In most ISAs except MIPS and Power, this was implemented as
inst->advancePC(). It works just fine to call this function all the
time, but the idea had originally been that for ISAs which could simply
advance the PC using the PC itself, they could save the virtual function
call. Since the only ISAs which could skip the call were MIPS and Power,
and neither is at the point where that level of performance tuning
matters, this function can be collapsed with little downside.
If this turns out to be a performance bottleneck in the future, the way
the PC is managed could be revisited to see if we can factor out this
trip to the instruction object in the first place.
Change-Id: I533d1ad316e5c936466c529b7f1238a9ab87bd1c
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/39335
Maintainer: Gabe Black <gabe.black@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Reviewed-by: Alex Dutu <alexandru.dutu@amd.com>
The only way to allocate fixed sized arrays which will definitely be big
enough for all source/destination registers for a given instruction is
to track the maximum number of each at compile time, and then size the
arrays appropriately. That creates a point of centralization which
prevents breaking up decoder and instruction definitions into more
modular pieces, and if multiple ISAs are ever built at once, would
require coordination between all ISAs, and wasting memory for most of
them.
The dynamic allocation overhead is minimized by allocating the storage
for all variable arrays in one chunk, and then placing the arrays there
using placement new. There is still some overhead, although less than it
might be otherwise.
Change-Id: Id2c42869cba944deb97da01ca9e0e70186e22532
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/38384
Reviewed-by: Jason Lowe-Power <power.jg@gmail.com>
Maintainer: Jason Lowe-Power <power.jg@gmail.com>
Tested-by: kokoro <noreply+kokoro@google.com>
The create() method on Params structs usually instantiate SimObjects
using a constructor which takes the Params struct as a parameter
somehow. There has been a lot of needless variation in how that was
done, making it annoying to pass Params down to base classes. Some of
the different forms were:
const Params &
Params &
Params *
const Params *
Params const*
This change goes through and fixes up every constructor and every
create() method to use the const Params & form. We use a reference
because the Params struct should never be null. We use const because
neither the create method nor the consuming object should modify the
record of the parameters as they came in from the config. That would
make consuming them not idempotent, and make it impossible to tell what
the actual simulation configuration was since it would change from any
user visible form (config script, config.ini, dot pdf output).
Change-Id: I77453cba52fdcfd5f4eec92dfb0bddb5a9945f31
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/35938
Reviewed-by: Gabe Black <gabeblack@google.com>
Reviewed-by: Daniel Carvalho <odanrc@yahoo.com.br>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
There were three different StaticInst flags for memory barriers,
IsMemBarrier, IsReadBarrier, and IsWriteBarrier. IsReadBarrier was never
used, and IsMemBarrier was for both loads and stores, so a composite of
IsReadBarrier and IsWriteBarrier.
This change gets rid of IsMemBarrier and replaces by setting
IsReadBarrier and IsWriteBarrier at the same time. An isMemBarrier
accessor is left, but is now implemented by checking if both of the
other flags are set, and renamed to isFullMemBarrier to make it clear
that it's checking both for both types of barrier, not one or the other.
Change-Id: I702633a047f4777be4b180b42d62438ca69f52ea
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/33743
Reviewed-by: Gabe Black <gabeblack@google.com>
Maintainer: Gabe Black <gabeblack@google.com>
Tested-by: kokoro <noreply+kokoro@google.com>
This patch adds support for pinning registers for a certain number of
consecutive writes. This is only relevant for timing CPU models
(functional-only models are unaffected), and it is primarily needed to
provide a realistic execution model for micro-coded operations whose
microops can write to non-overlapping portions of a destination
register, e.g. vector gather loads. In those cases, this mechanism
can disable renaming for a sequence of consecutive writes, thus making
the resulting execution more efficient: allocating a new physical
register for each microop would introduce a read-modify-write chain of
dependencies, while with these modifications the microops can write
back in parallel.
Please note that this new feature is only leveraged by O3CPU for the
time being.
Additional authors:
- Gabor Dozsa <gabor.dozsa@arm.com>
Change-Id: I07eb5fdbd1fa0b748c9bdc1174d9f330fda34f81
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/13520
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Tested-by: kokoro <noreply+kokoro@google.com>
This patch enables all 4 CPU models (AtomicSimpleCPU, TimingSimpleCPU,
MinorCPU and DerivO3CPU) to issue atomic memory (AMO) requests to memory
system.
