xref: /freebsd/sys/contrib/zstd/lib/dictBuilder/fastcover.c (revision 37f1f2684f2670b204080ef2d6c303becd28545f)
1*37f1f268SConrad Meyer /*
2*37f1f268SConrad Meyer  * Copyright (c) 2018-2020, Facebook, Inc.
3*37f1f268SConrad Meyer  * All rights reserved.
4*37f1f268SConrad Meyer  *
5*37f1f268SConrad Meyer  * This source code is licensed under both the BSD-style license (found in the
6*37f1f268SConrad Meyer  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7*37f1f268SConrad Meyer  * in the COPYING file in the root directory of this source tree).
8*37f1f268SConrad Meyer  * You may select, at your option, one of the above-listed licenses.
9*37f1f268SConrad Meyer  */
10*37f1f268SConrad Meyer 
110f743729SConrad Meyer /*-*************************************
120f743729SConrad Meyer *  Dependencies
130f743729SConrad Meyer ***************************************/
140f743729SConrad Meyer #include <stdio.h>  /* fprintf */
150f743729SConrad Meyer #include <stdlib.h> /* malloc, free, qsort */
160f743729SConrad Meyer #include <string.h> /* memset */
170f743729SConrad Meyer #include <time.h>   /* clock */
180f743729SConrad Meyer 
19*37f1f268SConrad Meyer #include "../common/mem.h" /* read */
20*37f1f268SConrad Meyer #include "../common/pool.h"
21*37f1f268SConrad Meyer #include "../common/threading.h"
220f743729SConrad Meyer #include "cover.h"
23*37f1f268SConrad Meyer #include "../common/zstd_internal.h" /* includes zstd.h */
240f743729SConrad Meyer #ifndef ZDICT_STATIC_LINKING_ONLY
250f743729SConrad Meyer #define ZDICT_STATIC_LINKING_ONLY
260f743729SConrad Meyer #endif
270f743729SConrad Meyer #include "zdict.h"
280f743729SConrad Meyer 
290f743729SConrad Meyer 
300f743729SConrad Meyer /*-*************************************
310f743729SConrad Meyer *  Constants
320f743729SConrad Meyer ***************************************/
33a0483764SConrad Meyer #define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
340f743729SConrad Meyer #define FASTCOVER_MAX_F 31
350f743729SConrad Meyer #define FASTCOVER_MAX_ACCEL 10
360f743729SConrad Meyer #define DEFAULT_SPLITPOINT 0.75
370f743729SConrad Meyer #define DEFAULT_F 20
380f743729SConrad Meyer #define DEFAULT_ACCEL 1
390f743729SConrad Meyer 
400f743729SConrad Meyer 
410f743729SConrad Meyer /*-*************************************
420f743729SConrad Meyer *  Console display
430f743729SConrad Meyer ***************************************/
440f743729SConrad Meyer static int g_displayLevel = 2;
450f743729SConrad Meyer #define DISPLAY(...)                                                           \
460f743729SConrad Meyer   {                                                                            \
470f743729SConrad Meyer     fprintf(stderr, __VA_ARGS__);                                              \
480f743729SConrad Meyer     fflush(stderr);                                                            \
490f743729SConrad Meyer   }
500f743729SConrad Meyer #define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
510f743729SConrad Meyer   if (displayLevel >= l) {                                                     \
520f743729SConrad Meyer     DISPLAY(__VA_ARGS__);                                                      \
530f743729SConrad Meyer   } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
540f743729SConrad Meyer #define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
550f743729SConrad Meyer 
560f743729SConrad Meyer #define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
570f743729SConrad Meyer   if (displayLevel >= l) {                                                     \
580f743729SConrad Meyer     if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
590f743729SConrad Meyer       g_time = clock();                                                        \
600f743729SConrad Meyer       DISPLAY(__VA_ARGS__);                                                    \
610f743729SConrad Meyer     }                                                                          \
620f743729SConrad Meyer   }
630f743729SConrad Meyer #define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
640f743729SConrad Meyer static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
