From 96860647bc4a883db81da9194ee9048556fae314 Mon Sep 17 00:00:00 2001
From: DouZhichong <douzhichong@inspur.com>
Date: Fri, 26 Sep 2025 17:10:05 +0800
Subject: [PATCH] add blis test cases

---
 .../blis/tc_blis_basic_fun001.yaml            | 47 +++++++++++++
 .../blis/tc_blis_devel_fun002.yaml            | 42 +++++++++++
 .../blis/tc_blis_mixed_precision_fun009.yaml  | 41 +++++++++++
 .../system_base/blis/tc_blis_numa_fun010.yaml | 42 +++++++++++
 .../blis/tc_blis_openmp64_fun004.yaml         | 38 ++++++++++
 .../blis/tc_blis_openmp_fun003.yaml           | 40 +++++++++++
 .../blis/tc_blis_serial64_fun005.yaml         | 40 +++++++++++
 .../blis/tc_blis_special_fun008.yaml          | 37 ++++++++++
 .../blis/tc_blis_threads64_fun007.yaml        | 41 +++++++++++
 .../blis/tc_blis_threads_fun006.yaml          | 40 +++++++++++
 .../system_base/blis/tc_blis_basic_fun001.py  | 39 +++++++++++
 .../system_base/blis/tc_blis_devel_fun002.py  | 70 +++++++++++++++++++
 .../blis/tc_blis_mixed_precision_fun009.py    | 39 +++++++++++
 tests/system_base/blis/tc_blis_numa_fun010.py | 40 +++++++++++
 .../blis/tc_blis_openmp64_fun004.py           | 43 ++++++++++++
 .../system_base/blis/tc_blis_openmp_fun003.py | 43 ++++++++++++
 .../blis/tc_blis_serial64_fun005.py           | 39 +++++++++++
 .../blis/tc_blis_special_fun008.py            | 36 ++++++++++
 .../blis/tc_blis_threads64_fun007.py          | 43 ++++++++++++
 .../blis/tc_blis_threads_fun006.py            | 42 +++++++++++
 20 files changed, 842 insertions(+)
 create mode 100644 docs/system_base/blis/tc_blis_basic_fun001.yaml
 create mode 100644 docs/system_base/blis/tc_blis_devel_fun002.yaml
 create mode 100644 docs/system_base/blis/tc_blis_mixed_precision_fun009.yaml
 create mode 100644 docs/system_base/blis/tc_blis_numa_fun010.yaml
 create mode 100644 docs/system_base/blis/tc_blis_openmp64_fun004.yaml
 create mode 100644 docs/system_base/blis/tc_blis_openmp_fun003.yaml
 create mode 100644 docs/system_base/blis/tc_blis_serial64_fun005.yaml
 create mode 100644 docs/system_base/blis/tc_blis_special_fun008.yaml
 create mode 100644 docs/system_base/blis/tc_blis_threads64_fun007.yaml
 create mode 100644 docs/system_base/blis/tc_blis_threads_fun006.yaml
 create mode 100644 tests/system_base/blis/tc_blis_basic_fun001.py
 create mode 100644 tests/system_base/blis/tc_blis_devel_fun002.py
 create mode 100644 tests/system_base/blis/tc_blis_mixed_precision_fun009.py
 create mode 100644 tests/system_base/blis/tc_blis_numa_fun010.py
 create mode 100644 tests/system_base/blis/tc_blis_openmp64_fun004.py
 create mode 100644 tests/system_base/blis/tc_blis_openmp_fun003.py
 create mode 100644 tests/system_base/blis/tc_blis_serial64_fun005.py
 create mode 100644 tests/system_base/blis/tc_blis_special_fun008.py
 create mode 100644 tests/system_base/blis/tc_blis_threads64_fun007.py
 create mode 100644 tests/system_base/blis/tc_blis_threads_fun006.py

diff --git a/docs/system_base/blis/tc_blis_basic_fun001.yaml b/docs/system_base/blis/tc_blis_basic_fun001.yaml
new file mode 100644
index 00000000..6a641d29
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_basic_fun001.yaml
@@ -0,0 +1,47 @@
+作者: douzhichong
