1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
|
RISCV32 illegal instruction exception
I'm running a machine learning model on qemu riscv32 and I ran into illegal instruction exception for some reason. I wasn't sure if this is a bug and if so whether it is related to zephyr or qemu, however I'll try to provide as much as information to get a better understanding.
Here is the assembly code that I'm running:
0x8000bd74 <+0>: lw a4,0(a0)
0x8000bd76 <+2>: lw a5,8(a0)
0x8000bd78 <+4>: lw a0,0(a4)
0x8000bd7a <+6>: lw a1,0(a5)
0x8000bd7c <+8>: li a3,0
0x8000bd7e <+10>: j 0x8000bda4 <fused_nn_pad_layout_transform+48>
0x8000bd80 <+12>: addi a5,a3,-2
0x8000bd84 <+16>: li a2,27
0x8000bd86 <+18>: bgeu a2,a5,0x8000bdae <fused_nn_pad_layout_transform+58>
=> 0x8000bd8a <+22>: fmv.w.x fa5,zero
0x8000bd8e <+26>: slli a5,a3,0x5
0x8000bd92 <+30>: add a5,a5,a4
0x8000bd94 <+32>: slli a5,a5,0x2
0x8000bd96 <+34>: add a5,a5,a1
0x8000bd98 <+36>: fsw fa5,0(a5)
0x8000bd9a <+38>: addi a4,a4,1
0x8000bd9c <+40>: li a5,31
0x8000bd9e <+42>: bge a5,a4,0x8000bd80 <fused_nn_pad_layout_transform+12>
0x8000bda2 <+46>: addi a3,a3,1
0x8000bda4 <+48>: li a5,31
0x8000bda6 <+50>: blt a5,a3,0x8000bde0 <fused_nn_pad_layout_transform+108>
0x8000bdaa <+54>: li a4,0
0x8000bdac <+56>: j 0x8000bd9c <fused_nn_pad_layout_transform+40>
0x8000bdae <+58>: li a5,1
0x8000bdb0 <+60>: bge a5,a4,0x8000bdd4 <fused_nn_pad_layout_transform+96>
0x8000bdb4 <+64>: li a5,29
0x8000bdb6 <+66>: blt a5,a4,0x8000bdda <fused_nn_pad_layout_transform+102>
0x8000bdba <+70>: li a5,28
0x8000bdbc <+72>: mul a5,a3,a5
0x8000bdc0 <+76>: add a5,a5,a4
0x8000bdc2 <+78>: lui a2,0x40000
0x8000bdc6 <+82>: addi a2,a2,-58 # 0x3fffffc6
0x8000bdca <+86>: add a5,a5,a2
0x8000bdcc <+88>: slli a5,a5,0x2
0x8000bdce <+90>: add a5,a5,a0
0x8000bdd0 <+92>: flw fa5,0(a5)
0x8000bdd2 <+94>: j 0x8000bd8e <fused_nn_pad_layout_transform+26>
0x8000bdd4 <+96>: fmv.w.x fa5,zero
0x8000bdd8 <+100>: j 0x8000bd8e <fused_nn_pad_layout_transform+26>
0x8000bdda <+102>: fmv.w.x fa5,zero
0x8000bdde <+106>: j 0x8000bd8e <fused_nn_pad_layout_transform+26>
0x8000bde0 <+108>: li a0,0
0x8000bde2 <+110>: ret
My code breaks on line 0x8000bd8a and then the mcause register is loaded with value 0x02 which translates to illegal instruction. Please let me know if you need more information about this.
I also posted this on ZephyrProject in case it is related to the Zephyr: https://github.com/zephyrproject-rtos/zephyr/issues/34026
I have tested on both QEMU 5.1.0 and 5.2.0 versions. I ran the same code on qemu riscv64 and didn't have the same problem. Here is the assembly code that is generated for the same operation:
=> 0x000000008000b446 <+0>: ld a4,0(a0)
0x000000008000b448 <+2>: ld a5,8(a0)
0x000000008000b44a <+4>: ld a0,0(a4)
0x000000008000b44c <+6>: ld a1,0(a5)
0x000000008000b44e <+8>: li a3,0
0x000000008000b450 <+10>: j 0x8000b476 <fused_nn_pad_layout_transform+48>
0x000000008000b452 <+12>: addiw a5,a3,-2
0x000000008000b456 <+16>: li a2,27
0x000000008000b458 <+18>: bgeu a2,a5,0x8000b480 <fused_nn_pad_layout_transform+58>
0x000000008000b45c <+22>: li a2,0
0x000000008000b460 <+26>: slliw a5,a3,0x5
0x000000008000b464 <+30>: addw a5,a5,a4
0x000000008000b466 <+32>: slli a5,a5,0x2
0x000000008000b468 <+34>: add a5,a5,a1
0x000000008000b46a <+36>: sw a2,0(a5)
0x000000008000b46c <+38>: addiw a4,a4,1
0x000000008000b46e <+40>: li a5,31
0x000000008000b470 <+42>: bge a5,a4,0x8000b452 <fused_nn_pad_layout_transform+12>
0x000000008000b474 <+46>: addiw a3,a3,1
0x000000008000b476 <+48>: li a5,31
0x000000008000b478 <+50>: blt a5,a3,0x8000b4ac <fused_nn_pad_layout_transform+102>
0x000000008000b47c <+54>: li a4,0
0x000000008000b47e <+56>: j 0x8000b46e <fused_nn_pad_layout_transform+40>
0x000000008000b480 <+58>: li a5,1
0x000000008000b482 <+60>: bge a5,a4,0x8000b4a0 <fused_nn_pad_layout_transform+90>
0x000000008000b486 <+64>: li a5,29
0x000000008000b488 <+66>: blt a5,a4,0x8000b4a6 <fused_nn_pad_layout_transform+96>
0x000000008000b48c <+70>: li a5,28
0x000000008000b48e <+72>: mulw a5,a5,a3
0x000000008000b492 <+76>: addw a5,a5,a4
0x000000008000b494 <+78>: addiw a5,a5,-58
0x000000008000b498 <+82>: slli a5,a5,0x2
0x000000008000b49a <+84>: add a5,a5,a0
0x000000008000b49c <+86>: lw a2,0(a5)
0x000000008000b49e <+88>: j 0x8000b460 <fused_nn_pad_layout_transform+26>
0x000000008000b4a0 <+90>: li a2,0
0x000000008000b4a4 <+94>: j 0x8000b460 <fused_nn_pad_layout_transform+26>
0x000000008000b4a6 <+96>: li a2,0
0x000000008000b4aa <+100>: j 0x8000b460 <fused_nn_pad_layout_transform+26>
0x000000008000b4ac <+102>: li a0,0
0x000000008000b4ae <+104>: ret
|