X-Git-Url: http://mj.ucw.cz/gitweb/?a=blobdiff_plain;f=lib%2Fsorter%2Fsorter.h;h=971e6e67b91d9102d837f4aa06ff13afc70ab887;hb=a1f74b39f32118bc29bd83e73564143000aeb951;hp=fe6cefc11f28b67df6c4d92e9f95704441fccce6;hpb=5ff0a5b047624c8f215da9cd6c7182d609fcbd01;p=libucw.git

diff --git a/lib/sorter/sorter.h b/lib/sorter/sorter.h
index fe6cefc1..971e6e67 100644
--- a/lib/sorter/sorter.h
+++ b/lib/sorter/sorter.h
@@ -46,34 +46,42 @@
  *			is supplied automatically and the sorting function gets an extra
  *			parameter specifying a range of the integers. The better the range
  *			fits, the faster we sort. Sets up SORT_HASH_xxx automatically.
+ *  SORT_INT64(key)	the same for 64-bit integers.
  *
  *  Hashing (optional, but it can speed sorting up):
  *
  *  SORT_HASH_BITS	signals that a monotone hashing function returning a given number of
  *			bits is available. Monotone hash is a function f such that f(x) < f(y)
  *			implies x < y and which is approximately uniformly distributed.
- *  uns PREFIX_hash(SORT_KEY *a, SORT_KEY *b)
+ *  uns PREFIX_hash(SORT_KEY *a)
  *
  *  Unification:
  *
- *  SORT_MERGE		merge items with identical keys, needs the following functions:
- *  void PREFIX_write_merged(struct fastbuf *f, SORT_KEY **keys, uns n, byte *buf)
+ *  SORT_UNIFY		merge items with identical keys, needs the following functions:
+ *  void PREFIX_write_merged(struct fastbuf *f, SORT_KEY **keys, void **data, uns n, void *buf)
  *			takes n records in memory with keys which compare equal and writes
- *			a single record to the given fastbuf. Data for each key can
- *			be accessed by the SORT_GET_DATA(*key) macro. `buf' points
- *			to a buffer which is guaranteed to hold all given records.
+ *			a single record to the given fastbuf. `buf' points to a buffer which
+ *			is guaranteed to hold the sum of workspace requirements (see below)
+ *			over all given records. The function is allowed to modify all its inputs.
  *  void PREFIX_copy_merged(SORT_KEY **keys, struct fastbuf **data, uns n, struct fastbuf *dest)
  *			takes n records with keys in memory and data in fastbufs and writes
- *			a single record.
+ *			a single record. Used only if SORT_DATA_SIZE or SORT_UNIFY_WORKSPACE is defined.
+ *  SORT_UNIFY_WORKSPACE(key)
+ *			gets a key and returns the amount of workspace required when merging
+ *			the given record. Defaults to 0.
  *
  *  Input (choose one of these):
  *
  *  SORT_INPUT_FILE	file of a given name
- *  SORT_INPUT_FB	fastbuf stream
- *  SORT_INPUT_PRESORT	custom presorter: call function PREFIX_presorter (see below)
- *			to get successive batches of pre-sorted data as temporary
- *			fastbuf streams or NULL if no more data is available.
+ *  SORT_INPUT_FB	seekable fastbuf stream
+ *  SORT_INPUT_PIPE	non-seekable fastbuf stream
+ *  SORT_INPUT_PRESORT	custom presorter. Calls function
+ *  int PREFIX_presort(struct fastbuf *dest, void *buf, size_t bufsize)
+ *			to get successive batches of pre-sorted data.
  *			The function is passed a page-aligned presorting buffer.
+ *			It returns 1 on success or 0 on EOF.
+ *  SORT_DELETE_INPUT	A C expression, if true, then the input files are deleted
+ *			as soon as possible.
  *
  *  Output (chose one of these):
  *
@@ -85,14 +93,9 @@
  *
  *  SORT_UNIQUE		all items have distinct keys (checked in debug mode)
  *
- *  FIXME: Maybe implement these:
- *  ??? SORT_DELETE_INPUT	a C expression, if true, the input files are
- *			deleted as soon as possible
- *  ??? SORT_ALIGNED
- *
  *  The function generated:
  *
- *  <outfb> PREFIX_SORT(<in>, <out> [,<range>]), where:
+ *  <outfb> PREFIX_sort(<in>, <out> [,<range>]), where:
  *			<in> = input file name/fastbuf or NULL
  *			<out> = output file name/fastbuf or NULL
  *			<range> = maximum integer value for the SORT_INT mode
@@ -125,6 +128,14 @@ typedef SORT_KEY P(key);
 #error Missing definition of sorting key.
 #endif
 
+#ifdef SORT_INT64
+typedef u64 P(hash_t);
+#define SORT_INT SORT_INT64
+#define SORT_LONG_HASH
+#else
+typedef uns P(hash_t);
+#endif
+
 #ifdef SORT_INT
 static inline int P(compare) (P(key) *x, P(key) *y)
 {
@@ -136,14 +147,14 @@ static inline int P(compare) (P(key) *x, P(key) *y)
 }
 
 #ifndef SORT_HASH_BITS
-static inline int P(hash) (P(key) *x)
+static inline P(hash_t) P(hash) (P(key) *x)
 {
   return SORT_INT((*x));
 }
 #endif
 #endif
 
-#ifdef SORT_MERGE
+#ifdef SORT_UNIFY
 #define LESS <
 #else
 #define LESS <=
@@ -154,18 +165,48 @@ static inline int P(hash) (P(key) *x)
 #define SORT_ASSERT_UNIQUE
 #endif
 