+优先级: P0
+支持架构: noarch
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis, basic
+用例描述: 验证BLIS基础向量和矩阵运算功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis python3-numpy
+- 软件要求:
+  - BLIS基础库
+  - NumPy ≥ 1.19 (稳定类型支持)
+
+测试步骤:
+- ## 向量运算测试
+  - 数据类型:
+    - float32: 单精度浮点
+    - float64: 双精度浮点
+    - complex64: 单精度复数
+    - complex128: 双精度复数
+  - 操作:
+    - 向量加法 [1,2,3] + [4,5,6]
+  - 验证:
+    - 结果 == [5,7,9]
+    - 误差容限 < 数据类型epsilon的10倍
+
+- ## 矩阵运算测试
+  - 数据类型: 同向量测试
+  - 输入矩阵:
+    - A: [[1,2],[3,4]]
+    - B: [[5,6],[7,8]]
+  - 操作: 矩阵乘法 (A @ B)
+  - 验证:
+    - 结果 == [[19,22],[43,50]]
+    - 误差容限 < 数据类型epsilon的100倍
+
+期望结果:
+- 所有数据类型运算结果正确
+- 向量运算误差 < 1e-6 (float32)
+- 矩阵运算误差 < 1e-12 (float64)
+- 复数运算保持共轭对称性
+- 无类型提升或截断错误
+
+环境清理:
+- 无持久化资源需要清理
diff --git a/docs/system_base/blis/tc_blis_devel_fun002.yaml b/docs/system_base/blis/tc_blis_devel_fun002.yaml
new file mode 100644
index 00000000..b7e97721
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_devel_fun002.yaml
@@ -0,0 +1,42 @@
+作者: douzhichong
+优先级: P1
+支持架构: noarch
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis-devel, blas
+用例描述: 验证BLIS开发包BLAS Level 3函数功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis-devel python3-scipy python3-numpy
+- 环境要求:
+  - BLIS开发头文件存在
+  - SciPy ≥ 1.6 (稳定BLAS接口)
+  - NumPy ≥ 1.20 (稳定F-order支持)
+
+测试步骤:
+- ## DGEMM测试(实数双精度)
+  - 输入矩阵:
+    - A: [[1,2],[3,4]] (float64, F-order)
+    - B: [[5,6],[7,8]] (float64, F-order)
+  - 运算: dgemm(1.0, A, B)
+  - 验证:
+    - 结果值匹配: [[19,22],[43,50]]
+    - 数值误差 < 1e-8
+
+- ## ZGEMM测试(复数双精度)
+  - 输入矩阵:
+    - A: [[1+2j,2+3j],[3+4j,4+5j]] (complex128, F-order)
+    - B: [[5+6j,6+7j],[7+8j,8+9j]] (complex128, F-order)
+  - 运算: zgemm(1.0, A, B)
+  - 验证:
+    - 结果与手动计算值一致
+    - 最大元素误差 < 1e-8
+    - 复数分量误差独立控制
+
+期望结果:
+- 所有BLAS函数返回正确结果
+- 实数运算误差 < 1e-8
+- 复数运算误差 < 1e-8
+- 内存布局(F-order)正确处理
+- 无内存泄漏或段错误
diff --git a/docs/system_base/blis/tc_blis_mixed_precision_fun009.yaml b/docs/system_base/blis/tc_blis_mixed_precision_fun009.yaml
new file mode 100644
index 00000000..d6a36249
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_mixed_precision_fun009.yaml
@@ -0,0 +1,41 @@
+作者: douzhichong
+优先级: P1
+支持架构: noarch
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis-openmp, mixed-precision
+用例描述: 测试BLIS混合精度矩阵运算功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis-openmp python3-numpy
+- 软件要求:
+  - BLIS ≥ 3.0 (支持自动类型提升)
+  - NumPy ≥ 1.20 (稳定混合精度支持)
+
+测试步骤:
+- ## 浮点混合精度测试
+  - 输入矩阵:
+    - A: float32 (100×100)
+    - B: float64 (100×100)
+  - 运算: 矩阵乘法 (A @ B)
+  - 验证:
+    - 结果类型为float64
+    - 数值精度误差 < 1e-8
+
+- ## 复数-实数混合测试
+  - 输入矩阵:
+    - A: complex64 (10×10)
+    - B: float32 (10×10)
+  - 运算: 矩阵乘法 (A @ B)
+  - 验证:
+    - 结果类型为complex64
+    - 数值精度误差 < 1e-6
+
+期望结果:
+- 自动完成精度提升:
+  - float32 ⊗ float64 → float64
+  - complex64 ⊗ float32 → complex64
+- 无数值溢出或精度损失
+- 支持OpenMP并行计算
+
diff --git a/docs/system_base/blis/tc_blis_numa_fun010.yaml b/docs/system_base/blis/tc_blis_numa_fun010.yaml
new file mode 100644
index 00000000..1ac9b81e
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_numa_fun010.yaml
@@ -0,0 +1,42 @@
+# tc_blis_numa_fun010.yaml
+作者: douzhichong
+优先级: P1
