2005-05-27 05:30:12 +02:00
|
|
|
/*
|
|
|
|
|
* Copyright (c) 2004-2005 The Regents of The University of Michigan
|
|
|
|
|
* All rights reserved.
|
|
|
|
|
*
|
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
|
* modification, are permitted provided that the following conditions are
|
|
|
|
|
* met: redistributions of source code must retain the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer;
|
|
|
|
|
* redistributions in binary form must reproduce the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
|
* documentation and/or other materials provided with the distribution;
|
|
|
|
|
* neither the name of the copyright holders nor the names of its
|
|
|
|
|
* contributors may be used to endorse or promote products derived from
|
|
|
|
|
* this software without specific prior written permission.
|
|
|
|
|
*
|
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
|
|
|
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
|
|
|
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
|
|
|
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
|
|
|
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
|
|
|
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
|
|
|
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
|
|
|
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
|
|
|
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
|
|
|
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
2006-06-01 01:26:56 +02:00
|
|
|
*
|
|
|
|
|
* Authors: Kevin Lim
|
2005-05-27 05:30:12 +02:00
|
|
|
*/
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
|
2009-09-23 17:34:21 +02:00
|
|
|
#include <algorithm>
|
|
|
|
|
|
2006-08-15 11:07:15 +02:00
|
|
|
#include "arch/types.hh"
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
#include "arch/utility.hh"
|
2006-09-01 02:51:30 +02:00
|
|
|
#include "arch/isa_traits.hh"
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
#include "base/trace.hh"
|
2005-05-27 05:30:12 +02:00
|
|
|
#include "base/traceflags.hh"
|
2009-09-23 17:34:21 +02:00
|
|
|
#include "config/the_isa.hh"
|
2005-06-05 02:50:10 +02:00
|
|
|
#include "cpu/o3/bpred_unit.hh"
|
2008-08-11 21:22:16 +02:00
|
|
|
#include "params/DerivO3CPU.hh"
|
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template<class Impl>
|
2008-08-11 21:22:16 +02:00
|
|
|
BPredUnit<Impl>::BPredUnit(DerivO3CPUParams *params)
|
2009-03-07 23:30:54 +01:00
|
|
|
: _name(params->name + ".BPredUnit"),
|
|
|
|
|
BTB(params->BTBEntries,
|
|
|
|
|
params->BTBTagSize,
|
|
|
|
|
params->instShiftAmt)
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
{
|
2006-05-25 23:01:48 +02:00
|
|
|
// Setup the selected predictor.
|
|
|
|
|
if (params->predType == "local") {
|
|
|
|
|
localBP = new LocalBP(params->localPredictorSize,
|
|
|
|
|
params->localCtrBits,
|
|
|
|
|
params->instShiftAmt);
|
|
|
|
|
predictor = Local;
|
|
|
|
|
} else if (params->predType == "tournament") {
|
|
|
|
|
tournamentBP = new TournamentBP(params->localPredictorSize,
|
|
|
|
|
params->localCtrBits,
|
|
|
|
|
params->localHistoryTableSize,
|
|
|
|
|
params->localHistoryBits,
|
|
|
|
|
params->globalPredictorSize,
|
|
|
|
|
params->globalHistoryBits,
|
|
|
|
|
params->globalCtrBits,
|
|
|
|
|
params->choicePredictorSize,
|
|
|
|
|
params->choiceCtrBits,
|
|
|
|
|
params->instShiftAmt);
|
|
|
|
|
predictor = Tournament;
|
|
|
|
|
} else {
|
|
|
|
|
fatal("Invalid BP selected!");
|
|
|
|
|
}
|
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
for (int i=0; i < Impl::MaxThreads; i++)
|
|
|
|
|
RAS[i].init(params->RASSize);
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
}
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
2006-05-25 23:01:48 +02:00
|
|
|
BPredUnit<Impl>::regStats()
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
{
|
|
|
|
|
lookups
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".lookups")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of BP lookups")
|
|
|
|
|
;
|
|
|
|
|
|
|
|
|
|
condPredicted
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".condPredicted")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of conditional branches predicted")
|
|
|
|
|
;
|
|
|
|
|
|
|
|
|
|
condIncorrect
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".condIncorrect")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of conditional branches incorrect")
|
|
|
|
|
;
|
|
|
|
|
|
|
|
|
|
BTBLookups
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".BTBLookups")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of BTB lookups")
|
|
|
|
|
;
|
|
|
|
|
|
|
|
|
|
BTBHits
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".BTBHits")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of BTB hits")
|
|
|
|
|
;
|
|
|
|
|
|
|
|
|
|
BTBCorrect
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".BTBCorrect")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of correct BTB predictions (this stat may not "
|
|
|
|
|
"work properly.")
