Code issues in read_buffer.py

[Mo-guan]

In line 334-337 of read_buffer.py,

        start_idx = self.next_line_prefetch_idx
        num_lines = math.ceil(self.prefetch_buf_size / self.req_gen_bandwidth)
        end_idx = start_idx + num_lines
        requested_data_size = num_lines * self.req_gen_bandwidth

requested_data_size is computed based on self.prefetch_buf_size, then in line 360-364,

        if requested_data_size > self.active_buf_size:
            valid_cols = int(self.active_buf_size % self.req_gen_bandwidth)
            row = prefetch_requests.shape[0] - 1
            for col in range(valid_cols, self.req_gen_bandwidth):
                prefetch_requests[row][col] = -1

it is compared to the self.active_buf_size, and due to the significant difference in buffer capacity, the code seems to be useless, maybe requested_data_size should be compared with self.prefetch_buf_size considering self.next_col_prefetch_idx

[ExcitingYi]

I also think there's something wrong. Strictly, the requested_data cannot be larger than prefetch_buf_size. Also, self.next_col_prefetch_idx should be updated in the func. Besides, the code in the following lines

        if requested_data_size > self.active_buf_size:
            self.next_line_prefetch_idx = num_lines % self.fetch_matrix.shape[0]
        else:
            self.next_line_prefetch_idx = (num_lines + 1) % self.fetch_matrix.shape[1]

seems to be wrong. I think it should be

        if requested_data_size > self.active_buf_size:
            self.next_line_prefetch_idx = (self.next_line_prefetch_idx + num_lines) % self.fetch_matrix.shape[0]
        else:
            self.next_line_prefetch_idx = (self.next_line_prefetch_idx + num_lines + 1) % self.fetch_matrix.shape[0]

But they don't affect the results. I think we don't need to care about the "fake data" from DRAM. As the stall cycle is decided by hashed_buf (whether the addr in buf) and num_lines (the size of prefetched buf).