1.vivado hls file (AMD)
https://docs.amd.com/r/zh-CN/ug1399-vitis-hls/%E5%88%9B%E5%BB%BA-HLS-%E7%BB%84%E4%BB%B6
2.get liscence to open vitis hls (2024)
https://blog.csdn.net/weixin_74013425/article/details/142458347
3.Overlay Tutorial (AMD)
https://pynq.readthedocs.io/en/v2.5.1/overlay_design_methodology/overlay_tutorial.html
4.Overlay Tutorial in Chinese (1)
https://blog.csdn.net/CSD_N_csdn/article/details/105669069
5.Overlay Tutorial in Chinese (2)
https://blog.csdn.net/CSD_N_csdn/article/details/105691939
6.Overlay Tutorial in Chinese (Filter)
https://blog.csdn.net/CSD_N_csdn/article/details/105894172?sharetype=blogdetail&shareId=105894172&sharerefer=APP&sharesource=Azhuwizard&sharefrom=link

internet scan:

sudo apt install nmap
nmap -sn 192.168.3.0/24

adder.cpp #HLS

void adder(int a, int b, int& c) {
#pragma HLS INTERFACE ap_ctrl_none port=return
#pragma HLS INTERFACE s_axilite port=a
#pragma HLS INTERFACE s_axilite port=b
#pragma HLS INTERFACE s_axilite port=c

	c = a + b;
}

adder.tcl #PYNQ-Z2

# Open a new project
open_project my_adder_project

# Add the design source file
add_files adder.cpp

# Create and open a solution with the specified target device
open_solution -reset solution1

# Replace with your FPGA part number
set_part xc7z020clg400-1  

# Set the top-level function
set_top adder

# Run synthesis
csynth_design

# Export the design as an IP
export_design -format ip_catalog -rtl verilog

# Print completion message
puts "IP core has been successfully generated and stored in the 'impl/ip' directory."

vitis_hls -f adder.tcl

/home/esslab/Pynq_adder/my_adder_project/solution1/impl/misc/drivers
/home/esslab/Pynq_adder/my_adder_project/solution1/impl/verilog

.bit .hw .tcl

import time
from pynq import Overlay, DefaultIP

# 自定義驅動程式
class AddDriver(DefaultIP):
    def __init__(self, description):
        super().__init__(description=description)

    bindto = ['xilinx.com:hls:add:1.0']  # 與 HLS 定義的名稱匹配

    def add(self, a, b):
        self.write(0x10, a)
        self.write(0x18, b)
        return self.read(0x20)

# 加載比特流
overlay = Overlay('adder.bit')

# 從 ip_dict 中獲取描述資訊
adder_description = overlay.ip_dict['adder_0']  # 'adder_0' 為 IP 名稱

# 手動將 adder_0 綁定到 AddDriver
overlay.adder_0 = AddDriver(adder_description)

# 測試數據
a, b = 123, 456

# 計算使用 IP 的運算時間
start_time_ip = time.time()
ip_result = overlay.adder_0.add(a, b)
end_time_ip = time.time()
time_ip = end_time_ip - start_time_ip
print(f"IP Result: {ip_result}, Time taken with IP: {time_ip:.6f} seconds")

# 計算使用純 Python 的運算時間
start_time_python = time.time()
python_result = a + b
end_time_python = time.time()
time_python = end_time_python - start_time_python
print(f"Python Result: {python_result}, Time taken with Python: {time_python:.6f} seconds")

# 驗證結果一致性
if ip_result == python_result:
    print("Results are consistent.")
else:
    print("Results are inconsistent.")

file

sudo chmod 777 /media/esslab/root/home/xilinx/jupyter_notebooks/matrix

3*3矩陣乘法版本

#include <ap_int.h>
#include <hls_stream.h>

#define MATRIX_SIZE 3

void matrix(
    int A[MATRIX_SIZE][MATRIX_SIZE],
    int B[MATRIX_SIZE][MATRIX_SIZE],
    int C[MATRIX_SIZE][MATRIX_SIZE]
) {
    // 使用標準控制介面
#pragma HLS INTERFACE s_axilite     port=return   bundle=CTRL
#pragma HLS INTERFACE s_axilite     port=A        bundle=CTRL
#pragma HLS INTERFACE s_axilite     port=B        bundle=CTRL
#pragma HLS INTERFACE s_axilite     port=C        bundle=CTRL

    // 初始化 C
    for (int i = 0; i < MATRIX_SIZE; i++) {
        for (int j = 0; j < MATRIX_SIZE; j++) {
#pragma HLS PIPELINE II=1
            C[i][j] = 0;
        }
    }

