Fixed: Compiler not thread-safe...

STOP/RESTART operates on a global without any lock...
Fixed: By making a thread-specific compiler disabled flag.


Fixed: (TODO: Implemented concurrent compilation and make sure no race condition)


- Bug: jit input code array allocated in jit_compile.c without using
  SableJIT class loader interface.  Same thing for the exceptionHandlingOffsets
  array and maybe others....

- Bug: some local vars in jit_sablejit.h CALLCODE and maybe elsewhere are
  unnecessarily reread.  For each var after compiled_code() call, check if
  needs to be reread.



--- stack_offsets_diff issue

Solution to implement: Remove stack_offsets_diff from sequence struct.
Store in vm->instructions instead if inst affect stack in a fix way, otherwise
store MAX_INT.



BUGS - fix this before getting into trouble...

jint *stack_offsets_diff; is added to the sequence data structure.
stack_offsets_diff[j] is inst[j + 1].stack_size - inst[j].stack_size
delta stack size...


NOTE: To search: Can these be kept in vm-\>instructions?  Can some
instruction affect the stack differently depending on their arguments?
I guess the invokes?  But is current implementation always correct????
if instruction stack size depends on arguments, our implementation is
incorrect...  If these instruction may be inlined... oh oh


INVOKE*
MULTIANEWARRAY
affect the size depending on their args...

ok because marked as uninlineable...

TODO: store how instructions that affect the stack in a fix way in
vm->instructions[i]...  Use a dummy value (ex: MAX_INT) if does it
variably (like the INVOKE etc...).

stack_offsets_diff will be no longer required.  Also, can use an:
assert(inst->stack_delta != MAX_INT) as safety check when inlined
stack offsets array is constructed.

---------------------------







makeCheckpoint (see comment in code)


Fixed: maxUsed not resetted...

Fixed:
If CFLAGS variable defined (even to empty) when inlined/ppc is built,
SableVM will segfault.


Fixed:
_sjf_callback_NATIVE_STATIC_METHOD_compiled_code
_sjf_callback_NATIVE_NONSTATIC_METHOD_compiled_code
  Do not undo compiled code assumptions
  (init locals to NULL, sync code etc.)

Fixed:
inlined-jit-verbose (with signals for exceptions)
- sigseg for test cases...

Fixed:
jit signals does not work in debug mode (need to set env->sigsegv_expected etc.)


Fixed: (some regs was not properly freed in div implementation)
OutOfReg exceptions - on some x86 benchmarks...
compress direct/jit (does not happens if --enable-sablejit-testing???)
maybe memory corruption or "lucky" for boolean array???