# CUDA_Samples

**Repository Path**: xdltek/CUDA_Samples

## Basic Information

- **Project Name**: CUDA_Samples
- **Description**: This repository provides a compact set of CUDA-style samples for RPP. 
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-07-10
- **Last Updated**: 2026-07-10

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

<a id="english"></a>

# CUDA Sample Demo on RPP

Language switch: click [English](#english) or [中文](#chinese) to jump to the corresponding section.

![XDL Logo](docs/logo/logo_color_horizontal.png)

This repository provides a compact set of CUDA-style samples for RPP. It is intended to help users build, run, benchmark, and inspect representative examples for basic kernels, SRAM data movement, common algorithms, and RPP library APIs.

**Table Of Contents**

Click a section link below to jump within this README:

- [Quick Start](#quick-start)
- [Repository Layout](#repository-layout)
- [Sample Guides](#sample-guides)
- [Documentation](#documentation)
- [Prerequisites](#prerequisites)
- [How to Build](#how-to-build)
- [How to Run](#how-to-run)
- [Benchmark](#benchmark)
- [Direct Binary Usage](#direct-binary-usage)
- [Code Execution Workflow](#code-execution-workflow)
- [Key Source Files](#key-source-files)
- [How to Adapt This Demo](#how-to-adapt-this-demo)

## Quick Start

Run from the repository root:

```shell
scripts/build.sh
scripts/run.sh
```

Successful execution should print each sample name and a PASS/FAIL-style result or the sample's own correctness output.

The standard scripts are:

| Script | Purpose |
|---|---|
| `scripts/build.sh` | Cleans previous build outputs, configures CMake, and builds all samples. |
| `scripts/run.sh` | Runs the default sample set, one category, or one selected sample. It also protects the `rpp_server` startup flow when the server binary exists. |
| `scripts/benchmark.sh` | Runs samples through `scripts/run.sh` and saves a timestamped console log under `build/logs/`. |

Common first checks:

```shell
scripts/run.sh basic
scripts/run.sh library_api
scripts/benchmark.sh
```

`scripts/run.sh` resolves binaries from `build/bin/`. If `/usr/local/rpp/bin/rpp_server` exists, the script ensures it is running before launching a sample and writes the server log to `build/logs/rpp_server.log`.

`scripts/benchmark.sh` saves the full console log to `build/logs/benchmark_<target>_<timestamp>.log`.

## Repository Layout

- `CMakeLists.txt`
  Root build entry. Configures the RPP toolchain and adds all sample targets.
- [`samples/basic/`](samples/basic/)
  Minimal kernel and select examples.
- [`samples/memory_tensor/`](samples/memory_tensor/)
  SRAM and 2D load/store examples.
- [`samples/algorithm/`](samples/algorithm/)
  Reduction and prefix sum examples.
- [`samples/library_api/`](samples/library_api/)
  RPP FFT and RPP BLAS GEMM examples.
- `docs/logo/`
  Logo assets used by this README.
- `scripts/`
  Standard build, run, and benchmark command entry points.

## Sample Guides

The root README is the single documentation entry. The following list points to the main source file of each sample:

- [vector_add](samples/basic/vector_add/src/vector_add.cu): minimal vector addition sample.
- [vec_select](samples/basic/vec_select/src/vec_select.cu): predicate/select sample.
- [vec_sram_fuse](samples/memory_tensor/vec_sram_fuse/src/vec_sram_fuse.cu): SRAM fusion sample.
- [vec_load2d](samples/memory_tensor/vec_load2d/src/vec_load2d.cu): 2D load/store sample.
- [vec_reduce_max](samples/algorithm/vec_reduce_max/src/vec_reduce_max.cu): reduction sample.
- [prefix_sum](samples/algorithm/prefix_sum/src/prefix_sum.cu): prefix sum sample.
- [rpp_fft_v4](samples/library_api/rpp_fft_v4/src/simpleRPPFFT_v4.cpp): RPP FFT API sample.
- [rpp_gemm](samples/library_api/rpp_gemm/src/simpleGemm.cpp): RPP BLAS GEMM sample.