Atomic memory instruction is treated as a special store instruction in
all CPU models.
In simple CPUs, an AMO request with an associated AtomicOpFunctor is
simply sent to L1 dcache.
In MinorCPU, an AMO request bypasses store buffer and waits for any
conflicting store request(s) currently in the store buffer to retire
before the AMO request is sent to the cache. AMO requests are not buffered
in the store buffer, so their effects appear immediately in the cache.
In DerivO3CPU, an AMO request is inserted in the store buffer so that it
is delivered to the cache only after all previous stores are issued to
the cache. Data forwarding between between an outstanding AMO in the
store buffer and a subsequent load is not allowed since the AMO request
does not hold valid data until it's executed in the cache.
This implementation assumes that a target ISA implementation must insert
enough memory fences as micro-ops around an atomic instruction to
enforce a correct order of memory instructions with respect to its
memory consistency model. Without extra memory fences, this implementation
can allow AMOs and other memory instructions that do not conflict
(i.e., not target the same address) to reorder.
This implementation also assumes that atomic instructions execute within
a cache line boundary since the cache for now is not able to execute an
operation on two different cache lines in one single step. Therefore,
ISAs like x86 that require multi-cache-line atomic instructions need to
either use a pair of locking load and unlocking store or change the
cache implementation to guarantee the atomicity of an atomic
instruction.
Change-Id: Ib8a7c81868ac05b98d73afc7d16eb88486f8cf9a
Reviewed-on: https://gem5-review.googlesource.com/c/8188
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
When a thread executed an exit syscall in SE mode, the thread context
was removed immediately in the same cycle, which left inflight squash
operations and trap event incomplete. The problem happened when a new
thread was assigned to the CPU later. The new thread started with some
incomplete transactions of the previous thread (e.g., squashing). This
problem could cause incorrect execution flow for the new thread (i.e.,
pc was not reset properly at the exit point), deadlock (i.e., some
stage-to-stage signals were not reset) and incorrect rename map between
logical and physical registers.
This patch adds a new state called 'Halting' to the thread context and
defers removing thread context from a CPU until a trap event initiated
by an exit syscall execution is processed. This patch also makes sure
that the removal of a thread context happens after all inflight
transactions of the to-be-removed thread in the pipeline complete.
Change-Id: If7ef1462fb8864e22b45371ee7ae67e2a5ad38b8
Reviewed-on: https://gem5-review.googlesource.com/c/8184
Reviewed-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
This patch does a large modification of the LSQ in the O3 model. The
main goal of the patch is to remove the 'an operation can be served with
one or two memory requests' assumption that is present in the LSQ
and the instruction with the req, reqLow, reqHigh triplet, and
generalising it to operations that can be addressed with one request,
and operations that require many requests, embodied in the
SingleDataRequest and the SplitDataRequest.
This modification has been done mimicking the minor model to an extent,
shifting the responsibilities of dealing with VtoP translation and
tracking the status and resources from the DynInst to the LSQ via the
LSQRequest. The LSQRequest models the information concerning the
operation, handles the creation of fragments for translation and request
as well as assembling/splitting the data accordingly.
With this modifications, the implementation of vector ISAs, particularly
on the memory side, become more rich, as the new model permits a
dissociation of the ISA characteristics as vector length, from the
microarchitectural characteristics that govern how contiguous loads are
executing, allowing exploration of different LSQ to DL1 bus widths to
understand the tradeoffs in complexity and performance.
Part of the complexities introduced stem from the fact that gem5 keeps a
large amount of metadata regarding, in particular, memory operations,
thus, when an instruction is squashed while some operation as TLB lookup
or cache access is ongoing, when the relevant structure communicates to
the LSQ that the operation is over, it tries to access some pieces of
data that should have died when the instruction is squashed, leading to
asserts, panics, or memory corruption. To ensure the correct behaviour,
the LSQRequest rely on assesing who is their owner, and self-destroying
if they detect their owner is done with the request, and there will be
no subsequent action. For example, in the case of an instruction
squashed whal the TLB is doing a walk to serve the translation, when the
translation is served by the TLB, the LSQRequest detects that the
instruction was squashed, and as the translation is done, no one else
expect to access its information, and therefore, it self-destructs.