|
|
|
|
|
;
|
|
|
|
|
|
|
|
|
|
usedRAS
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".usedRAS")
|
2006-04-23 00:26:48 +02:00
|
|
|
.desc("Number of times the RAS was used to get a target.")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
;
|
|
|
|
|
|
|
|
|
|
RASIncorrect
|
2009-03-07 23:30:54 +01:00
|
|
|
.name(name() + ".RASInCorrect")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of incorrect RAS predictions.")
|
|
|
|
|
;
|
|
|
|
|
}
|
|
|
|
|
|
2006-05-04 17:36:20 +02:00
|
|
|
template <class Impl>
|
|
|
|
|
void
|
2006-05-25 23:01:48 +02:00
|
|
|
BPredUnit<Impl>::switchOut()
|
2006-05-04 17:36:20 +02:00
|
|
|
{
|
2006-05-25 23:01:48 +02:00
|
|
|
// Clear any state upon switch out.
|
2006-05-04 17:36:20 +02:00
|
|
|
for (int i = 0; i < Impl::MaxThreads; ++i) {
|
2006-05-25 23:01:48 +02:00
|
|
|
squash(0, i);
|
2006-05-04 17:36:20 +02:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
2006-05-25 23:01:48 +02:00
|
|
|
BPredUnit<Impl>::takeOverFrom()
|
2006-05-04 17:36:20 +02:00
|
|
|
{
|
2006-05-25 23:01:48 +02:00
|
|
|
// Can reset all predictor state, but it's not necessarily better
|
|
|
|
|
// than leaving it be.
|
2006-05-11 21:39:02 +02:00
|
|
|
/*
|
2006-05-04 17:36:20 +02:00
|
|
|
for (int i = 0; i < Impl::MaxThreads; ++i)
|
|
|
|
|
RAS[i].reset();
|
|
|
|
|
|
|
|
|
|
BP.reset();
|
|
|
|
|
BTB.reset();
|
2006-05-11 21:39:02 +02:00
|
|
|
*/
|
2006-05-04 17:36:20 +02:00
|
|
|
}
|
|
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
template <class Impl>
|
|
|
|
|
bool
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
BPredUnit<Impl>::predict(DynInstPtr &inst, TheISA::PCState &pc, ThreadID tid)
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
{
|
|
|
|
|
// See if branch predictor predicts taken.
|
|
|
|
|
// If so, get its target addr either from the BTB or the RAS.
|
|
|
|
|
// Save off record of branch stuff so the RAS can be fixed
|
|
|
|
|
// up once it's done.
|
|
|
|
|
|
|
|
|
|
bool pred_taken = false;
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
TheISA::PCState target = pc;
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
|
|
|
|
++lookups;
|
|
|
|
|
|
2006-05-25 23:01:48 +02:00
|
|
|
void *bp_history = NULL;
|
|
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
if (inst->isUncondCtrl()) {
|
2006-07-23 19:39:42 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Unconditional control.\n", tid);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
pred_taken = true;
|
2006-05-25 23:01:48 +02:00
|
|
|
// Tell the BP there was an unconditional branch.
