Oh?? sorry, i got the point. that's the optional args.
------------------ ???????? ------------------
??????:
"The gem5 Developer List"
<gem5-dev@gem5.org>;
????????: 2023??4??28??(??????) ????7:47
??????: "The gem5 Users mailing list"<gem5-us...@gem5.org>;
????: "The gem5 Developer
List"<gem5-dev@gem5.org>;"gem5-users"<gem5-us...@gem5.org>;"Matt
Sinclair"<mattdsinclair.w...@gmail.com>;"429442672"<429442...@qq.com>;
????: [gem5-dev] ??????[gem5-users] Re: Hi, i'm new to gem5. is there a
way to make simulation for a CPU-GPU heterogeneous computing??
hi,
Thank you so much for replying me.
I tried to run DNNMark, following the doc from
https://resources.gem5.org/resources/dnn-mark. However, when i ran this code :
To generate the MIOpen kernels:
cd src/gpu/DNNMark docker run --rm -v ${PWD}:${PWD}
-v${PWD}/cachefiles:/root/.cache/miopen/2.9.0 -w ${PWD}
gcr.io/gem5-test/gcn-gpu:v22-1 python3 generate_cachefiles.py cachefiles.csv
[--gfx-version={gfx801,gfx803}] [--num-cus=N]
I got result like this.
tang@tang-Matrimax-PC:~/gem5-resources/src/gpu/DNNMark$ sudo docker run --rm -v
${PWD}:${PWD} -v${PWD}/cachefiles:/root/.cache/miopen/2.9.0 -w ${PWD} gcn-gpu
python3 generate_cachefiles.py cachefiles.csv [--gfx-version={gfx801,gfx803}]
[--num-cus=N]
usage: generate_cachefiles.py [-h] [--num-cus NUM_CUS]
[--gfx-version {gfx801,gfx803,gfx900}]
csv_file
generate_cachefiles.py: error: unrecognized arguments: [--gfx-version=gfx801]
[--gfx-version=gfx803] [--num-cus=N]
tang@tang-Matrimax-PC:~/gem5-resources/src/gpu/DNNMark$ sudo docker run --rm -v
${PWD}:${PWD} -v${PWD}/cachefiles:/root/.cache/miopen/2.9.0 -w ${PWD} gcn-gpu
python3 generate_cachefiles.py cachefiles.csv [--gfx-version={gfx801,gfx803}]
[--num-cus=4]
usage: generate_cachefiles.py [-h] [--num-cus NUM_CUS]
[--gfx-version {gfx801,gfx803,gfx900}]
csv_file
generate_cachefiles.py: error: unrecognized arguments: [--gfx-version=gfx801]
[--gfx-version=gfx803] [--num-cus=4]
tang@tang-Matrimax-PC:~/gem5-resources/src/gpu/DNNMark$ sudo docker run --rm -v
${PWD}:${PWD} -v${PWD}/cachefiles:/root/.cache/miopen/2.9.0 -w ${PWD} gcn-gpu
python3 generate_cachefiles.py [--gfx-version={gfx801,gfx803}] [--num-cus=4]
cachefiles.csv
usage: generate_cachefiles.py [-h] [--num-cus NUM_CUS]
[--gfx-version {gfx801,gfx803,gfx900}]
csv_file
generate_cachefiles.py: error: unrecognized arguments: [--gfx-version=gfx803]
[--num-cus=4] cachefiles.csv
May i ask for the correct usage?
Thank you!
------------------ ???????? ------------------
??????:
"The gem5 Users mailing list"
<gem5-us...@gem5.org>;
????????: 2023??4??25??(??????) ????8:53
??????: "429442672"<429442...@qq.com>;
????: "The gem5 Developer
List"<gem5-dev@gem5.org>;"gem5-users"<gem5-us...@gem5.org>;"Matt
Sinclair"<mattdsinclair.w...@gmail.com>;
????: [gem5-users] Re: [gem5-dev] Hi, i'm new to gem5. is there a way to
make simulation for a CPU-GPU heterogeneous computing??
1. I am not personally aware of any CPU-FPGA support, but maybe others on
this mailing list can chime in if they are aware of it.
2. gem5 already has support for some small ML workloads. For
example DNNMark (https://resources.gem5.org/resources/dnn-mark) is already
integrated into gem5-resources. I know DeepBench works too, although my
group has not had time to integrate them into gem5-resources yet.
However, larger workloads (e.g., ResNet, AlexNet) do not work yet. We've
spent a lot of time getting things like that running in SE mode, but since the
ROCm versions keep changing, we decided to invest our time in getting GPU FS
mode support working first, so we didn't have to get new syscalls working over
and over. For example, in ROCm 1.6, I was able to get ResNet18 to run into the
18th layer, but then there were a number of additional failures that needed
tending to, which we haven't gotten around to fixing.
Matt
On Sun, Apr 23, 2023 at 8:42?6?2PM 429442672 <429442...@qq.com> wrote:
Thank you so much, i will follow your guide to make some trials.
By the way, i have a few more questions:
1. Does current Gem5 support simulating CPU-FPGA heterogeneous computing, if it
is possible, could you please show me a way or url?
2. Is Gem5 able to simulate deep learning workload?
------------------ ???????? ------------------
??????:
"Matt Sinclair"
<mattdsinclair.w...@gmail.com>;
????????: 2023??4??20??(??????) ????12:04
??????: "The gem5 Developer List"<gem5-dev@gem5.org>;
????: "gem5-users"<gem5-us...@gem5.org>;"429442672"<429442...@qq.com>;
????: Re: [gem5-dev] Hi, i'm new to gem5. is there a way to make
simulation for a CPU-GPU heterogeneous computing??
Yes, all of the ??GPU?? examples posted on gem5-resources do this, for both
GCN3 and Vega models. For example, I usually recommend people start with
square: https://resources.gem5.org/resources/square
You can find many more examples of this on the homepage of gem5 resources too:
https://resources.gem5.org/
Hope this helps,
Matt
On Wed, Apr 19, 2023 at 10:58?6?2PM 429442672 via gem5-dev
<gem5-dev@gem5.org> wrote:
i'm new to gem5. May i ask is there a way to make simulation for a CPU-GPU
heterogeneous computing?? the gem5-gpu is too old and poorly maintained, so it
is better to use GCN3, in ES mode.
For example, i want to simulate that:
1.CPU load several data from CPU memory and handle them.
2.CPU send them to GPU memory
3.GPU fetch data from GPU memory and handle them.
4.GPU write the data back to CPU memory
Is there a way to achieve this in Gem5?
Sincerely ask for help.
I find no example about CPU-GPU heterogeneous computing. Is there any example
here?
_______________________________________________
gem5-dev mailing list -- gem5-dev@gem5.org
To unsubscribe send an email to gem5-dev-le...@gem5.org_______________________________________________
gem5-dev mailing list -- gem5-dev@gem5.org
To unsubscribe send an email to gem5-dev-le...@gem5.org