## Documentation

Public usage documentation is kept in this root README. The `docs/logo/` directory only stores README display assets.

## Prerequisites

Before building this demo, make sure the following dependencies are available:

- Linux development environment
- CMake `>= 3.10`
- `/usr/local/bin/clang++`
- RPP SDK installed under `/usr/local/rpp`
- Runtime libraries required by the target RPP accelerator environment

## How to Build

Run the following command from the repository root:

```shell
scripts/build.sh
```

`scripts/build.sh` removes the previous build directory before configuring CMake, then builds all samples.

Manual build:

```shell
mkdir -p build
cd build
cmake ..
make -j8
```

After the build completes successfully, executables are generated under:

```text
build/bin/
```

## How to Run

Run samples from the repository root after building the project:

```shell
scripts/run.sh
```

By default, `scripts/run.sh` runs the basic and library API samples. Use a category or sample name when running a narrower check:

```shell
scripts/run.sh basic
scripts/run.sh memory_tensor
scripts/run.sh algorithm
scripts/run.sh library_api
scripts/run.sh vector_add
scripts/run.sh samples/library_api/rpp_gemm
```

Each script supports `-h` and `--help` where applicable.

## Benchmark

Benchmark runs use the same sample selection style as `scripts/run.sh`:

```shell
scripts/benchmark.sh
scripts/benchmark.sh library_api
scripts/benchmark.sh samples/library_api/rpp_gemm
```

Benchmark logs are saved under `build/logs/`. Benchmark output records sample-defined timing and validation output; this README does not state fixed performance numbers.

## Direct Binary Usage

The scripts are the recommended first-use entry point because they resolve paths and protect the `rpp_server` startup flow. After building, users can also run binaries directly:

```shell
build/bin/vector_add
build/bin/vec_select
build/bin/simpleRPPFFT_v4
build/bin/simpleGemm -m 128 -n 128 -k 1024 -bf16
```

When running binaries directly, make sure `rpp_server` is already running if the installed RPP environment requires it.

## Code Execution Workflow

```text
+--------------------+
| Prepare host input |
+--------------------+
          |
          v
+--------------------+
| Copy input to DDR  |
+--------------------+
          |
          v
+--------------------+
| Move data to SRAM  |
+--------------------+
          |
          v
+--------------------+
| Run kernel/API     |
+--------------------+
          |
          v
+--------------------+
| Copy result back   |
+--------------------+
          |
          v
+--------------------+
| Verify output      |
+--------------------+
```

## Key Source Files

- `samples/basic/vector_add/src/vector_add.cu`
  Minimal RPP CUDA-style kernel sample.
- `samples/memory_tensor/vec_sram_fuse/src/vec_sram_fuse.cu`
  SRAM data movement and fused compute sample.
- `samples/algorithm/vec_reduce_max/src/vec_reduce_max.cu`
  Reduction sample.
- `samples/library_api/rpp_gemm/src/simpleGemm.cpp`
  RPP BLAS GEMM API sample.
- `samples/library_api/rpp_fft_v4/src/simpleRPPFFT_v4.cpp`
  RPP FFT API sample.
- `scripts/build.sh`
  Cleans previous build outputs, configures CMake, and builds the repository.
- `scripts/run.sh`
  Runs all, category, or single sample targets.
- `scripts/benchmark.sh`
  Runs samples and saves benchmark logs.

## How to Adapt This Demo

1. Copy a sample under the closest `samples/<category>/` directory.
2. Rename the sample directory and target name.
3. Update the source logic, input size, kernel, or RPP library API call.
4. Add the new sample directory to the category `CMakeLists.txt`.
5. Add a run mapping in `scripts/run.sh`.
6. Build, run, benchmark, and check correctness output.

