[Benchmark] Detect XPU runner #7629
Conversation
Signed-off-by: Huy Do <[email protected]>
meta-cla
bot
added
the
CLA Signed
|
The latest updates on your projects. Learn more about Vercel for GitHub. |
Signed-off-by: Huy Do <[email protected]>
| elif [[ "${DEVICE_NAME}" == "cpu" ]]; then | ||
| DEVICE_TYPE="$(lscpu | grep "Model name" | sed -E 's/.*Model name:[[:space:]]*//; s/Intel\(R\)//g; s/\(R\)//g; s/\(TM\)//g; s/CPU//g; s/Processor//g; s/[[:space:]]+/ /g; s/^ //; s/ $//; s/ /_/g')_$(awk -F: '/Core\(s\) per socket/ {c=$2} /Socket\(s\)/ {s=$2} END {gsub(/ /,"",c); gsub(/ /,"",s); printf "%sc", c*s}' < <(lscpu))" | ||
| elif [[ "${DEVICE_NAME}" == "xpu" ]]; then | ||
| DEVICE_TYPE=$(xpu-smi discovery -d 0 -j | jq -r '.[0].device_name') |
There was a problem hiding this comment.
do we expect this be different? what DEVICE_TYPE we expect here? in FE we assume there is arch name, if this has some value, we might need to update FE query logics for compipler
Source:
mapping device and arch
https://github.com/pytorch/test-infra/blob/main/torchci/components/benchmark/compilers/common.tsx#L82
used in compiler query converter before query data
https://github.com/pytorch/test-infra/blob/main/torchci/components/benchmark_v3/configs/teams/compilers/config.ts#L158
There was a problem hiding this comment.
The convention in the upload script is to set DEVICE_NAME to a generic name like cuda, rocm, xpu while the DEVICE_TYPE holds the specific device name. xpu-smi discovery returns something like https://github.com/pytorch/pytorch/actions/runs/20727627657/job/59509816791#step:3:42
+-----------+--------------------------------------------------------------------------------------+
| Device ID | Device Information |
+-----------+--------------------------------------------------------------------------------------+
| 0 | Device Name: Intel(R) Data Center GPU Max 1100 |
| | Vendor Name: Intel(R) Corporation |
| | SOC UUID: 00000000-0000-001a-0000-002f0bda8086 |
| | PCI BDF Address: 0000:1a:00.0 |
| | DRM Device: /dev/dri/card3 |
| | Function Type: physical |
+-----------+--------------------------------------------------------------------------------------+
So, the DEVICE_TYPE here will be Intel(R) Data Center GPU Max 1100
There was a problem hiding this comment.
From what I see, we can keep arch as empty for XPU device for now https://github.com/pytorch/test-infra/blob/main/torchci/components/benchmark/compilers/common.tsx#L82
|
test failed pytorch/pytorch#171731 |
Darn, Claude lets me down, I don't have an XPU device to test this locally, thus the need to test it out on CI |
Signed-off-by: Huy Do <[email protected]>
Signed-off-by: Huy Do <[email protected]>
Signed-off-by: Huy Do <[email protected]>
This should fix the issue where XPU results were confused as CPU because the device name wasn't set correctly
Testing
pytorch/pytorch#171731
cc @chuanqi129