[TOC]

# 1. Parity Generator

## Top Function (Kernel)

### In HLS

// ap_uint<W> a  means "a" is a W-bit unsigned integer
bool parity_generator(ap_uint<W> a) {
#pragma HLS INTERFACE ap_none      port=a
#pragma HLS INTERFACE ap_ctrl_none port=return
    
    bool parity = 0;
    
    for (int i = 0; i < W; i++){
#pragma HLS UNROLL
        parity = parity ^ a[i];
    }
    return parity;
}

#### Without UNROLL

Following is the logic if we don't add `#pragma HLS UNROLL` to unroll the loop.


- A circuit with feedback incorporates some form of memory or state storage, which allows it to consider past input or outputs when generating the current output.
- These circuits are known as sequential logic circuits.

Following are the utilization and the dataflow if we don't unroll the loop.


#### Loop Unrolling

We can unroll the loop manually


- The green code is one way of manually unrolling the loop.
- Unroll loops can create multiple independent operations rather than a single collection of operations.

#### **#pragma HLS UNROLL**

We can unroll the loop automatically by adding `pragma HLS UNROLL`,


and it will create a balanced tree structure as follow:


- Transforms loops by creating multiple copies of the loop body in RTL design, which allows some or all loop iterations to occur in parallel.

#### .v file in HLS:

- As we've added `#pragma HLS UNROLL` in our design, HLS creates the balanced structure as I mentioned before:



- Following are the utilization and the dataflow if we added `#pragma HLS UNROLL` in our design.


#### Parity Generator in Testbench(golden in HLS)

bool parity_generator_golden(ap_uint<W> a) {
    bool parity = 0;
    
    for (int i = 0; i < W; i++) {
        parity = parity ^ a[i];
    }
    return parity;
}

#### Result:

- Which generates even parity.


### In Verilog

- Here is the kernel which writes with Verilog:

module parity_generator(
    input  wire [15:0] a,
    output reg  [0:0]  parity
    );
    
    integer i;

    always @* begin
        parity = 1'b0;
        for (i = 0; i < 16; i = i + 1) begin
            parity = parity ^ a[i];
        end
    end
    
endmodule

## Testbench

- Testbench in Verilog

module vic_group3_testbench();
    wire [0:0]  parity_hw;
    reg  [0:0]  parity_sw;
    reg  [15:0] a_hw;
    
    // Design under test
    design_1_wrapper DUT0 (.a(a_hw), .parity(parity_hw));
    // parity_generator DUT1 (.a(a_hw), .parity(parity_hw));
    
    reg  [15:0] tmp;
    
    integer i, j;

- Golden parity generator in the testbench

    task parity_generator_sw;
        input  [15:0] a;
        output [0:0]  parity_sw;
        begin
            parity_sw = 1'b0;
            for (i = 0; i < 16; i = i+1) begin
                parity_sw = parity_sw ^ a[i];
            end
        end
    endtask

- Main function of testbench

    initial begin
        tmp = 16'd0; // initialize the value
        for (j = 0; j < 65536; j = j + 1) begin
            a_hw = tmp;
            // Get the value of golden
            parity_generator_sw(tmp, parity_sw);
            // Compare the value after the short delay
            #5 if (parity_hw != parity_sw) begin
                $display("Error at %b, parity_sw = %b, parity_hw = %b", tmp, parity_sw, parity_hw);
            end
            else begin
                // $display("%b, parity_sw = %b, parity_hw = %b", tmp, parity_sw, parity_hw);
            end
            // tmp + 1 after a short delay to avoid the mismatch of output
            #5 tmp = tmp + 1'b1;
        end    
    end

endmodule

### DUT (IP)

- Export RTL IP and Create Block Design in Vivado


#### Utilization


- Only used 3 LUT.

#### Waveform


- The values are all the same

### DUT (Verilog)

#### Utilization


#### Waveform


-------------------------------------------------------

# 2. Leading One

## Top Function (Kernel)

### In HLS


#### Utilization & Dataflow in HLS

- If we use the clock period 10ns, it may transfer to sequential.


- Here are the utilization and dataflow:




- **So, the solution uses a longer clock period to synthesize.**


- Then, the utilization and dataflow in HLS:


- We can see the design only using LUT(Combinational).


- Following is the golden `leading_one` in the testbench.


