GPUFS+KVM simulations automatically enable AVX. This commit adds a
command line option to disable AVX if its not needed for a GPUFS
simulation.
Change-Id: Ic22592767dbdca86f3718eca9c837a8e29b6b781
During checkpoint restoration, the unserialize() function writes rptr,
wptr, and indirect buffer rptr, wptr to PM4 queue's rptr, wptr fields.
This commit updates this to write only the relevant pointers to the
queue structure. If indirect buffers are used, then it writes only the
indirect buffer pointers to the queue. If they are not used, then it
writes rptr, wptr values to the queue.
Change-Id: Iedb25a726112e1af99cc1e7bc012de51c4ebfd45
Previously, the cache recorder used the Sequencer to issue flush
requests and cache warmup requests. The GPU however uses GPUCoalescer to
access the cache, and not the Sequencer. This commit adds a GPUCoalescer
map to the cache recorder and uses it to send flushes and cache warmup
requests to any GPU caches in the system
Change-Id: I10490cf5e561c8559a98d4eb0550c62eefe769c9
This commit adds flush support to the GPU VIPER coherence protocol. The
L1 cache will now initiate a flush request if the packet it receives
is of type RubyRequestType_FLUSH. During the flush process, the L1 cache
will a request to L2 if its in either V or I state. L2 will issue a
flush request to the directory if its cache line is in the valid
state before invalidating its copy. The directory, on receiving this
request, writes data to memory and sends an ack back to the L2. L2
forwards this ack back to the L1, which then ends the flush by calling
the write callback
Change-Id: I9dfc0c7b71a1e9f6d5e9e6ed4977c1e6a3b5ba46
The GPU Coalescer does not contain cache cooldown and warmup support.
This commit updates the coalsecer to support cache cooldown during flush
and warmup during checkpoint restore.
Change-Id: I5459471dec20ff304fd5954af1079a7486ee860a
GPUFS uses aql information from PM4 queues to initialize doorbells. This
commit adds aql information to the checkpoint so that it can be used
during restoration to correctly initialize all doorbells. Additionally,
this commit also sets the hsa queue correctly during checkpoint-restoration
Change-Id: Ief3ef6dc973f70f27255234872a12c396df05d89
This allows us to generate stubs for the modules in gem5. The output
will be a "typings" directory which can be used by Pylance (Python
IntelliSense) to infer typings in Visual Studio Code.
Note: A "typings" directory in the root of the workspace is the default
location for Pylance to look for typings. This can be changed via
`python.analysis.stubPath` in "settings.json".
Usage
=====
```
pip3 install -r requirements.txt
scons build/ALL/gem5.opt -j$(nproc)
./build/ALL/gem5.opt util/gem5-stubgen.py
```
The jal and jalr share the same instruction format JumpConstructor,
which sets the IsCall and IsReturn flags by the register ID. However, it
may cause wrong instruction flags set for jal because the section
"handle the 'Jalr' instruction" misses the opcode checking. The PR fix
the issue to ensure the IsReturn can be only set in Jalr.
It is possible to execute a GPU atomic instruction using a memory
address that is in the host memory space (e.g, HMM, __managed__,
hipHostMalloc'd address). Since these are in host memory they are passed
to the SystemHub DmaDevice. However, this currently executes as a write
packet without modifying data. This leads to hangs in applications that
use atomics for forward progress (e.g., HeteroSync).
It is not clear where these are handled on a real GPU, but they are
certainly not handled by the software stack nor driver, so they must be
handled in hardware and therefore implemented in gem5. Handling for
atomics in the SystemHub makes the most sense.
To make atomics work a few extra changes need to be made to the
SystemHub. (1) The atomic is implemented as a host memory read, followed
by calling the AtomicOpFunctor, followed by a write. This requires a
second event to handle read response, performing atomic, and issuing a
write. (2) Atomics must be serialized otherwise two atomics might return
the same value which is incorrect. This patch adds serialization logic
for all request types to the same address to handle this. (3) With the
added complexity of the SystemHub, a new debug flag explicitly for
SystemHub is added.
Testing done: The heterosync application with input "sleepMutex 10 16 4"
previously hung before this patch. It passes with the patch applied.
This application tests both (1) and (2) above, as it allocates locks
with hipHostMalloc and has multiple workgroups sending an atomic request
in the same Tick, verifying the serialization mechanism.
The implementation of the x86 PACK micro-op had a logical bug that
caused the `PACKSSWB` and `PACKSSDW` instructions to produce incorrect
results. Specifically, due to a signedness error, the overflow check for
negative integers being packed always evaluated to true, resulting in
all negative integers being packed as -1 in the output.
This patch fixes the signedness error that causes the bug.
