The Minor and o3 cpu models share the branch prediction
code. Minor relies on the BPredUnit::squash() function
to update the branch predictor tables on a branch mispre-
diction. This is fine because Minor executes in-order, so
the update is on the correct path. However, this causes the
branch predictor to be updated on out-of-order branch
mispredictions when using the o3 model, which should not
be the case.
This patch guards against speculative update of the branch
prediction tables. On a branch misprediction, BPredUnit::squash()
calls BpredUnit::update(..., squashed = true). The underlying
branch predictor tests against the value of squashed. If it is
true, it restores any speculatively updated internal state
it might have (e.g., global/local branch history), then returns.
If false, it updates its prediction tables. Previously, exist-
ing predictors did not test against the "squashed" parameter.
To accomodate for this change, the Minor model must now call
BPredUnit::squash() then BPredUnit::update(..., squashed = false)
on branch mispredictions. Before, calling BpredUnit::squash()
performed the prediction tables update.
The effect is a slight MPKI improvement when using the o3
model. A further patch should perform the same modifications
for the indirect target predictor and BTB (less critical).
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
The tournament predictor is presented as doing speculative
update of the global history and non-speculative update
of the local history used to generate the branch prediction.
However, the code does speculative update of both histories.
Signed-off-by: Jason Lowe-Power <jason@lowepower.com>
This function was used by the now-defunct InOrderCPU model. Since this
model is no longer in gem5, this function was not called from anywhere in
the code.
Fixing an issue with regStats not calling the parent class method
for most SimObjects in Gem5. This causes issues if one adds new
stats in the base class (since they are never initialized properly!).
Change-Id: Iebc5aa66f58816ef4295dc8e48a357558d76a77c
Reviewed-by: Andreas Sandberg <andreas.sandberg@arm.com>
This patch adds a configurable indirect branch predictor that can be indexed
by a combination of GHR and path history hashes. Implements the functionality
described in:
"Target prediction for indirect jumps" by Chang, Hao, and Patt
http://dl.acm.org/citation.cfm?id=264209
This is a re-spin of fb9d142 after the revert (bd1c6789).
The extant BTB code doesn't hash on the thread id but does check the
thread id for 'btb hits'. This results in 1-thread of a multi-threaded
workload taking a BTB entry, and all other threads missing for the same branch
missing.
The following patches had unexpected interactions with the current
upstream code and have been reverted for now:
e07fd01651f3: power: Add support for power models
831c7f2f9e39: power: Low-power idle power state for idle CPUs
4f749e00b667: power: Add power states to ClockedObject
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
--HG--
extra : amend_source : 0b6fb073c6bbc24be533ec431eb51fbf1b269508
This patch adds a configurable indirect branch predictor that can be indexed
by a combination of GHR and path history hashes. Implements the functionality
described in:
"Target prediction for indirect jumps" by Chang, Hao, and Patt
http://dl.acm.org/citation.cfm?id=264209
The extant BTB code doesn't hash on the thread id but does check the
thread id for 'btb hits'. This results in 1-thread of a multi-threaded
workload taking a BTB entry, and all other threads missing for the same branch
missing.
This patch adds explicit overrides as this is now required when using
"-Wall" with clang >= 3.5, the latter now part of the most recent
XCode. The patch consequently removes "virtual" for those methods
where "override" is added. The latter should be enough of an
indication.
As part of this patch, a few minor issues that clang >= 3.5 complains
about are also resolved (unused methods and variables).
This patch moves away from using M5_ATTR_OVERRIDE and the m5::hashmap
(and similar) abstractions, as these are no longer needed with gcc 4.7
and clang 3.1 as minimum compiler versions.
