Add the %map() function which can apply arguments to a macro from a
list.
Allow the user to specify the desired radix for an evaluated
parameter. It doesn't make any direct difference, but can be nice for
debugging or turning into strings.
As part of this, split expand_one_smacro() into two parts: parameter
parsing and macro expansion. This is a very straightforward splitting
of two mostly unrelated pieces of functionality.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
NASM would try to "eat the comma token" in db expressions, even for
cases where the token was not a comma. Fix that and error out
properly.
To give better error messages, track where in the input string a token
starts or ends. This information is only valid as long as the input
string is kept, but that is just fine for error messages during
parsing.
Reported-by: Peter Cordes <pcordes@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
For VEX instructions created *after* the corresponding EVEX
instructions, we need the user to either explicitly declare them {vex}
or specifying "cpu latevex".
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Instead of handling conditional instructions ad hoc, generate
individual instruction patterns as normal. This simplifies the code
and makes CMPccXADD support simpler (otherwise it would be necessary
to hack in the handling of a condition code in the middle of an
instruction.)
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The FP16 patch had a case of bit overlap. Clean up the handling of
broadcast flags a little in the process.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Add support for AVX512-FP16 instructions and the associated
handling. Allow "mapN" syntax as well as "mN" syntax to match the
documentation.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
while (nbs--) { ... } ends with nbs == -1. Rather than a minimal fix,
introduce mempset() to make these kinds of errors less likely in the
future.
Fixes: https://bugzilla.nasm.us/show_bug.cgi?id=3392815
Reported-by: <13579and24680@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Add a {rex} prefix to force REX encoding (typically a redundant 40h
prefix).
For prefix parsing, we can use t_inttwo to encode the prefix slot
number.
Give more verbose error messages for encoding mismatches.
These two symbolic names were reversed. In practice, the code uses '('
and ')' instead, so it wasn't a problem, but still very confusing.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
We need the ability to produce consistent output for our own tests,
anyway, so make this a user-accessible feature. This was requested in
BR 3392635.
This obsoletes the NASM_TEST_RUN environment variable; simply use the
normal NASMENV environment variable instead.
The .obj tests in travis needed to be updated in order to remove the
rather pointless suffix " CONST" from the NASM signatures.
Reported-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
NASM 2.15.04
Conflicts:
asm/listing.h
asm/pptok.pl
asm/preproc.c
version
This doesn't pass travis test 3392711, which is using an extremely odd
construct of %?? in the middle of an argument sequence for an smacro
while not being in a macro itself, and expecting it to expand to the
macro name. This seems to *really* confuse the master branch.
Resolve this later...
At least DWARF can encode C-style macros. In doing so, it wants the
file include hierarchy, so give the debug format backend the option of
receiving that information from the preprocessor.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
BNDMK, BNDLDX, and BNDSTX are split-SIB (MIB) instructions, but do
*not* require a SIB encoding. However, TILELOAD* and TILESTORE* *do*
require a SIB in all cases. Split the MIB flag into MIB (split
address) and SIB (SIB required) flags.
This fixes travis test mpx.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Add a new macro vprintf_func() for vprintf-style functions, and add
printf_func() and vprintf_func() attribute arguments whereever
meaningful.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Support generating bfloat16 constants. This is a bit awkward, as "DW"
already generates IEEE half precision constants; therefore there is no
longer a single floating-point format for each size. This requires
some replumbing.
Fortunately bfloat16 fits in 64 bits, so support generating them with
a macro that uses __?bfloat16?__() to convert to integers first before
passing them to DW.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The different token codes between the preprocessor and the assembler
is a completely unnecessary headache. Furthermore, lumping all the
operators under TOK_OTHER in the preprocessor causes a whole bunch of
unnecessary headaches.
In combining them, the only tricky part is that PP_CONCAT_MASK() is no
longer usable, as the range of token codes is too large. Replace with
dedicated category masks.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
ERR_PASS1 only remains in three places:
a. Unterminated %! string, an error
- Should be signalled no matter which pass it is encountered in
b. Two cases of map file problems in outbin
- The buffered warning system take care of that issue
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Fold the prepreprocessor and the nop preprocessor into the main
preprocessor. This means handling # cpp-like lines and TASM
compatibility tokens in the preprocessor proper, but that is really
not very hard to do.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Signed and unsigned are really two flags; might as well allow this
field to contain additional flags.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
If the segment number changes, we also need to invoke dfmt->linenum(),
as a .nolist macro may end up emitting to more than one section.
