├─prj
│ └─pipline
│-riscvpipeline.sv
|
|-build.bat
|
|-testbench.sv
|
|-top_hardware.sv
|
|-M
|
└─README.assets
prj存放vivado工程文件 -v2021.2
开发板 Genesys2
仿真编译工具 iverilog
xdc文件见文件目录
riscvpipeline.sv --> 面向仿真测试
build.bat --> win下iverilog编译仿真
testbench.sv --> 仿真文件
top_hardware.sv --> 上板sv
M文件夹存放对RV32M指令集支持的五级流水CPU，并对MUL和DIV指令进一步的测试
PCSrcE --> pc跳转选择信号 --> +4 or jump
（参考）
1.俩类数据冒险：
sub x2, x1, x3
and x12, x2, x5
or x13, x6, x2
-> sub & and == 1a
-> sub & or == 2b
sub x2,x1,x3 机器码：
0100000_00011_00001_000_00010_0110011
and x12, x2, x5 机器码：
0000000_00101_00010_111_01100_0110011
or x13, x6, x2
0000000_00010_00110_110_01101_0110011
addi x1, x0, 5
addi x3, x0, 3
addi x5, x0, 3
addi x6, x0, 5
0000_0000_0101_00000_000_00001_0010011
0000_0000_0011_00000_000_00011_0010011
0000_0000_0011_00000_000_00101_0010011
0000_0000_0101_00000_000_00110_0010011
x2 = 2 -> 0010 x6 = 5 -> 0101
0010 0010
& 0011 | 0101
----------- ----------
0010 0111
验证检查：
x12 = 2
x13 = 7
00000000010100000000000010010011 ->00500093
00000000001100000000000110010011 ->00300193
00000000001100000000001010010011 ->00300293
00000000010100000000001100010011 ->00500313
01000000001100001000000100110011 ->40308133
00000000010100010111011000110011 ->00517633
00000000001000110110011010110011 ->002366B3
整体汇编如下：
addi x1, x0, 5
addi x3, x0, 3
addi x5, x0, 3
addi x6, x0, 5
sub x2, x1, x3
and x12, x2, x5
or x13, x6, x2
测试： PASS
2.基于此解决流水线过程中，regfile上升沿写的冒险
上述汇编 addi x3,x0,3 和 sub x2,x1,x3 在同一时钟周期对x3读和写
可以看到RD2E并没有将数据传递下去。目前思路是直接将写回值传给读出值。但一定要注意逻辑的先后顺序
测试： PASS
结构冒险解决图
addi x3, x
lw x2, 96(x0)
add x9, x2, x5
# RISC-V Assembly Description Address Machine Code
init: addi t5, x0, 2047 # count_max, t5 = 2047 0 0x7FF00F13 # 0111_1111_1111_00000_000_11110_0010011
add t5, t5, t5 # count_max, t5 = 2047*2 4 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*4 8 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*8 12 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*16 16 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*32 20 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*64 24 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*128 28 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*256 32 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*512 36 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*1024 40 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*2048 44 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*4096 48 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
add t5, t5, t5 # count_max, t5 = 2047*8192 52 0x01EF0F33 # 0000_000_11110_11110_000_11110_0110011
/for sim addi t5, x0, 4 56 0x00200F13 # 0000_0000_0010_00000_000_11110_0010011
addi x2, x0, 1 60 0x00100113 # 0000_0000_0001_00000_000_00010_0010011
addi t6, x0, 256 # led_max, t6 = 256 64 0x20000f93 # 0001_0000_0000_00000_000_11111_0010011
clean_led: addi t4, x0, 1 # init led value, t4 = 1 68 0x00100e93 # 0000_0000_0001_00000_000_11101_0010011
clean_cnt: addi t3, x0, 0 # clear count(t3). 72 0x00000e13 # 0000_0000_0000_00000_000_11100_0010011
count: addi t3, t3, 1 # count++ 76 0x001E0E13 # 0000_0000_0001_11100_000_11100_0010011
beq t3, t5, mov_led pc +8 80 0x01EE0463 # 0000_000_11110_11100_000_01000_1100011
jal count pc -8 84 0xFF9FF06F # 1_1111111100_1_11111111_00000_1101111
mov_led add t4, t4, t4 88 0x01DE8EB3 # 0000_000_11101_11101_000_11101_0110011
beq t4, t6, right pc +8 92 0x01FE8463 # 0000_000_11111_11101_000_01000_1100011
jal clean_cnt pc-24 96 0xFE9FF06F # 1_1111110100_1_11111111_00000_1101111
right srl t4, t4, x2 100 0x002EDEB3 # 0000000_00010_11101_101_11101_0110011
beq t4, x2, clean_led pc- 4*9 104 0xFC2E8EE3 # 1111_110_00010_11101_000_11101_1100011
clean_cnt2: addi t3, x0, 0 108 0x00000e13 # 0000_0000_0000_00000_000_11100_0010011
count2: addi t3, t3, 1 112 0x001E0E13 # 0000_0000_0001_11100_000_11100_0010011
beq t3, t5, right pc- 4*4 116 0xFFEE08E3 # 1111_111_11110_11100_000_10001_1100011
jal count2 pc-8 120 0xFF9FF06F # 1_1111111100_1_11111111_00000_1101111
Note:可以上板验证，为了简化仿真，添加仿真指令addi t5, x0, 4,防止仿真时间过长。
实际上板就去掉 for sim 汇编
ToDo
- 完善对M文件夹的排版和注释