2005-05-27 05:30:12 +02:00
|
|
|
/*
|
2011-01-18 23:30:05 +01:00
|
|
|
* Copyright (c) 2010 ARM Limited
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* The license below extends only to copyright in the software and shall
|
|
|
|
* not be construed as granting a license to any other intellectual
|
|
|
|
* property including but not limited to intellectual property relating
|
|
|
|
* to a hardware implementation of the functionality of the software
|
|
|
|
* licensed hereunder. You may use the software subject to the license
|
|
|
|
* terms below provided that you ensure that this notice is replicated
|
|
|
|
* unmodified and in its entirety in all distributions of the software,
|
|
|
|
* modified or unmodified, in source code or in binary form.
|
|
|
|
*
|
2006-05-16 20:06:35 +02:00
|
|
|
* Copyright (c) 2004-2006 The Regents of The University of Michigan
|
2005-05-27 05:30:12 +02:00
|
|
|
* 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
|
2006-06-16 04:01:28 +02:00
|
|
|
* Korey Sewell
|
2005-05-27 05:30:12 +02:00
|
|
|
*/
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
#ifndef __CPU_O3_COMMIT_HH__
|
|
|
|
#define __CPU_O3_COMMIT_HH__
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2011-04-15 19:44:14 +02:00
|
|
|
#include <queue>
|
|
|
|
|
2005-02-11 23:54:33 +01:00
|
|
|
#include "base/statistics.hh"
|
2006-04-23 00:26:48 +02:00
|
|
|
#include "cpu/exetrace.hh"
|
2006-05-19 21:53:17 +02:00
|
|
|
#include "cpu/inst_seq.hh"
|
2011-04-15 19:44:06 +02:00
|
|
|
#include "cpu/timebuf.hh"
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2012-01-31 18:05:52 +01:00
|
|
|
struct DerivO3CPUParams;
|
2008-08-11 21:22:16 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
template <class>
|
2012-01-31 18:05:52 +01:00
|
|
|
struct O3ThreadState;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/**
|
2006-05-16 20:06:35 +02:00
|
|
|
* DefaultCommit handles single threaded and SMT commit. Its width is
|
|
|
|
* specified by the parameters; each cycle it tries to commit that
|
|
|
|
* many instructions. The SMT policy decides which thread it tries to
|
|
|
|
* commit instructions from. Non- speculative instructions must reach
|
|
|
|
* the head of the ROB before they are ready to execute; once they
|
|
|
|
* reach the head, commit will broadcast the instruction's sequence
|
|
|
|
* number to the previous stages so that they can issue/ execute the
|
|
|
|
* instruction. Only one non-speculative instruction is handled per
|
|
|
|
* cycle. Commit is responsible for handling all back-end initiated
|
|
|
|
* redirects. It receives the redirect, and then broadcasts it to all
|
|
|
|
* stages, indicating the sequence number they should squash until,
|
|
|
|
* and any necessary branch misprediction information as well. It
|
|
|
|
* priortizes redirects by instruction's age, only broadcasting a
|
|
|
|
* redirect if it corresponds to an instruction that should currently
|
|
|
|
* be in the ROB. This is done by tracking the sequence number of the
|
|
|
|
* youngest instruction in the ROB, which gets updated to any
|
|
|
|
* squashing instruction's sequence number, and only broadcasting a
|
|
|
|
* redirect if it corresponds to an older instruction. Commit also
|
|
|
|
* supports multiple cycle squashing, to model a ROB that can only
|
2006-05-19 21:53:17 +02:00
|
|
|
* remove a certain number of instructions per cycle.
|
2006-04-23 00:26:48 +02:00
|
|
|
*/
|
2004-08-20 20:54:07 +02:00
|
|
|
template<class Impl>
|
2006-04-23 00:26:48 +02:00
|
|
|
class DefaultCommit
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
public:
|
|
|
|
// Typedefs from the Impl.
