Dump run flag into files in new benchmark suite

[pzread]

Dump run flags of iree-benchmark-module when building e2d test artifacts.

It should help users find and manually run benchmarks with benchmark ID from the IREE perf dashboard, as the benchmark ID (e2e model run ID) is part of the path of its run flagfile.

Example run directory:

ls e2e_test_artifacts/iree_EfficientNetV2STF_run_974086586c080164bf56972646f76a8f116cd849d51713ca3d9b380366c8d305

run.flagfile

Example dump run flagfile:

--module=iree_EfficientNetV2STF_module_c8949024e2472bec7b18c4e3757412715c248273005ca6f8d5769656ed425a84/module.vmfb
--function=forward
--input=1x384x384x3xf32=0
--device_allocator=caching
--task_topology_group_count=8
--device=local-task

Each commit should be able to be reviewed independently as separate PR.

Based on: #12387

benchmarks: x86_64, cuda

[github-actions]

Abbreviated Benchmark Summary

@ commit 877576ccaa510d939267035c0defdf5f45b8f316 (vs. base d6f8f151bd633a586d5afecf88bd8c64ea96f16c)

Improved Latencies 🎉

Benchmark Name	Average Latency (ms)	Median Latency (ms)	Latency Standard Deviation (ms)
MobileNetV3Small\_fp32 [fp32,imagenet] (exported\_tflite) [experimental-flags,fuse-padding][1-thread,full-inference,default-flags] with IREE-LLVM-CPU @ GCP-c2-standard-16 (CPU-x86\_64-CascadeLake)	3.923 (vs. 4.274, 8.21%↓)	3.920	0.023
MobileSSD\_fp32 [fp32] (exported\_tflite) [experimental-flags,fuse-padding][full-inference,default-flags] with IREE-LLVM-CPU-Sync @ GCP-c2-standard-16 (CPU-x86\_64-CascadeLake)	31.555 (vs. 34.041, 7.30%↓)	31.574	0.126
MobileNetV1\_fp32 [fp32,imagenet] (exported\_tflite) [experimental-flags,fuse-padding][1-thread,full-inference,default-flags] with IREE-LLVM-CPU @ GCP-c2-standard-16 (CPU-x86\_64-CascadeLake)	23.571 (vs. 25.344, 7.00%↓)	23.497	0.217

[Top 3 out of 8 results showed]

No improved or regressed compilation metrics 🏖️

For more information:

• Full benchmark result tables
• Build

Source Workflow Run