fixed point multiplier and divider
shwetika
My program deals with real numbers all of which are 32 bits long.I
have several functions like square root,exponential,cosine and sine in
my program.I need to convert all my results to fixed point.
My questions are:
1.Should i be using the floating point notation for all the 32 bit
numbers in my program wherein the exponent part is 8 bits and the
fractional part is 23 bits?
2.How do i make the following multiply code work for fixed point
numbers?
module multiply(ready,product,multiplier,multiplicand,sign,clk);
input clk;
input sign;
input [31:0] multiplier, multiplicand;
output [63:0] product;
output ready;
reg [63:0] product, product_temp;
reg [31:0] multiplier_copy;
reg [63:0] multiplicand_copy;
reg negative_output;
reg [5:0] bit;
wire ready = !bit;
initial bit = 0;
initial negative_output = 0;
always @( posedge clk )
if( ready ) begin
bit = 6'd32;
product = 0;
product_temp = 0;
multiplicand_copy = (!sign || !multiplicand[31]) ?
{ 32'd0, multiplicand } :
{ 32'd0, ~multiplicand + 1'b1};
multiplier_copy = (!sign || !multiplier[31]) ?
multiplier :
~multiplier + 1'b1;
negative_output = sign &&
((multiplier[31] && !multiplicand[31])
||(!multiplier[31] && multiplicand[31]));
end
else if ( bit > 0 ) begin
if( multiplier_copy[0] == 1'b1 ) product_temp = product_temp +
multiplicand_copy;
product = (!negative_output) ?
product_temp :
~product_temp + 1'b1;
multiplier_copy = multiplier_copy >> 1;
multiplicand_copy = multiplicand_copy << 1;
bit = bit - 1'b1;
end
endmodule
Thanks
Shwetika
glen herrmannsfeldt
> My program deals with real numbers all of which are 32 bits long.I
> have several functions like square root,exponential,cosine and sine in
> my program.I need to convert all my results to fixed point.
Are you planning for synthesis?
I would suggest if it can be done in fixed point, that would
be a better choice. You have to figure out where the binary
point should be for each, not necessarily the same for
all signals.
-- glen
kit
wrote:
Hi Shwetika,
This is what I understand from your concerns (correct me if I am
wrong):
You want to treat a 32 bit Fixed-Point number as a Floating Point
number: (Treating the bits [30:23] as the Exponent and [22:0] as the
Mantissa) If this is the case then you have to go through the basics
of IEEE Standard 754 for FP Numbers. As for as I know, this is totally
not possible!
Secondly, conversion from Fixed-Point to Floating Point is no easy
task if you are aiming at a synthesiable module. I feel it is beyond
the scope of your project.
So, as the above person's succinct suggestion I would go with Fixed
Point Numbers.
I saw your code, using procedural assignments, "=", inside an 'always'
block is not a good practice. Simulation and Synthesis might not give
the same results.
All the best.
Krishna Praveen M. R.