|
2006-06-16 23:08:47 +02:00
|
|
|
typedef typename Impl::O3CPU O3CPU;
|
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
|
|
|
typedef typename Impl::DynInstPtr DynInstPtr;
|
|
|
|
typedef typename Impl::CPUPol CPUPol;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
typedef typename CPUPol::RenameMap RenameMap;
|
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
|
|
|
typedef typename CPUPol::ROB ROB;
|
2004-08-20 20:54:07 +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
|
|
|
typedef typename CPUPol::TimeStruct TimeStruct;
|
2006-04-23 00:26:48 +02:00
|
|
|
typedef typename CPUPol::FetchStruct FetchStruct;
|
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
|
|
|
typedef typename CPUPol::IEWStruct IEWStruct;
|
|
|
|
typedef typename CPUPol::RenameStruct RenameStruct;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
typedef typename CPUPol::Fetch Fetch;
|
2006-04-23 00:26:48 +02:00
|
|
|
typedef typename CPUPol::IEW IEW;
|
|
|
|
|
|
|
|
typedef O3ThreadState<Impl> Thread;
|
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Event class used to schedule a squash due to a trap (fault or
|
|
|
|
* interrupt) to happen on a specific cycle.
|
|
|
|
*/
|
2006-04-23 00:26:48 +02:00
|
|
|
class TrapEvent : public Event {
|
|
|
|
private:
|
|
|
|
DefaultCommit<Impl> *commit;
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
public:
|
2009-05-26 18:23:13 +02:00
|
|
|
TrapEvent(DefaultCommit<Impl> *_commit, ThreadID _tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
void process();
|
2008-02-06 22:32:40 +01:00
|
|
|
const char *description() const;
|
2006-04-23 00:26:48 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
/** Overall commit status. Used to determine if the CPU can deschedule
|
|
|
|
* itself due to a lack of activity.
|
|
|
|
*/
|
|
|
|
enum CommitStatus{
|
|
|
|
Active,
|
|
|
|
Inactive
|
|
|
|
};
|
|
|
|
|
|
|
|
/** Individual thread status. */
|
|
|
|
enum ThreadStatus {
|
2004-08-20 20:54:07 +02:00
|
|
|
Running,
|
|
|
|
Idle,
|
|
|
|
ROBSquashing,
|
2006-04-23 00:26:48 +02:00
|
|
|
TrapPending,
|
|
|
|
FetchTrapPending
|
|
|
|
};
|
|
|
|
|
|
|
|
/** Commit policy for SMT mode. */
|
|
|
|
enum CommitPolicy {
|
|
|
|
Aggressive,
|
|
|
|
RoundRobin,
|
|
|
|
OldestReady
|
2004-08-20 20:54:07 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
private:
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Overall commit status. */
|
|
|
|
CommitStatus _status;
|
|
|
|
/** Next commit status, to be set at the end of the cycle. */
|
|
|
|
CommitStatus _nextStatus;
|
|
|
|
/** Per-thread status. */
|
|
|
|
ThreadStatus commitStatus[Impl::MaxThreads];
|
|
|
|
/** Commit policy used in SMT mode. */
|
|
|
|
CommitPolicy commitPolicy;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
|
|
|
public:
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Construct a DefaultCommit with the given parameters. */
|
2008-08-11 21:22:16 +02:00
|
|
|
DefaultCommit(O3CPU *_cpu, DerivO3CPUParams *params);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/** Returns the name of the DefaultCommit. */
|
|
|
|
std::string name() const;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Registers statistics. */
|
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
|
|
|
void regStats();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Sets the list of threads. */
|
|
|
|
void setThreads(std::vector<Thread *> &threads);
|
|
|
|
|
|
|
|
/** Sets the main time buffer pointer, used for backwards communication. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr);
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
void setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr);
|
|
|
|
|
|
|
|
/** Sets the pointer to the queue coming from rename. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void setRenameQueue(TimeBuffer<RenameStruct> *rq_ptr);
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Sets the pointer to the queue coming from IEW. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void setIEWQueue(TimeBuffer<IEWStruct> *iq_ptr);
|
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Sets the pointer to the IEW stage. */
|
2006-04-23 00:26:48 +02:00
|
|
|
void setIEWStage(IEW *iew_stage);
|
|
|
|
|
2006-07-27 00:47:06 +02:00
|
|
|
/** Skid buffer between rename and commit. */
|
|
|
|
std::queue<DynInstPtr> skidBuffer;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
/** The pointer to the IEW stage. Used solely to ensure that
|
|
|
|
* various events (traps, interrupts, syscalls) do not occur until
|
|
|
|
* all stores have written back.
