【文档修复】快速入门文档修复-asc-devkit-AtomGit

已合并

【文档修复】快速入门文档修复 #3587

YANXI_ZHAO创建于 6月25日从YANXI_ZHAO/asc-doc： master 合入到 cann/asc-devkit： master

【文档修复】快速入门文档修复 #3587

已合并

YANXI_ZHAO创建于 6月25日

从YANXI_ZHAO/asc-doc master合入到cann/asc-devkit master

共 2 个文件变更+31-25

Mdocs/guide/入门教程/快速入门/基于SIMD编程/Add算子快速入门.md+20-18

	@@ -117,38 +117,40 @@
117		117
118	- Host端代码实现：	118	- Host端代码实现：
119		119
120	- Host端通过`<<<>>>`语法糖调用Device端核函数，示例代码如下：	120	+ Host端通过`<<<>>>`语法糖调用Device端核函数，示例代码片段如下：
121	```cpp	121	```cpp
122	int32_t main(int argc, char const *argv[])	122	int32_t main(int argc, char const *argv[])
123	{	123	{
124	...	124	...
125	- constexpr uint32_t num_blocks = 8;	125	+ constexpr uint32_t numBlocks = 8;
126	...	126	...
		127	+
		128	+ // Create runtime stream by aclrtCreateStream API
		129	+ aclrtStream stream = nullptr;
		130	+ aclrtCreateStream(&stream);
		atomgit-bot6月25日 🔵 Low Priority 变更在第 128-130 行新增了 `aclrtCreateStream(&stream)` 调用，但在整个代码片段中（至第 156 行的 `}` 结束）未出现对应的 `aclrtDestroyStream(stream)` 调用。该文档作为入门教程，用户会直接复制示例代码；缺少销毁操作会示范一个资源泄漏模式，在生产代码中持续创建 stream 而不销毁将导致运行时资源耗尽。触发条件：用户按照此示例代码在循环或多次调用场景中创建 stream。失败模式：stream 句柄泄漏，最终可能导致 stream 资源耗尽。建议：在示例代码末尾（`aclrtMemcpy` 结果回传之后）增加 stream 销毁调用：`aclrtDestroyStream(stream);`，或者在 `...` 省略处附近添加注释提示用户需要销毁 stream。
127		131
128	// Allocate host and device memory, and copy input data from host to device	132	// Allocate host and device memory, and copy input data from host to device
129	- aclrtMallocHost((void **)(&z_host), total_byte_size);	133	+ aclrtMalloc((void**)&xDevice, totalByteSize, ACL_MEM_MALLOC_HUGE_FIRST);
130	- aclrtMalloc((void **)&x_device, total_byte_size, ACL_MEM_MALLOC_HUGE_FIRST);	134	+ aclrtMalloc((void**)&yDevice, totalByteSize, ACL_MEM_MALLOC_HUGE_FIRST);
131	- aclrtMalloc((void **)&y_device, total_byte_size, ACL_MEM_MALLOC_HUGE_FIRST);	135	+ aclrtMalloc((void**)&zDevice, totalByteSize, ACL_MEM_MALLOC_HUGE_FIRST);
132	- aclrtMalloc((void **)&z_device, total_byte_size, ACL_MEM_MALLOC_HUGE_FIRST);	136	+ aclrtMallocHost((void**)&zHost, totalByteSize);
133	- aclrtMemcpy((uint8_t*)x_device, total_byte_size, x_host, total_byte_size, ACL_MEMCPY_HOST_TO_DEVICE);	137	+
134	- aclrtMemcpy((uint8_t*)y_device, total_byte_size, y_host, total_byte_size, ACL_MEMCPY_HOST_TO_DEVICE);	138	+ aclrtMemcpy(xDevice, totalByteSize, x.data(), totalByteSize, ACL_MEMCPY_HOST_TO_DEVICE);
		139	+ aclrtMemcpy(yDevice, totalByteSize, y.data(), totalByteSize, ACL_MEMCPY_HOST_TO_DEVICE);
135		140
136	// Launch kernel <<<numBlocks, dynUBufSize, stream>>>	141	// Launch kernel <<<numBlocks, dynUBufSize, stream>>>
137	// numBlocks : Number of blocks. Default to 8 in this example.	142	// numBlocks : Number of blocks. Default to 8 in this example.
138	// dynUBufSize : Dynamic unified buffer size. Default to 0 in this example.	143	// dynUBufSize : Dynamic unified buffer size. Default to 0 in this example.
139	- // nullptr : Runtime stream. Uses default stream in this example.	144	+ // stream : Runtime stream.
140		145
141	- // Example One：Call add kernel which is implemented by SIMD C API	146	+ // Example：Call add kernel which is implemented by SIMD C++ Basic API
142	- add_custom<<<num_blocks, 0>>>(xDevice, yDevice, zDevice);	147	+ add_custom<blockLength><<<numBlocks, 0, stream>>>(xDevice, yDevice, zDevice);
143	-
144	- // Example Two：Call add kernel which is implemented by SIMD C++ Basic API
145	- // add_custom<blockLength><<<numBlocks, 0, stream>>>(xDevice, yDevice, zDevice);
146		148
147	// Wait for the add_custom kernel to complete	149	// Wait for the add_custom kernel to complete
148	- aclrtSynchronizeDevice();	150	+ aclrtSynchronizeStream(stream);
149		151
150	// Copy the result from device memory to host memory	152	// Copy the result from device memory to host memory
151	- aclrtMemcpy(z_host, total_byte_size, (uint8_t*)z_device, total_byte_size, ACL_MEMCPY_DEVICE_TO_HOST);	153	+ aclrtMemcpy(zHost, totalByteSize, zDevice, totalByteSize, ACL_MEMCPY_DEVICE_TO_HOST);
152	...	154	...
153	}	155	}
154	```	156	```
	@@ -156,7 +158,7 @@
156	- 算子编译与运行：	158	- 算子编译与运行：
157		159
158	CMake配置文件示例：	160	CMake配置文件示例：
159	- ```	161	+ ```cmake
160	cmake_minimum_required(VERSION 3.16)	162	cmake_minimum_required(VERSION 3.16)
161		163
162	find_package(ASC REQUIRED)	164	find_package(ASC REQUIRED)
	@@ -178,7 +180,7 @@
178	)	180	)
179	```	181	```
180	编译与执行示例：	182	编译与执行示例：
181	- ```	183	+ ```bash
182	mkdir -p build && cd build; # 创建并进入build目录	184	mkdir -p build && cd build; # 创建并进入build目录
183	cmake ..;make -j; # 编译工程	185	cmake ..;make -j; # 编译工程
184	./c_api_add_example # 运行样例	186	./c_api_add_example # 运行样例