+支持架构: x86_64,aarch64
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis-threads, numa
+用例描述: 测试BLIS在NUMA架构下的内存分配策略
+修改人: douzhichong
+
+前置条件:
+- 执行命令:
+  - rpm -q blis-threads python3-numpy numactl
+  - numactl --hardware
+- 硬件要求:
+  - NUMA节点数 ≥ 2
+  - 每个节点内存 ≥ 4GB
+  - CPU核心数 ≥ 8
+
+测试步骤:
+- ## NUMA交错分配测试
+  - 设置: BLIS_NUMA="--interleave=all"
+  - 操作: 2000×2000矩阵乘法
+  - 验证:
+    - 结果矩阵与输入矩阵数值相等(误差<1e-10)
+    - 内存均匀分布在所有NUMA节点
+
+- ## NUMA节点绑定测试
+  - 设置: BLIS_NUMA="--cpunodebind=0"
+  - 操作: 相同矩阵乘法
+  - 验证:
+    - 计算结果正确
+    - 内存分配限制在节点0
+
+期望结果:
+- 两种NUMA策略下计算结果均正确
+- 无内存分配错误
+- 无跨节点访问性能惩罚
+
+环境清理:
+- 删除临时环境变量BLIS_NUMA
+- 释放矩阵内存
diff --git a/docs/system_base/blis/tc_blis_openmp64_fun004.yaml b/docs/system_base/blis/tc_blis_openmp64_fun004.yaml
new file mode 100644
index 00000000..7d21769a
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_openmp64_fun004.yaml
@@ -0,0 +1,38 @@
+作者: douzhichong
+优先级: P1
+支持架构: x86_64,aarch64
+执行方式: 自动
+测试类型: 性能测试
+通用标签: local, blis-openmp64, large-matrix, parallel
+用例描述: 测试64位BLIS OpenMP大矩阵并行计算功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis-openmp64 python3-numpy
+- 系统要求:
+  - 内存 ≥ 32GB (5000×5000双精度矩阵约需200MB×3=600MB)
+  - CPU核心数 ≥ 4
+  - 64位操作系统
+  - 禁用CPU节能模式
+
+测试步骤:
+- ## 内存分配验证
+  - 操作: 分配5000×5000零矩阵和单位矩阵
+  - 验证: 内存分配成功或优雅跳过测试
+
+- ## 并行矩阵乘法
+  - 设置: OMP_NUM_THREADS=4
+  - 操作: 零矩阵与单位矩阵乘法
+  - 验证:
+    - 结果矩阵与输入矩阵二进制相等(np.array_equal)
+    - 无内存错误或线程竞争
+
+- ## 并行性能监控
+  - 记录: 计算耗时和内存使用峰值
+  - 阈值: 4线程计算时间 < 15秒 (参考值)
+
+期望结果:
+- 成功完成大矩阵并行计算
+- 结果精度完全匹配(二进制相等)
+- 线程利用率 ≥ 90% (4线程时)
+
diff --git a/docs/system_base/blis/tc_blis_openmp_fun003.yaml b/docs/system_base/blis/tc_blis_openmp_fun003.yaml
new file mode 100644
index 00000000..0dae97f4
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_openmp_fun003.yaml
@@ -0,0 +1,40 @@
+作者: douzhichong
+优先级: P1
+支持架构: noarch
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis-openmp, parallel
+用例描述: 测试BLIS OpenMP多线程矩阵乘法功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis-openmp python3-numpy
+- 环境准备:
+  - 支持OpenMP 4.0+
+  - CPU核心数 ≥ 4
+  - 内存 ≥ 4GB
+
+测试步骤:
+- ## 单线程测试
+  - 设置: OMP_NUM_THREADS=1
+  - 操作: 1000×1000随机矩阵乘法
+  - 验证: 结果矩阵维度正确(1000×1000)
+
+- ## 双线程测试
+  - 设置: OMP_NUM_THREADS=2
+  - 操作: 相同矩阵乘法
+  - 验证: 结果正确且无线程竞争
+
+- ## 四线程测试
+  - 设置: OMP_NUM_THREADS=4
+  - 操作: 相同矩阵乘法
+  - 验证: 结果正确且无内存错误
+
+期望结果:
+- 所有线程配置下计算正确
+- 结果矩阵精度误差 < 1e-12
+- 线程数增加时无资源竞争
+
+环境清理:
+- 恢复OMP_NUM_THREADS环境变量
+- 释放矩阵内存
diff --git a/docs/system_base/blis/tc_blis_serial64_fun005.yaml b/docs/system_base/blis/tc_blis_serial64_fun005.yaml
new file mode 100644
index 00000000..503c4229
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_serial64_fun005.yaml
@@ -0,0 +1,40 @@
+作者: douzhichong
+优先级: P1
+支持架构: x86_64,aarch64
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis-serial64, large-matrix
+用例描述: 测试64位BLIS串行大矩阵乘法功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis-serial64 python3-numpy
+- 系统要求:
+  - 内存 ≥ 16GB (5000×5000双精度矩阵约需200MB×3=600MB)
+  - 64位操作系统
+  - 禁用CPU节能模式
+
+测试步骤:
+- ## 内存分配验证
+  - 操作: 分配5000×5000零矩阵和单位矩阵
+  - 验证: 内存分配成功或优雅跳过测试
+
+- ## 大矩阵乘法
+  - 操作: 执行零矩阵与单位矩阵乘法
+  - 验证:
+    - 结果矩阵与输入矩阵完全相等(np.array_equal)
