https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117831
Bug ID: 117831
Summary: [RISCV] A dead loop occurs when calculating the
multiplication of two uint64 integers under rv32 -Os
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: mumuxi_ll at outlook dot com
Target Milestone: ---
I wrote a library function to calculate the multiplication of two uint64
integers under rv32. But after compiling with '-march=rv32ic -mabi=ilp32 -Os',
calling this function will lead to an endless loop.
function :
============
typedef unsigned long long uint64_t ;
uint64_t mul_u64(uint64_t a, uint64_t b)
{
uint64_t Product;
Product = 0;
while (a) {
if (a & 1) {
Product += b;
}
a >>= 1;
b <<= 1;
}
return Product;
}
============
assembly:
============
uint64_t mul_u64(uint64_t a, uint64_t b)
{
0: 1101 addi sp,sp,-32
2: cc22 sw s0,24(sp)
4: ca26 sw s1,20(sp)
6: c84a sw s2,16(sp)
8: c64e sw s3,12(sp)
a: c452 sw s4,8(sp)
c: c256 sw s5,4(sp)
e: ce06 sw ra,28(sp)
10: 842a mv s0,a0
12: 892e mv s2,a1
14: 89b2 mv s3,a2
16: 84b6 mv s1,a3
uint64_t Product;
Product = 0;
18: 4a81 li s5,0
1a: 4a01 li s4,0
0000001c <.L2>:
while (a) {
1c: 012467b3 or a5,s0,s2
20: ef81 bnez a5,38 <.L3>
a >>= 1;
b <<= 1;
}
return Product;
}
22: 40f2 lw ra,28(sp)
24: 4462 lw s0,24(sp)
00000026 <.LVL2>:
26: 44d2 lw s1,20(sp)
00000028 <.LVL3>:
28: 4942 lw s2,16(sp)
2a: 49b2 lw s3,12(sp)
2c: 8556 mv a0,s5
2e: 85d2 mv a1,s4
30: 4a92 lw s5,4(sp)
00000032 <.LVL4>:
32: 4a22 lw s4,8(sp)
00000034 <.LVL5>:
34: 6105 addi sp,sp,32
36: 8082 ret
00000038 <.L3>:
if (a & 1) {
38: 864e mv a2,s3
3a: 86a6 mv a3,s1
3c: 00147513 andi a0,s0,1
40: 4581 li a1,0
42: 00000097 auipc ra,0x0
46: 000080e7 jalr ra # 42 <.L3+0xa>
0000004a <.LVL7>:
4a: 9556 add a0,a0,s5
4c: 015537b3 sltu a5,a0,s5
50: 9a2e add s4,s4,a1
00000052 <.LVL8>:
52: 9a3e add s4,s4,a5
00000054 <.LVL9>:
a >>= 1;
54: 8005 srli s0,s0,0x1
00000056 <.LVL10>:
56: 01f91793 slli a5,s2,0x1f
5a: 943e add s0,s0,a5
b <<= 1;
5c: 0486 slli s1,s1,0x1
0000005e <.LVL11>:
5e: 01f9d793 srli a5,s3,0x1f
if (a & 1) {
62: 8aaa mv s5,a0
a >>= 1;
64: 00195913 srli s2,s2,0x1
00000068 <.LVL12>:
b <<= 1;
68: 94be add s1,s1,a5
6a: 0986 slli s3,s3,0x1
6c: bf45 j 1c <.L2>
============
But at other optimization levels, it is correct.
