Parallel Case

Symptom

There are 128 physical cores but only one parallel thread. The program running takes a long time.

Tuning Strategy

Use the basic OpenMP programming method to implement parallel computing. In a multi-core environment, improving the degree of parallelism is a direct and effective tuning measure.

Procedure

Figure 1 Parallel case code

1. Run the parallel_matmult case whose matrix determinant is 2048.

   1	./matmul 2048 1

   Command output:

   1 2 3

   Size is 2048, Matrix multiplication method is: 1, Check correctness is: 0 Initialization time = 0.175117s Matrix multiplication time = 2.971563s

   When the matrix determinant is 2048, the parallel computing takes approximately 3s.

2. Create a roofline task for the parallel_matmult case whose matrix determinant is 2048.

   1	devkit tuner roofline -o parallel_matmult_2048 -m region ./matmul 2048 1

   Command output:

   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

   Note: 1. Roofline task is currently only supported on the 920 platform. 2. The application must be a binary file in ELF format, and read permissions are required to detect the format of the application. 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. 5. Roofline analysis is available only on physical machines. 6. You can learn about the roofline profiling method by looking at document /usr/local/devkit/tuner/docs/ROOFLINE_KNOW_HOW.MD RFCOLLECT: Start collection for ./matmul RFCOLLECT: Launch application to collect performance metrics of ./matmul Size is 2048, Matrix multiplication method is: 1, Check correctness is: 0 Initialization time = 0.174164s ROOFLINE_EVENTS are initialized. Matrix multiplication time = 2.996051s RFCOLLECT: Launch application to do binary instrumentation of ./matmul Size is 2048, Matrix multiplication method is: 1, Check correctness is: 0 Initialization time = 0.523171s Matrix multiplication time = 3.427321s RFCOLLECT: Launch benchmarks for measuring roofs RFCOLLECT: Processing all collected data RFCOLLECT: Result is captured at /matrix_multiplication/rfcollect-20240506-154009.json RFCOLLECT: Run "rfreport /matrix_multiplication/rfcollect-20240506-154009.json" to get report. Get roofline report ... The roofline json report: /matrix_multiplication/parallel_matmult_2048.json The roofline html report: /matrix_multiplication/parallel_matmult_2048.html

3. View the parallel_matmult_2048.html report.
   Figure 2 parallel_matmult_2048.html
   

   In this case, Parallel Threads of roofs is 128, Elapsed Time is 2.953 seconds, GFLOP Count is 17.18, and Performance is 5.818 GFLOPS.

Tuning Result

In a multi-core environment, improving the degree of parallelism is a direct and effective tuning measure.