The -ffunction-sections and -fdata-sections, if supported, can enable
the linker to do some amount of dead code elimination. Basically a
very basic form of LTO.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Default to -O3 instead of -O2. The code gets somewhat bigger, but
nasm is full of tiny functions called from all over the place.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Put in the necessary machinery to compile with gcc link-time
optimization. This means compiling and linking with -flto, as well as
using gcc-ar and gcc-ranlib instead of the normal tools.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Most of the time we don't need the dumping code, so move it to a
separate C file to the linker can exclude it.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
There is no fundamental reason for all objects that don't contain a
main() function to not be part of libnasm.a; this allows the linker to
do its job optimally, especially in the presence of debugging code
which may not be needed under normal conditions.
If we do end up with function name conflicts the library might have to
be split, but it would be better to simply avoid that case.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
There is absolutely no reason to call reloc_value() twice while
processing an immediate. Doing so is both expensive and unnecessary.
Make some more deliberate decorations to try to avoid miscompilation
on OpenWatcom, but I'm not sure this will help enough.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
HAVE_DECL_* are 0/1 not ifdef; use HAVE_DECL_STRNLEN to see if we need
to declare this, lest stdlib/strnlen.c fails to compile.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Move memory-mapping functions from file.c into a separate mmap.c.
This will be cleaner especially once (if) we end up doing a Windows
implementation, which is likely to look entirely different.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Use PRId64 instead of %ld to print a 64-bit number. This is debugging
code, but as it is frequently useful I included it in mainline code
and, well, then it has to compile everywhere...
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Avoid type promotion due to ?:, and put in a comment to explain what
the heck the purpose is of wrapping default in a macro.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Use sizeof() to prevent the compiler from invoking
watcom_switch_hack() unnecessarily. Hopefully the optimizer would be
smart enough to recognize this inherently, but this is an old compiler
we are dealing with, so make life a little easier for it.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
It is kind of sad and impressive at the same time how little of the
2.13 changes are actually user-visible at this point.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The UD0 and UD1 opcodes are now officially documented, with UD1
officially documented to take a modr/m. Still permit the "UD2B" and
argument-less aliases, but not as preferred.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
We can remove OUT_ADDRESS relocations for absolute addresses (NO_SEG),
but for OUT_RELADDR relocations we can remove them if they point into
*our own segment*, not NO_SEG.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
For expressions like [foo - $] or [bar - $$] our relocation base is
not the same as the end of the instruction. Make that explicit.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Handle, hopefully correctly, self-relative expressions (that is,
expressions of the form X - Y where Y is a symbol in the current
segment, possibly $ or $$) used as offsets or immediates, as opposed
to arguments to Dx statements (which have already been supported for a
while.)
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
OpenWatcom still doesn't have proper support for 64-bit switch
statements. Hack around it in a truly vile way.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
This fixes disassembly of instructions like jmp/call when target address
is larger than 2^32-1, and also printing of current address when it's
large.
After this change ndisasm still assumes that the files to disassemble
will never reach 4GiB: only offsets are made 64 bit, but not lengths.
https://bugzilla.nasm.us/show_bug.cgi?id=3392349
Signed-off-by: Ruslan Kabatsayev <b7.10110111@gmail.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Use an ugly hack to make nasm_delete() side effect free. This assumes
all pointers have the same internal NULL pointer representation as
void *, however, we already assume zero-initialized memory will
represent a NULL pointer, so hopefully this is okay on any platform we
actually care about.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
I have not figured out a way to make nasm_delete() side effect safe
without using compiler-specific hacks, which would defeat the whole
purpose.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
We do have a number of places where we had problems doing things like:
memset(foo, 0, sizeof foo) instead of sizeof *foo. Add a helper macro
nasm_zero() to the list of (sadly not yet well used) pointer-safe
helper macros.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
http://www.drdobbs.com/compile-time-assertions/184401873 describes a
number of possible implementations of static_assert() on compilers
that do not support it natively. Use their best recommendation.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
"Assertion failed" is likely to be redundant with static_assert().
__attribute__((error)) is only guaranteed to work while optimizing, so
do not use it unless __OPTIMIZE__ is defined.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Type long is not necessarily long enough to be valid to hold a 64-bit
number (e.g. 32-bit platforms or Win64), however, the output variable
newaddr is uint64_t.
Cc: Martin Lindhe <martin-commit@ubique.se>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
In commits e1f985c167 and f7606613d0
we reordered macros handling, but ofmt specific
macros are to be handled after standart ones
are processed.
In particular __SECT__ handling must not change,
so the order of inclusion does matter.
https://bugzilla.nasm.us/show_bug.cgi?id=3392376
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
fixes pvs-studio error 'V530 The return value of function 'freopen'
is required to be utilized.'
Signed-off-by: Martin Lindhe <martin-commit@ubique.se>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
fixes pvs-studio error 'V707 Giving short names to global variables
is considered to be bad practice. It is suggested to rename 'sb' variable.
gorcunov@: Use local non-static variable.
Signed-off-by: Martin Lindhe <martin-commit@ubique.se>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
fixes pvs-studio error 'V707 Giving short names to global variables
is considered to be bad practice. It is suggested to rename 'i' variable.
gorcunov@: Simply define it as local variable.
Signed-off-by: Martin Lindhe <martin-commit@ubique.se>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
