AVX-512 introduced new syntax using braces for decorators.
Opmask, broadcat, rounding control use this new syntax.
http://software.intel.com/sites/default/files/319433-015.pdf
Signed-off-by: Jin Kyu Song <jin.kyu.song@intel.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
There are two instructions (VGATHERQPS, VPGATHERQD) where the only
separation between two forms is the vector length given to the vector
SIB. This means the *matcher* has to be able to distinguish
instructions by vector SIB length and the matcher only operates on the
operands and the instruction flags, not on the bytecode.
Export the vector index-ness into the operand flags and add to the
matcher.
This resolves BR 3392260.
Reported-by: Agner <agner@anger.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
It doesn't seem worth >200 lines of C and Perl to save ~50 lines in insns.dat.
In order to make this work I had to rename sbyte16/sbyte32 so that
they can take an ordinary size suffix (their size suffix was formerly
treated specially).
This fixes one disassembly bug: 48C7C000000080 disassembles to mov
rax,0x80000000, which reassembles to B800000080, which loads a
different value.
Signed-off-by: Ben Rudiak-Gould <benrudiak@gmail.com>
Acked-by: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Soon we will need to encode 512 bits values
thus there is no space left in our opflags_t
which is 32 bitfield.
Extend it to 64 bits width.
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
This implements the mechanism for XACQUIRE/XRELEASE. It does not
include the necessary annotations in insns.dat.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
insn->prefixes might contain not only values from
'enum prefixes' but from 'enum reg_enum' as well so
make it generic 'int' instead.
This calms down the compiler about enum's mess and
eliminates a wrong assumption that we always have
values by particular type in this field.
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
For big numbers it's easier to estimate bytes count
if they are written in hex notation. Let it be so.
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Handle immediate-size optimization for "mov r64,imm" -- reduce it to
"mov r32,imm32" or "mov r64,imm32" as appropriate.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Two fixes:
1. Optimization of [bx+0xFFFF] etc
0xFFFF is an sbyte under 16-bit semantics,
so make sure to check it right.
2. Don't optimize displacements in -O0
Displacements that fit into an sbyte or
can be removed should *not* be optimized in -O0.
Implicit zero displacements are still optimized, e.g.:
[eax] -> 0 bit displacement, [ebp] -> 8 bit displacement.
However explicit displacements are not optimized:
[eax+0] -> 32 bit displacement, [ebp+0] -> 32 bit displacement.
Because #2 breaks compatibility with 0.98,
I introduced a new optimization level: -OL, legacy.
This reverts commit 41208028ff52d190044ee7532bf14c5aca0f899a.
|
| From: "H. Peter Anvin" <hpa@zytor.com>
|
| The tests for overflow_signed() are wrong too. Those are relevant for if a
| warning should be issued, but not for how the value should be encoded in
| the byte stream.
|
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Instead of manipulating stdscan buffer pointer directly
we switch to a routine interface.
This allow us to unify stdscan access: ie caller should
"talk" to stdscan via stdscan_get/set routines.
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
We should use overflow_ helpers before check if an immediate
operand fit a type range, otherwise we may loose high bits.
For example when we assemble the following instruction
imul eax,eax,0x10000
with -Ox passed we optimze it up to imm8 though it must be imm32.
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Consolidate I_none opcode to be used everywhere
instead of mix (-1,I_none).
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
We never invoked the parser with anything but the default definitions
for evaluate and efunc. Unlike the backend, though, we sometimes
invoke parse_line() with redefine_label() instead of define_label().
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
We never set ofmt and errfunc to anything but the global values.
Dropping them from the label definition function command line
simplifies the code.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
*To the best of my knowledge*, we now have authorization from everyone
who has significantly contributed to NASM in the past. As such,
change the license to the 2-clause BSD license.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Add copyright headers to the *.c/*.h files in the main directory. For
files where I'm sure enough that we have all the approvals, I have
given them the 2-BSD license, the others have been given the "LGPL for
now" license header. Most of them can probably be changed after
auditing.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
WAIT is technically an instruction, but from an assembler standpoint
it behaves as if it had been a prefix. In particular, it has to be
ordered *before* any real hardware prefixes.
Users who wish to control the level of optimization can
continue to specify -O0, -O1, or -Ox,
where x can be the letter itself, or any number > 1.
However, even with optimization turned off,
NASM will always make enough passes to resolve
forward references. As a result, INCBIN is now the only
item left in the critical expressions list, although TIMES
still has its own constant value check.
Make strings a proper, first-class token type, instead of relying on
the "TOKEN_NUM with tv_charptr" hack. Only convert a string to a
number if requested in an expression context; this also makes it
possible to actually issue a warning when it overflows.
In the prior code, enabling optimization effectively diabled
critical expression checking, because the optimization passes
took place before checking was enabled.
Clean up remaining build warnings. None of this should affect code
operations. The only warnings which were actually relevant might have
been the ones in ldrdf.c, but it's not clear if anyone ever uses that.
Permit opcode names to be used as labels if and only if they are
succeeded by a colon. Opcode names occurring when parsing expressions
are all treated as labels; a leading colon occurred when parsing an
instruction forces a parser restart with the instruction forcibly
treated as an identifier.
Use a 32-bit limb size ("like a digit, but bigger") for floating-point
conversion. This cuts the number of multiplications per constant by a
factor of four.
This means supporting fractional-limb-sized numbers, so while we're at
it, add support for 8-bit floating point numbers (apparently used in
graphics and in audio compression applications.)