+#ifdef SORT_KEY_SIZE
+#define SORT_VAR_KEY
+#else
+#define SORT_KEY_SIZE(key) sizeof(key)
+#endif
+
+#ifdef SORT_DATA_SIZE
+#define SORT_VAR_DATA
+#else
+#define SORT_DATA_SIZE(key) 0
+#endif
+
 static inline void P(copy_data)(P(key) *key, struct fastbuf *in, struct fastbuf *out)
 {
-  bwrite(out, key, sizeof(P(key)));
-#ifdef SORT_DATA_SIZE
+  P(write_key)(out, key);
+#ifdef SORT_VAR_DATA
   bbcopy(in, out, SORT_DATA_SIZE(*key));
 #else
   (void) in;
 #endif
 }
 
+#if defined(SORT_UNIFY) && !defined(SORT_VAR_DATA) && !defined(SORT_UNIFY_WORKSPACE)
+static inline void P(copy_merged)(P(key) **keys, struct fastbuf **data UNUSED, uns n, struct fastbuf *dest)
+{
+  P(write_merged)(dest, keys, NULL, n, NULL);
+}
+#endif
+
+#if defined(SORT_HASH_BITS) || defined(SORT_INT)
+#define SORT_INTERNAL_RADIX
+#include "lib/sorter/s-radix.h"
+#endif
+
+#if defined(SORT_VAR_KEY) || defined(SORT_VAR_DATA) || defined(SORT_UNIFY_WORKSPACE)
 #include "lib/sorter/s-internal.h"
+#else
+#include "lib/sorter/s-fixint.h"
+#endif
+
 #include "lib/sorter/s-twoway.h"
+#include "lib/sorter/s-multiway.h"
 
 static struct fastbuf *P(sort)(
 #ifdef SORT_INPUT_FILE
@@ -179,7 +220,7 @@ static struct fastbuf *P(sort)(
 			       struct fastbuf *out
 #endif
 #ifdef SORT_INT
-			       , uns int_range
+			       , u64 int_range
 #endif
 			       )
 {
@@ -187,15 +228,25 @@ static struct fastbuf *P(sort)(
   bzero(&ctx, sizeof(ctx));
 
 #ifdef SORT_INPUT_FILE
-  ctx.in_fb = bopen(in, O_RDONLY, sorter_stream_bufsize);
+  ctx.in_fb = bopen_file(in, O_RDONLY, &sorter_fb_params);
+  ctx.in_size = bfilesize(ctx.in_fb);
 #elif defined(SORT_INPUT_FB)
   ctx.in_fb = in;
+  ctx.in_size = bfilesize(in);
+#elif defined(SORT_INPUT_PIPE)
+  ctx.in_fb = in;
+  ctx.in_size = ~(u64)0;
 #elif defined(SORT_INPUT_PRESORT)
   ASSERT(!in);
-  ctx.custom_presort = P(presorter);
+  ctx.custom_presort = P(presort);
+  ctx.in_size = ~(u64)0;
 #else
 #error No input given.
 #endif
+#ifdef SORT_DELETE_INPUT
+  if (SORT_DELETE_INPUT)
+    bconfig(ctx.in_fb, BCONFIG_IS_TEMP_FILE, 1);
+#endif
 
 #ifdef SORT_OUTPUT_FB
   ASSERT(!out);
@@ -209,34 +260,40 @@ static struct fastbuf *P(sort)(
 
 #ifdef SORT_HASH_BITS
   ctx.hash_bits = SORT_HASH_BITS;
+  ctx.radix_split = P(radix_split);
 #elif defined(SORT_INT)
   ctx.hash_bits = 0;
-  while (ctx.hash_bits < 32 && (int_range >> ctx.hash_bits))
+  while (ctx.hash_bits < 64 && (int_range >> ctx.hash_bits))
     ctx.hash_bits++;
+  ctx.radix_split = P(radix_split);
 #endif
 
   ctx.internal_sort = P(internal);
+  ctx.internal_estimate = P(internal_estimate);
   ctx.twoway_merge = P(twoway_merge);
+  ctx.multiway_merge = P(multiway_merge);
 
   sorter_run(&ctx);
 
 #ifdef SORT_OUTPUT_FILE
-  if (rename(ctx.out_fb->name, out) < 0)
-    die("Cannot rename %s to %s: %m", ctx.out_fb->name, out);
-  bconfig(ctx.out_fb, BCONFIG_IS_TEMP_FILE, 0);
-  bclose(ctx.out_fb);
+  bfix_tmp_file(ctx.out_fb, out);
   ctx.out_fb = NULL;
 #endif
   return ctx.out_fb;
 }
 
+#undef SORT_PREFIX
 #undef SORT_KEY
 #undef SORT_KEY_REGULAR
 #undef SORT_KEY_SIZE
 #undef SORT_DATA_SIZE
+#undef SORT_VAR_KEY
+#undef SORT_VAR_DATA
 #undef SORT_INT
+#undef SORT_INT64
 #undef SORT_HASH_BITS
-#undef SORT_MERGE
+#undef SORT_UNIFY
+#undef SORT_UNIFY_WORKSPACE
 #undef SORT_INPUT_FILE
 #undef SORT_INPUT_FB
 #undef SORT_INPUT_PRESORT
@@ -245,6 +302,9 @@ static struct fastbuf *P(sort)(
 #undef SORT_OUTPUT_THIS_FB
 #undef SORT_UNIQUE
 #undef SORT_ASSERT_UNIQUE
+#undef SORT_DELETE_INPUT
+#undef SORT_INTERNAL_RADIX
+#undef SORT_LONG_HASH
 #undef SWAP
 #undef LESS
 #undef P