compiler generate invalid arm code for function with large stack and callee_saved resgiter

[bgregoir]

param int N = 1024;
param int T = 10;

export fn main() -> reg u32 {
stack u32[N] s;
reg ptr u32[N] ps;
reg u32 i n d;
inline int j;

i = 0;
n = N;
ps = s;
while(i < n) {
ps[i] = i;
i += 1;
}

reg u32[T] t;
for j = 0 to T {
t[j] = ps[j];
}
i = T;
n = (N / T) * T;
while (i < n) {
for j = 0 to T {
d = ps[i];
t[j] += d;
i += 1;
}
}
n = N;
while (i < n) {
d = ps[i];
t[0] += d;
i += 1;
}

for j = 1 to T {
t[0] += t[j];
}

i = t[0];
return i;

}