Hardware: MIMXRT1060-EVK.
I used modifed evkmimxrt1060_gpt_timer example (MCUXpresso IDE v10.3.0) to get coremark score.
CPU clock is changed to 528MHz. Got following results:
| Executed from | Address | Coremark score | Optimization |
| OCRAM (BOARD_SDRAM) | 0x20200000 | 2127 | O3 |
| ITCM | 0 | 2069 | O3 |
| DTCM | 0x20000000 | 2083 | O3 |
| SDRAM | 0x80000000 | 2143 | O3 |
| SDRAM | 0x80000000 | 425 | O0 |
| SDRAM | 0x80000000 | 1515 | O1 |
| SDRAM | 0x80000000 | 1515 | O2 |
FlexSPI/ QSPI flash
(BOARD_FLASH)
Advanced High-performance Bus? | 0x60000000 | 2149 | O3 |
|
|
|
|
For everything but SDRAM CMSIS-DAP was used, and for SDRAM - a J-LInk
probe with evkmimxrt1060_sdram_init.jlinkscript (with defined SKIP_SYSCLK_INIT).
A few questions/observations based on the table above:
- the scores are way below 3000 even for the reduced clock frequency
- is it really possible to execute code from DTCM?
- why QSPI flash produces the best performance?
- why TCM performance is worse than SDRAM?
