These are similar to the structures TLM's DMI mechanism uses. Instead
of having an invalidation broadcast which propogates backwards up the
port hierarchy, this mechanism tracks a set of callbacks which are
triggered when a back door is invalidated to let other holders clean
up their bookkeeping.
Change-Id: If24489258dcaee14d7b6e5b996dfb1c2636f26ab
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17589
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Gabe Black <gabeblack@google.com>
For atomic RMW instructions that go directly to memory, we want to put
them on the write queue instead of the read queue. Swap the if/else
condition to accomplish this.
Note: This is ignoring the read latency of the RMW, but these
instructions should usually be handled in caches anyway.
Change-Id: I62dbfff3a16ac470f1ebdb489abe878962b20bb6
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17828
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Clang with -Wconstant-conversion is _very_ restrictive on casting.
The shift operator results in an incorrect promotion.
This patch add a compile-time static cast that remove the error
when clang is used.
Change-Id: I3aa1e77da2565799feadc32317d5faa111b2de86
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17308
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
These objects expose a standard TLM initiator or target socket with
width 64, and a gem5 slave or master port. What goes in one type of
port comes out the other with the appropriate conversion applied.
Change-Id: I65e07f746d46d3db0197968b78fffc5ddaede9bf
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17232
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
This is a slightly mangled version of the existing bridge code in
util/tlm/src/. The changes fix some small style issues, change to gem5
specific include paths, and removes the Gem5SimControl code. That code
coordinates gem5 with the external systemc kernel, and in this usage
there's no external kernel.
The code imported here compiles, but it isn't yet expected to work.
Change-Id: I9c593a52e2554534720d21cd31a03e543ad897ad
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17231
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Current operand generation is not providing VecElems with the right
vector index and element index.
The bug was covered when registers were 128 bit wide, but with SVE we
have augmented the vector register size and the bug has been exposed.
E.g. With dest = 2,
FpDestP2 = (vec_index = 0, elem_index = 4)
whereas it should be
FpDestP2 = (vec_index = 1, elem_index = 0)
Change-Id: Iad02fb477afd0d3dd3d437bf2ca4338fbd142107
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Ciro Santilli <ciro.santilli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17710
Get rid of comments which just restate the code, get rid of redundant
"virtual" keywords, add "override"s, fix style, and get rid of
xbar::init which was empty and hiding the parent class init.
Change-Id: I8ce20abee340baa88084d142f2fb8c633ee54ba9
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17592
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
The erase() method only accepts regular iterators which is consistent
with the normal STL map, but the existing find() only returns const
iterators. The STL container can return either depending on if "this"
is const.
Unfortunately there isn't a great way to have only one find
implementation which returns the right type of iterator under the right
conditions. Also, it's not possible to turn a const_iterator into an
iterator, but it is possible to go the other way. This change
duplicates very short functions which return iterators, and for find
does the only thing I could find which avoids having to copy that
whole large function.
Change-Id: I2f789b5d0881feb9adff9978bd40e31731c6a688
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17588
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
When adding multiple SimObjects to --debug-ignore, either separating the values with
a colon or adding multiple --debug-ignore flags, the previous code only ignored the
last SimObject in the list. This changeset adds and uses new `ObjectMatch::add` and
`Logger::addIgnore` methods to make the functionality of the flag consistent with
its description.
Change-Id: Ib6967a48611ea59a211f81af2a970c4de429b1be
Signed-off-by: Isaac Sánchez Barrera <isaac.sanchez@bsc.es>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17488
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
The Power ISA specification lists the Program Counter (PC) and
the Next Program Counter (NPC) registers as Current Instruction
Address (CIA) and Next Instruction Address (NIA). This applies
the ISA naming convention for these two registers.
Change-Id: I8b9094ab1c809f4dfdb4d7330c17f360adf063e9
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/16603
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Currently, 'sq' and 'uq' are used to represent signed and
unsigned doublewords respectively. Since all recent Power
ISA specifications list 128-bit quadwords as a valid data
type, it may be misleading to use the current terminology
in case support for such operands are added in the future.
So, to simplify this, 'sd' and 'ud' are used to represent
signed and unsigned doublewords respectively.
Change-Id: Ie7831c596fc8f9ddfdf3b652c37cfe26484ebe01
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/16602
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
PerfKvmCounter::attach fails if the user doesn't have privileges to make
the perf_event_open syscall. This is the default privilege setting since
kernel 4.6. I've seen some users in the mailing list resort to running
as root; changing the perf_event_paranoid setting is an alternative.
