串行示例

运行最基本的串行示例，由于串行示例耗时较长，选取行列大小均为2048的矩阵进行任务分析。

该部分代码通过串行方式实现矩阵相乘。

图1 串行示例代码

1. 运行矩阵行列大小为2048的base_matmult的示例。

   1	./matmul 2048 0

   返回信息如下：

   1 2 3

   Size is 2048, Matrix multiplication method is: 0, Check correctness is: 0 Initialization time = 0.174492s Matrix multiplication time = 62.657254s

   矩阵行列大小为2048情况下，串行计算耗时62秒左右。

2. 创建矩阵行列大小为2048的base_matmult的Roofline任务。

   使用命令行工具进行roofline任务分析。

   1	devkit tuner roofline -o base_matmult_2048 -m region ./matmul 2048 0

   

   示例中均使用鲲鹏DevKit命令行模式，也可使用Web模式的Roofline任务进行分析。

   返回信息如下：

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

   Note: 1. Roofline task is currently only supported on the 920 platform. 2. The application must be a binary file in ELF format. 3. Roofline task collection needs to ensure the application has finished running. 4. The estimated time of roofline collection is about 3 * application estimated time. RFCOLLECT: Start collection for ./matmul RFCOLLECT: Launch application to collect performance metrics of ./matmul Size is 2048, Matrix multiplication method is: 0, Check correctness is: 0 Initialization time = 0.174628s ROOFLINE_EVENTS are initialized. Matrix multiplication time = 62.718666s RFCOLLECT: Launch application to do binary instrumentation of ./matmul Size is 2048, Matrix multiplication method is: 0, Check correctness is: 0 Initialization time = 0.528283s Matrix multiplication time = 85.328236s RFCOLLECT: Launch benchmarks for measuring roofs RFCOLLECT: Processing all collected data RFCOLLECT: Result is captured at /matrix_multiplication/rfcollect-20240506-151117.json RFCOLLECT: Run "rfreport /matrix_multiplication/rfcollect-20240506-151117.json" to get report. Get roofline report ... The roofline json report: /matrix_multiplication/base_matmult_2048.json The roofline html report: /matrix_multiplication/base_matmult_2048.html

3. 查看base_matmult_2048_html报告。
   图2 base_matmult_2048_html报告
   

   此时获取的roofs的并行度为1(即串行)，获取到Elapsed Time 62.699s，GFLOP Count 17.18，Performance 0.274 GFLOPS。

   根据Roofline分析，由于物理内核是128个，而并行线程只有1个，因此可以增加并行数来实现调优。