+    - 无内存泄漏或计算错误
+
+- ## 串行性能验证(可选)
+  - 记录: 计算耗时
+  - 阈值: 单线程计算时间 < 60秒 (参考值)
+
+期望结果:
+- 成功完成大矩阵乘法运算
+- 结果精度完全匹配(二进制相等)
+- 严格单线程执行(无并行加速)
+
+环境清理:
+- 释放大矩阵内存
+- 保持环境变量不变
+
diff --git a/docs/system_base/blis/tc_blis_special_fun008.yaml b/docs/system_base/blis/tc_blis_special_fun008.yaml
new file mode 100644
index 00000000..85a5a448
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_special_fun008.yaml
@@ -0,0 +1,37 @@
+作者: douzhichong
+优先级: P1
+支持架构: noarch
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis, special-matrix
+用例描述: 测试BLIS特殊矩阵运算功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis python3-numpy
+- 软件依赖:
+  - BLIS基础库
+  - NumPy ≥ 1.16
+
+测试步骤:
+- ## 对称矩阵测试
+  - 测试矩阵: [[1, 2], [2, 3]]
+  - 运算类型: 自乘(S×S)
+  - 预期结果: [[5, 8], [8, 13]]
+  - 误差容忍: 1e-10
+
+- ## 三角矩阵测试
+  - 下三角矩阵: [[1, 0], [2, 3]]
+  - 上三角矩阵: [[1, 2], [0, 3]]
+  - 运算类型: 乘法(L×U)
+  - 预期结果: [[1, 2], [2, 13]]
+  - 误差容忍: 1e-10
+
+期望结果:
+- 对称矩阵运算结果精确匹配
+- 三角矩阵乘法结果正确
+- 无数值计算溢出
+- 支持所有浮点精度模式
+
+环境清理:
+- 无持久化资源需要清理
\ No newline at end of file
diff --git a/docs/system_base/blis/tc_blis_threads64_fun007.yaml b/docs/system_base/blis/tc_blis_threads64_fun007.yaml
new file mode 100644
index 00000000..57a7337e
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_threads64_fun007.yaml
@@ -0,0 +1,41 @@
+# tc_blis_threads64_fun007.yaml
+作者: douzhichong
+优先级: P1
+支持架构: x86_64,aarch64
+执行方式: 自动
+测试类型: 性能测试
+通用标签: local, blis-threads64, large-matrix
+用例描述: 测试64位BLIS多线程大矩阵乘法功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis-threads64 python3-numpy
+- 系统要求:
+  - 内存 ≥ 16GB (5000×5000双精度矩阵约需200MB×3=600MB)
+  - 64位操作系统
+  - CPU支持AVX/NEON指令集
+
+测试步骤:
+- ## 内存预分配检查
+  - 操作: 尝试分配5000×5000零矩阵
+  - 验证: 内存分配成功或优雅跳过
+
+- ## 多线程矩阵乘法
+  - 设置环境变量: BLIS_NUM_THREADS=4
+  - 执行操作: 5000×5000单位矩阵乘法
+  - 验证:
+    - 结果矩阵与输入矩阵完全相等
+    - 无内存错误或线程崩溃
+
+- ## 性能监控(可选)
+  - 记录: 计算耗时和内存使用峰值
+  - 阈值: 计算时间 < 30秒 (参考值)
+
+期望结果:
+- 成功完成大矩阵乘法运算
+- 结果精度完全匹配(误差<1e-12)
+- 多线程工作正常无竞争
+
+环境清理:
+- 释放大矩阵内存
+- 恢复BLIS_NUM_THREADS环境变量
diff --git a/docs/system_base/blis/tc_blis_threads_fun006.yaml b/docs/system_base/blis/tc_blis_threads_fun006.yaml
new file mode 100644
index 00000000..1a8c52ea
--- /dev/null
+++ b/docs/system_base/blis/tc_blis_threads_fun006.yaml
@@ -0,0 +1,40 @@
+# tc_blis_threads_fun006.yaml
+作者: douzhichong
+优先级: P1
+支持架构: noarch
+执行方式: 自动
+测试类型: 功能测试
+通用标签: local, blis-threads, parallel
+用例描述: 测试BLIS多线程矩阵乘法功能
+修改人: douzhichong
+
+前置条件:
+- 执行命令: rpm -q blis-threads python3-numpy
+- 环境准备:
+  - 确保系统支持多线程
+  - 内存可用空间 ≥ 2GB
+
+测试步骤:
+- ## 测试单线程模式
+  - 设置环境变量: BLIS_NUM_THREADS=1
+  - 执行操作: 1000×1000 矩阵乘法
+  - 验证: 结果矩阵形状正确 (1000×1000)
+
+- ## 测试双线程模式
+  - 设置环境变量: BLIS_NUM_THREADS=2
+  - 执行操作: 1000×1000 矩阵乘法
+  - 验证: 结果矩阵形状正确 (1000×1000)
+
+- ## 测试四线程模式
+  - 设置环境变量: BLIS_NUM_THREADS=4
+  - 执行操作: 1000×1000 矩阵乘法
+  - 验证: 结果矩阵形状正确 (1000×1000)
+
+期望结果:
+- 所有线程配置下矩阵乘法运算正常完成
+- 结果矩阵维度与输入矩阵匹配
+- 无内存泄漏或线程竞争问题
+
+环境清理:
+- 恢复原始线程设置: BLIS_NUM_THREADS=原值
+- 释放矩阵内存
diff --git a/tests/system_base/blis/tc_blis_basic_fun001.py b/tests/system_base/blis/tc_blis_basic_fun001.py
new file mode 100644
index 00000000..46a3e732
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_basic_fun001.py