#### Result


### In Verilog

module leading_one(
    input  wire [8:0] a,
    output reg  [4:0] one
    );

    always @* begin
        if (a[8] == 1'b1) begin
            one = 5'd8;
        end
        else if (a[7] == 1'b1) begin
            one = 5'd7;
        end
        else if (a[6] == 1'b1) begin
            one = 5'd6;
        end
        else if (a[5] == 1'b1) begin
            one = 5'd5;
        end
        else if (a[4] == 1'b1) begin
            one = 5'd4;
        end
        else if (a[3] == 1'b1) begin
            one = 5'd3;
        end
        else if (a[2] == 1'b1) begin
            one = 5'd2;
        end
        else if (a[1] == 1'b1) begin
            one = 5'd1;
        end
        else if (a[0] == 1'b1) begin
            one = 5'd0;
        end
        else
            one = 5'd0 - 1'b1;
    end
    
endmodule

## Testbench

- Testbench in Verilog

module vic_group3_testbench();
    wire [4:0] one_hw;
    reg  [4:0] one_sw;
    reg  [8:0] a_hw;
    
    // design under test
    design_1_wrapper DUT0 (.a(a_hw), .leading_one(one_hw));
    // leading_one DUT1 (.a(a_hw), . one(one_hw));
    
    reg  [8:0] tmp;
    
    integer i;

- Write a golden leading_one_sw in the testbench

    task leading_one_sw;
        input  [8:0] a;
        output [4:0] leading_one_sw;
        begin
            if (a[8] == 1'b1) begin
                leading_one_sw = 5'd8;
            end
            else if (a[7] == 1'b1) begin
                leading_one_sw = 5'd7;
            end
            else if (a[6] == 1'b1) begin
                leading_one_sw = 5'd6;
            end
            else if (a[5] == 1'b1) begin
                leading_one_sw = 5'd5;
            end
            else if (a[4] == 1'b1) begin
                leading_one_sw = 5'd4;
            end
            else if (a[3] == 1'b1) begin
                leading_one_sw = 5'd3;
            end
            else if (a[2] == 1'b1) begin
                leading_one_sw = 5'd2;
            end
            else if (a[1] == 1'b1) begin
                leading_one_sw = 5'd1;
            end
            else if (a[0] == 1'b1) begin
                leading_one_sw = 5'd0;
            end
            else begin
                leading_one_sw = 5'd0 - 1'b1;
            end
        end
    endtask

- Then, the main function of the testbench.

    initial begin
        for (i = 0; i < 512; i = i + 1) begin
            a_hw = tmp;
            // get the value from golden
            leading_one_sw(tmp, one_sw);
            // compare the value after a short delay in order to get the right value
            #5 if (one_hw != one_sw) begin
                $display("Error at %b, leading_one_sw = %b, leading_one_hw = %b", tmp, one_sw, one_hw);
            end
            else begin
                $display("%b, leading_one_sw = %b, leading_one_hw = %b", tmp, one_sw, one_hw);
            end
            // update the tmp after a short delay to avoid the mismatch of output and input
            #5 tmp = tmp + 1'b1;
        end
    end
    
endmodule

### DUT (IP)

- Export RTL IP and Create Block Design in Vivado


#### Utilization


#### Waveform


### DUT (Verilog)

#### Utilization


#### Waveform


----------------------------------------------------

# 3. Integer Division&Modulus

## Top Function (Kernel)

### In HLS

const int n = 1234101;

int divbyconstant(int a) {
#pragma HLS INLINE off
    
    return a / n;
}

void integer_division_modulus(int a, int &r) {
#pragma HLS INTERFACE ap_ctrl_none port=return
#pragma HLS INTERFACE ap_none      port=a
#pragma HLS INTERFACE ap_none      port=r
    
    r = a - n * divbyconstant(a);
}

- It generates the remainder of a.

#### Utilization in HLS


- The operation is still combinational since it only uses DSP and LUT.
- There are 17 FFs for registers.

