AscendC Operator Testcase Gen Skill

Generate verification testcases and benchmarks based on the operator design document (design.md) for subsequent use by the ascendc-operator-ut-gen, ascendc-operator-precision-eval, and ascendc-operator-performance-eval skills.

Usage Scenarios

Call this skill after creating

design.md

ascendc-operator-design

and before invoking

ascendc-operator-ut-gen

ascendc-operator-precision-eval

ascendc-operator-performance-eval

Calling Interface

Required Parameters

When calling this skill, you must explicitly provide the following parameters:

Parameter 1: Design Document

Design testcases based on the
```
design.md
```
provided by the user
Must include operator logic, input/output, constraints, and supported data types
If not specified, the default is design.md in the project path

Design Process

1. Operator Design Document Parsing

Read the [Operator Interface], [Computation Logic], and [Reference Implementation] sections in the design document
Generate operator benchmarks based on the actual computation logic of the operator, existing PyTorch implementations, reference documents, etc., to understand the mutual constraints on the shapes of various inputs and outputs of the operator.
Based on the constraints, generate typical testcases with common shapes when the operator is used in model scenarios, as well as generalized testcases generated according to boundary conditions.

2. Testcase Generation

MANDATORY: Before generating testcases, you must read the following reference documents:

Required Reading:

```
templates/test-cases-template.md
```
— Testcase format reference
```
design.md
```
— Operator design document provided by the user

2.1 Design Principles

Full dtype Coverage: Traverse all dtypes supported by the operator for each shape / each boundary value.
```
SUPPORTED_DTYPES
```
must include all data types supported in the [Parameter Description] section of
```
design.md
```
Shape Determined by Operator: Select appropriate shapes based on the dimensions supported by the operator; do not use fixed dimensions
Avoid Oversized Shapes: Control the number of elements in a single testcase within a reasonable range to avoid unnecessary large tensors
Total Number of Testcases = (len(TEST_SHAPES) + len(GENERAL_SHAPES)) × len(SUPPORTED_DTYPES) ≥ 30

Part A: Regular Shape Testing (TEST_SHAPES)

Select suitable shapes from the following dimension pool based on the dimensions supported by the operator and operator type to form the

TEST_SHAPES

list.

MUST select based on the actual dimensions supported by the operator; do not select unsupported dimensions.

Shape Selection Reference Pool

Dimension	Recommended Shape	Applicable Operator Type
1D	(128,), (1024,), (4096,), (8192,)	elementwise, reduction
2D	(32, 512), (64, 768), (128, 1024)	elementwise, matmul, linear
3D	(8, 16, 64), (4, 128, 256)	elementwise, attention, conv1d
4D	(4, 8, 32, 16), (2, 64, 32, 32)	conv2d, elementwise
5D	(2, 3, 4, 5, 6)	conv3d, elementwise

Avoid Oversized Shapes: It is recommended that the number of elements in a single shape ≤ 200K.

TEST_SHAPES Format

python

TEST_SHAPES = [
    ("category_name", "description", (dim0, dim1, ...)),
    # ...
]

Category names can be customized, and it is recommended to name them by dimension or scenario. Example:

python

# elementwise operator (supports any dimension)
TEST_SHAPES = [
    ("1D",         "128 elements",         (128,)),
    ("1D",         "1024 elements",        (1024,)),
    ("1D",         "4096 elements",        (4096,)),
    ("1D",         "8192 elements",        (8192,)),
    ("2D",         "32x512",              (32, 512)),
    ("3D",         "8x16x64",             (8, 16, 64)),
    ("3D",         "4x128x256",           (4, 128, 256)),
    ("4D",         "4x8x32x16",           (4, 8, 32, 16)),
]

Part B: Generalized Shape Testing (GENERAL_SHAPES)

Select suitable shapes from the following dimension pool based on the dimensions supported by the operator and operator type to form the

GENERAL_SHAPES

list.

MUST select based on the actual dimensions supported by the operator; do not select unsupported dimensions.