@@ -0,0 +1,39 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_basic_fun001.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+from common.basetest import LocalTest
+
+class Test(LocalTest):
+    """
+    See tc_blis_basic_fun001.yaml for details
+
+    :avocado: tags=P0,noarch,local,blis
+    """
+    PARAM_DIC = {"pkg_name": "blis"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+        self.cmd("rpm -q blis")
+
+    def test(self):
+
+        for dtype in [np.float32, np.float64, np.complex64, np.complex128]:
+            v1 = np.array([1, 2, 3], dtype=dtype)
+            v2 = np.array([4, 5, 6], dtype=dtype)
+            assert np.allclose(v1 + v2, [5, 7, 9]), f"{dtype} vector addition failed"
+
+            m1 = np.array([[1, 2], [3, 4]], dtype=dtype)
+            m2 = np.array([[5, 6], [7, 8]], dtype=dtype)
+            assert np.allclose(m1 @ m2, [[19, 22], [43, 50]]), f"{dtype} matrix multiplication failed"
+
+    def tearDown(self):
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_devel_fun002.py b/tests/system_base/blis/tc_blis_devel_fun002.py
new file mode 100644
index 00000000..dcc9ac49
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_devel_fun002.py
@@ -0,0 +1,70 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_devel_fun002.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+from common.basetest import LocalTest
+
+
+class Test(LocalTest):
+    """
+    See tc_blis_devel_fun002.yaml for details
+
+    :avocado: tags=P1,noarch,local,blis-devel
+    """
+    PARAM_DIC = {"pkg_name": "blis-devel"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+
+        self.rtol = 1e-8
+        self.atol = 1e-8
+
+    def test(self):
+
+        from scipy.linalg import blas
+
+        A_real = np.array([[1, 2], [3, 4]], dtype=np.float64, order='F')
+        B_real = np.array([[5, 6], [7, 8]], dtype=np.float64, order='F')
+
+        C_real = blas.dgemm(1.0, A_real, B_real)
+        expected_real = np.array([[19, 22], [43, 50]])
+        assert np.allclose(C_real, expected_real, rtol=self.rtol, atol=self.atol), \
+            f"DGEMM failed:\nGot:\n{C_real}\nExpected:\n{expected_real}"
+
+        A_complex = np.array([
+            [1 + 2j, 2 + 3j],
+            [3 + 4j, 4 + 5j]
+        ], dtype=np.complex128, order='F')
+
+        B_complex = np.array([
+            [5 + 6j, 6 + 7j],
+            [7 + 8j, 8 + 9j]
+        ], dtype=np.complex128, order='F')
+
+        expected_complex = np.zeros((2, 2), dtype=np.complex128)
+        for i in range(2):
+            for j in range(2):
+                for k in range(2):
+                    expected_complex[i, j] += A_complex[i, k] * B_complex[k, j]
+
+        C_complex = blas.zgemm(1.0, A_complex, B_complex)
+
+        if not np.allclose(C_complex, expected_complex, rtol=self.rtol, atol=self.atol):
+            diff = np.abs(C_complex - expected_complex)
+            max_diff = np.max(diff)
+            self.log.error(f"ZGEMM validation failed. Max difference: {max_diff}")
+            self.log.debug("Actual result:\n" + str(C_complex))
+            self.log.debug("Expected result:\n" + str(expected_complex))
+            self.fail(f"ZGEMM failed with max difference {max_diff:.2e} "
