The buffering implementation here is inspired by the CAB described in [Ha et al. 09] (paper is in this repo).
TODO:
- need the blocking version in the instrumentation
For enqueing by the program under test:
void enqueue(u64 data) {
if (buf[i] != CLEAR)
block();
buf[i++] = data;
i = i & 0xffff; // wrap around if reached the end
}
void block() {
int poll_i = i + CHUNK_SIZE * 2;
while (buf[poll_i] != CLEAR) {
spin(n);
}
}For dequeing by the monitor:
void main() {
while (true) {
int i = ((chunk_num + 2) * CHUNK_SIZE) % buffer_size;
while (buf[i] == CLEAR)
spin_or_sleep();
consume_chunk(chunk_num++);
}
}
void consume_chunk(int chunk_num) {
5256
for (int i = 0; i < CHUNK_SIZE; ++i) {
int idx = chunk_num * CHUNK_SIZE + i;
u64 data = buf[idx];
f(data);
}
}