2 * UCW Library -- Universal Sorter: Governing Routines
4 * (c) 2007 Martin Mares <mj@ucw.cz>
6 * This software may be freely distributed and used according to the terms
7 * of the GNU Lesser General Public License.
11 #include "lib/fastbuf.h"
12 #include "lib/mempool.h"
13 #include "lib/sorter/common.h"
18 sorter_alloc(struct sort_context *ctx, uns size)
20 return mp_alloc_zero(ctx->pool, size);
24 sbuck_new(struct sort_context *ctx)
26 struct sort_bucket *b = sorter_alloc(ctx, sizeof(struct sort_bucket));
32 sbuck_drop(struct sort_bucket *b)
36 ASSERT(!(b->flags & SBF_DESTROYED));
41 b->flags = SBF_DESTROYED;
46 sbuck_size(struct sort_bucket *b)
48 if ((b->flags & SBF_OPEN_WRITE) && !(b->flags & SBF_SWAPPED_OUT))
55 sbuck_have(struct sort_bucket *b)
57 return b && sbuck_size(b);
61 sbuck_swap_in(struct sort_bucket *b)
63 if (b->flags & SBF_SWAPPED_OUT)
65 b->fb = bopen(b->filename, O_RDWR, sorter_stream_bufsize);
66 if (b->flags & SBF_OPEN_WRITE)
67 bseek(b->fb, 0, SEEK_END);
68 bconfig(b->fb, BCONFIG_IS_TEMP_FILE, 1);
69 b->flags &= ~SBF_SWAPPED_OUT;
70 SORT_XTRACE("Swapped in %s", b->filename);
75 sbuck_read(struct sort_bucket *b)
78 if (b->flags & SBF_OPEN_READ)
80 else if (b->flags & SBF_OPEN_WRITE)
82 b->size = btell(b->fb);
83 b->flags = (b->flags & ~SBF_OPEN_WRITE) | SBF_OPEN_READ;
92 sbuck_write(struct sort_bucket *b)
95 if (b->flags & SBF_OPEN_WRITE)
99 ASSERT(!(b->flags & (SBF_OPEN_READ | SBF_DESTROYED)));
100 b->fb = bopen_tmp(sorter_stream_bufsize);
101 b->flags |= SBF_OPEN_WRITE;
102 b->filename = mp_strdup(b->ctx->pool, b->fb->name);
108 sbuck_swap_out(struct sort_bucket *b)
110 if ((b->flags & (SBF_OPEN_READ | SBF_OPEN_WRITE)) && b->fb)
112 if (b->flags & SBF_OPEN_WRITE)
113 b->size = btell(b->fb);
114 bconfig(b->fb, BCONFIG_IS_TEMP_FILE, 0);
117 b->flags |= SBF_SWAPPED_OUT;
118 SORT_XTRACE("Swapped out %s", b->filename);
123 sorter_alloc_buf(struct sort_context *ctx)
127 u64 bs = MAX(sorter_bufsize/2, 1);
128 bs = ALIGN_TO(bs, (u64)CPU_PAGE_SIZE);
129 ctx->big_buf = big_alloc(2*bs);
130 ctx->big_buf_size = 2*bs;
131 ctx->big_buf_half = ((byte*) ctx->big_buf) + bs;
132 ctx->big_buf_half_size = bs;
133 SORT_XTRACE("Allocated sorting buffer (%jd bytes)", (uintmax_t) bs);
137 sorter_free_buf(struct sort_context *ctx)
141 big_free(ctx->big_buf, ctx->big_buf_size);
143 SORT_XTRACE("Freed sorting buffer");
146 static int sorter_presort(struct sort_context *ctx, struct sort_bucket *in, struct sort_bucket *out, struct sort_bucket *out_only)
148 /* FIXME: Mode with no presorting (mostly for debugging) */
149 sorter_alloc_buf(ctx);
150 if (in->flags & SBF_CUSTOM_PRESORT)
152 struct fastbuf *f = sbuck_write(out);
153 return ctx->custom_presort(f, ctx->big_buf, ctx->big_buf_size); // FIXME: out_only optimization?