### In Verilog

`define N 32'd1234101

module integer_division_modulus(
    input wire [31:0] a_in,
    output reg [31:0] r_out
    );
    
    reg [31:0] tmp;
    
    function [31:0] divbyconst;
        input [31:0] a_in;
        begin
            divbyconst = a_in / `N;
        end
    endfunction
    
    always @* begin
        tmp = divbyconst (a_in);
        r_out = a_in - (`N * tmp);
    end
    
endmodule

## Testbench

- Testbench in Verilog

`define N 32'd1234101

module VIC_TESTBENCH();
    reg         clk;          // 10ns clk;
    reg         rst;          // active high, synchronous reset
    reg  [31:0] a_in;         // input of kernel and golden
    wire [31:0] r_hw;         // remainder of Kernel
    
    reg  [31:0] r_sw;         // remainder of Golden

    reg  [0:0]  error_detect; // detect the error, if 1, error
    
    integer i;
    
    // design under test
    design_1_wrapper DUT (.ap_clk(clk),
                          .ap_rst(rst),
                          .a(a_in),
                          .r(r_hw));
    // integer_division_modulus (.a_in(a_in),
    //                           .r_out(r_hw));
    
    // 10ns clk generator
    initial begin
        clk = 1'b1;
        forever begin
            #5 clk = ~clk;
        end
    end
    
    // divide by constant
    task divbyconst;
        input  [31:0] a_in;
        output [31:0] d_out;
        begin
            d_out = a_in / `N;
        end
    endtask
    
    // golden integer_division_modulus
    task golden_integer_division_modulus;
        input  [31:0] a_in;    // the 32-bit input
        output [31:0] r_out;   // the 32-bit output remainder
        reg    [31:0] tmp; // temperate value which will be the output of divbyconst
        begin
            tmp = 32'd0; // initialize the tmp
            divbyconst(a_in, tmp);
            r_out = a_in - (`N * tmp);
        end
    endtask
    
    // main function of the testbench
    initial begin
        // initialize the value of reset, input, error_detect
        rst          = 1'b1;
        a_in         = `N;
        error_detect = 1'b0;
        
        // rst fall after 50ns, since the first 40ns of kernel don't have any output
        #50 @(posedge clk) rst = 1'b0;
        
        // Compare the value of kernel and golden
        for (i = 0; i < 3000; i = i + 1) begin
            @(posedge clk) golden_integer_division_modulus(a_in, r_sw);
            #5
            if (r_sw != r_hw) begin
                $display("Error at %d, r_sw = %d, r_hw = %d", a_in, r_sw, r_hw);
                error_detect = 1'b1;
            end
            else begin
                /* $display("%d, r_sw = %d, r_hw = %d", a_in, r_sw, r_hw); */
            end
            #4
            a_in = a_in + 1'b1;
        end
        
        // Check the error
        if (error_detect != 1'b0) begin
            $display("Failed!!!");
        end
        else begin
            $display("Test Passed!!");
        end  
    end
        
endmodule        

### DUT (IP)

- Export RTL IP and create block design in Vivado


#### Utilization


- 5 FFs for registers

#### Waveform


### DUT (Verilog)

#### Utilization


- It uses only LUT in Verilog design, not like HLS design contains FFs for registers.

#### Waveform


- We can know that the `r_hw` of Verilog design output is at the same time as the input `a_in`, since it is purely combinational logic with LUT.

--------------------------------------------------------
# 4. Binary-to-BCD (Division)

## Top Function (Kernel)

### In HLS

void get_digit(uint14 &a, uint4 &digit) {
    digit = a - 10 * (a / 10);
    a = a / 10;
}

void binary2bcd_div(uint14 in_binary, uint16 *packed_bcd) {
#pragma HLS INTERFACE ap_none      port=in_binary
#pragma HLS INTERFACE ap_none      port=packed_bcd
#pragma HLS INTERFACE ap_ctrl_none port=return
    
    uint14 a = in_binary;

    uint4 digit_0;
    uint4 digit_1;
    uint4 digit_2;
    uint4 digit_3;
    
    get_digit(a, digit_0);
    get_digit(a, digit_1);
    get_digit(a, digit_2);
    get_digit(a, digit_3);
    
    *packed_bcd = (digit_3, digit_2, digit_1, digit_0);
}

### In Verilog

module binary2bcd_div(
    input wire [13:0] in_binary,
    output reg [15:0] packed_bcd
    );
    
    reg [13:0] a;
    reg  [3:0] digit_0;
    reg  [3:0] digit_1;
    reg  [3:0] digit_2;
    reg  [3:0] digit_3;
    
    function [3:0] get_digit;
        input [13:0] a;
        begin
            get_digit = a - 10 * (a / 10);
        end
    endfunction
    
    always @* begin
        a = in_binary;
        
        digit_0 = get_digit(a);
        a = a / 10;
        digit_1 = get_digit(a);
        a = a / 10;
        digit_2 = get_digit(a);
        a = a / 10;
        digit_3 = get_digit(a);
        a = a / 10;
        
        packed_bcd = {digit_3, digit_2, digit_1, digit_0};
    end
    
endmodule

## Testbench

- Testbench in Verilog

module VIC_TESTBENCH();
    reg         clk;
    reg         rst;
    reg  [13:0] in_binary;
    wire [15:0] packed_bcd_hw;
    reg  [15:0] packed_bcd_sw;
    
    reg  [0:0]  error_detect;
    
    integer i;
    
