Require equal issue mode for R-GAT (#1968)

* Require equal issue mode for R-GAT

* Add equal issue note in readme

---------

Co-authored-by: Miro <mirhodak@amd.com>

graph/R-GAT/README.md

@@ -188,3 +188,5 @@ Go inside the root folder and run benchmark inside the docker container:
 cd root
 python3 main.py --dataset igbh-dgl --dataset-path igbh/ --profile rgat-dgl-full --device gpu [--model-path <path_to_ckpt>] [--in-memory] [--dtype <fp16 or fp32>] [--scenario <SingleStream, MultiStream, Server or Offline>]
 ```
+
+**NOTE:** For official submissions, this benchmark is required to run in equal issue mode. Please make sure that the flag `rgat.*.sample_concatenate_permutation` is set to one in the [mlperf.conf](../../loadgen/mlperf.conf) file when loadgen is built.

loadgen/mlperf.conf

@@ -42,6 +42,9 @@ retinanet.MultiStream.target_latency = 528
 # 3D-UNet uses equal issue mode because it has non-uniform inputs
 3d-unet.*.sample_concatenate_permutation = 1
 
+# R-GAT uses equal issue mode because it may have non-uniform inputs
+rgat.*.sample_concatenate_permutation = 1
+
 # LLM benchmarks have non-uniform inputs and outputs, and use equal issue mode for all latency scenario
 gptj.*.sample_concatenate_permutation = 1
 llama2-70b.*.sample_concatenate_permutation = 1

tools/submission/submission_checker.py

@@ -957,6 +957,7 @@ def requires_equal_issue(self, model, division):
                 "llama2-70b-99.9",
                 "mixtral-8x7b",
                 "llama3-405b",
+                "rgat",
             ]
             and self.version not in ["v4.0", "v4.1"]
         )