When a branch gets squashed, it's speculative branch predictor state should get
rolled back in squash(). However, only the globalHistory state was being
rolled back. This patch adds (at least some) support for rolling back the
local predictor state also.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
This changeset adds probe points that can be used to implement PMU
counters for branch predictor stats. The following probes are
supported:
* BPRedUnit::ppBranches / Branches
* BPRedUnit::ppMisses / Misses
This patch optimises the passing of StaticInstPtr by avoiding copying
the reference-counting pointer. This avoids first incrementing and
then decrementing the reference-counting pointer.
A small bug in the bimodal predictor caused significant degradation in
performance on some benchmarks. This was caused by using the wrong
globalHistoryReg during the update phase. This patches fixes the bug
and brings the performance to normal level.
When a branch mispredicted gem5 would squash all history after and including
the mispredicted branch. However, the mispredicted branch is still speculative
and its history is required to rollback state if another, older, branch
mispredicts. This leads to things like RAS corruption.
The branch predictor is normally only built when a CPU that uses a
branch predictor is built. The list of CPUs is currently incomplete as
the simple CPUs support branch predictors (for warming, branch stats,
etc). In practice, all CPU models now use branch predictors, so this
changeset removes the CPU model check and replaces it with a check for
the NULL ISA.
This patch does some minor house keeping of the branch predictor by
adopting STL containers, and shifting some iterator to use range-based
for loops.
The predictor history is also changed from a list to a deque as we
never to insertion/deletion other than at the front and back.
This patch contains a new CPU model named `Minor'. Minor models a four
stage in-order execution pipeline (fetch lines, decompose into
macroops, decompose macroops into microops, execute).
The model was developed to support the ARM ISA but should be fixable
to support all the remaining gem5 ISAs. It currently also works for
Alpha, and regressions are included for ARM and Alpha (including Linux
boot).
Documentation for the model can be found in src/doc/inside-minor.doxygen and
its internal operations can be visualised using the Minorview tool
utils/minorview.py.
Minor was designed to be fairly simple and not to engage in a lot of
instruction annotation. As such, it currently has very few gathered
stats and may lack other gem5 features.
Minor is faster than the o3 model. Sample results:
Benchmark | Stat host_seconds (s)
---------------+--------v--------v--------
(on ARM, opt) | simple | o3 | minor
| timing | timing | timing
---------------+--------+--------+--------
10.linux-boot | 169 | 1883 | 1075
10.mcf | 117 | 967 | 491
20.parser | 668 | 6315 | 3146
30.eon | 542 | 3413 | 2414
40.perlbmk | 2339 | 20905 | 11532
50.vortex | 122 | 1094 | 588
60.bzip2 | 2045 | 18061 | 9662
70.twolf | 207 | 2736 | 1036
having separate params for the local/globalHistoryBits and the
local/globalPredictorSize can lead to inconsistencies if they
are not carefully set. this patch dervies the number of bits
necessary to index into the local/global predictors based on
their size.
the value of the localHistoryTableSize for the ARM O3 CPU has been
increased to 1024 from 64, which is more accurate for an A15 based
on some correlation against A15 hardware.
This patch moves the branch predictor files in the o3 and inorder directories
to src/cpu/pred. This allows sharing the branch predictor across different
cpu models.
This patch was originally posted by Timothy Jones in July 2010
but never made it to the repository.
--HG--
rename : src/cpu/o3/bpred_unit.cc => src/cpu/pred/bpred_unit.cc
rename : src/cpu/o3/bpred_unit.hh => src/cpu/pred/bpred_unit.hh
rename : src/cpu/o3/bpred_unit_impl.hh => src/cpu/pred/bpred_unit_impl.hh
rename : src/cpu/o3/sat_counter.hh => src/cpu/pred/sat_counter.hh
globalHistoryBits, globalPredictorSize, and choicePredictorSize are decoupled.
globalHistoryBits controls how much history is kept, global and choice
predictor sizes control how much of that history is used when accessing
predictor tables. This way, global and choice predictors can actually be
different sizes, and it is no longer possible to walk off the predictor arrays
and cause a seg fault.
There are now individual thresholds for choice, global, and local saturating
counters, so that taken/not taken decisions are correct even when the
predictors' counters' sizes are different.