This also adds the source location explicitly to the output data
structure; the cost for that is minimal, and will enable a more
sophisticated debug backend to receive the entire data structure in
the future.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
list_reverse() used "next" as an argument, while also needing to
refer to the structure field "next". Furthermore, the two temp
variables can be made generic by making them void *, and as this is
not a loop construct this is doable by declaring them inside the macro
loop.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Sometimes we want to search for an exact key only, and reject the case
when tree->key < key. Add rb_search_exact() for this purpose, rather
than forcing the caller to perform the comparison in open code.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Collect macro call/nesting information for the benefit of the debug
back end. So far, the only backend for which this is provided is the
debug back end, to show what information is present.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Add operations to get the first and last entry in the tree,
respectively. Searching for 0 or ~UINT64_C(0) is not sufficient in the
presence of duplicated keys, and is more inefficient anyway.
rb_first() followed by rb_next() until NULL, or equivalently rb_last()
followed by rb_prev() until NULL, can be used to walk the tree in key
order (ascending or descending), including all duplicate key
entries.
Since this is a *threaded* tree now, this walk can safely free entires
as it goes along, as long as the whole tree is destroyed; once any one
entry has been freed, the tree is no longer valid for anything other
than proceeding with the same tree walk.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Change the left-leaning red-black (LLRB) trees into left-leaning
threaded red-black trees. Instead of NULL pointers at leaf nodes, use
the otherwise unused field as a pointer to the lexical predecessor
(left) or successor (right). This allows fast previous/next
interator operation without needing to keep track of the root of the
tree at all times.
The additional metadata that needs to be kept can be done for "free"
simply by changing "bool red" into a flag field.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Update the table used by pass_type() to give the name of the pass
type. It was not updated properly after PASS_PREPROC was added.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Still technically not defined behavior, but this *should* work on all
reasonable or semi-reasonable systems.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
If the C compiler has typeof(), we can use a safer implementation of
the offsetin() macro by using typeof(). Since typeof() isn't standard
(yet?) autoconf has to test for it.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
1. Error messages would issue with the line number of %endrep.
2. Debug line information would ignore both macros and reps.
This is doubly wrong; macros are semantically equivalent to
inline functions, and it is expected that debuggers trace
into these functions.
These changes finishes the last parts of moving all responsibility for
the listing enable/disable into the preprocessor, so remove the
way over-complicated macro inhibit facility from the listing module
entirely.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
In preproc-only mode, we only ever execute a single pass, so we need
to still issue error messages created during that pass, otherwise we
don't even generate %warning or %error messages...
Reported-by: Jason Hood <jadoxa@yahoo.com.au>
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Move the safe_alloc declaration after the * symbol to be consistent
with all other functions with this attribute. Just in case it matters
somewhere...
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
The autoconf process automatically generates macros for function
attributes, including empty placeholders. Said empty placeholders also
propagate automatically into config/unconfig.h for the compilers which
don't support autoconf.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
clang, unlike gcc, will warn on inline functions which are
unused. This can happen if a function is either intended to be used in
the future, or it is only used under certain config options. Mark
those functions with the "unused" attribute; not only does it quiet
the warning, but it also documents it for the user.
Shuffle around the warning options in configure and add a few more
that are specific to clang.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Don't issue smacro expansion warnings until we are sure we are
actually *done* with the smacro expansion. The last pass of
expand_smacro_noreset() gets to commit warnings.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
Set an expression descent limit to 8192, which is more reasonable to
expect to work on most platforms. Furthermore, if getrlimit() exists,
then try to use it to see if we need to further limit the size.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
If an include file exists, but cannot be opened, that is still a
critical error.
However, downgrade this from a fatal to a nonfatal error. There really
isn't any reason to stop cold here.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com>