650f743729SConrad Meyer static clock_t g_time = 0;
660f743729SConrad Meyer 
670f743729SConrad Meyer 
680f743729SConrad Meyer /*-*************************************
690f743729SConrad Meyer * Hash Functions
700f743729SConrad Meyer ***************************************/
710f743729SConrad Meyer static const U64 prime6bytes = 227718039650203ULL;
720f743729SConrad Meyer static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
730f743729SConrad Meyer static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
740f743729SConrad Meyer 
750f743729SConrad Meyer static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
760f743729SConrad Meyer static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
770f743729SConrad Meyer static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
780f743729SConrad Meyer 
790f743729SConrad Meyer 
800f743729SConrad Meyer /**
810f743729SConrad Meyer  * Hash the d-byte value pointed to by p and mod 2^f
820f743729SConrad Meyer  */
830f743729SConrad Meyer static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 h, unsigned d) {
840f743729SConrad Meyer   if (d == 6) {
850f743729SConrad Meyer     return ZSTD_hash6Ptr(p, h) & ((1 << h) - 1);
860f743729SConrad Meyer   }
870f743729SConrad Meyer   return ZSTD_hash8Ptr(p, h) & ((1 << h) - 1);
880f743729SConrad Meyer }
890f743729SConrad Meyer 
900f743729SConrad Meyer 
910f743729SConrad Meyer /*-*************************************
920f743729SConrad Meyer * Acceleration
930f743729SConrad Meyer ***************************************/
940f743729SConrad Meyer typedef struct {
950f743729SConrad Meyer   unsigned finalize;    /* Percentage of training samples used for ZDICT_finalizeDictionary */
960f743729SConrad Meyer   unsigned skip;        /* Number of dmer skipped between each dmer counted in computeFrequency */
970f743729SConrad Meyer } FASTCOVER_accel_t;
980f743729SConrad Meyer 
990f743729SConrad Meyer 
1000f743729SConrad Meyer static const FASTCOVER_accel_t FASTCOVER_defaultAccelParameters[FASTCOVER_MAX_ACCEL+1] = {
1010f743729SConrad Meyer   { 100, 0 },   /* accel = 0, should not happen because accel = 0 defaults to accel = 1 */
1020f743729SConrad Meyer   { 100, 0 },   /* accel = 1 */
1030f743729SConrad Meyer   { 50, 1 },   /* accel = 2 */
1040f743729SConrad Meyer   { 34, 2 },   /* accel = 3 */
1050f743729SConrad Meyer   { 25, 3 },   /* accel = 4 */
1060f743729SConrad Meyer   { 20, 4 },   /* accel = 5 */
1070f743729SConrad Meyer   { 17, 5 },   /* accel = 6 */
1080f743729SConrad Meyer   { 14, 6 },   /* accel = 7 */
1090f743729SConrad Meyer   { 13, 7 },   /* accel = 8 */
1100f743729SConrad Meyer   { 11, 8 },   /* accel = 9 */
1110f743729SConrad Meyer   { 10, 9 },   /* accel = 10 */
1120f743729SConrad Meyer };
1130f743729SConrad Meyer 
1140f743729SConrad Meyer 
1150f743729SConrad Meyer /*-*************************************
1160f743729SConrad Meyer * Context
1170f743729SConrad Meyer ***************************************/
1180f743729SConrad Meyer typedef struct {
1190f743729SConrad Meyer   const BYTE *samples;
1200f743729SConrad Meyer   size_t *offsets;
1210f743729SConrad Meyer   const size_t *samplesSizes;
1220f743729SConrad Meyer   size_t nbSamples;
1230f743729SConrad Meyer   size_t nbTrainSamples;
1240f743729SConrad Meyer   size_t nbTestSamples;
1250f743729SConrad Meyer   size_t nbDmers;
1260f743729SConrad Meyer   U32 *freqs;
1270f743729SConrad Meyer   unsigned d;
1280f743729SConrad Meyer   unsigned f;
1290f743729SConrad Meyer   FASTCOVER_accel_t accelParams;
1300f743729SConrad Meyer } FASTCOVER_ctx_t;
1310f743729SConrad Meyer 
1320f743729SConrad Meyer 
1330f743729SConrad Meyer /*-*************************************
1340f743729SConrad Meyer *  Helper functions
1350f743729SConrad Meyer ***************************************/
1360f743729SConrad Meyer /**
1370f743729SConrad Meyer  * Selects the best segment in an epoch.
1380f743729SConrad Meyer  * Segments of are scored according to the function:
1390f743729SConrad Meyer  *
1400f743729SConrad Meyer  * Let F(d) be the frequency of all dmers with hash value d.
1410f743729SConrad Meyer  * Let S_i be hash value of the dmer at position i of segment S which has length k.
