ascendc-operator-performance-eval

Name: ascendc-operator-performance-eval
Availability: InStock
Author: Ascend

Benchmark Ascend-C operators with torch_npu.profiler and output Markdown reports.

31|1|Updated Mar 28, 2026

Search Tags:#performance #benchmark #jsonl #profiler #torch_npu #ascend_kernel #op_statistic

Ready to run

npx skills add https://github.com/Ascend/agent-skills --skill ascendc-operator-performance-eval

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: ascendc-operator-performance-eval
Source: https://github.com/Ascend/agent-skills/tree/main/skills/ascendc-operator-performance-eval
Command: npx skills add https://github.com/Ascend/agent-skills --skill ascendc-operator-performance-eval

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Benchmark Ascend-C operator performance using torch_npu.profiler to enable deterministic dual-path comparisons (custom vs baseline) and automation of Markdown reports from profiler data.

Core Features & Use Cases

Dual-path comparison between custom operator implementation and a baseline (native) implementation using the profiler.
JSONL-based performance case generation derived from design.md and testcase-gen, ensuring full coverage across supported dtypes and shapes.
Fixed profiler schedule with warmup=5 and active=5, with results presented in Markdown and op_statistic-based metrics.

Quick Start

Run the profiler workflow on the prepared op to generate a Markdown performance report comparing custom vs baseline using a fixed warmup and active of 5.

Frequently Asked Questions about ascendc-operator-performance-eval

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I benchmark Ascend-C operator performance using torch_npu profiler?▼

Benchmark Ascend-C operator performance by applying torch_npu.profiler with a fixed schedule of warmup=5 and active=5 to quantify custom versus baseline implementations, generating a Markdown report and per-case JSONL results.

What is dual-path comparison for Ascend kernel performance evaluation?▼

Dual-path comparison for Ascend kernel performance evaluation measures custom operator implementations against native baselines using profiler metrics to quantify performance differences and output deterministic Markdown reports.

How do I generate JSONL test cases from design.md for Ascend operators?▼

Generate JSONL test cases from design.md by deriving supported dtypes and shapes, ensuring full coverage across execution modes to validate the operator performance test suite before running the profiler workflow.

Does the torch_npu profiler support fixed warmup and active iterations for benchmarking?▼

Yes, the torch_npu profiler supports fixed warmup and active iterations for benchmarking; this workflow enforces a specific schedule with warmup=5 and active=5 to ensure deterministic performance measurement across profiler runs.

What metrics are used for Ascend operator performance statistics?▼

Ascend operator performance statistics use op_statistic.csv metrics derived from torch_npu.profiler output, providing quantifiable data to compare custom and baseline implementations within the final Markdown report.