GitHub issue: https://github.com/gem5/gem5/issues/331
It is possible to execute a GPU atomic instruction using a memory
address that is in the host memory space (e.g, HMM, __managed__,
hipHostMalloc'd address). Since these are in host memory they are passed
to the SystemHub DmaDevice. However, this currently executes as a write
packet without modifying data. This leads to hangs in applications that
use atomics for forward progress (e.g., HeteroSync).
It is not clear where these are handled on a real GPU, but they are
certianly not handled by the software stack nor driver, so they must be
handled in hardware and therefore implemented in gem5. Handling for
atomics in the SystemHub makes the most sense.
To make atomics work a few extra changes need to be made to the
SystemHub. (1) The atomic is implemented as a host memory read, followed
by calling the AtomicOpFunctor, followed by a write. This requires a
second event to handle read response, performing atomic, and issuing a
write. (2) Atomics must be serialized otherwise two atomics might return
the same value which is incorrect. This patch adds serialization logic
for all request types to the same address to handle this. (3) With the
added complexity of the SystemHub, a new debug flag explicitly for
SystemHub is added.
Testing done: The heterosync application with input "sleepMutex 10 16 4"
previously hung before this patch. It passes with the patch applied.
This application tests both (1) and (2) above, as it allocates locks
with hipHostMalloc and has multiple workgroups sending an atomic request
in the same Tick, verifying the serialization mechanism.
Change-Id: Ife84b30037d1447dd384340cfeb06fdfd472fff9
The jal and jalr share the same instruction format JumpConstructor,
which sets the IsCall and IsReturn flags by the register ID.
However, it may cause wrong instruction flags set for jal because
the section "handle the 'Jalr' instruction" misses the opcode
checking. The PR fix the issue to ensure the IsReturn can be only
set in Jalr.
Change-Id: I9ad867a389256f9253988552e6567d2b505a6901
The implementation of the x86 PACK micro-op had a logical bug that
caused the `PACKSSWB` and `PACKSSDW` instructions to produce
incorrect results. Specifically, due to a signedness error, the
overflow check for negative integers being packed always evaluated
to true, resulting in all negative integers being packed as -1 in
the output.
This patch fixes the signedness error that causes the bug.
GitHub issue: https://github.com/gem5/gem5/issues/331
Change-Id: I44b7328a8ce31742a3c0dfaebd747f81751e8851
When there is race between FwdGetX
and PUTX on owner. Owner in this case hands off
ownership to GetX requestor and PUTX still goes
through. But since owner has changed, state should go back to M and PUTX
is essentially trashed.
An Unblock to the Directory in this case will give an undefined
transition. I have added transitions which indicate that when an Unblock
is served to the Directory, it means that some kind of ownership
transfer has happened while a PUTX/PUTO was in progress.
This allows users to obtain resources via the CLI instead of having to
write a python script to do so. It is essentially a nice CLI wrapper for
"gem5.resources.resource.obtain_resource"
## Usage
```sh
> scons build/ALL/gem5.opt -j `nproc`
> ./build/ALL/gem5.opt util/obtain-resource.py --help
usage: obtain-resource.py [-h] [-p PATH] [-q] id
positional arguments:
id The resource id to download.
options:
-h, --help show this help message and exit
-p PATH, --path PATH The path the resource is to be downloaded to. If not specified, the resource will be downloaded to the default
location in the gem5 local cache of resources
-q, --quiet Suppress output.
```
E.g.:
```sh
./build/ALL/gem5.opt util/obtain-resource.py arm-hello64-static -p arm-hello
```
Will download the resource with ID `arm-hello64-static` to `arm-hello`
in the CWD.
This allows for a user to specify the exact path they want a resource to
be downloaded to. This differs from 'resource_direcctory' in that a user
may specify the file/directory name of the resource (using just the
'resource_directory' will have the resource as its ID in that directory.
Change-Id: I887be6216c7607c22e49cf38226a5e4600f39057
Via this workflow we now can build and push our docker images to the
GitHub Docker container registry:
26a1ee4e61/.github/workflows/docker-build.yaml
GitHub does not charge for downloads to runners (hosted or self-hosted).
This can therefore save the project money if we download from GitHub's
Docker reigstry over Google Cloud's.
This is a test to ensure this works as intended.
The long/daily tests in "tests/gem5/gem5_library_tests" were running in
both the "testlib-long-tests" and the
"testlib-long-gem5_library_example_tests" job in the Daily tests
Workflow. The running in "testlib-long-tests" is removed in this PR.
This change, https://github.com/gem5/gem5/pull/205, mistakenly allocates
write buffer for clflush instruction when there's a cache miss. However,
clflush in gem5 is not a write instruction. Thus, the cache should
allocate miss buffer in this case.
Via this workflow we now can build and push our docker images to
the GitHub Docker container registry:
26a1ee4e61/.github/workflows/docker-build.yaml
GitHub does not charge for downloads to runners (hosted or self-hosted).
This can therefore save the project money if we download from GitHub's
Docker reigstry over Google Cloud's.
