2 * UCW Library -- Universal Sorter: Common Declarations
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.
10 #ifndef _UCW_SORTER_COMMON_H
11 #define _UCW_SORTER_COMMON_H
13 #include "ucw/clists.h"
15 /* Configuration variables */
16 extern uns sorter_trace, sorter_trace_array, sorter_stream_bufsize;
17 extern uns sorter_debug, sorter_min_radix_bits, sorter_max_radix_bits, sorter_add_radix_bits;
18 extern uns sorter_min_multiway_bits, sorter_max_multiway_bits;
19 extern uns sorter_threads;
20 extern u64 sorter_bufsize, sorter_small_input;
21 extern u64 sorter_thread_threshold, sorter_thread_chunk, sorter_radix_threshold;
22 extern struct fb_params sorter_fb_params, sorter_small_fb_params;
24 #define SORT_TRACE(x...) do { if (sorter_trace) msg(L_DEBUG, x); } while(0)
25 #define SORT_XTRACE(level, x...) do { if (sorter_trace >= level) msg(L_DEBUG, x); } while(0)
28 SORT_DEBUG_NO_PRESORT = 1,
29 SORT_DEBUG_NO_JOIN = 2,
30 SORT_DEBUG_KEEP_BUCKETS = 4,
31 SORT_DEBUG_NO_RADIX = 8,
32 SORT_DEBUG_NO_MULTIWAY = 16,
33 SORT_DEBUG_ASORT_NO_RADIX = 32,
34 SORT_DEBUG_ASORT_NO_THREADS = 64
40 struct fastbuf *in_fb;
41 struct fastbuf *out_fb;
44 struct fb_params *fb_params;
51 int (*custom_presort)(struct fastbuf *dest, void *buf, size_t bufsize);
53 // Take as much as possible from the source bucket, sort it in memory and dump to destination bucket.
54 // Return 1 if there is more data available in the source bucket.
55 int (*internal_sort)(struct sort_context *ctx, struct sort_bucket *in, struct sort_bucket *out, struct sort_bucket *out_only);
57 // Estimate how much input data from `b' will fit in the internal sorting buffer.
58 u64 (*internal_estimate)(struct sort_context *ctx, struct sort_bucket *b);
60 // Two-way split/merge: merge up to 2 source buckets to up to 2 destination buckets.
61 // Bucket arrays are NULL-terminated.
62 void (*twoway_merge)(struct sort_context *ctx, struct sort_bucket **ins, struct sort_bucket **outs);
64 // Multi-way merge: merge an arbitrary number of source buckets to a single destination bucket.
65 void (*multiway_merge)(struct sort_context *ctx, struct sort_bucket **ins, struct sort_bucket *out);
67 // Radix split according to hash function
68 void (*radix_split)(struct sort_context *ctx, struct sort_bucket *in, struct sort_bucket **outs, uns bitpos, uns numbits);
70 // State variables of internal_sort
75 timestamp_t start_time;
77 uns total_int_time, total_pre_time, total_ext_time;
80 void sorter_run(struct sort_context *ctx);
84 void *sorter_alloc(struct sort_context *ctx, uns size);
85 void sorter_prepare_buf(struct sort_context *ctx);
86 void sorter_alloc_buf(struct sort_context *ctx);
87 void sorter_free_buf(struct sort_context *ctx);
93 struct sort_context *ctx;
97 u64 size; // Size in bytes (not valid when writing)
98 uns runs; // Number of runs, 0 if not sorted
99 uns hash_bits; // Remaining bits of the hash function
100 byte *ident; // Identifier used in debug messages
103 enum sort_bucket_flags {
104 SBF_FINAL = 1, // This bucket corresponds to the final output file (always 1 run)
105 SBF_SOURCE = 2, // Contains the source file (always 0 runs)
106 SBF_CUSTOM_PRESORT = 4, // Contains source to read via custom presorter
107 SBF_OPEN_WRITE = 256, // We are currently writing to the fastbuf
108 SBF_OPEN_READ = 512, // We are reading from the fastbuf
109 SBF_DESTROYED = 1024, // Already done with, no further references allowed
110 SBF_SWAPPED_OUT = 2048, // Swapped out to a named file
113 struct sort_bucket *sbuck_new(struct sort_context *ctx);
114 void sbuck_drop(struct sort_bucket *b);
115 int sbuck_have(struct sort_bucket *b);
116 int sbuck_has_file(struct sort_bucket *b);
117 ucw_off_t sbuck_size(struct sort_bucket *b);
118 struct fastbuf *sbuck_read(struct sort_bucket *b);
119 struct fastbuf *sbuck_write(struct sort_bucket *b);
120 void sbuck_swap_out(struct sort_bucket *b);
122 /* Contexts and helper functions for the array sorter */
124 struct asort_context {
125 // Interface between generic code in array.c and functions generated by array.h
126 void *array; // Array to sort
127 void *buffer; // Auxiliary buffer (required when radix-sorting)
128 uns num_elts; // Number of elements in the array
129 uns elt_size; // Bytes per element
130 uns hash_bits; // Remaining bits of the hash function
131 uns radix_bits; // How many bits to process in a single radix-sort pass
132 void (*quicksort)(void *array_ptr, uns num_elts);
133 void (*quicksplit)(void *array_ptr, uns num_elts, int *leftp, int *rightp);
134 void (*radix_count)(void *src_ptr, uns num_elts, uns *cnt, uns shift);
135 void (*radix_split)(void *src_ptr, void *dest_ptr, uns num_elts, uns *ptrs, uns shift);
137 // Used internally by array.c
138 struct rs_work **rs_works;
139 struct work_queue *rs_work_queue;
140 struct eltpool *eltpool;
142 // Configured limits translated from bytes to elements
143 uns thread_threshold;
148 void asort_run(struct asort_context *ctx);
149 void asort_start_threads(uns run);
150 void asort_stop_threads(void);