The default `is_cpu_op` value causes the COMM node to be skipped by the Workload layer.
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 196
- Forks
- 84
- PR merge metrics
- No merged PRs in 30d
Description
I converted ASTRA-sim 1.0 text input files Resnet50_DataParallel.txt into Chakra traces. However, I observed that the COMM node was skipped by the workload layer of ASTRA-sim. This means that ASTRA-sim only output the time consumed by computation, without considering communication.
This issue was caused by the text converter not setting the is_cpu_op attribute of the COMM node to false. The default value of true causes the workload to skip the node, leading to the anomaly as shown in the following code (Workload.cc):
else if (
node->is_cpu_op() ||
(!node->is_cpu_op() && node->type() == ChakraNodeType::COMP_NODE)) {
if ((node->runtime() == 0) && (node->num_ops() == 0)) {
skip_invalid(node);
}
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the text converter that produces Chakra traces from ASTRA-sim 1.0 text inputs, then inspect the COMM node handling alongside the condition shown from Workload.cc. Ensure converted COMM nodes have is_cpu_op set to false, and verify that the workload layer accounts for their communication time rather than skipping them.
Written by the indexing model from the issue text.
Assessment
- Domain
- backend
- Issue type
- Bug
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 45/100