This is a test to ensure this works as intended.
Change-Id: Iccdb1b7a912f1e0a0d82b7f888694958099315b3
The long/daily tests in "tests/gem5/gem5_library_tests" were running in
both the "testlib-long-tests" and the
"testlib-long-gem5_library_example_tests" job in the Daily tests
Workflow. The running in "testlib-long-tests" is removed in this patch.
Change-Id: I1c665529e3dcb594ffb7f6e2224077ae366772d6
When there is race between FwdGetX
and PUTX on owner. Owner in this case hands off
ownership to GetX requestor and PUTX still goes
through. But since owner has changed, state should
go back to M and PUTX is essentially trashed.
An Unblock to the Directory in this case will give an undefined
transition. I have added transitions which indicate that when
an Unblock is served to the Directory, it means that some kind
of ownership transfer has happened while a PUTX/PUTO was in
progress.
Change-Id: I37439b5a363417096030a0875a51c605bd34c127
https://github.com/gem5/gem5/actions/runs/6114221855 failure was due to
to running the actions inside our 22.04-all-dependencies container. This
container does not contain docker. We must therefore run this action
outside of the container. However, due to our policy of checking out the
code within this container, we must split this into two jobs and use the
artifact upload and download to get the resources we want.
Links to #293
After calling m5_dump_reset_stats(0,0) in a test program, some
statistics like
l1_controllers.L1Dcache.m_demand_hits,
l1_controllers.L1Dcache.m_demand_misses,
l1_controllers.L1Dcache.m_demand_accesses
were not getting reset in the newer stat dumps.
This one line patch fixes that. Changes were tested with calling two
m5_dump_reset_stats(0,0) in a row for a system with 1 core, tested on
both SE and FS.
Credits: @MeatBoy106
The x87 FPU tag word (FTW) was not explicitly initialized in
{X86_64,i386}Process::initState(), resulting in holding an initial value
of zero, resulting in an invalid x87 FPU state. This commit initializes
FTW to 0xFFFF, indicating the FPU is empty at program start during
syscall emulation.
The 16-bit FTW register was also incorrectly masked down to 8-bits in
X86ISA::ISA::setMiscRegNoEffect(), leading to an invalid X87 FPU state
that later caused crashes in the X86KvmCPU. This commit corrects the
bitwidth of the mask to 16.
GitHub issue: https://github.com/gem5/gem5/issues/303
A recent PR [1] moved the TraceCPU away from the BaseCPU hierarchy.
While the common etrace_replayer.py has been amended, I missed these
hybrid TLM + TraceCPU example scripts.
[1]: https://github.com/gem5/gem5/pull/302
After calling m5_dump_reset_stats(0,0) in a test program,
some statistics like
l1_controllers.L1Dcache.m_demand_hits,
l1_controllers.L1Dcache.m_demand_misses,
l1_controllers.L1Dcache.m_demand_accesses
were not getting reset in the newer stat dumps.
This one line patch fixes that. Changes were tested with
calling two m5_dump_reset_stats(0,0) in a row for a system
with 1 core, tested on both SE and FS.
Credits to Gabriel Busnot for finding the fix.
Change-Id: I19d75996fa53d31ef20f7b206024fd38dbeac643
A recent PR [1] moved the TraceCPU away from the BaseCPU hierarchy.
While the common etrace_replayer.py has been amended, I missed these
hybrid TLM + TraceCPU example scripts.
[1]: https://github.com/gem5/gem5/pull/302
Change-Id: I7e9bc9a612d2721d72f5881ddb2fb4d9ee011587
Signed-off-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
This is used by Pylance IntelliSense to infer gem5 typing.
See "util/gem5-stubgen.py" for generating this directory.
Change-Id: Ie39762c718e5392f6194ff7c8238bd0cd677f486
Python 3's `-P` flag, when set, means `sys.path` is not prepended with
potentially unsafe paths:
https://docs.python.org/3/using/cmdline.html#cmdoption-P
This patch allows gem5 to mimic this. This is necesssary when using
`mypy.stubgen` as it expects the Python Interpreter to have the `-P`
flag.
Change-Id: I456c8001d3ee1e806190dc37142566d50d54cc90
The x87 FPU tag word (FTW) was not explicitly initialized in
{X86_64,i386}Process::initState(), resulting in holding an initial
value of zero, resulting in an invalid x87 FPU state. This commit
initializes FTW to 0xFFFF, indicating the FPU is empty at program
start during syscall emulation.
The 16-bit FTW register was also incorrectly masked down to 8-bits
in X86ISA::ISA::setMiscRegNoEffect(), leading to an invalid X87 FPU
state that later caused crashes in the X86KvmCPU. This commit
corrects the bitwidth of the mask to 16.
GitHub issue: https://github.com/gem5/gem5/issues/303
Change-Id: I97892d707998a87c1ff8546e08c15fede7eed66f