|
2006-04-23 00:26:48 +02:00
|
|
|
*/
|
|
|
|
IEW *iewStage;
|
|
|
|
|
|
|
|
/** Sets pointer to list of active threads. */
|
2009-05-26 18:23:13 +02:00
|
|
|
void setActiveThreads(std::list<ThreadID> *at_ptr);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/** Sets pointer to the commited state rename map. */
|
|
|
|
void setRenameMap(RenameMap rm_ptr[Impl::MaxThreads]);
|
|
|
|
|
|
|
|
/** Sets pointer to the ROB. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void setROB(ROB *rob_ptr);
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Initializes stage by sending back the number of free entries. */
|
|
|
|
void initStage();
|
|
|
|
|
2006-07-06 19:59:02 +02:00
|
|
|
/** Initializes the draining of commit. */
|
2006-07-06 23:57:20 +02:00
|
|
|
bool drain();
|
2006-07-06 19:59:02 +02:00
|
|
|
|
|
|
|
/** Resumes execution after draining. */
|
|
|
|
void resume();
|
2006-05-04 17:36:20 +02:00
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Completes the switch out of commit. */
|
2006-07-06 19:59:02 +02:00
|
|
|
void switchOut();
|
2006-05-16 20:06:35 +02:00
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Takes over from another CPU's thread. */
|
2006-05-04 17:36:20 +02:00
|
|
|
void takeOverFrom();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Ticks the commit stage, which tries to commit instructions. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void tick();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Handles any squashes that are sent from IEW, and adds instructions
|
|
|
|
* to the ROB and tries to commit instructions.
|
|
|
|
*/
|
2004-08-20 20:54:07 +02:00
|
|
|
void commit();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Returns the number of free ROB entries for a specific thread. */
|
2009-05-26 18:23:13 +02:00
|
|
|
size_t numROBFreeEntries(ThreadID tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Generates an event to schedule a squash due to a trap. */
|
2009-05-26 18:23:13 +02:00
|
|
|
void generateTrapEvent(ThreadID tid);
|
2006-05-31 17:45:02 +02:00
|
|
|
|
|
|
|
/** Records that commit needs to initiate a squash due to an
|
2006-06-06 23:32:21 +02:00
|
|
|
* external state update through the TC.
|
2006-05-31 17:45:02 +02:00
|
|
|
*/
|
2009-05-26 18:23:13 +02:00
|
|
|
void generateTCEvent(ThreadID tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
private:
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Updates the overall status of commit with the nextStatus, and
|
2006-05-31 17:45:02 +02:00
|
|
|
* tell the CPU if commit is active/inactive.
|
|
|
|
*/
|
2006-04-23 00:26:48 +02:00
|
|
|
void updateStatus();
|
|
|
|
|
|
|
|
/** Sets the next status based on threads' statuses, which becomes the
|
|
|
|
* current status at the end of the cycle.
|
|
|
|
*/
|
|
|
|
void setNextStatus();
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Checks if the ROB is completed with squashing. This is for the case
|
|
|
|
* where the ROB can take multiple cycles to complete squashing.
|
|
|
|
*/
|
|
|
|
bool robDoneSquashing();
|
|
|
|
|
|
|
|
/** Returns if any of the threads have the number of ROB entries changed
|
|
|
|
* on this cycle. Used to determine if the number of free ROB entries needs
|
|
|
|
* to be sent back to previous stages.
|
|
|
|
*/
|
|
|
|
bool changedROBEntries();
|
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Squashes all in flight instructions. */
|
2009-05-26 18:23:13 +02:00
|
|
|
void squashAll(ThreadID tid);
|
2006-05-16 20:06:35 +02:00
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Handles squashing due to a trap. */
|
2009-05-26 18:23:13 +02:00
|
|
|
void squashFromTrap(ThreadID tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-06-06 23:32:21 +02:00
|
|
|
/** Handles squashing due to an TC write. */
|
2009-05-26 18:23:13 +02:00
|
|
|
void squashFromTC(ThreadID tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2010-12-08 01:19:57 +01:00
|
|
|
/** Handles squashing from instruction with SquashAfter set.