Change-Id: I2bc6f76abb6e97bf34b408a611f64b1910f50a43
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17508
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Reference:
Stephen Somogyi, Thomas F. Wenisch, Anastasia Ailamaki, and
Babak Falsafi. 2009. Spatio-temporal memory streaming.
In Proceedings of the 36th annual international symposium on
Computer architecture (ISCA '09). ACM, New York, NY, USA, 69-80.
Change-Id: I58cea1a7faa9391f8aa4469eb4973feabd31097a
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/16423
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
This retrieves ports using the getPort method, and connects them using
the bind method on the ports themselves. Any smarts as far as what type
of peers are allowed to connect or how they connect is left up to the
individual bind methods.
Change-Id: Ic640d1fce8af1bed411116e5830edc4a8a0f9d66
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17039
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
These are now pure virtual methods which more specialized port
subclasses will need to implement. The SlavePort class implements them
by ignoring them and then providing parallel functions for the
MasterPort to call. The MasterPort's methods do basically what they
did before, except now bind() uses dynamic cast to check if its peer
is of the appropriate type and also to convert it into that type before
connecting to it.
Change-Id: I0948799bc954acaebf371e6b6612cee1d3023bc4
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17038
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
The connectPorts function currently checks if *either* of the peers in
a port connection are a MessageBuffer, and if so will ignore the
connection. This CL changes that || into a && so that *both* of the
peers need to be a Ruby types (either a MessageBuffer or Network) for
the connection to be ignored. That makes it easier to contain that
abnormal behavior to those types instead of having it apply even when
other types of port owners are involved.
Unfortunately the number of interesting Ruby types is unbounded, but
these are the types with ports as of today. This mechanism will
hopefully be replacedall together so this should be a temporary issue.
Change-Id: I140498770e5d37eb2abd3d99261d47e111f1c8ab
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17031
Reviewed-by: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
When setting up a SimObject's Param structure, gem5 will autogenerate
a header file which attempts to declare the SimObject's C++ type. It
has had at least some level of sophistication there where it would
pull off the namespaces ahead of the class name and handle them
properly, but it didn't know how to handle templates.
This change improves that handling in two ways. First, it adds a new
magical SimObject attribute called 'cxx_template_params' which is used
to specify what the template parameters are as a list. For instance, if
your SimObject was a template which took an integer constant as its
first parameter and a type as its second, this attribute could look
like the following:
cxx_template_params = [ 'int FOO', 'class Bar' ]
Importantly, if there are any default values for these template
parameters, they should *not* be included here, they should be
specified where the class is later defined.
The second new mechanism is to add an internal CxxClass in the
SimObject.cxx_param_decl method. This class accepts the class signature
in the cxx_class attribute and the cxx_template_params and does two
things. First, it strips off namespaces like in the old implementation.
Second, it extracts and processes any template arguments attached to
the class. If these are constants (as determined by the contents of
cxx_template_params), then they are stored verbatim. If they're types,
then they're recursively expanded into a CxxClass and stored that way.
Note that these are the *values* of the template arguments, where as
cxx_template_params lists the *types* and *names* of those arguments.
In our earlier example, if cxx_class was:
cxx_class = 'CoolClasses::ClassName<12, Fruit::Apple>'
Then CxxClass would extract the namespace 'CoolClasses', the class
name 'ClassName', the argument '12', and the argument 'Fruit::Apple'.
That second argument would be expanded into a CxxClass with the
namespace 'Fruit' and the class name 'Apple'.
Importantly here, because there were no default arguments given in
cxx_template_params, all "hidden" arguments which would fall through
to their defaults need to be fully specified in cxx_class.
The CxxClass has a method called declare() which uses the information
extracted earlier to output all of the "stuff" necessary for declaring
the given class, including opening any containing namespaces and
putting template<...> ahead of the actual class declaration with the
template parameters specified.
If any of the template arguments are themselves CxxClass instances,
then they'll be recursively declared immediately before the current
class is.
An alternative solution to this problem might be to include the header
file which actually defines the cxx_class type to avoid having to
come up with a declaration. Unfortunately this doesn't work since it
can set up include loops where the SimObject C++ header file includes
the param header to get access to the Param type, but that includes
the C++ header to get access to the SimObject type.
This also makes it harder for SimObjects to refer to each other, since
they rely on the declaration in the params header files when declaring
a member pointer to that type in their own Param structures.
Change-Id: I68cfc36ddff6d789eb4cdef5178c4619ac2cc8b1
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17228
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Gabe Black <gabeblack@google.com>
