Also add macros which define the old names to the new names for
backwards compatibility.
This change also takes the opportunity to rename the M5_ATTR_PACKED
macro to GEM5_PACKED, dropping the ATTR. None of the other attribute
based macros bother to specify that they stand for an attribute, making
this one inconsistent and overly verbose.
Similarly, it renames M5_NODISCARD to GEM5_NO_DISCARD to be consistent
with other macros which separate out individual words like GEM5_VAR_USED
and GEM5_NO_INLINE
Change-Id: I22d404492faf28b79a8247f869f14af21c9cf967
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/45230
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Rather than have a function with a if which checks against each
supported size of a type T which calls the right version of
swap_bytesNN, this change uses SFINAE and type traits to select the
function, and adds the requirement that the type being swapped is
actually convertible to the unsigned integer type of the appropriate
size.
This has the nice benefit of also avoiding hard coding which sizes can
be supported by the base swap_bytes, although that's probably not much a
limit in practice, avoids forcing all types to be convertible to all
sizes of uint*_ts for unused branches of the if, and removes a
dependence on base/logging.hh for the panic.
Change-Id: I2267f69a2868fcba2318c7562a49d27fef785395
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/45286
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
This should be either the fixed index if there are no predicated
operands (ie operands which may not be used), and an auto incrementing
index otherwise. This is still not bulletproof since the auto
incrementing index is just code which ++-es the index, and so the index
will be different and incremented each time that value is substituted
in.
Also, the mixture of predicated operands and the vector operands is
broken and will not generate compilable code, but I'm not going to try
to fix that here.
Change-Id: I1ceae519649762e54eaa019610e51bb8c21d28d6
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/42970
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
These are currently only used in the reg, float and media ops, but
should be expanded to the other types eventually. This is part of an
attempt to use common machinery for the microops instead of having
each type recreate the mechanisms it needs locally.
Change-Id: I67f521b6b161310ba50a30ac4c73305cc75a63d5
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/42969
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Instead of having a bunch of implicit integer registers which are
identified only by number, except in the ISA description where there are
operands hard coded to certain numeric offsets into the implicit
registers, and having an independent count of those implicit registers,
just create a set of named indices for the registers we need.
Redefine NUM_INTREGS to be the total number of registers, and
NUM_ARCH_INTREGS to be the number of architectural integer registers.
Instead of using NUM_INTREGS as a coincidental short hand for the first
microop register (since that comes after the architectural ones), and
implicitly knowing that that's the zero register, define an INTREG_T0
constant which at least makes it clear what register we're talking
about, even if it's not clear that the semantics of that register make
it mostly a placeholder.
Change-Id: I5fa41169b9619ea68a50d6d5241ff9a07440bceb
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/42523
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
A comment in that section of SConstruct talked about how ParseConfig
couldn't understand -Xlinker and -export-dynamic, and how that meant
they needed to parse the output of python-config manually. Some
searching around turned up this:
https://bugs.python.org/issue36508
which seems to suggest that exactly those two options weren't actually
supposed to be returned by python-config in the first place, and are
really for python's own internal build.
This change switches over to using ParseConfig, but to avoid bad options
sneaking through, it uses an option argument to ParseConfig to intervene
before the options are consumed and filter out everything except
required libraries, and library and header search directories.
Change-Id: I1a50eb030019a447b919381c32168fb3bb27b89b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/40971
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Reviewed-by: Bobby R. Bruce <bbruce@ucdavis.edu>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
These are HDF5, PNG, FENV, and TUNTAP support, all of which add
capabilities to gem5 which can be ignored if not wanted. It could be
argued that FENV changes behavior because it makes setting the FP
rounding mode work or not as used by SPARC, but since the difference is
trivial and in a niche area, that (along with the other options) doesn't
seem to justify having a top level control in the build system.
Since these are no longer options which say whether to *use* a
particular feature, and are instead flags which say whether we *have* a
particular feature, change their names from USE_* to HAVE_*, to stay
consistent with other variables.
Most of the remaining USE_* flags, KVM, FASTMODEL, SYSTEMC, and
(indirectly) USE_PYTHON, toggle on and off major systems which can have
a significant effect on boot time, or, in the case of FASTMODEL, even
consume external resources which may not be available and which may
break the build.
USE_POSIX_TIMER was also left alone since it selects between two
implementations of some functions. By forcing it to be on or off
depending on the host, we would be forcing some code to be excluded in
either case. That would make that other code impossible to test without
hacking up scons or modifying the host machine.
Change-Id: I0b03f23e65478caefd50cd3516974386e3dbf0db
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/40964
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Re-upload of https://gem5-review.googlesource.com/c/public/gem5/+/41214
This provides a layer of indirection where the rounding mode
setting/getting code will do nothing if fenv.h isn't available. At build
time, if fenv.h can't be found, a warning is printed.
Also, the include for fenv.h was guarded in the includes in the ISA
header, but the functions from it weren't guarded in the actual code.
Finally, the code was setting the rounding mode, but not setting it
back. That would mean running these instructions would set the rounding
mode in gem5 as a whole, affecting its other behaviors and any other
instructions that might expect the default rounding mode.
Change-Id: I3ff2b97189487579554aae890e14889bd63461d2
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/44085
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Compound debug flags are intended to be a way to enable or disable a
group of simple debug flags at once, so that you don't need to enumerate
every more specialized flag in an area to get a broad amount of
debugging, nor do you give up the ability to select a general area
easily by making more specific flags.