Having destroyed the LSQRequest earlier, would lead to wrong behaviour
as the TLB walk may access some fields of it.
Additional authors:
- Gabor Dozsa <gabor.dozsa@arm.com>
Change-Id: I9578a1a3f6b899c390cdd886856a24db68ff7d0c
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/13516
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
The value that is not initialized has a bogus value that manifests when
using some debug-flags what makes the usage of tracediff a bit more
challenging.
In addition, while debugging with other techniques, it introduces the
problem of understanding if the value of a field is 'intended' or just
an effect of the lack of initialisation.
Change-Id: Ied88caa77479c6f1d5166d80d1a1a057503cb106
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/13125
Maintainer: Nikos Nikoleris <nikos.nikoleris@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Summary: Usage of const DynInstPtr& when possible and introduction of
move operators to RefCountingPtr.
In many places, scoped references to dynamic instructions do a copy of
the DynInstPtr when a reference would do. This is detrimental to
performance. On top of that, in case there is a need for reference
tracking for debugging, the redundant copies make the process much more
painful than it already is.
Also, from the theoretical point of view, a function/method that
defines a convenience name to access an instruction should not be
considered an owner of the data, i.e., doing a copy and not a reference
is not justified.
On a related topic, C++11 introduces move semantics, and those are
useful when, for example, there is a class modelling a HW structure that
contains a list, and has a getHeadOfList function, to prevent doing a
copy to an internal variable -> update pointer, remove from the list ->
update pointer, return value making a copy to the assined variable ->
update pointer, destroy the returned value -> update pointer.
Change-Id: I3bb46c20ef23b6873b469fd22befb251ac44d2f6
Signed-off-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/13105
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
There are cases where the IEW adds a non-speculative instruction to
the IQ twice. This can happen if an instruction is flagged as
IsMemBarrier and IsNonSpeculative. Avoid adding non-speculative
instructions in the IEW to the IQ by checking if it has been added
already.
Change-Id: Ifcff676a451b57b2406ce00ed8dae19ed399515f
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Javier Setoain <javier.setoain@arm.com>
Reviewed-by: Giacomo Gabrielli <giacomo.gabrielli@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/8374
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
With the hierarchical RegId there are a lot of functions that are
redundant now.
The idea behind the simplification is that instead of having the regId,
telling which kind of register read/write/rename/lookup/etc. and then
the function panic_if'ing if the regId is not of the appropriate type,
we provide an interface that decides what kind of register to read
depending on the register type of the given regId.
Change-Id: I7d52e9e21fc01205ae365d86921a4ceb67a57178
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
[ Fix RISCV build issues ]
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2702
Mimic the changes done on the architectural register indexes on the
physical register indexes. This is specific to the O3 model. The
structure, called PhysRegId, contains a register class, a register
index and a flat register index. The flat register index is kept
because it is useful in some cases where the type of register is not
important (dependency graph and scoreboard for example). Instead
of directly using the structure, most of the code is working with
a const PhysRegId* (typedef to PhysRegIdPtr). The actual PhysRegId
objects are stored in the regFile.
Change-Id: Ic879a3cc608aa2f34e2168280faac1846de77667
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2701
Reviewed-by: Anthony Gutierrez <anthony.gutierrez@amd.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
The drain did not wait until stages were ready again. Therefore, as a
result of messages in the TimeBuffer being drain, the state after the
drain was not consistent and asserts fired in some places when the
draining happened after a stage got blocked, but before the notification
arrived to the previous stages.
Change-Id: Ib50b3b40b7f745b62c1eba2931dec76860824c71
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
This patch adds probe points in Fetch, IEW, Rename and Commit stages as follows.
A probe point is added in the Fetch stage for probing when a fetch request is
sent. Notify is fired on the probe point when a request is sent succesfully in
the first attempt as well as on a retry attempt.
Probe points are added in the IEW stage when an instruction begins to execute
and when execution is complete. This points can be used for monitoring the
execution time of an instruction.
Probe points are added in the Rename stage to probe renaming of source and
destination registers and when there is squashing. These probe points can be
used to track register dependencies and remove when there is squashing.
A probe point for squashing is added in Commit to probe squashed instructions.
