Skip to content

GCC vs CCRX

In this study, we look for the performance differences between a publicly available compiler (GCC) and a proprietary industrial compiler (CCRX) for the Renesas RX65N board.

FC and CNN models converted by Renesas eAI Translator were used in our experiments.

No Flash Size or RAM Usage

Since we could not obtain the flash size and RAM usage for the CCRX compiler, we have omitted these metrics from the study.


Model Type:

Models

FC parameters and MACs FC parameters and MACs
FC parameters and MACs
CNN parameters and MACs CNN parameters and MACs
CNN parameters and MACs

Error

FC - Renesas RX65N FC - Renesas RX65N
FC - Renesas RX65N
CNN - Renesas RX65N CNN - Renesas RX65N
CNN - Renesas RX65N

Execution Time

FC - Renesas RX65N FC - Renesas RX65N
FC - Renesas RX65N
CNN - Renesas RX65N CNN - Renesas RX65N
CNN - Renesas RX65N

Summary

  • Model Correctness: As a result of using the eAI Translator, we have seen that some models fail to output correct results using GCC. The same applies to CCRX, and even the int8 only version of CNN_4 which previously had acceptable error, now has a high error rate. Other than these cases, the two have the same error rates.

  • Execution Time: The execution time of GCC is better than CCRX for all models.

  • Conclusion: GCC is a bit more reliable than CCRX and executes the models faster.