Dear all,
We're testing the DSP for data processing on a DM37x module from LogicPD. The module runs Linux. Running the test of matrix product from TI forums (http://e2e.ti.com/support/dsp/omap_applications_processors/f/447/p/70317/782636.aspx), we detect that the performance of ARM is higher that DSP. These are the results:
- Matrix 800x800 in DSP with no change of configuration in platform.tci = 34s
- Matriu 800x800 in DSP while activating data cache in platform.tci = 25s
- Matriu 800x800 in ARM = 18s.
We're exploring if the behavior could be a problem of DSP clock:
1 - The frequency of ARM processor could run at the frequencies 300, 600, 800 MHz or 1 GHz and the DSP, at 260, 520, 660 and 800 MHz. The frequency of the ARM could be configured from OS, Linux, it is possible to configure the frequency of DSP or it changes according to the frequency of ARM processor.
2 - Assuming that from ARM sends a 'long' task to the DSP. While the ARM is waiting for the result, is it possible that Linux decreases the frequency of ARM and consequently, it also decrease the frequency of DSP?
On the other hand, the libraries used to run the sample are:
- C6Run=0_98_03_03:
- DSPLINK_VERSION=1_65_01_06
- LPM_VERSION=1_24_03_11
- BIOS_VERSION=5_41_03_17
- XDCTOOLS_VERSION=3_16_03_36
- LINUXUTILS_VERSION=2_26_02_05
- SYSLINK_VERSION=2_00_02_80
- SYSBIOS_VERSION=6_32_01_38
- IPC_VERSION=1_23_03_31
- C6000 compiler: 6.1.4
Could you suggest anything else to test revise? Which is the optimal configuration for DSP?
Thanks and Best Regards,
Joaquim Duran