The interface for localPredictorSize has been removed from TournamentBP because
the value can be calculated from localHistoryBits.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
1. Updates the Branch Predictor correctly to the state
just after a mispredicted branch, if a squash occurs.
2. If a BTB does not find an entry, the branch is predicted not taken.
The global history is modified to correctly reflect this prediction.
3. Local history is now updated at the fetch stage instead of
execute stage.
4. In the Update stage of the branch predictor the local predictors are
now correctly updated according to the state of local history during
fetch stage.
This patch also improves performance by as much as 17% on some benchmarks
Branch predictor could not predict a branch in a nested loop because:
1. The global history was not updated after a mispredict squash.
2. The global history was updated in the fetch stage. The choice predictors
that were updated used the changed global history. This is incorrect, as
it incorporates the state of global history after the branch in
encountered. Fixed update to choice predictor using the global history
state before the branch happened.
3. The global predictor table was also updated using the global history state
before the branch happened as above.
Additionally, parameters to initialize ctr and history size were reversed.
At the same time, rename the trace flags to debug flags since they
have broader usage than simply tracing. This means that
--trace-flags is now --debug-flags and --trace-help is now --debug-help
This change is a low level and pervasive reorganization of how PCs are managed
in M5. Back when Alpha was the only ISA, there were only 2 PCs to worry about,
the PC and the NPC, and the lsb of the PC signaled whether or not you were in
PAL mode. As other ISAs were added, we had to add an NNPC, micro PC and next
micropc, x86 and ARM introduced variable length instruction sets, and ARM
started to keep track of mode bits in the PC. Each CPU model handled PCs in
its own custom way that needed to be updated individually to handle the new
dimensions of variability, or, in the case of ARMs mode-bit-in-the-pc hack,
the complexity could be hidden in the ISA at the ISA implementation's expense.
Areas like the branch predictor hadn't been updated to handle branch delay
slots or micropcs, and it turns out that had introduced a significant (10s of
percent) performance bug in SPARC and to a lesser extend MIPS. Rather than
perpetuate the problem by reworking O3 again to handle the PC features needed
by x86, this change was introduced to rework PC handling in a more modular,
transparent, and hopefully efficient way.
PC type:
Rather than having the superset of all possible elements of PC state declared
in each of the CPU models, each ISA defines its own PCState type which has
exactly the elements it needs. A cross product of canned PCState classes are
defined in the new "generic" ISA directory for ISAs with/without delay slots
and microcode. These are either typedef-ed or subclassed by each ISA. To read
or write this structure through a *Context, you use the new pcState() accessor
which reads or writes depending on whether it has an argument. If you just
want the address of the current or next instruction or the current micro PC,
you can get those through read-only accessors on either the PCState type or
the *Contexts. These are instAddr(), nextInstAddr(), and microPC(). Note the
move away from readPC. That name is ambiguous since it's not clear whether or
not it should be the actual address to fetch from, or if it should have extra
bits in it like the PAL mode bit. Each class is free to define its own
functions to get at whatever values it needs however it needs to to be used in
ISA specific code. Eventually Alpha's PAL mode bit could be moved out of the
PC and into a separate field like ARM.
These types can be reset to a particular pc (where npc = pc +
sizeof(MachInst), nnpc = npc + sizeof(MachInst), upc = 0, nupc = 1 as
appropriate), printed, serialized, and compared. There is a branching()
function which encapsulates code in the CPU models that checked if an
instruction branched or not. Exactly what that means in the context of branch
delay slots which can skip an instruction when not taken is ambiguous, and
ideally this function and its uses can be eliminated. PCStates also generally
know how to advance themselves in various ways depending on if they point at
an instruction, a microop, or the last microop of a macroop. More on that
later.
