fix jit i386 with long double

[rurban]

on i386 Configure.pl --floatval='long double' leads to jit compilation problems, mostly just missing the interp argument. Note: This is NUMVAL_SIZE == 12

I have a patch, but this leads to:

../../parrot.exe ../../runtime/parrot/library/PGE/Perl6Grammar.pir  --output=PGE/builtins_gen.pir PGE/builtins.pg
make[1]: *** [PGE.pbc] Segmentation fault (core dumped)

I also see a potential problem in

/* numvar <- ST(i) */
#  define jit_emit_mov_mr_n(pc, d, r) { \
    emitm_fld((pc), (r)); \
    jit_emit_fstore_m_n((pc), (d)); \
}

jit_emit_fstore_m_n requires interp, but jit_emit_mov_mr_n does not give it to us. Apparently jit_emit_mov_mr_n is not used at build-time.

[rurban]

fixed jit_emit_mov_mr_n, but unused

[rurban]

The pge crash is caused by:

'I' is not a valid sprintf format

Note: This is NUMVAL_SIZE=12

$ ../../parrot.exe -Dff -dff ../../runtime/parrot/library/PGE/Perl6Grammar.pir --output=PGE/builtins_gen.pir PGE/builtins.pg

[rurban]

still mapping failures

[rurban]

failing tests with tt352-jit-i386-ld.5.patch:

$ TEST_PROG_ARGS=-j prove t/op/jit.t
t/op/jit....ok
$ TEST_PROG_ARGS=-j prove t/op/jitn.t
t/op/jitn (Wstat: 1792 Tests: 14 Failed: 7)
  Failed tests:  5-7, 9-11, 14
  Non-zero exit status: 7
$ TEST_PROG_ARGS=-j prove t/op/trans.t
t/op/trans (Wstat: 4096 Tests: 22 Failed: 16)
  Failed tests:  1-5, 9-12, 14-20
  Non-zero exit status: 16

looks like a mapping issue.

[rurban]

until it is fixed, disable jit with long double

[rurban]

tt352-disable-ld.patch applied as r37231.

[bacek]

Current JIT is deprecated.