|
|
|
|
* This differs from the other squashes as it squashes following
|
|
|
|
* instructions instead of the current instruction and doesn't
|
|
|
|
* clean up various status bits about traps/tc writes pending.
|
|
|
|
*/
|
2011-03-18 01:20:19 +01:00
|
|
|
void squashAfter(ThreadID tid, DynInstPtr &head_inst,
|
|
|
|
uint64_t squash_after_seq_num);
|
2010-12-08 01:19:57 +01:00
|
|
|
|
2007-03-23 18:13:10 +01:00
|
|
|
/** Handles processing an interrupt. */
|
|
|
|
void handleInterrupt();
|
2011-01-18 23:30:01 +01:00
|
|
|
|
|
|
|
/** Get fetch redirecting so we can handle an interrupt */
|
|
|
|
void propagateInterrupt();
|
2007-03-23 18:13:10 +01:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Commits as many instructions as possible. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void commitInsts();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Tries to commit the head ROB instruction passed in.
|
|
|
|
* @param head_inst The instruction to be committed.
|
|
|
|
*/
|
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
|
|
|
bool commitHead(DynInstPtr &head_inst, unsigned inst_num);
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Gets instructions from rename and inserts them into the ROB. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void getInsts();
|
|
|
|
|
2006-07-27 00:47:06 +02:00
|
|
|
/** Insert all instructions from rename into skidBuffer */
|
|
|
|
void skidInsert();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Marks completed instructions using information sent from IEW. */
|
2004-08-20 20:54:07 +02:00
|
|
|
void markCompletedInsts();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Gets the thread to commit, based on the SMT policy. */
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID getCommittingThread();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/** Returns the thread ID to use based on a round robin policy. */
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID roundRobin();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/** Returns the thread ID to use based on an oldest instruction policy. */
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID oldestReady();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2005-05-19 07:28:25 +02:00
|
|
|
public:
|
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
|
|
|
/** Reads the PC of a specific thread. */
|
|
|
|
TheISA::PCState pcState(ThreadID tid) { return pc[tid]; }
|
2005-05-19 07:28:25 +02:00
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Sets the PC of a specific thread. */
|
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
|
|
|
void pcState(const TheISA::PCState &val, ThreadID tid)
|
|
|
|
{ pc[tid] = val; }
|
2006-04-23 00:26:48 +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
|
|
|
/** Returns the PC of a specific thread. */
|
|
|
|
Addr instAddr(ThreadID tid) { return pc[tid].instAddr(); }
|
2006-06-16 04:01:28 +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
|
|
|
/** Returns the next PC of a specific thread. */
|
|
|
|
Addr nextInstAddr(ThreadID tid) { return pc[tid].nextInstAddr(); }
|
2007-04-14 19:13:18 +02:00
|
|
|
|
|
|
|
/** Reads the micro PC of a specific thread. */
|
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
|
|
|
Addr microPC(ThreadID tid) { return pc[tid].microPC(); }
|
2006-06-16 04:01:28 +02:00
|
|
|
|
2005-05-19 07:28:25 +02:00
|
|
|
private:
|
2004-08-20 20:54:07 +02:00
|
|
|
/** Time buffer interface. */
|
|
|
|
TimeBuffer<TimeStruct> *timeBuffer;
|
|
|
|
|
|
|
|
/** Wire to write information heading to previous stages. */
|
|
|
|
typename TimeBuffer<TimeStruct>::wire toIEW;
|
|
|
|
|
|
|
|
/** Wire to read information from IEW (for ROB). */
|
|
|
|
typename TimeBuffer<TimeStruct>::wire robInfoFromIEW;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
TimeBuffer<FetchStruct> *fetchQueue;
|
|
|
|
|
|
|
|
typename TimeBuffer<FetchStruct>::wire fromFetch;
|
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
/** IEW instruction queue interface. */
|
|
|
|
TimeBuffer<IEWStruct> *iewQueue;
|
|
|
|
|
|
|
|
/** Wire to read information from IEW queue. */
|
|
|
|
typename TimeBuffer<IEWStruct>::wire fromIEW;
|
|
|
|
|
|
|
|
/** Rename instruction queue interface, for ROB. */
|
|
|
|
TimeBuffer<RenameStruct> *renameQueue;
|
|
|
|
|
|
|
|
/** Wire to read information from rename queue. */
|
|
|
|
typename TimeBuffer<RenameStruct>::wire fromRename;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
public:
|
2004-08-20 20:54:07 +02:00
|
|
|
/** ROB interface. */
|
|
|
|
ROB *rob;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
private:
|
2006-06-16 23:08:47 +02:00
|
|
|
/** Pointer to O3CPU. */
|
|
|
|
O3CPU *cpu;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Vector of all of the threads. */
|
2006-04-23 00:26:48 +02:00
|
|
|
std::vector<Thread *> thread;
|
|
|
|
|
|
|
|
/** Records that commit has written to the time buffer this cycle. Used for
|
|
|
|
* the CPU to determine if it can deschedule itself if there is no activity.