<a id="chinese"></a>

# RPP CUDA Sample 示例仓库

语言切换：点击 [English](#english) 或 [中文](#chinese) 跳转到对应章节。

![XDL Logo](docs/logo/logo_color_horizontal.png)

本仓库提供一组精简的 RPP CUDA-style sample，用于帮助用户构建、运行、benchmark 和阅读基础 kernel、SRAM 数据搬移、常见算法以及 RPP 库 API 的代表性示例。

**目录**

点击以下链接可在 README 内跳转：

- [快速开始](#快速开始)
- [仓库结构](#仓库结构)
- [Sample 说明](#sample-说明)
- [项目文档](#项目文档)
- [环境依赖](#环境依赖)
- [构建方式](#构建方式)
- [运行方式](#运行方式)
- [Benchmark](#benchmark-1)
- [直接运行二进制](#直接运行二进制)
- [代码执行流程](#代码执行流程)
- [关键源码](#关键源码)
- [如何适配自己的示例](#如何适配自己的示例)

## 快速开始

在仓库根目录执行：

```shell
scripts/build.sh
scripts/run.sh
```

执行成功后，程序会打印每个 sample 名称，并输出 PASS/FAIL 风格结果或 sample 自身的正确性校验信息。

标准脚本如下：

| 脚本 | 用途 |
|---|---|
| `scripts/build.sh` | 清理上一次构建产物，执行 CMake 配置，并编译全部 sample。 |
| `scripts/run.sh` | 运行默认 sample 集合、指定分类或指定单个 sample；当 `rpp_server` 存在时，会保护 server 启动流程。 |
| `scripts/benchmark.sh` | 通过 `scripts/run.sh` 运行 sample，并将带时间戳的完整控制台日志保存到 `build/logs/`。 |

常用首次检查命令：

```shell
scripts/run.sh basic
scripts/run.sh library_api
scripts/benchmark.sh
```

`scripts/run.sh` 会从 `build/bin/` 解析可执行文件。如果 `/usr/local/rpp/bin/rpp_server` 存在，脚本会在启动 sample 前保证该服务已运行，并将 server 日志写入 `build/logs/rpp_server.log`。

`scripts/benchmark.sh` 会将完整控制台日志保存到 `build/logs/benchmark_<target>_<timestamp>.log`。

## 仓库结构

- `CMakeLists.txt`
  根构建入口。负责配置 RPP 工具链并纳入所有 sample target。
- [`samples/basic/`](samples/basic/)
  最小 kernel 和 select 示例。
- [`samples/memory_tensor/`](samples/memory_tensor/)
  SRAM 和二维 load/store 示例。
- [`samples/algorithm/`](samples/algorithm/)
  reduction 和 prefix sum 示例。
- [`samples/library_api/`](samples/library_api/)
  RPP FFT 和 RPP BLAS GEMM 示例。
- `docs/logo/`
  根 README 使用的 logo 资源。
- `scripts/`
  标准构建、运行和 benchmark 命令入口。

## Sample 说明

根 README 是当前唯一文档入口。以下列表指向每个 sample 的主源码文件：

- [vector_add](samples/basic/vector_add/src/vector_add.cu)：最小向量加法 sample。
- [vec_select](samples/basic/vec_select/src/vec_select.cu)：predicate/select sample。
- [vec_sram_fuse](samples/memory_tensor/vec_sram_fuse/src/vec_sram_fuse.cu)：SRAM 融合计算 sample。
- [vec_load2d](samples/memory_tensor/vec_load2d/src/vec_load2d.cu)：二维 load/store sample。
- [vec_reduce_max](samples/algorithm/vec_reduce_max/src/vec_reduce_max.cu)：归约 sample。
- [prefix_sum](samples/algorithm/prefix_sum/src/prefix_sum.cu)：前缀和 sample。
- [rpp_fft_v4](samples/library_api/rpp_fft_v4/src/simpleRPPFFT_v4.cpp)：RPP FFT API sample。
- [rpp_gemm](samples/library_api/rpp_gemm/src/simpleGemm.cpp)：RPP BLAS GEMM sample。