This patch enables instructions in LSQ to track two physical addresses for
corresponding two split requests. Later, the information is used in
checksnoop() to search for/invalidate the corresponding LD instructions.
The current implementation has kept track of only the physical address that is
referenced by the first split request. Thus, for checksnoop(), the line
accessed by the second request has not been considered, causing potential
correctness issues.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
The Request::UNCACHEABLE flag currently has two different
functions. The first, and obvious, function is to prevent the memory
system from caching data in the request. The second function is to
prevent reordering and speculation in CPU models.
This changeset gives the order/speculation requirement a separate flag
(Request::STRICT_ORDER). This flag prevents CPU models from doing the
following optimizations:
* Speculation: CPU models are not allowed to issue speculative
loads.
* Write combining: CPU models and caches are not allowed to merge
writes to the same cache line.
Note: The memory system may still reorder accesses unless the
UNCACHEABLE flag is set. It is therefore expected that the
STRICT_ORDER flag is combined with the UNCACHEABLE flag to prevent
this behavior.
IEW did not check the instQueue and memDepUnit to ensure
they were drained. This caused issues when drainSanityCheck()
did check those structures after asserting IEW was drained.
This patch fixes the load blocked/replay mechanism in the o3 cpu. Rather than
flushing the entire pipeline, this patch replays loads once the cache becomes
unblocked.
Additionally, deferred memory instructions (loads which had conflicting stores),
when replayed would not respect the number of functional units (only respected
issue width). This patch also corrects that.
Improvements over 20% have been observed on a microbenchmark designed to
exercise this behavior.
The o3 pipeline interlock/stall logic is incorrect. o3 unnecessicarily stalled
fetch and decode due to later stages in the pipeline. In general, a stage
should usually only consider if it is stalled by the adjacent, downstream stage.
Forcing stalls due to later stages creates and results in bubbles in the
pipeline. Additionally, o3 stalled the entire frontend (fetch, decode, rename)
on a branch mispredict while the ROB is being serially walked to update the
RAT (robSquashing). Only should have stalled at rename.
As highlighed on the mailing list gem5's writeback modeling can impact
performance. This patch removes the limitation on maximum outstanding issued
instructions, however the number that can writeback in a single cycle is still
respected in instToCommit().
Check for free entries in Load Queue and Store Queue separately to
avoid cases when load cannot be renamed due to full Store Queue and
vice versa.
This work was done while Binh was an intern at AMD Research.
Using '== true' in a boolean expression is totally redundant,
and using '== false' is pretty verbose (and arguably less
readable in most cases) compared to '!'.
It's somewhat of a pet peeve, perhaps, but I had some time
waiting for some tests to run and decided to clean these up.
Unfortunately, SLICC appears not to have the '!' operator,
so I had to leave the '== false' tests in the SLICC code.
O3CPU has a compile-time maximum width set in o3/impl.hh, but checking
the configuration against this limit was not implemented anywhere
except for fetch. Configuring a wider pipe than the limit can silently
cause various issues during the simulation. This patch adds the proper
checking in the constructor of the various pipeline stages.
A number of calls to isEmpty() and numFreeEntries()
should be thread-specific.
In cpu.cc, the fact that tid is /*commented*/ out is a bug. Say the rob
has instructions from thread 0 (isEmpty() returns false), and none from
thread 1. If we are trying to squash all of thread 1, then
readTailInst(thread 1) will be called because rob->isEmpty() returns
false. The result is end_it is not in the list and the while
statement loops indefinitely back over the cpu's instList.
In iew_impl.hh, all threads are told they have the entire remaining IQ, when
each thread actually has a certain allocation. The result is extra stalls at
the iew dispatch stage which the rename stage usually takes care of.
In commit_impl.hh, rob->readHeadInst(thread 1) can be called if the rob only
contains instructions from thread 0. This returns a dummyInst (which may work
since we are trying to squash all instructions, but hardly seems like the right
way to do it).
In rob_impl.hh this fix skips the rest of the function more frequently and is
more efficient.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
The probe patch is motivated by the desire to move analytical and trace code
away from functional code. This is achieved by the probe interface which is
essentially a glorified observer model.
What this means to users:
* add a probe point and a "notify" call at the source of an "event"
* add an isolated module, that is being used to carry out *your* analysis (e.g. generate a trace)
* register that module as a probe listener
Note: an example is given for reference in src/cpu/o3/simple_trace.[hh|cc] and src/cpu/SimpleTrace.py
What is happening under the hood:
* every SimObject maintains has a ProbeManager.