    // 執行乘法
    for (int i = 0; i < MATRIX_SIZE; i++) {
        for (int j = 0; j < MATRIX_SIZE; j++) {
            for (int k = 0; k < MATRIX_SIZE; k++) {
#pragma HLS PIPELINE II=1
                C[i][j] += A[i][k] * B[k][j];
            }
        }
    }
}

import time
from pynq import Overlay, DefaultIP
import numpy as np

class MatrixDriver(DefaultIP):
    # Match your IP’s "bindto" definition here
    bindto = ['xilinx.com:hls:matrix:1.0']  
    
    def __init__(self, description):
        super().__init__(description=description)

    def multiply(self, matrix_a, matrix_b):
        """
        Send two matrices to the hardware for multiplication and retrieve the result.
        """
        # 1) Write size if required (assume offset 0x10)
        size = matrix_a.shape[0]
        self.write(0x10, size)

        # 2) Write matrix A
        for i in range(size):
            for j in range(size):
                addrA = 0x40 + 4 * (i * size + j)
                self.write(addrA, int(matrix_a[i][j]))

        # 3) Write matrix B
        for i in range(size):
            for j in range(size):
                addrB = 0x80 + 4 * (i * size + j)
                self.write(addrB, int(matrix_b[i][j]))

        # 4) Start the IP
        control_val = self.read(0x00)
        self.write(0x00, (control_val & 0x80) | 0x01)

        # 5) Poll for ap_done
        while True:
            status = self.read(0x00)
            if (status & 0x2) != 0:  # bit 1 for ap_done
                break

        # 6) Read matrix C
        result = np.zeros((size, size), dtype=int)
        for i in range(size):
            for j in range(size):
                addrC = 0xC0 + 4 * (i * size + j)
                result[i][j] = self.read(addrC)

        return result


# --------------------------------------------------------------------
# Now load the bitstream, create the driver instance, and test
# --------------------------------------------------------------------

# 1) Load the bitstream
overlay = Overlay('matrix.bit')

# 2) Bind the IP description to the custom driver
overlay.matrix_0 = MatrixDriver(overlay.ip_dict['matrix_0'])

# 3) Test with 3x3 matrices
matrix_a = np.array([[1, 2, 3], 
                     [4, 5, 6],
                     [7, 8, 9]], dtype=int)

matrix_b = np.array([[9, 8, 7], 
                     [6, 5, 4],
                     [3, 2, 1]], dtype=int)

# 4) Check expected result via numpy
expected_result = np.dot(matrix_a, matrix_b)
print("Expected result (calculated with numpy):")
print(expected_result)

# 5) Multiply with hardware
print("\nPerforming multiplication with hardware...")
start_time_ip = time.time()
ip_result = overlay.matrix_0.multiply(matrix_a, matrix_b)
end_time_ip = time.time()

# 6) Display the result and check correctness
print("\nHardware Result:")
print(ip_result)
print(f"\nTime taken with IP: {end_time_ip - start_time_ip:.6f} seconds")

if np.array_equal(ip_result, expected_result):
    print("Verification successful: Results match.")
else:
    print("Verification failed: Results do NOT match.")

12*12矩陣

#include <ap_int.h>
#include <hls_stream.h>

#define MATRIX_SIZE 12

void matrix(
    int A[MATRIX_SIZE][MATRIX_SIZE],  // 輸入矩陣 A
    int B[MATRIX_SIZE][MATRIX_SIZE],  // 輸入矩陣 B
    int C[MATRIX_SIZE][MATRIX_SIZE]   // 輸出矩陣 C
) {
// AXI4-Lite interface for control signals
#pragma HLS INTERFACE s_axilite port=return bundle=CTRL
#pragma HLS INTERFACE s_axilite port=A bundle=CTRL
#pragma HLS INTERFACE s_axilite port=B bundle=CTRL
#pragma HLS INTERFACE s_axilite port=C bundle=CTRL

    // 初始化 C 矩陣
    for (int i = 0; i < MATRIX_SIZE; i++) {
        for (int j = 0; j < MATRIX_SIZE; j++) {
#pragma HLS UNROLL  // 完全展開以加速初始化
            C[i][j] = 0;
        }
    }

    // 執行矩陣乘法
    for (int i = 0; i < MATRIX_SIZE; i++) {
        for (int j = 0; j < MATRIX_SIZE; j++) {
#pragma HLS UNROLL // 完全展開以加速初始化
            for (int k = 0; k < MATRIX_SIZE; k++) {
#pragma HLS PIPELINE II=1  // 每次迴圈迭代佔用 1 個時鐘週期
                C[i][j] += A[i][k] * B[k][j];
            }
        }
    }
}