|
|
|
|
|
BPUncond(bp_history);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
} else {
|
|
|
|
|
++condPredicted;
|
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
pred_taken = BPLookup(pc.instAddr(), bp_history);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Branch predictor predicted %i "
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
"for PC %s\n",
|
|
|
|
|
tid, pred_taken, inst->pcState());
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
|
2009-04-18 16:42:29 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: [sn:%i] Creating prediction history "
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
"for PC %s\n",
|
|
|
|
|
tid, inst->seqNum, inst->pcState());
|
2009-04-18 16:42:29 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
PredictorHistory predict_record(inst->seqNum, pc.instAddr(),
|
|
|
|
|
pred_taken, bp_history, tid);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
|
|
|
|
// Now lookup in the BTB or RAS.
|
|
|
|
|
if (pred_taken) {
|
|
|
|
|
if (inst->isReturn()) {
|
|
|
|
|
++usedRAS;
|
|
|
|
|
|
|
|
|
|
// If it's a function return call, then look up the address
|
|
|
|
|
// in the RAS.
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
TheISA::PCState rasTop = RAS[tid].top();
|
|
|
|
|
target = TheISA::buildRetPC(pc, rasTop);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
|
|
|
|
// Record the top entry of the RAS, and its index.
|
|
|
|
|
predict_record.usedRAS = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
predict_record.RASIndex = RAS[tid].topIdx();
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
predict_record.RASTarget = rasTop;
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
assert(predict_record.RASIndex < 16);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
RAS[tid].pop();
|
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Instruction %s is a return, "
|
|
|
|
|
"RAS predicted target: %s, RAS index: %i.\n",
|
|
|
|
|
tid, inst->pcState(), target, predict_record.RASIndex);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
} else {
|
|
|
|
|
++BTBLookups;
|
|
|
|
|
|
|
|
|
|
if (inst->isCall()) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
RAS[tid].push(pc);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
|
|
|
|
// Record that it was a call so that the top RAS entry can
|
|
|
|
|
// be popped off if the speculation is incorrect.
|
|
|
|
|
predict_record.wasCall = true;
|
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Instruction %s was a "
|
|
|
|
|
"call, adding %s to the RAS index: %i.\n",
|
|
|
|
|
tid, inst->pcState(), pc, RAS[tid].topIdx());
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
if (BTB.valid(pc.instAddr(), tid)) {
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
++BTBHits;
|
|
|
|
|
|
2006-05-25 23:01:48 +02:00
|
|
|
// If it's not a return, use the BTB to get the target addr.
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
target = BTB.lookup(pc.instAddr(), tid);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Instruction %s predicted"
|
|
|
|
|
" target is %s.\n", tid, inst->pcState(), target);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
|
|
|
|
} else {
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: BTB doesn't have a "
|
|
|
|
|
"valid entry.\n",tid);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
pred_taken = false;
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
TheISA::advancePC(target, inst->staticInst);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
}
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
} else {
|
|
|
|
|
TheISA::advancePC(target, inst->staticInst);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
pc = target;
|
2006-12-28 20:29:17 +01:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
predHist[tid].push_front(predict_record);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2009-04-18 16:42:29 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: [sn:%i]: History entry added."
|
|
|
|
|
"predHist.size(): %i\n", tid, inst->seqNum, predHist[tid].size());
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
|
|
|
|
return pred_taken;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
BPredUnit<Impl>::update(const InstSeqNum &done_sn, ThreadID tid)
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
{
|
2009-04-18 16:42:29 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Committing branches until "
|
2006-07-23 19:39:42 +02:00
|
|
|
"[sn:%lli].\n", tid, done_sn);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
while (!predHist[tid].empty() &&
|
|
|
|
|
predHist[tid].back().seqNum <= done_sn) {
|
|
|
|
|
// Update the branch predictor with the correct results.