+                      f"(tolerance rtol={self.rtol}, atol={self.atol})")
+
+    def tearDown(self):
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_mixed_precision_fun009.py b/tests/system_base/blis/tc_blis_mixed_precision_fun009.py
new file mode 100644
index 00000000..c64b757f
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_mixed_precision_fun009.py
@@ -0,0 +1,39 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_mixed_precision_fun009.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+from common.basetest import LocalTest
+
+class Test(LocalTest):
+    """
+    See tc_blis_mixed_precision_fun009.yaml for details
+    
+    :avocado: tags=P1,noarch,local,blis-openmp
+    """
+    PARAM_DIC = {"pkg_name": "blis-openmp"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+
+    def test(self):
+
+        A = np.random.rand(100, 100).astype(np.float32)
+        B = np.random.rand(100, 100).astype(np.float64)
+        C = A @ B  
+        assert C.dtype == np.float64, "Mixed precision failed"
+
+        A = np.random.rand(10, 10).astype(np.complex64)
+        B = np.random.rand(10, 10).astype(np.float32)
+        C = A @ B
+        assert C.dtype == np.complex64, "Complex-real mixed failed"
+
+    def tearDown(self):
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_numa_fun010.py b/tests/system_base/blis/tc_blis_numa_fun010.py
new file mode 100644
index 00000000..be97d6ba
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_numa_fun010.py
@@ -0,0 +1,40 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_numa_fun010.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+import os
+from common.basetest import LocalTest
+
+
+class Test(LocalTest):
+    """
+    See tc_blis_numa_fun010.yaml for details
+    
+    :avocado: tags=P1,arch,local,blis-threads
+    """
+    PARAM_DIC = {"pkg_name": "blis-threads"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+        self.cmd("numactl --hardware")  
+
+    def test(self):
+        n = 2000
+        A = np.ones((n, n))
+        B = np.eye(n)
+
+        for policy in ["--interleave=all", "--cpunodebind=0"]:
+            os.environ['BLIS_NUMA'] = policy
+            C = A @ B
+            assert np.allclose(C, A), f"NUMA policy {policy} failed"
+
+    def tearDown(self):
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_openmp64_fun004.py b/tests/system_base/blis/tc_blis_openmp64_fun004.py
new file mode 100644
index 00000000..61d26edb
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_openmp64_fun004.py
@@ -0,0 +1,43 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_openmp64_fun004.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+import os
+from common.basetest import LocalTest
+
+class Test(LocalTest):
+    """
+    See tc_blis_openmp64_fun004.yaml for details
+
+    :avocado: tags=P1,arch,local,blis-openmp64
+    """
+    PARAM_DIC = {"pkg_name": "blis-openmp64"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+        self.cmd("rpm -q blis-openmp64")
+        self.orig_threads = os.environ.get('OMP_NUM_THREADS', '1')
+
+    def test(self):
+
+        n = 5000
+        try:
+            A = np.zeros((n, n))
