GitLab

**Description**
Add barrier before 'destroy_process_group' to resolve the bug due to when multi models in one model benchmark, some processes haven't finished the previous process group while others failed to initialize new process group for the next model on rocm4.x when running bert_models.

**Major Revision**
-  Add barrier before 'destroy_process_group'.

Benchmarks: Code Revision - Revise arguments of nccl/rccl to support mpi mode and rename metric (#189)

**Description**
Revise arguments of nccl/rccl to support mpi mode for (mpi can not run in nccl/rccl due to multiple operators run in sequence without barrier) and rename metric .

**Major Revision**
- revise argument operators to be a single one

**Minor Revision**
- rename metric to remove benchmark name info
- change argument ngpus default value to be 1

**Description**
This commit fixes error of missing key 'percentile' in parsing FIO result.

**Description**
Revise the DockerBenchmark base to support image pull, image rm etc.

**Major Revision**
- image pull in _preprocess()
- image clean in _postprocess()
- execute customized commands in _benchmark()
- add unit tests

Benchmarks: Code Revision - Revise metric name generation and default config for disk performance benchmark (#175)

**Description**
This commit revises disk performance benchmark, including:
1) Add missing benchmark name in default config;
2) Avoid using reserved character ':' in metric name.

**Description**
Package frequently-used subprocess invoke into function.

**Description**
This commit adds gpu_sm_copy benchmark and related tests.

**Description**
Remove IB device port info in command to fix bug of IB loopback.

**Major Revision**
- Remove IB device port info in command to fix bug of IB loopback 

**Description**
Add gemm flops microbenchmark for amd.

**Major Revision**
- Add gemm flops microbenchmark for amd.
- Add related example and test file.

**Description**
Extract base class for gemm flops microbenchmark.

**Major Revision**
- extract base class for gemm flops microbenchmark and add related test.
- revise gemm_flops_performance for cuda.

**Description**
Rename `kernel_launch_overhead_event` to `event_overhead`, `kernel_launch_overhead_wall` to `wall_overhead`.

**Description**
Add memory bus bandwidth performance microbenchmark for amd.

**Major Revision**
- Add memory bus bandwidth performance microbenchmark for amd.
- Add related example and test file.

**Description**
This commit adds the benchmark program for GPU-initiated data transfer benchmark.

Benchmarks: Fix Bug - fix bug of microbenmark building cublas and cudnn for amd in build pipeline (#166)

**Description**
Fix bug of microbenmark building cublas and cudnn for amd

**Major Revision**
- remove cuda LANGUAGES in project()
- check CUDAToolkit quiet and then build if found

**Description**
Rename computation_communication_overlap microbenchmark metric .

**Major Revision**
- remove rank info in metric.
- simplify and rename metric.

**Description**
extract base class for memory bandwidth microbenchmark.

**Major Revision**
- revise and optimize cuda_memory_bandwidth_performance
- extract base class for memory bandwidth microbenchmark
- add test for base class

**Description**
This commit adds readwrite I/O pattern for FIO benchmark. Read/write ratio is fixed at 4:1.

**Description**
Generate the summarized output files from all nodes. For each metric, do the reduce operation according to the `reduce_op`

**Major Revision**
- Generate the summarized json file per node:
For microbenchmark, the format is `{benchmark_name}/[{run_count}/]{metric_name}[:rank]`
For modelbenchmark, the format is `{benchmark_name}/{sub_benchmark_name}/[{run_count}/]{metric_name}`
`[]` means optional.
```
{
  "kernel-launch/overhead_event:0": 0.00583,
  "kernel-launch/overhead_event:1": 0.00545,
  "kernel-launch/overhead_event:2": 0.00581,
  "kernel-launch/overhead_event:3": 0.00572,
  "kernel-launch/overhead_event:4": 0.00559,
  "kernel-launch/overhead_event:5": 0.00591,
  "kernel-launch/overhead_event:6": 0.00562,
  "kernel-launch/overhead_event:7": 0.00586,
  "resnet_models/pytorch-resnet50/steptime-train-float32": 544.0827468410134,
  "resnet_models/pytorch-resnet50/throughput-train-float32": 353.7607016465773,
  "resnet_models/pytorch-resnet50/steptime-train-float16": 425.40482617914677,
  "resnet_models/pytorch-resnet50/throughput-train-float16": 454.0142363793973,
  "pytorch-sharding-matmul/0/allreduce": 10.561786651611328,
  "pytorch-sharding-matmul/1/allreduce": 10.561786651611328,
  "pytorch-sharding-matmul/0/allgather": 10.088025093078613,
  "pytorch-sharding-matmul/1/allgather": 10.088025093078613
}
```
- Generate the summarized jsonl file for all nodes, each line is the result from one node in json format.

**Description**
Change the field name `reduce` to `reduce_op`.

**Description**
Set reduce type for current benchmarks' metrics, including model benchmarks and ShardingMatmul.

**Description**
Replace `sum(results) / len(results)` with `statistics.mean(results)`

**Description**
Add reduce function support for output summary.

**Major Revision**
- Add reducer class to maintain all reduce functions.
- Save reduce type of each metric into `BenchmarkResult`
- Fix UT.

**Description**
Add rccl bandwidth microbenchmark for rocm.

**Major Revision**
- Register rccl-bw benchmark.

__Description__
Cherry-pick bug fixes from v0.2.1 to main.

__Major Revisions__
* Fix bug of VGG models failed on A100 GPU with batch_size=128.
* Fix Ansible connection issue when running in localhost.
* Update version in packages and docs.

**Description**
Add the source code of rocm kernel launch overhead benchmark. 

**Major Revision**
- Revise cmake build logic to support both cuda and rocm

**Description**
Support rocm cmake build. 

**Major Revision**
- Add  some envs in rocm_common.cmake to support rocm cmake build.

**Description**
Add NCCL performance microbenchmark.

**Major Revision**
- Add microbenchmark, example, test, config for NCCL

**Description**
Add RDMA Loopback performance microbenchmark.

**Major Revision**
- Add microbenchmark, example, test, config for RDMA Loopback

**Description**
Add disk performance microbenchmark.

**Major Revision**
- Add microbenchmark, example, test, config for disk performance.

**Minor Revision**
- Fix bugs in executor unit test related to default enabled tests.

Add microbenchmark, example, test, config for cuda memory performance and Add cuda-samples(tag with cuda version) as git submodule and update related makefile

* fix bug for nvidia v100
* hard code the supported dict for different arch.

* add config file for ndv4.

* replace torch.optim.AdamW with transformers.AdamW.

Benchmarks: Add Feature - Add DistributedImpl and DistributedBackend arguments for micro benchmark. (#100)

rename bin name and result metric of cublas and cudnn microbenchmark

Fix bugs and refine log in single GPU benchmarks:

* Fix none framework issue
* Fix empty parameter bug
* Remove missed mobilenet_v3 models
* Change benchmark registration log to debug level
* Add pid in logging
* Add missing benchmarks in default config
* Fix deprecated logging warn

* Clean up the cache.

* fix return code reset issue

* Change default shape of sharding-matmul.