15.4.23. Dual Memory (DualMem)

DSP Builder for Intel® FPGAs (Advanced Blockset): Handbook

Download PDF

ID 683337

Date 12/12/2022

Version

Public

A newer version of this document is available. Customers should click here to go to the newest version.

Visible to Intel only — GUID: hco1423077144404

Ixiasoft

View Details

15.4.23. Dual Memory (DualMem)

The DSP Builder DualMem block models a dual interface memory structure. You can read or write the first data interface (inputs d, a, and w). The second data interface (specified on the second address input) is read only by default, but can be read and write if you turn on Allow write on both ports. The memory size is inferred from the size of the initialization array.

The behavior of read during write cycles of the memories depends on the interface to which you read:

Reading from q1 while writing to interface 1 outputs the new data on q1 (write first behavior).
Reading from q2 while writing to interface 1 outputs the old data on q2 (read first behavior).

Turning on DONT_CARE may give a higher f_MAX for your design, especially if you implement the memory as a MLAB. When this option is on, the output is not double-registered (and therefore, in the case of MLAB implementation, uses fewer external registers), and you gain an extra half-cycle on the output. The word don’t care overlaid on the block symbol indicates the current setting is DON’T CARE. The default is off, which outputs old data for read-during-write.

Table 189. Parameters for the DualMem Block
Parameter	Description
Output data type mode	Determines how the block sets its output data type: Inherit via internal rule: the number of integer and fractional bits is the maximum of the number of bits in the input data types. Specify via dialog: you can set the output type of the block explicitly using additional fields that are available when this option is selected. This option reinterprets the output bit pattern from the LSB up according to the specified type. Boolean: the output type is Boolean.
Output data type	Specifies the output data type. For example, sfix(16), uint(8).
Output scaling value	Specifies the output scaling value. For example, 2^-15.
Initial contents	Specifies the initialization data. The size of the 1-D array determines the memory size. This parameter may also be a fi object when specifying data of arbitrarily high precision.
Use DONT_CARE when reading from and writing to the same address	Turn this option on to produce faster hardware ( a higher f_MAX) but with uncertain read data in hardware if you are simultaneously reading from and writing to the same address. Ensure that you do not read from or write to the same address at the same time to guarantee valid read data. Intel Hyperflex architectures restrict permissible configurations of the DualMem block. You might need to turn on this option for a valid configuration, if DSP Builder gives a warning.To avoid this restriction, implement a simple dual-port RAM, with port 1 as write-only (e.g. connect q1 read on port 1 to a Simulink terminator block), and port 2 as read-only (with separate addressing). When you turn on this option and you have a write on one port, a read on another port, and both have the same address, the read data is undefined. Simulink simulations represent these undefined values as zeros; the ModelSim simulation shows Xs. This difference in representation may cause simulation mismatches if you allow such undefined values to be generated. To prevent simulation mismatches: Either avoid generating accesses that cause undefined values or detect the conditions of address equality and a write access at the input Do not propagate that output.
Allow write on both ports	Turn on to read and write on both ports.
Initialize Hardware Memory Blocks with Initial Data Contents	Turn on to initialize the memory with the specified initial data values. Turn off to use only the size of the initial data for the size of memory. The Uninitialized means parameter on the Control Block Optimization tab determines the content when the memory is uninitialized..
Provide read-access inputs	Turn on to expose the memory read-access input control ('r1', 'r2') for each output port ('q1', 'q2'). You can use a read-access input to indicate when the corresponding output is not used. Disabling the output when the read-access input is '0', offers potential power-saving benefits and avoids write-read contention. When a read-access input is '0', you cannot assume the value of the corresponding output. The hardware response to read-access input '0' depends on the underlying physical memory's capabilities and design configuration. Turn off (default) to expose no read-access input controls. This setting is equivalent to permanently enabling read access to the outputs by driving the inputs with constant '1'.

You can specify the contents of the DualMem block in one of the following ways:

Use a single row or column vector to specify table contents. The length of the 1D row or column vector determines the number of addressable entries in the table. If DSP Builder reads vector data from the table, all components of a given vector share the same value.
When a look-up table contains vector data, you can provide a matrix to specify the table contents. The number of rows in the matrix determines the number of addressable entries in the table. Each row specifies the vector contents of the corresponding table entry. The number of columns must match the vector length, otherwise DSP Builder issues an error.

Every Primitive library block accepts double-precision floating-point values when specifying mask parameters. This format limits precision to no more than 53 bits, which is more than sufficient for most of the blocks. For higher precision, the Const, DualMem, or LUT blocks optionally accept values using Simulink's Fixed Point data type. For example:

constValue = fi(0.142, 1, 16, 15)

vectorValue = fi(sin([0:10]'), 1, 18, 15)

To configure a Const, DualMem, or LUT with data of precision higher than IEEE double precision, create a MATLAB fi object of the required precision that contains the high precision data. Avoid truncation when creating this object. Use the fi object to specify the Value of the Const, the Initial Contents of the DualMem block, or the Output value map of the LUT block.

Table 190. Port Interface for the DualMem Block
Signal	Direction	Type	Description	Vector Data Support	Complex Data Support
d1	Input	Any	Data to write for interface 1	Yes	Yes
a1	Input	Unsigned integer	Address to read or write from for interface 1	Yes	No
w1	Input	Boolean	Write is enabled for interface 1 when 1	Yes	No
r1	Input	Boolean	Read-access control for q output of interface 1. Only present when you turn on Provide Read-Access Inputs	Yes	No
d2	Input	Any	Data to write for interface 2. Only present when you turn on Allow write on both ports	Yes	Yes
a2	Input	Unsigned integer	Address to read or write from for interface 2	Yes	No
w2	Input	Boolean	Write is enabled for interface 2 when 1. Only present when you turn on Allow write on both ports	Yes	No
r2	Input	Boolean	Read-Access control for q output of interface 1. Only present when you turn on Provide Read-Access Inputs	Yes	No
q1	Output	Any	Data out from interface 1. (1)	Yes	Yes
q2	Output	Any	Data out from interface 2. (1)	Yes	Yes
Note: If the address for interface 1 exceeds the memory size, q1 is not defined. If the address for interface 2 exceeds the memory size, q2 is not defined.

Select Your Language

Using Intel.com Search

Quick Links

Recent Searches

Advanced Search

Only search in

DSP Builder for Intel® FPGAs (Advanced Blockset): Handbook

15.4.23. Dual Memory (DualMem)