+            I = np.eye(n)
+            os.environ['OMP_NUM_THREADS'] = '4'
+            C = A @ I
+            assert np.array_equal(C, A), "Large OpenMP matrix multiply failed"
+        except MemoryError:
+            self.skipTest("Not enough memory for large matrix test")
+
+    def tearDown(self):
+        os.environ['OMP_NUM_THREADS'] = self.orig_threads
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_openmp_fun003.py b/tests/system_base/blis/tc_blis_openmp_fun003.py
new file mode 100644
index 00000000..cbcaf249
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_openmp_fun003.py
@@ -0,0 +1,43 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_openmp_fun003.py
+@Time:      2025/09/26 10:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+import os
+from common.basetest import LocalTest
+
+
+class Test(LocalTest):
+    """
+    See tc_blis_openmp_fun003.yaml for details
+
+    :avocado: tags=P1,noarch,local,blis-openmp
+    """
+    PARAM_DIC = {"pkg_name": "blis-openmp"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+        self.cmd("rpm -q blis-openmp")
+        self.orig_threads = os.environ.get('OMP_NUM_THREADS', '1')
+
+    def test(self):
+        # Test thread scaling
+        n = 1000
+        A = np.random.rand(n, n)
+        B = np.random.rand(n, n)
+
+        for threads in [1, 2, 4]:
+            os.environ['OMP_NUM_THREADS'] = str(threads)
+            C = A @ B
+            assert C.shape == (n, n), f"OpenMP with {threads} threads failed"
+
+    def tearDown(self):
+        os.environ['OMP_NUM_THREADS'] = self.orig_threads
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_serial64_fun005.py b/tests/system_base/blis/tc_blis_serial64_fun005.py
new file mode 100644
index 00000000..a1713f7d
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_serial64_fun005.py
@@ -0,0 +1,39 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_serial64_fun005.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+from common.basetest import LocalTest
+
+class Test(LocalTest):
+    """
+    See tc_blis_serial64_fun005.yaml for details
+
+    :avocado: tags=P1,arch,local,blis-serial64
+    """
+    PARAM_DIC = {"pkg_name": "blis-serial64"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+        self.cmd("rpm -q blis-serial64")
+
+    def test(self):
+
+        n = 5000
+        try:
+            A = np.zeros((n, n))
+            I = np.eye(n)
+            C = A @ I 
+            assert np.array_equal(C, A), "Large serial matrix multiply failed"
+        except MemoryError:
+            self.skipTest("Not enough memory for large matrix test")
+
+    def tearDown(self):
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_special_fun008.py b/tests/system_base/blis/tc_blis_special_fun008.py
new file mode 100644
index 00000000..4f3a994d
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_special_fun008.py
@@ -0,0 +1,36 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_special_fun008.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+from common.basetest import LocalTest
+
+class Test(LocalTest):
+    """
+    See tc_blis_special_fun008.yaml for details
+
+    :avocado: tags=P1,noarch,local,blis
+    """
