22.4.4.3 eDMA performance example
Consider a system with the following characteristics:
• Internal SRAM can be accessed with one wait-state when viewed from the system
bus data phase
• All internal peripheral bus reads require two wait-states, and internal peripheral bus
writes three wait-states viewed from the system bus data phase
• System operates at 150 MHz
For an SRAM to internal peripheral bus transfer,
PEAKreq = 150 MHz / [ 4 + (1 + 1) + (1 + 3) + 3 ] cycles = 11.5 Mreq/sec
For an internal peripheral bus to SRAM transfer,
PEAKreq = 150 MHz / [ 4 + (1 + 2) + (1 + 1) + 3 ] cycles = 12.5 Mreq/sec
Assuming an even distribution of the two transfer types, the average peak request rate
would be:
PEAKreq = (11.5 Mreq/sec + 12.5 Mreq/sec) / 2 = 12.0 Mreq/sec
The minimum number of cycles to perform a single read/write, zero wait states on the
system bus, from a cold start where no channel is executing and eDMA is idle are:
• 11 cycles for a software, that is, a TCDn_CSR[START] bit, request
• 12 cycles for a hardware, that is, an eDMA peripheral request signal, request
Two cycles account for the arbitration pipeline and one extra cycle on the hardware
request resulting from the internal registering of the eDMA peripheral request signals.
For the peak request rate calculations above, the arbitration and request registering is
absorbed in or overlaps the previous executing channel.
Note
When channel linking or scatter/gather is enabled, a two cycle
delay is imposed on the next channel selection and startup. This
allows the link channel or the scatter/gather channel to be
eligible and considered in the arbitration pool for next channel
selection.
Chapter 22 Enhanced Direct Memory Access (eDMA)
K22F Sub-Family Reference Manual , Rev. 3, 7/2014
Freescale Semiconductor, Inc.
485