1420f743729SConrad Meyer  *
1430f743729SConrad Meyer  *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
1440f743729SConrad Meyer  *
1452b9c00cbSConrad Meyer  * Once the dmer with hash value d is in the dictionary we set F(d) = 0.
1460f743729SConrad Meyer  */
1470f743729SConrad Meyer static COVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx,
1480f743729SConrad Meyer                                               U32 *freqs, U32 begin, U32 end,
1490f743729SConrad Meyer                                               ZDICT_cover_params_t parameters,
1500f743729SConrad Meyer                                               U16* segmentFreqs) {
1510f743729SConrad Meyer   /* Constants */
1520f743729SConrad Meyer   const U32 k = parameters.k;
1530f743729SConrad Meyer   const U32 d = parameters.d;
1540f743729SConrad Meyer   const U32 f = ctx->f;
1550f743729SConrad Meyer   const U32 dmersInK = k - d + 1;
1560f743729SConrad Meyer 
1570f743729SConrad Meyer   /* Try each segment (activeSegment) and save the best (bestSegment) */
1580f743729SConrad Meyer   COVER_segment_t bestSegment = {0, 0, 0};
1590f743729SConrad Meyer   COVER_segment_t activeSegment;
1600f743729SConrad Meyer 
1610f743729SConrad Meyer   /* Reset the activeDmers in the segment */
1620f743729SConrad Meyer   /* The activeSegment starts at the beginning of the epoch. */
1630f743729SConrad Meyer   activeSegment.begin = begin;
1640f743729SConrad Meyer   activeSegment.end = begin;
1650f743729SConrad Meyer   activeSegment.score = 0;
1660f743729SConrad Meyer 
1670f743729SConrad Meyer   /* Slide the activeSegment through the whole epoch.
1680f743729SConrad Meyer    * Save the best segment in bestSegment.
1690f743729SConrad Meyer    */
1700f743729SConrad Meyer   while (activeSegment.end < end) {
1710f743729SConrad Meyer     /* Get hash value of current dmer */
172a0483764SConrad Meyer     const size_t idx = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.end, f, d);
1730f743729SConrad Meyer 
1742b9c00cbSConrad Meyer     /* Add frequency of this index to score if this is the first occurrence of index in active segment */
175a0483764SConrad Meyer     if (segmentFreqs[idx] == 0) {
176a0483764SConrad Meyer       activeSegment.score += freqs[idx];
1770f743729SConrad Meyer     }
1780f743729SConrad Meyer     /* Increment end of segment and segmentFreqs*/
1790f743729SConrad Meyer     activeSegment.end += 1;
180a0483764SConrad Meyer     segmentFreqs[idx] += 1;
1810f743729SConrad Meyer     /* If the window is now too large, drop the first position */
1820f743729SConrad Meyer     if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
1830f743729SConrad Meyer       /* Get hash value of the dmer to be eliminated from active segment */
1840f743729SConrad Meyer       const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
1850f743729SConrad Meyer       segmentFreqs[delIndex] -= 1;
1860f743729SConrad Meyer       /* Subtract frequency of this index from score if this is the last occurrence of this index in active segment */
1870f743729SConrad Meyer       if (segmentFreqs[delIndex] == 0) {
1880f743729SConrad Meyer         activeSegment.score -= freqs[delIndex];
1890f743729SConrad Meyer       }
1900f743729SConrad Meyer       /* Increment start of segment */
1910f743729SConrad Meyer       activeSegment.begin += 1;
1920f743729SConrad Meyer     }
1930f743729SConrad Meyer 
1940f743729SConrad Meyer     /* If this segment is the best so far save it */
1950f743729SConrad Meyer     if (activeSegment.score > bestSegment.score) {
1960f743729SConrad Meyer       bestSegment = activeSegment;
1970f743729SConrad Meyer     }
1980f743729SConrad Meyer   }
1990f743729SConrad Meyer 
2000f743729SConrad Meyer   /* Zero out rest of segmentFreqs array */
2010f743729SConrad Meyer   while (activeSegment.begin < end) {
2020f743729SConrad Meyer     const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
2030f743729SConrad Meyer     segmentFreqs[delIndex] -= 1;
2040f743729SConrad Meyer     activeSegment.begin += 1;
2050f743729SConrad Meyer   }
2060f743729SConrad Meyer 
2070f743729SConrad Meyer   {
2080f743729SConrad Meyer     /*  Zero the frequency of hash value of each dmer covered by the chosen segment. */
2090f743729SConrad Meyer     U32 pos;
2100f743729SConrad Meyer     for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