    // design under test
    design_1_wrapper DUT0 (.ap_clk(clk),
                           .ap_rst(rst),
                           .in_binary(in_binary),
                           .packed_bcd(packed_bcd_hw));
    /*
    binary2bcd_div DUT1 (.in_binary(in_binary),
                         .packed_bcd(packed_bcd_hw));
    */
    
    // 10ns clk generator
    initial begin
        clk = 1'b1;
        forever begin
            #5 clk = ~clk;
        end
    end
    
    // Golden get_digit
    task get_digit;
        input [13:0] a;
        output [3:0] digit;
        begin
            digit = a - 10 * (a / 10);
        end
    endtask
    
    // golden
    task binary2bcd_sw;
        input  [13:0] in_binary;
        output [15:0] packed_bcd;
        reg    [13:0] a;
        reg     [3:0] digit_0;
        reg     [3:0] digit_1;
        reg     [3:0] digit_2;
        reg     [3:0] digit_3;
        begin
            a = in_binary;
            
            get_digit (a, digit_0);
            a = a / 10;
            get_digit (a, digit_1);
            a = a / 10;
            get_digit (a, digit_2);
            a = a / 10;
            get_digit (a, digit_3);
            a = a / 10;
            
            packed_bcd = {digit_3, digit_2, digit_1, digit_0};
        end
    endtask
    
    //main function of the testbench
    initial begin
        rst          = 1'b1;
        in_binary    = 14'd0;
        error_detect = 1'b0;
        
        #30 @(posedge clk) rst = 1'b0;
        
        for (i = 0; i < 9999; i = i + 1) begin
            @(posedge clk)
                binary2bcd_sw (in_binary, packed_bcd_sw);
            #80
            if (packed_bcd_sw != packed_bcd_hw) begin
                $display("Error at %d, bcd_sw = %b, bcd_hw = %b", in_binary, packed_bcd_sw, packed_bcd_hw);
                error_detect = 1'b1;
            end
            #0 in_binary = in_binary + 1'b1;
        end
        
        #5
        if (error_detect == 1'b0) begin
            $display("Test Passed");
        end
        else begin
            $display("Error!!!");
        end
    end
    
endmodule

### DUT (IP)

- Export RTL IP and create a block design


#### Utilization


- There are 9 FFs for registers.

#### Waveform

1. This is the waveform of IP if I change the `in_binary` every 50ns, which clock period is 10ns.


- We can see that the longest period appears every 4 `packed_bcd_hw`, which is longer than others(40ns longer).

2. Then, I change the `in_binary` every 90ns.


- We can see that `packed_bcd_hw` and `packed_bcd_sw` match each other in every 4 times we change the `in_binary`.

3. Both 1, and 2 are correct.

- 

### DUT (Verilog)

#### Utilization


- Verilog design only uses LUTs, unlike the IP using FFs for registers.

#### Waveform



---------------------------------------------------
# Binary-to-BCD (Double Dabble)

## Top Function (Kernel)

### In HLS

void double_dabble(uint16 &scratch_pad) {
#pragma HLS INLINE
    scratch_pad = scratch_pad << 1;
    if (scratch_pad(11, 8) > 4)
        scratch_pad(11, 8) = scratch_pad(11, 8) + 3;
}

void binary2bcd_double_dabble(uint8 in_binary, uint16 *unpacked_bcd, uint8 *packed_bcd) {
#pragma HLS INTERFACE ap_none      port=in_binary
#pragma HLS INTERFACE ap_none      port=unpacked_bcd
#pragma HLS INTERFACE ap_none      port=packed_bcd
#pragma HLS INTERFACE ap_ctrl_none port=return
    uint16 scratch_pad = in_binary;
    uint4  zero_4      = 0b0000;
    
    double_dabble(scratch_pad);
    double_dabble(scratch_pad);
    double_dabble(scratch_pad);
    double_dabble(scratch_pad);

    double_dabble(scratch_pad);
    double_dabble(scratch_pad);
    double_dabble(scratch_pad);
    scratch_pad = scratch_pad << 1;


