Forum Discussion
Confusion regarding Stratix 10 OpenCL compilation report
- 6 years ago
Yes, the numbers in "acl_quartus_report.txt" is the entire chip usage. I just did a comparison between the HTML report on Arria 10 and Stratix 10 and I think I now see what you mean. On Arria 10 reports, after place and routing, a new section is added to the summary tab of the report named " Quartus Fit Resource Utilization Summary" which has the post-place-and-route resource usage of the kernel alone. However, this does not exist in the report for Stratix 10.
Looking through the compilation files, it seems these numbers are extracted from the biggest "*.fit.rpt". There is a table called "Fitter Resource Utilization by Entity" in the file and the entity called "freeze_wrapper_inst" seems to refer to the whole dynamic partition that includes the OpenCL kernel instance and its interfaces to the BSP. The same table also exists in the compilation folder for Stratix 10, it is just that it is located in "build/output_files/". I believe the same entity in that report also refers to the dynamic partition of the design and its area usage should represent the post-place-and-route area usage of the kernel, which is what you are looking for.
The correct place to look for post-place-and-route area usage and frequency, regardless of target device, is the "acl_quartus_report.txt" file. Any number that you see in the HTML report is an "estimation" (as also mentioned in the report itself). Moreover, it seems you are looking at the area usage of the kernel alone. The final area usage will be the area usage of the kernel plus the area usage of the OpenCL BSP, which, in the case of Stratix 10, is quite large.
Thanks for your response!
I had been given to believe that the report.html file was only an estimate *if* a full compile hadn't been run, but that after a full compile, the report reflected real numbers. I'm surprised that's not true; that's very useful to know, thank you.
Thanks also about the caution regarding kernel size vs. full-design size. I am aware of the distinction. I am specifically paying attention to kernel size in my spreadsheets because my kernel is small relative to the BSP, and I'm trying to estimate how many copies of it I can fit in the full design.