+    PARAM_DIC = {"pkg_name": "blis"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+
+    def test(self):
+
+        S = np.array([[1, 2], [2, 3]], dtype=np.float64)
+        assert np.allclose(S @ S, [[5, 8], [8, 13]]), "Symmetric matrix failed"
+
+        L = np.array([[1, 0], [2, 3]], dtype=np.float64)
+        U = np.array([[1, 2], [0, 3]], dtype=np.float64)
+        assert np.allclose(L @ U, [[1, 2], [2, 13]]), "Triangular matrix failed"
+
+    def tearDown(self):
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_threads64_fun007.py b/tests/system_base/blis/tc_blis_threads64_fun007.py
new file mode 100644
index 00000000..b76fced6
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_threads64_fun007.py
@@ -0,0 +1,43 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_threads64_fun007.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+import os
+from common.basetest import LocalTest
+
+class Test(LocalTest):
+    """
+    See tc_blis_threads64_fun007.yaml for details
+
+    :avocado: tags=P1,arch,local,blis-threads64
+    """
+    PARAM_DIC = {"pkg_name": "blis-threads64"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+        self.cmd("rpm -q blis-threads64")
+        self.orig_threads = os.environ.get('BLIS_NUM_THREADS', '1')
+
+    def test(self):
+
+        n = 5000  
+        try:
+            A = np.zeros((n, n))
+            I = np.eye(n)
+            os.environ['BLIS_NUM_THREADS'] = '4'
+            C = A @ I  
+            assert np.array_equal(C, A), "Large threaded matrix multiply failed"
+        except MemoryError:
+            self.skipTest("Not enough memory for large matrix test")
+
+    def tearDown(self):
+        os.environ['BLIS_NUM_THREADS'] = self.orig_threads
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
diff --git a/tests/system_base/blis/tc_blis_threads_fun006.py b/tests/system_base/blis/tc_blis_threads_fun006.py
new file mode 100644
index 00000000..1ea0cb8c
--- /dev/null
+++ b/tests/system_base/blis/tc_blis_threads_fun006.py
@@ -0,0 +1,42 @@
+# -*- encoding: utf-8 -*-
+
+"""
+@File:      tc_blis_threads_fun006.py
+@Time:      2025/09/26 17:00:00
+@Author:    douzhichong
+@Version:   1.0
+@Contact:   douzhichong@inspur.com
+@License:   Mulan PSL v2
+"""
+
+import numpy as np
+import os
+from common.basetest import LocalTest
+
+
+class Test(LocalTest):
+    """
+    See tc_blis_threads_fun006.yaml for details
+
+    :avocado: tags=P1,noarch,local,blis-threads
+    """
+    PARAM_DIC = {"pkg_name": "blis-threads"}
+
+    def setUp(self):
+        super().setUp(self.PARAM_DIC)
+        self.cmd("rpm -q blis-threads")
+        self.orig_threads = os.environ.get('BLIS_NUM_THREADS', '1')
+
+    def test(self):
+        n = 1000
+        A = np.random.rand(n, n)
+        B = np.random.rand(n, n)
+
+        for threads in [1, 2, 4]:
+            os.environ['BLIS_NUM_THREADS'] = str(threads)
+            C = A @ B
+            assert C.shape == (n, n), f"Threaded matrix multiply with {threads} threads failed"
+
+    def tearDown(self):
+        os.environ['BLIS_NUM_THREADS'] = self.orig_threads
+        super().tearDown(self.PARAM_DIC)
\ No newline at end of file
-- 
Gitee