Shape Selection Reference Pool

Small Shape Scenarios (Boundary Testing, Minimum Values, Misalignment Testing)

Dimension	Small Shape	Testing Purpose
1D	(1,), (2,), (4,), (8,)	Minimum elements, boundary values
1D	(3,), (5,), (7,)	Misaligned lengths
2D	(1, 1), (2, 2), (4, 4)	Tiny 2D tensor
2D	(1, 128), (2, 256)	Single row/column scenario
3D	(1, 1, 1), (2, 2, 2)	Tiny 3D tensor
3D	(1, 8, 64), (2, 16, 128)	Single batch scenario

Large Shape Scenarios (Production Environment, Stress Testing, Typical Model Scenarios)

Dimension	Large Shape	Applicable Model	Number of Elements
1D	(3072,), (4096,)	BERT FFN middle layer	3K-4K
1D	(5120,), (6400,)	GPT-2 FFN middle layer	5K-6K
2D	(512, 768)	BERT-base full sequence	393K
2D	(512, 1024)	BERT-large full sequence	524K
2D	(1024, 768)	GPT-2 sequence	786K
2D	(1024, 1024)	GPT-2 medium sequence	1M
2D	(1024, 1600)	GPT-2 XL sequence	1.6M
3D	(8, 512, 768)	BERT-base batch	3.1M
3D	(8, 197, 768)	ViT-base batch	1.2M
3D	(16, 1024, 1024)	GPT-2 large batch	16.7M

Notes:

Recommended shapes for production environments and typical model scenarios should vary according to the specific application scenario of each operator

Small shapes are used for boundary testing to ensure the operator works correctly with minimal inputs

Large shapes are used for performance testing and verification of correctness with large-scale data

Shapes with more than 200K elements are only used for generalized testing and not in regular testing

GENERAL_SHAPES Format

python

GENERAL_SHAPES = [
    ("Small", "description", (dim0, dim1, ...)),
    ("Large", "description", (dim0, dim1, ...)),
    # ...
]

It is recommended to use "Small" and "Large" as categories to distinguish scenarios. Example:

python

# elementwise operator (supports any dimension)
GENERAL_SHAPES = [
    # Small Shape Scenarios
    ("Small", "single element",         (1,)),
    ("Small", "tiny vector 2",          (2,)),
    ("Small", "tiny vector 4",          (4,)),
    ("Small", "unaligned length 3",     (3,)),
    ("Small", "unaligned length 5",     (5,)),
    ("Small", "2x2 matrix",             (2, 2)),
    ("Small", "1x128 single row",       (1, 128)),
    ("Small", "1x1x1 scalar",           (1, 1, 1)),
    ("Small", "1x8x64 single batch",    (1, 8, 64)),
    
    # Large Shape Scenarios (Production Environment)
    ("Large", "BERT-base FFN 3072",     (3072,)),
    ("Large", "BERT-large FFN 4096",    (4096,)),
    ("Large", "BERT-base 512x768",      (512, 768)),
    ("Large", "BERT-large 512x1024",    (512, 1024)),
    ("Large", "GPT-2 1024x768",         (1024, 768)),
    ("Large", "GPT-2 1024x1024",        (1024, 1024)),
    ("Large", "ViT-base 8x197x768",     (8, 197, 768)),
]

2.2 Design Testcase Output

Based on the collected information, read the

templates/test-cases-template.md

template, fill in all sections, and output to

csrc/ops/[op-name]/test/[op-name]-test-cases.md

Output Location:

ascend-kernel/csrc/ops/[op-name]/test/[op-name]-test-cases.md

Notes

Strictly Follow Template Sections: Only output sections defined in the template (operator benchmark, typical testcases, generalized testcases, usage instructions)
Prohibit Output of Irrelevant Content:
- ❌ Precision verification standards (handled by
```
ascendc-operator-precision-eval
```
  )
- ❌ Performance verification standards (handled by
```
ascendc-operator-performance-eval
```
  )
- ❌ UT test code (handled by other skills)
- ❌ Any content beyond the sections defined in the template

Output File Location:

ascend-kernel/csrc/ops/[op-name]/test/[op-name]-test-cases.md

ascendc-operator-testcase-gen

NPX Install

Tags

SKILL.md Content (Chinese)

AscendC Operator Testcase Gen Skill

Usage Scenarios

Calling Interface

Required Parameters

Design Process

1. Operator Design Document Parsing

2. Testcase Generation

2.1 Design Principles

Part A: Regular Shape Testing (TEST_SHAPES)

Shape Selection Reference Pool

TEST_SHAPES Format

Part B: Generalized Shape Testing (GENERAL_SHAPES)

Shape Selection Reference Pool

GENERAL_SHAPES Format

2.2 Design Testcase Output

Notes