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
BPUpdate(predHist[tid].back().pc,
|
2006-05-25 23:01:48 +02:00
|
|
|
predHist[tid].back().predTaken,
|
|
|
|
|
predHist[tid].back().bpHistory);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
predHist[tid].pop_back();
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
BPredUnit<Impl>::squash(const InstSeqNum &squashed_sn, ThreadID tid)
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
{
|
2006-04-23 00:26:48 +02:00
|
|
|
History &pred_hist = predHist[tid];
|
|
|
|
|
|
|
|
|
|
while (!pred_hist.empty() &&
|
|
|
|
|
pred_hist.front().seqNum > squashed_sn) {
|
2006-05-25 23:01:48 +02:00
|
|
|
if (pred_hist.front().usedRAS) {
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Restoring top of RAS to: %i,"
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
" target: %s.\n", tid,
|
|
|
|
|
pred_hist.front().RASIndex, pred_hist.front().RASTarget);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
|
RAS[tid].restore(pred_hist.front().RASIndex,
|
|
|
|
|
pred_hist.front().RASTarget);
|
|
|
|
|
} else if (pred_hist.front().wasCall) {
|
2006-05-25 23:01:48 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Removing speculative entry "
|
|
|
|
|
"added to the RAS.\n",tid);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
RAS[tid].pop();
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
|
2006-05-25 23:01:48 +02:00
|
|
|
// This call should delete the bpHistory.
|
|
|
|
|
BPSquash(pred_hist.front().bpHistory);
|
|
|
|
|
|
2009-04-18 16:42:29 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Removing history for [sn:%i] "
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
"PC %s.\n", tid, pred_hist.front().seqNum,
|
|
|
|
|
pred_hist.front().pc);
|
2009-04-18 16:42:29 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
pred_hist.pop_front();
|
2009-04-18 16:42:29 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Fetch, "[tid:%i]: predHist.size(): %i\n",
|
|
|
|
|
tid, predHist[tid].size());
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
2006-04-23 00:26:48 +02:00
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
2006-05-25 23:01:48 +02:00
|
|
|
BPredUnit<Impl>::squash(const InstSeqNum &squashed_sn,
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
const TheISA::PCState &corrTarget,
|
2009-05-26 18:23:13 +02:00
|
|
|
bool actually_taken,
|
|
|
|
|
ThreadID tid)
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
{
|
|
|
|
|
// Now that we know that a branch was mispredicted, we need to undo
|
|
|
|
|
// all the branches that have been seen up until this branch and
|
|
|
|
|
// fix up everything.
|
2009-04-18 16:42:29 +02:00
|
|
|
// NOTE: This should be call conceivably in 2 scenarios:
|
|
|
|
|
// (1) After an branch is executed, it updates its status in the ROB
|
|
|
|
|
// The commit stage then checks the ROB update and sends a signal to
|
|
|
|
|
// the fetch stage to squash history after the mispredict
|
|
|
|
|
// (2) In the decode stage, you can find out early if a unconditional
|
|
|
|
|
// PC-relative, branch was predicted incorrectly. If so, a signal
|
|
|
|
|
// to the fetch stage is sent to squash history after the mispredict
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
History &pred_hist = predHist[tid];
|
|
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
++condIncorrect;
|
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Squashing from sequence number %i, "
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
"setting target to %s.\n",
|
|
|
|
|
tid, squashed_sn, corrTarget);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2009-04-18 16:42:29 +02:00
|
|
|
// Squash All Branches AFTER this mispredicted branch
|
2006-05-25 23:01:48 +02:00
|
|
|
squash(squashed_sn, tid);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
// If there's a squash due to a syscall, there may not be an entry
|
|
|
|
|
// corresponding to the squash. In that case, don't bother trying to
|
|
|
|
|
// fix up the entry.
