Efficient Packing - Analog Devices VisualDSP++ 3.5 Manual

Table 3-4. Unpacked Order vs. Native Storage Order
Unpacked Order:
Two 24-Bit Internal Words
B3, B2, B1 B6, B5, B4
word1
Because the order of unpacked bytes does not match the requisite transfer
order, the linker must reorder the bytes into their unpacked order.
Depending on how the memory interfaces to the PMD (program memory
bus) is programmed, one command for doing so is:
PACKING(6 B2 B6 B1 B4 B6 B5);
Each set of 6 bytes would then be reordered as shown above.
The
PACKING()
fer data. In the previous example, the target must handle each 16-bit
memory read (set of three) differently. This task is computationally
expensive.

Efficient Packing

The packing (in
location is used in its entirety to build two-thirds of an instruction word
on the target. As a benefit, this scheme limits program size at the target
and, perhaps, decreases download time.
Conversely, it implies programming overhead and transfer latency con-
straints as each address in a set of three memory reads must be handled
differently. For this reason, the ordering shown above is possible rather
than obligatory. You can program the port to handle any reordering.
VisualDSP++ 3.5 Linker and Utilities Manual
for 16-Bit Processors
word2
order must match how the target is programmed to trans-
Table 3-4 on page
Linker Description File
Native Storage Order
16-bit External Memory
B2, B1 B4, B3
addr[n] addr[n+1]
3-35) is efficient. Each 16-bit memory
B6, B5
addr[n+2]
3-35