## 项目文档

公开使用说明统一保留在根 README 中。`docs/logo/` 目录仅保存 README 展示所需资源。

## 环境依赖

构建前请确认环境中已具备：

- Linux 开发环境
- CMake `>= 3.10`
- `/usr/local/bin/clang++`
- 安装在 `/usr/local/rpp` 下的 RPP SDK
- 目标 RPP 加速器运行所需的运行时库

## 构建方式

在仓库根目录执行：

```shell
scripts/build.sh
```

`scripts/build.sh` 会先删除上一次构建目录，再执行 CMake 配置并编译全部 sample。

手动构建方式：

```shell
mkdir -p build
cd build
cmake ..
make -j8
```

构建完成后，可执行文件默认生成在：

```text
build/bin/
```

## 运行方式

构建完成后，在仓库根目录运行 sample：

```shell
scripts/run.sh
```

默认情况下，`scripts/run.sh` 会运行基础样例和库 API 样例。需要缩小检查范围时，可以指定分类或 sample 名称：

```shell
scripts/run.sh basic
scripts/run.sh memory_tensor
scripts/run.sh algorithm
scripts/run.sh library_api
scripts/run.sh vector_add
scripts/run.sh samples/library_api/rpp_gemm
```

每个脚本在适用场景下支持 `-h` 和 `--help`。

## Benchmark

Benchmark 使用与 `scripts/run.sh` 相同的 sample 选择方式：

```shell
scripts/benchmark.sh
scripts/benchmark.sh library_api
scripts/benchmark.sh samples/library_api/rpp_gemm
```

Benchmark 日志会保存到 `build/logs/`。Benchmark 输出记录 sample 自身定义的耗时和校验信息；本文档不声明固定性能指标。

## 直接运行二进制

推荐第一次使用时通过脚本运行，因为脚本会统一解析路径并保护 `rpp_server` 启动流程。构建完成后，也可以直接运行二进制：

```shell
build/bin/vector_add
build/bin/vec_select
build/bin/simpleRPPFFT_v4
build/bin/simpleGemm -m 128 -n 128 -k 1024 -bf16
```

直接运行二进制时，如果当前 RPP 环境需要 `rpp_server`，请先确认该服务已在后台运行。

## 代码执行流程

```text
+--------------------+
| 准备 host 输入     |
+--------------------+
          |
          v
+--------------------+
| 拷贝输入到 DDR     |
+--------------------+
          |
          v
+--------------------+
| 搬移数据到 SRAM    |
+--------------------+
          |
          v
+--------------------+
| 执行 kernel/API    |
+--------------------+
          |
          v
+--------------------+
| 拷贝结果回 host    |
+--------------------+
          |
          v
+--------------------+
| 校验输出结果       |
+--------------------+
```

## 关键源码

- `samples/basic/vector_add/src/vector_add.cu`
  最小 RPP CUDA-style kernel sample。
- `samples/memory_tensor/vec_sram_fuse/src/vec_sram_fuse.cu`
  SRAM 数据搬移和融合计算 sample。
- `samples/algorithm/vec_reduce_max/src/vec_reduce_max.cu`
  归约 sample。
- `samples/library_api/rpp_gemm/src/simpleGemm.cpp`
  RPP BLAS GEMM API sample。
- `samples/library_api/rpp_fft_v4/src/simpleRPPFFT_v4.cpp`
  RPP FFT API sample。
- `scripts/build.sh`
  清理上一次构建产物，执行 CMake 配置并编译仓库。
- `scripts/run.sh`
  运行全部、分类或单个 sample target。
- `scripts/benchmark.sh`
  运行 sample 并保存 benchmark 日志。

## 如何适配自己的示例

1. 在最接近的 `samples/<category>/` 目录下复制一个 sample。
2. 修改 sample 目录名和 target 名称。
3. 更新源码逻辑、输入规模、kernel 或 RPP 库 API 调用。
4. 将新的 sample 目录加入对应分类的 `CMakeLists.txt`。
5. 在 `scripts/run.sh` 中添加运行映射。
6. 构建、运行、benchmark，并检查正确性输出。