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

#include "sorters.h"
#include "common.h"
#include "counts.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)
{
	struct counts snapshot = counts[CURRENT];

	for (char *cur = tail(base, bcount, bwidth); --bcount; cur -= bwidth)
		if (cmp(cur - width, cur) > 0) break;

	add_counts(counts + LAST_OVERLAP, &snapshot);
	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)
{
	struct counts snapshot = counts[CURRENT];

	size_t skipa;
	for (skipa = 0; cmp(a, b) <= 0; skipa++) a += width;
	anmel -= skipa;

	swap(buf, a, anmel * width);
	char *tmp = buf;
	buf = a;
	a   = tmp;

	while (anmel && bnmel) {
		size_t count;

		do {
			for (count = 0; bnmel && count < anmel && cmp(a, b + count*width) > 0; count++) bnmel--;
			swap(buf, b, count * width);
			b   += count * width;
			buf += count * width;
		} while (count == anmel);

		for (count = 0; anmel && cmp(a + count*width, b) <= 0; count++) anmel--;
		swap(buf, a, count * width);
		a   += count * width;
		buf += count * width;
	}

	swap(buf, a, anmel * width);

	add_counts(counts + MERGE, &snapshot);
	return skipa;
}

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 *buf = base;
	char *a   = buf + bufnmel * width;
	char *b   = a   + acount * bwidth;

	grailsort(buf, acount * blknmel + bufnmel, width, cmp);
	grailsort(b,   bcount * blknmel,           width, cmp);

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

		if (!overlap) break;

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

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

	grailsort(buf, bufnmel, width, cmp);
	distribute_buffer(buf, bufnmel, nmel - bufnmel, width, cmp);

#if 0
	assert_sorted(base, nmel, width, cmp);
#endif
}