#include <math.h>
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include <limits.h>

#include "sorters.h"
#include "common.h"

static inline void *tail(char *base, size_t nmel, size_t width)
{
	return base + (nmel-1) * width;
}

/* get index of last block whose head is less than the previous block's tail */
static size_t last_overlap(char *base, size_t bcount, size_t bwidth, size_t width, cmpfun cmp)
{
	for (char *cur = tail(base, bcount, bwidth); --bcount; cur -= bwidth)
		if (cmp(cur - width, cur) > 0) break;
	return bcount;
}

size_t merge(char *buf, char *a, char *b, size_t bufnmel, size_t anmel, size_t bnmel, size_t width, cmpfun cmp)
{
	while (bnmel) {
		char *a_tail   = tail(a,   anmel,   width);
		char *b_tail   = tail(b,   bnmel,   width);
		char *buf_tail = tail(buf, bufnmel, width);

		if (cmp(a_tail, b_tail) <= 0) {
			swap(buf_tail, b_tail, width);
			bnmel--;
		} else {
			swap(buf_tail, a_tail, width);
			anmel--;
		}
		bufnmel--;
	}
	return anmel;
}

void grailsort(void *unsorted, size_t nmel, size_t width, cmpfun cmp)
{
	char *base = unsorted;

	if (nmel <= MAX_SORTNET) {
		sorting_network(base, nmel, width, cmp);
		return;
	}

	size_t blknmel = sqrt(nmel);               /* elements in a block */
	size_t bufnmel = blknmel + nmel % blknmel; /* elements in the buffer */
	size_t bwidth  = blknmel * width;          /* size of a block in bytes */
	size_t blocks  = nmel / blknmel - 1;       /* number of blocks in a + b */
	size_t acount  = blocks / 2;
	size_t bcount  = blocks - acount;

	char *a   = base;
	char *b   = a + acount * bwidth;
	char *buf = b + bcount * bwidth;

	grailsort(a, acount * blknmel, width, cmp);
	grailsort(b, bcount * blknmel, width, cmp);

	/* sort all the a and b blocks together by their head elements */
	grailsort(base, blocks, bwidth, cmp);

	/* merge, starting from the end and working towards the beginning */
	size_t last_nmel = blknmel;
	while (blocks > 1) {
		size_t overlap = last_overlap(base, blocks, bwidth, width, cmp);

		if (!overlap) break;

		char *block_a = base + bwidth * (overlap-1);
		char *block_b = base + bwidth *  overlap;
		size_t bnmel  = (blocks - overlap - 1) * blknmel + last_nmel;

		last_nmel = merge(buf, block_a, block_b, bufnmel, blknmel, bnmel, width, cmp);
		buf       = block_a + last_nmel * width;
		blocks    = overlap;
	}

	rotate(base, buf-base + bufnmel*width, buf-base);
	grailsort(base, bufnmel, width, cmp);
	distribute_buffer(base, bufnmel, nmel - bufnmel, width, cmp);
}