|
|
|
|
*/
|
|
|
|
bool wroteToTimeBuffer;
|
|
|
|
|
|
|
|
/** Records if the number of ROB entries has changed this cycle. If it has,
|
|
|
|
* then the number of free entries must be re-broadcast.
|
|
|
|
*/
|
|
|
|
bool changedROBNumEntries[Impl::MaxThreads];
|
|
|
|
|
|
|
|
/** A counter of how many threads are currently squashing. */
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID squashCounter;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/** Records if a thread has to squash this cycle due to a trap. */
|
|
|
|
bool trapSquash[Impl::MaxThreads];
|
|
|
|
|
|
|
|
/** Records if a thread has to squash this cycle due to an XC write. */
|
2006-06-06 23:32:21 +02:00
|
|
|
bool tcSquash[Impl::MaxThreads];
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/** Priority List used for Commit Policy */
|
2009-05-26 18:23:13 +02:00
|
|
|
std::list<ThreadID> priority_list;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
/** IEW to Commit delay, in ticks. */
|
|
|
|
unsigned iewToCommitDelay;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Commit to IEW delay, in ticks. */
|
|
|
|
unsigned commitToIEWDelay;
|
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
/** Rename to ROB delay, in ticks. */
|
|
|
|
unsigned renameToROBDelay;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
unsigned fetchToCommitDelay;
|
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
/** Rename width, in instructions. Used so ROB knows how many
|
|
|
|
* instructions to get from the rename instruction queue.
|
|
|
|
*/
|
|
|
|
unsigned renameWidth;
|
|
|
|
|
|
|
|
/** Commit width, in instructions. */
|
|
|
|
unsigned commitWidth;
|
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
|
|
|
/** Number of Reorder Buffers */
|
|
|
|
unsigned numRobs;
|
|
|
|
|
|
|
|
/** Number of Active Threads */
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID numThreads;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-07-06 19:59:02 +02:00
|
|
|
/** Is a drain pending. */
|
|
|
|
bool drainPending;
|
2006-05-31 17:45:02 +02:00
|
|
|
|
|
|
|
/** Is commit switched out. */
|
2006-05-04 17:36:20 +02:00
|
|
|
bool switchedOut;
|
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** The latency to handle a trap. Used when scheduling trap
|
|
|
|
* squash event.
|
|
|
|
*/
|
2012-08-28 20:30:31 +02:00
|
|
|
uint trapLatency;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-11-13 08:49:03 +01:00
|
|
|
/** The interrupt fault. */
|
|
|
|
Fault interrupt;
|
|
|
|
|
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
|
|
|
/** The commit PC state of each thread. Refers to the instruction that
|
2007-04-14 19:13:18 +02:00
|
|
|
* is currently being processed/committed.
