User manual

Table Of Contents
Zynq-7000 AP SoC Technical Reference Manual www.xilinx.com 135
UG585 (v1.11) September 27, 2016
Chapter 5: Interconnect
5.3.7 Transaction Types
Table 5-8 summarizes the command types issued to the high performance AXI interface from the PL,
and the command modifications that occur.
5.3.8 Command Interleaving and Re-Ordering
When multi-threaded commands are used in AXI, there is the potential for commands being
processed out-of-order as well as interleaving of data beats.
The DDR controller guarantees that all read commands are completed continuously, that is, it does
not interleave read data onto its external AXI ports. It does however take advantage of re-ordering of
read and write commands, to perform internal optimizations. It is therefore expected that read and
write commands issued to the DDR controller are sometimes completed in a different order from
which they were issued.
Read data interleaving is not supported by either the DDR or the OCM. However, the interconnect
might introduce read data interleaving into the system when a single PL port issues multi-threaded
read commands to both the DDR and OCM memories.
From the high performance AXI interface perspective:
Both read and write commands can be re-ordered.
Read data interleaving might occur.
Table 5-8: High Performance AXI Interface Command Types
No. Mode Command Type Translation Comments
1 64-bit 64-bit reads all burst types None Best optimization possible
2 64-bit Narrow read None
Because no upsizing is performed, the
narrower the width, the more inefficient the
transaction.
3 64-bit 64-bit write all burst types None Best optimization possible
4 64-bit Narrow write None
Because no upsizing is performed, the
narrower the width, the more inefficient the
transaction.
532-bit
32-bit INCR read aligned to
64-bit even burst multiples
Upsized
to 64-bits
Best 32-bit mode optimization possible
6 32-bit All other 32-bit read commands
Expanded
to 64-bits
Each read command is blocked until all
previous read commands are completed.
Extremely inefficient.
732-bit
32-bit INCR write aligned to
64-bit even burst multiples
Upsized
to 64-bits
Best 32-bit mode optimization possible
8 32-bit All other 32-bit write commands
Expanded
to 64-bits
Relatively inefficient because no upsizing is
performed. No blocking occurs for writes.