2110f743729SConrad Meyer       const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, f, d);
2120f743729SConrad Meyer       freqs[i] = 0;
2130f743729SConrad Meyer     }
2140f743729SConrad Meyer   }
2150f743729SConrad Meyer 
2160f743729SConrad Meyer   return bestSegment;
2170f743729SConrad Meyer }
2180f743729SConrad Meyer 
2190f743729SConrad Meyer 
2200f743729SConrad Meyer static int FASTCOVER_checkParameters(ZDICT_cover_params_t parameters,
2210f743729SConrad Meyer                                      size_t maxDictSize, unsigned f,
2220f743729SConrad Meyer                                      unsigned accel) {
2230f743729SConrad Meyer   /* k, d, and f are required parameters */
2240f743729SConrad Meyer   if (parameters.d == 0 || parameters.k == 0) {
2250f743729SConrad Meyer     return 0;
2260f743729SConrad Meyer   }
2270f743729SConrad Meyer   /* d has to be 6 or 8 */
2280f743729SConrad Meyer   if (parameters.d != 6 && parameters.d != 8) {
2290f743729SConrad Meyer     return 0;
2300f743729SConrad Meyer   }
2310f743729SConrad Meyer   /* k <= maxDictSize */
2320f743729SConrad Meyer   if (parameters.k > maxDictSize) {
2330f743729SConrad Meyer     return 0;
2340f743729SConrad Meyer   }
2350f743729SConrad Meyer   /* d <= k */
2360f743729SConrad Meyer   if (parameters.d > parameters.k) {
2370f743729SConrad Meyer     return 0;
2380f743729SConrad Meyer   }
2390f743729SConrad Meyer   /* 0 < f <= FASTCOVER_MAX_F*/
2400f743729SConrad Meyer   if (f > FASTCOVER_MAX_F || f == 0) {
2410f743729SConrad Meyer     return 0;
2420f743729SConrad Meyer   }
2430f743729SConrad Meyer   /* 0 < splitPoint <= 1 */
2440f743729SConrad Meyer   if (parameters.splitPoint <= 0 || parameters.splitPoint > 1) {
2450f743729SConrad Meyer     return 0;
2460f743729SConrad Meyer   }
2470f743729SConrad Meyer   /* 0 < accel <= 10 */
2480f743729SConrad Meyer   if (accel > 10 || accel == 0) {
2490f743729SConrad Meyer     return 0;
2500f743729SConrad Meyer   }
2510f743729SConrad Meyer   return 1;
2520f743729SConrad Meyer }
2530f743729SConrad Meyer 
2540f743729SConrad Meyer 
2550f743729SConrad Meyer /**
2560f743729SConrad Meyer  * Clean up a context initialized with `FASTCOVER_ctx_init()`.
2570f743729SConrad Meyer  */
2580f743729SConrad Meyer static void
2590f743729SConrad Meyer FASTCOVER_ctx_destroy(FASTCOVER_ctx_t* ctx)
2600f743729SConrad Meyer {
2610f743729SConrad Meyer     if (!ctx) return;
2620f743729SConrad Meyer 
2630f743729SConrad Meyer     free(ctx->freqs);
2640f743729SConrad Meyer     ctx->freqs = NULL;
2650f743729SConrad Meyer 
2660f743729SConrad Meyer     free(ctx->offsets);
2670f743729SConrad Meyer     ctx->offsets = NULL;
2680f743729SConrad Meyer }
2690f743729SConrad Meyer 
2700f743729SConrad Meyer 
2710f743729SConrad Meyer /**
2720f743729SConrad Meyer  * Calculate for frequency of hash value of each dmer in ctx->samples
2730f743729SConrad Meyer  */
2740f743729SConrad Meyer static void
2750f743729SConrad Meyer FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
2760f743729SConrad Meyer {
2770f743729SConrad Meyer     const unsigned f = ctx->f;
2780f743729SConrad Meyer     const unsigned d = ctx->d;
2790f743729SConrad Meyer     const unsigned skip = ctx->accelParams.skip;
2800f743729SConrad Meyer     const unsigned readLength = MAX(d, 8);
2810f743729SConrad Meyer     size_t i;
2820f743729SConrad Meyer     assert(ctx->nbTrainSamples >= 5);
2830f743729SConrad Meyer     assert(ctx->nbTrainSamples <= ctx->nbSamples);
2840f743729SConrad Meyer     for (i = 0; i < ctx->nbTrainSamples; i++) {
2850f743729SConrad Meyer         size_t start = ctx->offsets[i];  /* start of current dmer */
2860f743729SConrad Meyer         size_t const currSampleEnd = ctx->offsets[i+1];
2870f743729SConrad Meyer         while (start + readLength <= currSampleEnd) {
2880f743729SConrad Meyer             const size_t dmerIndex = FASTCOVER_hashPtrToIndex(ctx->samples + start, f, d);
2890f743729SConrad Meyer             freqs[dmerIndex]++;
2900f743729SConrad Meyer             start = start + skip + 1;
2910f743729SConrad Meyer         }
2920f743729SConrad Meyer     }
2930f743729SConrad Meyer }
2940f743729SConrad Meyer 
2950f743729SConrad Meyer 
2960f743729SConrad Meyer /**
2970f743729SConrad Meyer  * Prepare a context for dictionary building.