|
|
|
|
|
if (!pred_hist.empty()) {
|
2009-04-18 16:42:29 +02:00
|
|
|
|
|
|
|
|
HistoryIt hist_it = pred_hist.begin();
|
|
|
|
|
//HistoryIt hist_it = find(pred_hist.begin(), pred_hist.end(),
|
|
|
|
|
// squashed_sn);
|
|
|
|
|
|
|
|
|
|
//assert(hist_it != pred_hist.end());
|
|
|
|
|
if (pred_hist.front().seqNum != squashed_sn) {
|
|
|
|
|
DPRINTF(Fetch, "Front sn %i != Squash sn %i\n",
|
|
|
|
|
pred_hist.front().seqNum, squashed_sn);
|
|
|
|
|
|
|
|
|
|
assert(pred_hist.front().seqNum == squashed_sn);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if ((*hist_it).usedRAS) {
|
2006-04-23 00:26:48 +02:00
|
|
|
++RASIncorrect;
|
|
|
|
|
}
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
BPUpdate((*hist_it).pc, actually_taken,
|
2006-05-25 23:01:48 +02:00
|
|
|
pred_hist.front().bpHistory);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
BTB.update((*hist_it).pc, corrTarget, tid);
|
2009-04-18 16:42:29 +02:00
|
|
|
|
|
|
|
|
DPRINTF(Fetch, "BranchPred: [tid:%i]: Removing history for [sn:%i] "
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
"PC %s.\n", tid, (*hist_it).seqNum, (*hist_it).pc);
|
2009-04-18 16:42:29 +02:00
|
|
|
|
|
|
|
|
pred_hist.erase(hist_it);
|
|
|
|
|
|
|
|
|
|
DPRINTF(Fetch, "[tid:%i]: predHist.size(): %i\n", tid, predHist[tid].size());
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
2006-05-25 23:01:48 +02:00
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
|
|
|
|
BPredUnit<Impl>::BPUncond(void * &bp_history)
|
|
|
|
|
{
|
|
|
|
|
// Only the tournament predictor cares about unconditional branches.
|
|
|
|
|
if (predictor == Tournament) {
|
|
|
|
|
tournamentBP->uncondBr(bp_history);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
|
|
|
|
BPredUnit<Impl>::BPSquash(void *bp_history)
|
|
|
|
|
{
|
|
|
|
|
if (predictor == Local) {
|
|
|
|
|
localBP->squash(bp_history);
|
|
|
|
|
} else if (predictor == Tournament) {
|
|
|
|
|
tournamentBP->squash(bp_history);
|
|
|
|
|
} else {
|
|
|
|
|
panic("Predictor type is unexpected value!");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
bool
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
BPredUnit<Impl>::BPLookup(Addr instPC, void * &bp_history)
|
2006-05-25 23:01:48 +02:00
|
|
|
{
|
|
|
|
|
if (predictor == Local) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
return localBP->lookup(instPC, bp_history);
|
2006-05-25 23:01:48 +02:00
|
|
|
} else if (predictor == Tournament) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
return tournamentBP->lookup(instPC, bp_history);
|
2006-05-25 23:01:48 +02:00
|
|
|
} else {
|
|
|
|
|
panic("Predictor type is unexpected value!");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
BPredUnit<Impl>::BPUpdate(Addr instPC, bool taken, void *bp_history)
|
2006-05-25 23:01:48 +02:00
|
|
|
{
|
|
|
|
|
if (predictor == Local) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
localBP->update(instPC, taken, bp_history);
|
2006-05-25 23:01:48 +02:00
|
|
|
} else if (predictor == Tournament) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
tournamentBP->update(instPC, taken, bp_history);
|
2006-05-25 23:01:48 +02:00
|
|
|
} else {
|
|
|
|
|
panic("Predictor type is unexpected value!");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
|
void
|
|
|
|
|
BPredUnit<Impl>::dump()
|
|
|
|
|
{
|
2009-04-18 16:42:29 +02:00
|
|
|
HistoryIt pred_hist_it;
|
2006-05-25 23:01:48 +02:00
|
|
|
|
|
|
|
|
for (int i = 0; i < Impl::MaxThreads; ++i) {
|
|
|
|
|
if (!predHist[i].empty()) {
|
|
|
|
|
pred_hist_it = predHist[i].begin();
|
|
|
|
|
|
|
|
|
|
cprintf("predHist[%i].size(): %i\n", i, predHist[i].size());
|
|
|
|
|
|
|
|
|
|
while (pred_hist_it != predHist[i].end()) {
|
|
|
|
|
cprintf("[sn:%lli], PC:%#x, tid:%i, predTaken:%i, "
|
|
|
|
|
"bpHistory:%#x\n",
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
pred_hist_it->seqNum, pred_hist_it->pc,
|
|
|
|
|
pred_hist_it->tid, pred_hist_it->predTaken,
|
|
|
|
|
pred_hist_it->bpHistory);
|
2006-05-25 23:01:48 +02:00
|
|
|
pred_hist_it++;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cprintf("\n");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