    *packed_bcd   = scratch_pad(15, 8);
    *unpacked_bcd = (zero_4, scratch_pad(15, 12), zero_4, scratch_pad(11, 8));
    
}

### In Verilog

module binary2bcd_double_dabble(
    input wire  [7:0] in_binary,    // binary input
    output reg [15:0] unpacked_bcd, // 16-bit bcd output
    output reg  [7:0] packed_bcd    // 8-bit bcd output
    );
    
    reg [15:0] num_list;
    reg [16:0] tmp_list;
    reg  [7:0] bcd;
    
    integer i;
    
    always @* begin
        num_list = {8'd0, in_binary};
        // shift and compare 7 times
        for (i = 0; i < 7; i = i + 1) begin
            tmp_list = num_list * 2;
            bcd      = tmp_list[15:8];
            // whether > 4?
            if (bcd[3:0] > 3'd4) begin
                bcd      = bcd + 2'd3;
                num_list = {bcd, tmp_list[7:0]};
            end
            else begin
                num_list = {bcd, tmp_list[7:0]};
            end
        end
        // In the last time, only need to shift.
        tmp_list = num_list * 2;
        bcd      = tmp_list[15:8];
        num_list = {bcd, tmp_list[7:0]};
        // bcd output
        packed_bcd   = num_list[15:8];
        unpacked_bcd = {4'd0, num_list[15:12], 4'd0, num_list[11:8]};
    end
    
endmodule

## Testbench

- Testbench in Verilog

module VIC_TESTBENCH();
    reg   [7:0] in_binary; // binary input
    wire [15:0] up_bcd_hw; // unpacked bcd number
    wire  [7:0] p_bcd_hw;  // packed bcd number

    reg  [15:0] up_bcd_sw; // software
    reg   [7:0] p_bcd_sw;  // software
    
    reg   [0:0] error      // error detect
    
    // design under test
    design_1_wrapper DUT0 (.in_binary(in_binary),
                           .unpacked_bcd(up_bcd_hw),
                           .packed_bcd(p_bcd_hw));
    
    /*
    binary2bcd_double_dabble DUT1 (.in_binary(in_binary),
                                   .unpacked_bcd(up_bcd_hw),
                                   .packed_bcd(p_bcd_hw));
    */
    
    // double dabble in software
    task double_dabble;
        inout [15:0] scratch_pad;
        begin
            scratch_pad = scratch_pad * 2;
            if (scratch_pad[11:8] > 3'd4) begin
                scratch_pad[11:8] = scratch_pad[11:8] + 2'd3;
            end
        end
    endtask
    
    // golden
    task binary2bcd_double_dabble;
        input   [7:0] in_binary;
        output [15:0] unpacked_bcd;
        output  [7:0] packed_bcd;
        reg    [15:0] scratch_pad;
        integer i;
        begin
            scratch_pad = {8'd0, in_binary};
            for (i = 0; i < 7; i = i + 1) begin
                double_dabble (scratch_pad);
            end
            scratch_pad = scratch_pad * 2;
            
            packed_bcd = scratch_pad[15:8];
            unpacked_bcd = {4'd0, scratch_pad[15:12], 4'd0, scratch_pad[11:8]};
        end
        
        // main function of the testbench
        initial begin
            // initialize the value
            in_binary = 8'd0;
            error     = 1'b0;
            
            for (i = 0; i < 99; i = i + 1) begin
                binary2bcd_double_dabble (in_binary, up_bcd_sw, p_bcd_sw);
                #5
                if (up_bcd_sw != up_bcd_hw) begin
                    $display("Error at %d, bcd_sw = %b, bcd_hw = %b", in_binary, p_bcd_sw, p_bcd_hw);
                end
                #5
                in_binary = in_binary + 1'b1;
            end
            
            if (error == 1'b1) begin
                $display("Error!!");
            end
            else begin
                $display("Test Passed!!!");
            end
        end
endmodule

### DUT (IP)

- Export RTL IP and create a block design


#### Utilization


- Design only contains 5 LUTs.

#### Waveform



### DUT (Verilog)

#### Utilization


- Design contains 9 LUTs, more than IP.

#### Waveform