2980f743729SConrad Meyer  * The context is only dependent on the parameter `d` and can used multiple
2990f743729SConrad Meyer  * times.
3004d3f1eafSConrad Meyer  * Returns 0 on success or error code on error.
3010f743729SConrad Meyer  * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
3020f743729SConrad Meyer  */
3034d3f1eafSConrad Meyer static size_t
3040f743729SConrad Meyer FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
3050f743729SConrad Meyer                    const void* samplesBuffer,
3060f743729SConrad Meyer                    const size_t* samplesSizes, unsigned nbSamples,
3070f743729SConrad Meyer                    unsigned d, double splitPoint, unsigned f,
3080f743729SConrad Meyer                    FASTCOVER_accel_t accelParams)
3090f743729SConrad Meyer {
3100f743729SConrad Meyer     const BYTE* const samples = (const BYTE*)samplesBuffer;
3110f743729SConrad Meyer     const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
3120f743729SConrad Meyer     /* Split samples into testing and training sets */
3130f743729SConrad Meyer     const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
3140f743729SConrad Meyer     const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
3150f743729SConrad Meyer     const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
3160f743729SConrad Meyer     const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
3170f743729SConrad Meyer 
3180f743729SConrad Meyer     /* Checks */
3190f743729SConrad Meyer     if (totalSamplesSize < MAX(d, sizeof(U64)) ||
3200f743729SConrad Meyer         totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
3210f743729SConrad Meyer         DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
322a0483764SConrad Meyer                     (unsigned)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
3234d3f1eafSConrad Meyer         return ERROR(srcSize_wrong);
3240f743729SConrad Meyer     }
3250f743729SConrad Meyer 
3260f743729SConrad Meyer     /* Check if there are at least 5 training samples */
3270f743729SConrad Meyer     if (nbTrainSamples < 5) {
3280f743729SConrad Meyer         DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
3294d3f1eafSConrad Meyer         return ERROR(srcSize_wrong);
3300f743729SConrad Meyer     }
3310f743729SConrad Meyer 
3320f743729SConrad Meyer     /* Check if there's testing sample */
3330f743729SConrad Meyer     if (nbTestSamples < 1) {
3340f743729SConrad Meyer         DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
3354d3f1eafSConrad Meyer         return ERROR(srcSize_wrong);
3360f743729SConrad Meyer     }
3370f743729SConrad Meyer 
3380f743729SConrad Meyer     /* Zero the context */
3390f743729SConrad Meyer     memset(ctx, 0, sizeof(*ctx));
3400f743729SConrad Meyer     DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
341a0483764SConrad Meyer                     (unsigned)trainingSamplesSize);
3420f743729SConrad Meyer     DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
343a0483764SConrad Meyer                     (unsigned)testSamplesSize);
3440f743729SConrad Meyer 
3450f743729SConrad Meyer     ctx->samples = samples;
3460f743729SConrad Meyer     ctx->samplesSizes = samplesSizes;
3470f743729SConrad Meyer     ctx->nbSamples = nbSamples;
3480f743729SConrad Meyer     ctx->nbTrainSamples = nbTrainSamples;
3490f743729SConrad Meyer     ctx->nbTestSamples = nbTestSamples;
3500f743729SConrad Meyer     ctx->nbDmers = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
3510f743729SConrad Meyer     ctx->d = d;
3520f743729SConrad Meyer     ctx->f = f;
3530f743729SConrad Meyer     ctx->accelParams = accelParams;