Ideally, accessing all the PCs at once when setting them will improve
performance of M5 even though more data needs to be moved around. This is
because often all the PCs need to be manipulated together, and by getting them
all at once you avoid multiple function calls. Also, the PCs of a particular
thread will have spatial locality in the cache. Previously they were grouped
by element in arrays which spread out accesses.
Advancing the PC:
The PCs were previously managed entirely by the CPU which had to know about PC
semantics, try to figure out which dimension to increment the PC in, what to
set NPC/NNPC, etc. These decisions are best left to the ISA in conjunction
with the PC type itself. Because most of the information about how to
increment the PC (mainly what type of instruction it refers to) is contained
in the instruction object, a new advancePC virtual function was added to the
StaticInst class. Subclasses provide an implementation that moves around the
right element of the PC with a minimal amount of decision making. In ISAs like
Alpha, the instructions always simply assign NPC to PC without having to worry
about micropcs, nnpcs, etc. The added cost of a virtual function call should
be outweighed by not having to figure out as much about what to do with the
PCs and mucking around with the extra elements.
One drawback of making the StaticInsts advance the PC is that you have to
actually have one to advance the PC. This would, superficially, seem to
require decoding an instruction before fetch could advance. This is, as far as
I can tell, realistic. fetch would advance through memory addresses, not PCs,
perhaps predicting new memory addresses using existing ones. More
sophisticated decisions about control flow would be made later on, after the
instruction was decoded, and handed back to fetch. If branching needs to
happen, some amount of decoding needs to happen to see that it's a branch,
what the target is, etc. This could get a little more complicated if that gets
done by the predecoder, but I'm choosing to ignore that for now.
Variable length instructions:
To handle variable length instructions in x86 and ARM, the predecoder now
takes in the current PC by reference to the getExtMachInst function. It can
modify the PC however it needs to (by setting NPC to be the PC + instruction
length, for instance). This could be improved since the CPU doesn't know if
the PC was modified and always has to write it back.
ISA parser:
To support the new API, all PC related operand types were removed from the
parser and replaced with a PCState type. There are two warts on this
implementation. First, as with all the other operand types, the PCState still
has to have a valid operand type even though it doesn't use it. Second, using
syntax like PCS.npc(target) doesn't work for two reasons, this looks like the
syntax for operand type overriding, and the parser can't figure out if you're
reading or writing. Instructions that use the PCS operand (which I've
consistently called it) need to first read it into a local variable,
manipulate it, and then write it back out.
Return address stack:
The return address stack needed a little extra help because, in the presence
of branch delay slots, it has to merge together elements of the return PC and
the call PC. To handle that, a buildRetPC utility function was added. There
are basically only two versions in all the ISAs, but it didn't seem short
enough to put into the generic ISA directory. Also, the branch predictor code
in O3 and InOrder were adjusted so that they always store the PC of the actual
call instruction in the RAS, not the next PC. If the call instruction is a
microop, the next PC refers to the next microop in the same macroop which is
probably not desirable. The buildRetPC function advances the PC intelligently
to the next macroop (in an ISA specific way) so that that case works.
Change in stats:
There were no change in stats except in MIPS and SPARC in the O3 model. MIPS
runs in about 9% fewer ticks. SPARC runs with 30%-50% fewer ticks, which could
likely be improved further by setting call/return instruction flags and taking
advantage of the RAS.
TODO:
Add != operators to the PCState classes, defined trivially to be !(a==b).
Smooth out places where PCs are split apart, passed around, and put back
together later. I think this might happen in SPARC's fault code. Add ISA
specific constructors that allow setting PC elements without calling a bunch
of accessors. Try to eliminate the need for the branching() function. Factor
out Alpha's PAL mode pc bit into a separate flag field, and eliminate places
where it's blindly masked out or tested in the PC.
Suppose the saturating counters of a branch predictor contain n bits. When the
counter is between 0 and (2^(n-1) - 1), boundaries included, the branch is
predicted as not taken. When the counter is between 2^(n-1) and (2^n - 1),
boundaries included, the branch is predicted as taken.