Mdocs/guide/入门教程/快速入门/基于SIMD编程/HelloWorld.md+11-7

	@@ -22,24 +22,28 @@
22		22
23	- Host端代码实现：	23	- Host端代码实现：
24		24
25	- Host端通过<<<>>>语法糖调用Device端代码。	25	+ Host端通过<<<>>>语法糖调用Device端代码片段。
26	```cpp	26	```cpp
27	int main(int argc, char const* argv[])	27	int main(int argc, char const* argv[])
28	{	28	{
29	- aclrtSetDevice(0);	29	+ ...
		30	+ constexpr uint32_t numBlocks = 8;
		31	+ ...
		32	+
		33	+ aclrtStream stream = nullptr;
		34	+ aclrtCreateStream(&stream);
30	// Launch kernel <<<numBlocks, dynUBufSize, stream>>>	35	// Launch kernel <<<numBlocks, dynUBufSize, stream>>>
31	// numBlocks : Number of blocks. Default to 8 in this example.	36	// numBlocks : Number of blocks. Default to 8 in this example.
32	// dynUBufSize : Dynamic unified buffer size. Default to 0 in this example.	37	// dynUBufSize : Dynamic unified buffer size. Default to 0 in this example.
33	- // nullptr : Runtime stream. Uses default stream in this example.	38	+ // stream : Runtime stream. Uses stream created by aclrtCreateStream API in this example.
34	- hello_world<<<8, 0, nullptr>>>();	39	+ hello_world<<<numBlocks, 0, stream>>>();
35	- aclrtSynchronizeDevice();	40	+ ...
36	- return 0;
37	}	41	}
38	```	42	```
39		43
40	- 算子编译与运行：	44	- 算子编译与运行：
41		45
42	- ```	46	+ ```bash
43	bisheng hello_world.asc --npu-arch=dav-2201 -o demo	47	bisheng hello_world.asc --npu-arch=dav-2201 -o demo
44	./demo	48	./demo
45	```	49	```