Altera Corporation
2–17
May 2007
Stratix II Device Handbook, Volume 1
Stratix II Architecture
The Quartus II Compiler automatically creates carry chain logic during
design processing, or you can create it manually during design entry.
Parameterized functions such as LPM functions automatically take
advantage of carry chains for the appropriate functions.
The Quartus II Compiler creates carry chains longer than 16 (8 ALMs in
arithmetic or shared arithmetic mode) by linking LABs together
automatically. For enhanced fitting, a long carry chain runs vertically
allowing fast horizontal connections to TriMatrix memory and DSP
blocks. A carry chain can continue as far as a full column.
To avoid routing congestion in one small area of the device when a high
fan-in arithmetic function is implemented, the LAB can support carry
chains that only utilize either the top half or the bottom half of the LAB
before connecting to the next LAB. This leaves the other half of the ALMs
in the LAB available for implementing narrower fan-in functions in
normal mode. Carry chains that use the top four ALMs in the first LAB
carry into the top half of the ALMs in the next LAB within the column.
Carry chains that use the bottom four ALMs in the first LAB carry into the
bottom half of the ALMs in the next LAB within the column. Every other
column of LABs is top-half bypassable, while the other LAB columns are
bottom-half bypassable.
information on carry chain interconnect.
Shared Arithmetic Mode
In shared arithmetic mode, the ALM can implement a three-input add. In
this mode, the ALM is configured with four 4-input LUTs. Each LUT
either computes the sum of three inputs or the carry of three inputs. The
output of the carry computation is fed to the next adder (either to adder1
in the same ALM or to adder0 of the next ALM in the LAB) via a
dedicated connection called the shared arithmetic chain. This shared
arithmetic chain can significantly improve the performance of an adder
tree by reducing the number of summation stages required to implement
an adder tree.
Figure 2–13 shows the ALM in shared arithmetic mode.