intl/icu/source/tools/toolutil/denseranges.cpp

Wed, 31 Dec 2014 07:22:50 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Wed, 31 Dec 2014 07:22:50 +0100
branch
TOR_BUG_3246
changeset 4
fc2d59ddac77
permissions
-rw-r--r--

Correct previous dual key logic pending first delivery installment.

     1 /*
     2 *******************************************************************************
     3 *   Copyright (C) 2010, International Business Machines
     4 *   Corporation and others.  All Rights Reserved.
     5 *******************************************************************************
     6 *   file name:  denseranges.cpp
     7 *   encoding:   US-ASCII
     8 *   tab size:   8 (not used)
     9 *   indentation:4
    10 *
    11 *   created on: 2010sep25
    12 *   created by: Markus W. Scherer
    13 *
    14 * Helper code for finding a small number of dense ranges.
    15 */
    17 #include "unicode/utypes.h"
    18 #include "denseranges.h"
    20 // Definitions in the anonymous namespace are invisible outside this file.
    21 namespace {
    23 /**
    24  * Collect up to 15 range gaps and sort them by ascending gap size.
    25  */
    26 class LargestGaps {
    27 public:
    28     LargestGaps(int32_t max) : maxLength(max<=kCapacity ? max : kCapacity), length(0) {}
    30     void add(int32_t gapStart, int64_t gapLength) {
    31         int32_t i=length;
    32         while(i>0 && gapLength>gapLengths[i-1]) {
    33             --i;
    34         }
    35         if(i<maxLength) {
    36             // The new gap is now one of the maxLength largest.
    37             // Insert the new gap, moving up smaller ones of the previous
    38             // length largest.
    39             int32_t j= length<maxLength ? length++ : maxLength-1;
    40             while(j>i) {
    41                 gapStarts[j]=gapStarts[j-1];
    42                 gapLengths[j]=gapLengths[j-1];
    43                 --j;
    44             }
    45             gapStarts[i]=gapStart;
    46             gapLengths[i]=gapLength;
    47         }
    48     }
    50     void truncate(int32_t newLength) {
    51         if(newLength<length) {
    52             length=newLength;
    53         }
    54     }
    56     int32_t count() const { return length; }
    57     int32_t gapStart(int32_t i) const { return gapStarts[i]; }
    58     int64_t gapLength(int32_t i) const { return gapLengths[i]; }
    60     int32_t firstAfter(int32_t value) const {
    61         if(length==0) {
    62             return -1;
    63         }
    64         int32_t minValue=0;
    65         int32_t minIndex=-1;
    66         for(int32_t i=0; i<length; ++i) {
    67             if(value<gapStarts[i] && (minIndex<0 || gapStarts[i]<minValue)) {
    68                 minValue=gapStarts[i];
    69                 minIndex=i;
    70             }
    71         }
    72         return minIndex;
    73     }
    75 private:
    76     static const int32_t kCapacity=15;
    78     int32_t maxLength;
    79     int32_t length;
    80     int32_t gapStarts[kCapacity];
    81     int64_t gapLengths[kCapacity];
    82 };
    84 }  // namespace
    86 /**
    87  * Does it make sense to write 1..capacity ranges?
    88  * Returns 0 if not, otherwise the number of ranges.
    89  * @param values Sorted array of signed-integer values.
    90  * @param length Number of values.
    91  * @param density Minimum average range density, in 256th. (0x100=100%=perfectly dense.)
    92  *                Should be 0x80..0x100, must be 1..0x100.
    93  * @param ranges Output ranges array.
    94  * @param capacity Maximum number of ranges.
    95  * @return Minimum number of ranges (at most capacity) that have the desired density,
    96  *         or 0 if that density cannot be achieved.
    97  */
    98 U_CAPI int32_t U_EXPORT2
    99 uprv_makeDenseRanges(const int32_t values[], int32_t length,
   100                      int32_t density,
   101                      int32_t ranges[][2], int32_t capacity) {
   102     if(length<=2) {
   103         return 0;
   104     }
   105     int32_t minValue=values[0];
   106     int32_t maxValue=values[length-1];  // Assume minValue<=maxValue.
   107     // Use int64_t variables for intermediate-value precision and to avoid
   108     // signed-int32_t overflow of maxValue-minValue.
   109     int64_t maxLength=(int64_t)maxValue-(int64_t)minValue+1;
   110     if(length>=(density*maxLength)/0x100) {
   111         // Use one range.
   112         ranges[0][0]=minValue;
   113         ranges[0][1]=maxValue;
   114         return 1;
   115     }
   116     if(length<=4) {
   117         return 0;
   118     }
   119     // See if we can split [minValue, maxValue] into 2..capacity ranges,
   120     // divided by the 1..(capacity-1) largest gaps.
   121     LargestGaps gaps(capacity-1);
   122     int32_t i;
   123     int32_t expectedValue=minValue;
   124     for(i=1; i<length; ++i) {
   125         ++expectedValue;
   126         int32_t actualValue=values[i];
   127         if(expectedValue!=actualValue) {
   128             gaps.add(expectedValue, (int64_t)actualValue-(int64_t)expectedValue);
   129             expectedValue=actualValue;
   130         }
   131     }
   132     // We know gaps.count()>=1 because we have fewer values (length) than
   133     // the length of the [minValue..maxValue] range (maxLength).
   134     // (Otherwise we would have returned with the one range above.)
   135     int32_t num;
   136     for(i=0, num=2;; ++i, ++num) {
   137         if(i>=gaps.count()) {
   138             // The values are too sparse for capacity or fewer ranges
   139             // of the requested density.
   140             return 0;
   141         }
   142         maxLength-=gaps.gapLength(i);
   143         if(length>num*2 && length>=(density*maxLength)/0x100) {
   144             break;
   145         }
   146     }
   147     // Use the num ranges with the num-1 largest gaps.
   148     gaps.truncate(num-1);
   149     ranges[0][0]=minValue;
   150     for(i=0; i<=num-2; ++i) {
   151         int32_t gapIndex=gaps.firstAfter(minValue);
   152         int32_t gapStart=gaps.gapStart(gapIndex);
   153         ranges[i][1]=gapStart-1;
   154         ranges[i+1][0]=minValue=(int32_t)(gapStart+gaps.gapLength(gapIndex));
   155     }
   156     ranges[num-1][1]=maxValue;
   157     return num;
   158 }

mercurial