Consider this testcase:
target i686
function u0:0() -> i32 system_v {
ss0 = explicit_slot 0
block0:
v2 = stack_addr.i32 ss0
return v2
}
Before this commit, in 32-bit mode the x86 backend would generate
incorrect code for stack addresses:
0: 55 push ebp
1: 89 e5 mov ebp, esp
3: 83 ec 08 sub esp, 8
6: 8d 44 24 00 lea eax, [esp]
a: 00 00 add byte ptr [eax], al
c: 00 83 c4 08 5d c3 add byte ptr [ebx - 0x3ca2f73c], al
This happened because the ModRM byte indicated a disp8 encoding, but
the instruction actually used a disp32 encoding. After this commit,
correct code is generated:
0: 55 push ebp
1: 89 e5 mov ebp, esp
3: 83 ec 08 sub esp, 8
6: 8d 84 24 00 00 00 00 lea eax, [esp]
d: 83 c4 08 add esp, 8
10: 5d pop ebp
11: c3 ret
This crate contains the core Cranelift code generator. It translates code from an intermediate representation into executable machine code.