It doesn't, however, make a lot of sense to check the value of a
compound debug flag, since it could be enabled but then have individual
subflags disabled. Exactly whether that means the compound flag should
be enabled or not is not clear, and figuring it out takes a fair amount
of work since each member simple flag needs to be visited.
Also, by having different behavior depending on the flag type, the
"enabled" method needed to be virtual.
This change eliminates the virtual method and moves the _tracing bool
member into the base class. If a subclass (only SimpleFlag currently)
wants to start or stop tracing based on itself, it should set or clear
this flag. Also, the "enabled" method has been renamed to "tracing",
since that's actually what it tracked. Being enabled by itself is not
sufficient to be tracing since there is also a global enable.
Finally, rather than duplicate the logic to convert a flag to bool in
the python wrapper, we can just use a cast to bool and take advantage of
the version in C++.
Change-Id: I3dc64c2364f0239294093686ddac6fcc8441f306
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/45007
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Instead of gating the behavior of DTRACE based on TRACING_ON in the
preprocessor, move it to C++. Beyond being a little simpler, this
ensures that the value of Debug::x is always valid (the proper header is
included, x is spelled correctly, etc) even if TRACING_ON is false.
Change-Id: Ie0085c0f8753ad5283ef1850d493706b977c21a8
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/45006
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
The code in the body of a DPRINTF will always be compiled, even if it's
disabled. If TRACING_ON is false, the if around it will short circuit to
false without actually running any code to check the specified
condition, and the body of the if will be elided by the compiler as
unreachable code.
This creates a more consistent environment whether TRACING_ON is on or
not, so that variables which are only used in DPRINTF don't have to be
guarded by their own TRACING_ON #ifs at the call site. It also ensures
that the code inside DPRINTF is always checked to be valid code, even if
the DPRINTF itself will never go off. This helps avoid syntax errors,
etc, which aren't found because of the configuration of the build being
tested with.
Change-Id: Ia95ae229ebcd2fc9828f62e87f037f76b9279819
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/44988
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
GEM5_DEPRECATED(message) expands to the [[gnu::deprecated(message)]]
attribute on gnu compatible compilers and marks an entity (function,
variable, etc) as deprecated. The msg parameter should be used to tell
the user what they should do to move away from the deprecated entity.
GEM5_DEPRECATED_MACRO(name, definition, message) is used inside an
expression-like macro with definition "definition" to mark that macro as
deprecated with message "message". For instance, if there were macros
like this:
#define SUM(a, b) ((a) + (b))
#define ONE 1
We could mark that as deprecated like so:
#define SUM(a, b) GEM5_DEPRECATED_MACRO(SUM, (a) + (b), \
"The SUM macro is deprecated, use the + operator instead.")
#define ONE GEM5_DEPRECATED_MACRO(ONE, 1, "Use the literal 1.")
Note that this macro should not be used for macros which, for instance,
declare variables, since it assumes the body of the macro expands into
an expression and not a statement or statements.
Change-Id: I58f5e834cfeeb23e37a6548433dfe7e4f695a5ec
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/45086
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
These would extract particular numbered source or destination registers
from an instruction and feed them into a printReg method. Now that the
register operands know how to print themselves with printReg directly,
these methods are no longer used.
Change-Id: I4244cb2ee97015f51f4783f377282e2a1db5d69b
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/42363
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
These methods had looked up the register index using an index into the
src or dest index arrays. This level of indirection is less efficient
than using the index itself (which we know already), and also requires
that the array is layed out like how we think it is.
Before:
array idx => reg idx => folded?
After:
reg idx => folded?
Change-Id: Ice2262250ff77582ba5530fc52e6738ba1bebe18
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/42356
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
These were a weird holdover from when register indices were all squished
into a single scalar value, offset based on the register type. When that
change happened, the person who made it misunderstood what the
InstRegIndex type was for, and decided to build RegId into it for some
reason. The only purpose of InstRegIndex is to make sure everybody
agrees that the value being passed around is going to be used as a
register index and not a literal value, or vice versa. There is no type
associated with it as far as floating point, integer, or misc registers.
That gets applied at a different step, and actually can't be part of
InstRegIndex since the same base class may need to hold register indices
that are going to be treated as integer or floating point depending on
the subclass using them.
Also, since the values of the various constants in the DepdenceTags enum
where never actually added into register indices in the first place, the
code in the InstRegIndex constructor would never do anything. All
registers would be arbitrarily sorted into Int, FP, etc, and then when
they actually went to be used the category would be thrown away.
Change-Id: I8c4e8d6e9cdb53e298c00ad2f56e8c7304e51e96
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/40339
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Rather than force all x86 microops to have one destination and two
sources, the second of which is a register or immediate, make it
possible for these microops to pick any combination of those elements by
modularizing the operand aspects of the base class. This prevents having
a bunch of extra parameters and members of the classes, or having a lot
of explicitly laid out classes with various combinations.
This also improves the accuracy/usefulness of Exec traces since register
types and therefore names will be determined correctly.
Also, there was a branchTarget override added to all register uops which
would be used when the macroop was an direct control transfer
instruction. The assumption was that the immediate value of the whole
instruction would be the PC offset, which is not necessarily correct but
is probably a fairly safe assumption.
This override was only provided for all *register* uops though, and
there's nothing saying the last uop in a branch instruction has to be a
a register uop. This change moves that override to the uop base class so
that *any* uop can be last in the macroop and still support branchTarget
correctly (or at least as correctly as a register uop would).
Change-Id: I9d42d22609d511fa757a784c04a5a9874beca479
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/42343
Reviewed-by: Gabe Black <gabe.black@gmail.com>
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