155 return ctx->internal_sort(ctx, in, out, out_only);
158 static inline struct sort_bucket *
159 sbuck_join_to(struct sort_bucket *b)
161 struct sort_bucket *out = (struct sort_bucket *) b->n.prev; // Such bucket is guaranteed to exist
162 return (out->flags & SBF_FINAL) ? out : NULL;
166 sorter_join(struct sort_bucket *b)
168 struct sort_bucket *join = sbuck_join_to(b);
171 // FIXME: What if the final bucket doesn't contain any file yet?
173 SORT_TRACE("Copying %jd bytes to output file", (uintmax_t) sbuck_size(b));
174 struct fastbuf *src = sbuck_read(b);
175 struct fastbuf *dest = sbuck_write(join);
176 bbcopy(src, dest, ~0U);
181 sorter_twoway(struct sort_context *ctx, struct sort_bucket *b)
183 struct sort_bucket *ins[3] = { NULL }, *outs[3] = { NULL };
184 cnode *list_pos = b->n.prev;
185 struct sort_bucket *join = sbuck_join_to(b);
186 if (sorter_debug & SORT_DEBUG_NO_JOIN)
189 if (!(sorter_debug & SORT_DEBUG_NO_PRESORT) || (b->flags & SBF_CUSTOM_PRESORT))
191 SORT_TRACE("Presorting");
192 ins[0] = sbuck_new(ctx);
193 if (!sorter_presort(ctx, b, ins[0], join ? : ins[0]))
198 clist_insert_after(&ins[0]->n, list_pos);
203 ins[1] = sbuck_new(ctx);
205 while (sorter_presort(ctx, b, ins[i], ins[i]))
211 SORT_TRACE("Skipped presorting");
215 SORT_TRACE("Main sorting");
217 if (ins[0]->runs == 1 && ins[1]->runs == 1 && join)
219 // This is guaranteed to produce a single run, so join if possible
222 ctx->twoway_merge(ctx, ins, outs);
223 ASSERT(outs[0]->runs == 2);
225 SORT_TRACE("Pass done (joined final run)");
230 outs[0] = sbuck_new(ctx);
231 outs[1] = sbuck_new(ctx);
233 ctx->twoway_merge(ctx, ins, outs);
234 SORT_TRACE("Pass done (%d+%d runs, %jd+%jd bytes)", outs[0]->runs, outs[1]->runs, (uintmax_t) sbuck_size(outs[0]), (uintmax_t) sbuck_size(outs[1]));
237 memcpy(ins, outs, 3*sizeof(struct sort_bucket *));
238 } while (sbuck_have(ins[1]));
241 clist_insert_after(&ins[0]->n, list_pos);
245 sorter_run(struct sort_context *ctx)
247 ctx->pool = mp_new(4096);
248 clist_init(&ctx->bucket_list);
250 /* FIXME: There should be a way how to detect size of the input file */
251 /* FIXME: Remember to test sorting of empty files */
253 // Create bucket containing the source
254 struct sort_bucket *bin = sbuck_new(ctx);
255 bin->flags = SBF_SOURCE | SBF_OPEN_READ;
256 if (ctx->custom_presort)
257 bin->flags |= SBF_CUSTOM_PRESORT;
259 bin->fb = ctx->in_fb;
262 bin->hash_bits = ctx->hash_bits;
263 clist_add_tail(&ctx->bucket_list, &bin->n);
265 // Create bucket for the output
266 struct sort_bucket *bout = sbuck_new(ctx);
267 bout->flags = SBF_FINAL;
268 bout->fb = ctx->out_fb;
271 clist_add_head(&ctx->bucket_list, &bout->n);
273 struct sort_bucket *b;
274 while (b = clist_next(&ctx->bucket_list, &bout->n))
278 else if (b->runs == 1)
281 sorter_twoway(ctx, b);
284 sorter_free_buf(ctx);
285 SORT_XTRACE("Final size: %jd", (uintmax_t) sbuck_size(bout));
286 ctx->out_fb = sbuck_read(bout);