Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
15 years ago

Calculation of Number of embedded multiplier

Operation: z<= a*b;

Therer are 2 cases below and I need someone to explain on how the number below obtained. Pls explain the reason there are uses of mixed sign, unsigned and signed multiplier since this is just a signed operation. thanks

Case1:

input signed [31:0] b;

input signed [31:0] a;

(*multstyle = "dsp"*) output reg signed [63:0]z;

Simple Multipliers (9-bit) 0

Simple Multipliers (18-bit) 4

Embedded Multiplier Blocks --

Embedded Multiplier 9-bit elements 8

Signed Embedded Multipliers 1

Unsigned Embedded Multipliers 1

Mixed Sign Embedded Multipliers 2

Variable Sign Embedded Multipliers 0

Dedicated Input Shift Register Chains 0

Case2

input signed [31:0] b;

input signed [31:0] a;

(*multstyle = "dsp"*) output reg signed [31:0]z;

Simple Multipliers (9-bit) 0

Simple Multipliers (18-bit) 3

Embedded Multiplier Blocks --

Embedded Multiplier 9-bit elements 6

Signed Embedded Multipliers 0

Unsigned Embedded Multipliers 1

Mixed Sign Embedded Multipliers 2

Variable Sign Embedded Multipliers 0

Dedicated Input Shift Register Chains 0

1 Reply

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    Since the device's multipliers can only handle 18x18 multiplications, larger multiplications are implemented using several partial multiplications, like you learned in school.

    Thus, a full 32x32 multiplication requires 4 partial multiplcations. And since the signal is in the MSB, the other parts are handled as unsigned.

    1st a[17:0] * b [17:0] - unsigned * unsigned

    2nd a[31:18] * b [17:0] - signed * unsigned

    3rd a[17:0] * b [31:18] - unsigned * signed

    4th a[31:18] * b [31:18] - signed * signed

    Thus, the mix you're seeing in the report.

    As for case 2, since you're only keeping the result's 32 LSBs, the 4th term isn't needed.