup uses the R12 reference in order to set
the R2 (TOC register). This is only acknowledged by using LuaJIT as a
library, as torch uses on it qtlua/qlua subproject.

The POWER 64-bit LE ABI calls for floating point function
arguments beyond the 8th to be passed via floating point
registers and also to reserve a slot in the parameter save
area on the stack. The PPC CCALL_HANDLE_REGARG correctly
includes the values in FPRs, but neglects to reserve
corresponding slots when spilling into the save area.

ABI mandates 13 but only 8 was implemented.

RELEASE LuaJIT-2.1.0-beta3

Removed the complicated handling of lj_vm_ffi_call (it was a variable
size frame) and now backtrace works all over (e.g:)

 #0  0x00003fffb7d4875c in __libc_send (fd=32, buf=0x3fffb09a0028, len=8192, flags=0) at ../sysdeps/unix/sysv/linux/send.c:31
 #1  0x00003fffb7bea214 in socket_send (ps=0x3fffb7bc7778, data=0x3fffb09a0028 'A' <repeats 200 times>..., count=8192, sent=0x3fffffffee60, tm=0x3fffb7bc97d8) at usocket.c:205
 #2  0x00003fffb7be4ef8 in sendraw (buf=0x3fffb7bc77a0, data=0x3fffb09a0028 'A' <repeats 200 times>..., count=52428800, sent=0x3fffffffeee8) at buffer.c:176
 #3  0x00003fffb7be4960 in buffer_meth_send (L=0x3fffb7f6d280, buf=0x3fffb7bc77a0) at buffer.c:87
 #4  0x00003fffb7bec3f4 in meth_send (L=0x3fffb7f6d280) at tcp.c:130
 #5  0x0000000010042d44 in lj_BC_FUNCC ()
 #6  0x0000000010043f24 in lj_ff_coroutine_resume ()
 #7  0x000000001001d7d4 in lua_pcall (L=0x3fffb7f60378, nargs=0, nresults=-1, errfunc=2) at lj_api.c:1129
 #8  0x00000000100045e8 in docall (L=0x3fffb7f60378, narg=0, clear=0) at luajit.c:121
 #9  0x00000000100053ec in handle_script (L=0x3fffb7f60378, argx=0x3ffffffffa40) at luajit.c:291
 #10 0x0000000010006600 in pmain (L=0x3fffb7f60378) at luajit.c:551
 #11 0x0000000010042d44 in lj_BC_FUNCC ()
 #12 0x000000001001da40 in lua_cpcall (L=0x3fffb7f60378, func=0x10006334 <pmain>, ud=0x0) at lj_api.c:1153
 #13 0x00000000100067a4 in main (argc=2, argv=0x3ffffffffa38) at luajit.c:580

Despite unwind working for vm frames, the change sent before was not
fully correct. By analysing the DWARF code with
"readelf --debug-dump=frames luajit" I noticed it was not well formated.