SPRADG0A April   2024  – August 2024 AM62P , AM62P-Q1

 

  1.   Abstract
  2.   2
  3.   Trademarks
  4. 1Introduction
    1. 1.1 Change Cortex-A53 Clock Frequency
  5. 2Processor Core and Compute Benchmarks
    1. 2.1 Dhrystone
    2. 2.2 CoreMark-Pro
    3. 2.3 Fast Fourier Transform
    4. 2.4 Cryptographic Benchmarks
    5. 2.5 IPC Mailbox Latency
  6. 3Memory System Benchmarks
    1. 3.1 Memory Bandwidth and Latency
      1. 3.1.1 LMBench
      2. 3.1.2 STREAM
    2. 3.2 Critical Memory Access Latency
    3. 3.3 UDMA: DDR to DDR Data Copy
  7. 4Graphics Processing Unit Benchmarks
    1. 4.1 Glmark2
    2. 4.2 GFXBench5
  8. 5Video Codec
  9. 6References
  10. 7Revision History

CoreMark-Pro

CoreMark®-Pro tests the entire processor, adding comprehensive support for multi-core technology, a combination of integer and floating-point workloads, and data sets for utilizing larger memory subsystems. The components of CoreMark-Pro utilizes all levels of cache with an up to 3MB data memory footprint. Many, but not all of the tests, are also using P threads to allow utilization of multiple cores. The score scales with the number of cores but is always less than linear (dual core score is less than 2x single core).

CoreMark-Pro must not be confused with the smaller CoreMark which, like Dhrystone, is a microbenchmark contained in L1 caches of a modern processor.

CoreMark-Pro is not included in the SDK and can be downloaded from the official host website. In this tests, the code is directly cloned and built in the AM62Px EVM. Next are the steps to clone, build, and run CoreMark-Pro directly on the target:

  1. Clone the repository.
    root@am62pxx-evm:~# git clone https://github.com/eembc/coremark-pro.git
  2. Build CoreMark-Pro
    root@am62pxx-evm:~# cd coremark-pro/
    root@am62pxx-evm:~/coremark-pro# make TARGET=linux64 build-all
  3. Run CoreMark-Pro: use "certify-all" to run all 9 benchmarks of CoreMark-Pro and "XCMD" to set the number of cores.
    root@am62pxx-evm:~/coremark-pro# make TARGET=linux64 certify-all XCMD='-c4'

All official CoreMark-Pro rules have been satisfied such as making sure that the execution time of each workload is at least 1000 times the minimum timer resolution. Table 2-2 shows the CoreMark-Pro results for single, dual, and quad A53 cores at both 1.25GHz and 1.4GHz.

Table 2-2 CoreMark®-Pro Results
Arm-Cortex-A53
At 1.25GHz [iter/s]
Parallel Scaling Arm-Cortex-A53
At 1.4GHz [iter/s]
Parallel Scaling
Single Core 850 1 936 1
Dual Core 1,531 1.82 1,700 1.82
Quad Core 2,426 2.88 2,654 2.83