|
|
|
|
*/
|
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 pc[Impl::MaxThreads];
|
2007-04-14 19:13:18 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** The sequence number of the youngest valid instruction in the ROB. */
|
|
|
|
InstSeqNum youngestSeqNum[Impl::MaxThreads];
|
|
|
|
|
2011-01-18 23:30:05 +01:00
|
|
|
/** The sequence number of the last commited instruction. */
|
|
|
|
InstSeqNum lastCommitedSeqNum[Impl::MaxThreads];
|
|
|
|
|
2007-03-23 18:13:10 +01:00
|
|
|
/** Records if there is a trap currently in flight. */
|
|
|
|
bool trapInFlight[Impl::MaxThreads];
|
|
|
|
|
|
|
|
/** Records if there were any stores committed this cycle. */
|
|
|
|
bool committedStores[Impl::MaxThreads];
|
|
|
|
|
|
|
|
/** Records if commit should check if the ROB is truly empty (see
|
|
|
|
commit_impl.hh). */
|
|
|
|
bool checkEmptyROB[Impl::MaxThreads];
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Pointer to the list of active threads. */
|
2009-05-26 18:23:13 +02:00
|
|
|
std::list<ThreadID> *activeThreads;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
/** Rename map interface. */
|
|
|
|
RenameMap *renameMap[Impl::MaxThreads];
|
|
|
|
|
2012-02-10 15:37:30 +01:00
|
|
|
/** True if last committed microop can be followed by an interrupt */
|
|
|
|
bool canHandleInterrupts;
|
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Updates commit stats based on this instruction. */
|
2006-04-24 23:06:00 +02:00
|
|
|
void updateComInstStats(DynInstPtr &inst);
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Stat for the total number of committed instructions. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Scalar commitCommittedInsts;
|
2012-02-12 23:07:39 +01:00
|
|
|
/** Stat for the total number of committed ops. */
|
|
|
|
Stats::Scalar commitCommittedOps;
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Stat for the total number of squashed instructions discarded by commit.
|
|
|
|
*/
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Scalar commitSquashedInsts;
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Stat for the total number of times commit is told to squash.
|
|
|
|
* @todo: Actually increment this stat.
|
|
|
|
*/
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Scalar commitSquashEvents;
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Stat for the total number of times commit has had to stall due to a non-
|
|
|
|
* speculative instruction reaching the head of the ROB.
|
|
|
|
*/
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Scalar commitNonSpecStalls;
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Stat for the total number of branch mispredicts that caused a squash. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Scalar branchMispredicts;
|
2006-04-23 00:26:48 +02:00
|
|
|
/** Distribution of the number of committed instructions each cycle. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Distribution numCommittedDist;
|
2006-04-24 23:06:00 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
/** Total number of instructions committed. */
|
2012-02-12 23:07:39 +01:00
|
|
|
Stats::Vector instsCommitted;
|
|
|
|
/** Total number of ops (including micro ops) committed. */
|
|
|
|
Stats::Vector opsCommitted;
|
2006-05-16 20:06:35 +02:00
|
|
|
/** Total number of software prefetches committed. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Vector statComSwp;
|
2006-05-16 20:06:35 +02:00
|
|
|
/** Stat for the total number of committed memory references. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Vector statComRefs;
|
2006-05-16 20:06:35 +02:00
|
|
|
/** Stat for the total number of committed loads. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Vector statComLoads;
|
2006-05-16 20:06:35 +02:00
|
|
|
/** Total number of committed memory barriers. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Vector statComMembars;
|
2006-05-16 20:06:35 +02:00
|
|
|
/** Total number of committed branches. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Vector statComBranches;
|
2011-02-07 07:14:17 +01:00
|
|
|
/** Total number of floating point instructions */
|
|
|
|
Stats::Vector statComFloating;
|
|
|
|
/** Total number of integer instructions */
|
|
|
|
Stats::Vector statComInteger;
|
|
|
|
/** Total number of function calls */
|
|
|
|
Stats::Vector statComFunctionCalls;
|
2006-05-16 20:06:35 +02:00
|
|
|
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Number of cycles where the commit bandwidth limit is reached. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Scalar commitEligibleSamples;
|
2006-05-31 17:45:02 +02:00
|
|
|
/** Number of instructions not committed due to bandwidth limits. */
|
2009-03-06 04:09:53 +01:00
|
|
|
Stats::Vector commitEligible;
|
2004-08-20 20:54:07 +02:00
|
|
|
};
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
#endif // __CPU_O3_COMMIT_HH__
|