* during initialization (src/python/m5/simulate.py) first regProbePoints and
the regProbeListeners is called on each SimObject. this hooks up the probe
point notify calls with the listeners.
FAQs:
Why did you develop probe points:
* to remove trace, stats gathering, analytical code out of the functional code.
* the belief that probes could be generically useful.
What is a probe point:
* a probe point is used to notify upon a given event (e.g. cpu commits an instruction)
What is a probe listener:
* a class that handles whatever the user wishes to do when they are notified
about an event.
What can be passed on notify:
* probe points are templates, and so the user can generate probes that pass any
type of argument (by const reference) to a listener.
What relationships can be generated (1:1, 1:N, N:M etc):
* there isn't a restriction. You can hook probe points and listeners up in a
1:1, 1:N, N:M relationship. They become useful when a number of modules
listen to the same probe points. The idea being that you can add a small
number of probes into the source code and develop a larger number of useful
analysis modules that use information passed by the probes.
Can you give examples:
* adding a probe point to the cpu's commit method allows you to build a trace
module (outputting assembler), you could re-use this to gather instruction
distribution (arithmetic, load/store, conditional, control flow) stats.
Why is the probe interface currently restricted to passing a const reference:
* the desire, initially at least, is to allow an interface to observe
functionality, but not to change functionality.
* of course this can be subverted by const-casting.
What is the performance impact of adding probes:
* when nothing is actively listening to the probes they should have a
relatively minor impact. Profiling has suggested even with a large number of
probes (60) the impact of them (when not active) is very minimal (<1%).
This patch changes the IEW drain check to include the FU pool as there
can be instructions that are "stored" in FU completion events and thus
not covered by the existing checks. With this patch, we simply include
a check to see if all the FUs are considered non-busy in the next
tick.
Without this patch, the pc-switcheroo-full regression fails after
minor changes to the cache timing (aligning to clock edge).
Fixes the tick used from rename:
- previously this gathered the tick on leaving rename which was always 1 less
than the dispatch. This conflated the decode ticks when back pressure built
in the pipeline.
- now picks up tick on entry.
Added --store_completions flag:
- will additionally display the store completion tail in the viewer.
- this highlights periods when large numbers of stores are outstanding (>16 LSQ
blocking)
Allows selection by tick range (previously this caused an infinite loop)
Previously, the O3 CPU could stop in the middle of a microcode
sequence. This patch makes sure that the pipeline stops when it has
committed a normal instruction or exited from a microcode
sequence. Additionally, it makes sure that the pipeline has no
instructions in flight when it is drained, which should make draining
more robust.
Draining is controlled in the commit stage, which checks if the next
PC after a committed instruction is in microcode. If this isn't the
case, it requests a squash of all instructions after that the
instruction that just committed and immediately signals a drain stall
to the fetch stage. The CPU then continues to execute until the
pipeline and all associated buffers are empty.
The entire O3 pipeline used to be initialized from init(), which is
called before initState() or unserialize(). This causes the pipeline
to be initialized from an incorrect thread context. This doesn't
currently lead to correctness problems as instructions fetched from
the incorrect start PC will be squashed a few cycles after
initialization.
This patch will affect the regressions since the O3 CPU now issues its
first instruction fetch to the correct PC instead of 0x0.
Enables the CheckerCPU to be selected at runtime with the --checker option
from the configs/example/fs.py and configs/example/se.py configuration
files. Also merges with the SE/FS changes.
This patch continues the unification of how the different CPU models
create and share their instruction and data ports. Most importantly,
it forces every CPU to have an instruction and a data port, and gives
these ports explicit getters in the BaseCPU (getDataPort and
getInstPort). The patch helps in simplifying the code, make
assumptions more explicit, andfurther ease future patches related to
the CPU ports.
The biggest changes are in the in-order model (that was not modified
in the previous unification patch), which now moves the ports from the
CacheUnit to the CPU. It also distinguishes the instruction fetch and
load-store unit from the rest of the resources, and avoids the use of
indices and casting in favour of keeping track of these two units
explicitly (since they are always there anyways). The atomic, timing
and O3 model simply return references to their already existing ports.