3540f743729SConrad Meyer 
3550f743729SConrad Meyer     /* The offsets of each file */
3560f743729SConrad Meyer     ctx->offsets = (size_t*)calloc((nbSamples + 1), sizeof(size_t));
3570f743729SConrad Meyer     if (ctx->offsets == NULL) {
3580f743729SConrad Meyer         DISPLAYLEVEL(1, "Failed to allocate scratch buffers \n");
3590f743729SConrad Meyer         FASTCOVER_ctx_destroy(ctx);
3604d3f1eafSConrad Meyer         return ERROR(memory_allocation);
3610f743729SConrad Meyer     }
3620f743729SConrad Meyer 
3630f743729SConrad Meyer     /* Fill offsets from the samplesSizes */
3640f743729SConrad Meyer     {   U32 i;
3650f743729SConrad Meyer         ctx->offsets[0] = 0;
3660f743729SConrad Meyer         assert(nbSamples >= 5);
3670f743729SConrad Meyer         for (i = 1; i <= nbSamples; ++i) {
3680f743729SConrad Meyer             ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
3690f743729SConrad Meyer         }
3700f743729SConrad Meyer     }
3710f743729SConrad Meyer 
3720f743729SConrad Meyer     /* Initialize frequency array of size 2^f */
3730f743729SConrad Meyer     ctx->freqs = (U32*)calloc(((U64)1 << f), sizeof(U32));
3740f743729SConrad Meyer     if (ctx->freqs == NULL) {
3750f743729SConrad Meyer         DISPLAYLEVEL(1, "Failed to allocate frequency table \n");
3760f743729SConrad Meyer         FASTCOVER_ctx_destroy(ctx);
3774d3f1eafSConrad Meyer         return ERROR(memory_allocation);
3780f743729SConrad Meyer     }
3790f743729SConrad Meyer 
3800f743729SConrad Meyer     DISPLAYLEVEL(2, "Computing frequencies\n");
3810f743729SConrad Meyer     FASTCOVER_computeFrequency(ctx->freqs, ctx);
3820f743729SConrad Meyer 
3834d3f1eafSConrad Meyer     return 0;
3840f743729SConrad Meyer }
3850f743729SConrad Meyer 
3860f743729SConrad Meyer 
3870f743729SConrad Meyer /**
3880f743729SConrad Meyer  * Given the prepared context build the dictionary.
3890f743729SConrad Meyer  */
3900f743729SConrad Meyer static size_t
3910f743729SConrad Meyer FASTCOVER_buildDictionary(const FASTCOVER_ctx_t* ctx,
3920f743729SConrad Meyer                           U32* freqs,
3930f743729SConrad Meyer                           void* dictBuffer, size_t dictBufferCapacity,
3940f743729SConrad Meyer                           ZDICT_cover_params_t parameters,
3950f743729SConrad Meyer                           U16* segmentFreqs)
3960f743729SConrad Meyer {
3970f743729SConrad Meyer   BYTE *const dict = (BYTE *)dictBuffer;
3980f743729SConrad Meyer   size_t tail = dictBufferCapacity;
3992b9c00cbSConrad Meyer   /* Divide the data into epochs. We will select one segment from each epoch. */
4002b9c00cbSConrad Meyer   const COVER_epoch_info_t epochs = COVER_computeEpochs(
4012b9c00cbSConrad Meyer       (U32)dictBufferCapacity, (U32)ctx->nbDmers, parameters.k, 1);
4022b9c00cbSConrad Meyer   const size_t maxZeroScoreRun = 10;
4032b9c00cbSConrad Meyer   size_t zeroScoreRun = 0;
4040f743729SConrad Meyer   size_t epoch;
405a0483764SConrad Meyer   DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n",
4062b9c00cbSConrad Meyer                 (U32)epochs.num, (U32)epochs.size);
4070f743729SConrad Meyer   /* Loop through the epochs until there are no more segments or the dictionary
4080f743729SConrad Meyer    * is full.
4090f743729SConrad Meyer    */
4102b9c00cbSConrad Meyer   for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs.num) {
4112b9c00cbSConrad Meyer     const U32 epochBegin = (U32)(epoch * epochs.size);
4122b9c00cbSConrad Meyer     const U32 epochEnd = epochBegin + epochs.size;
4130f743729SConrad Meyer     size_t segmentSize;
4140f743729SConrad Meyer     /* Select a segment */
4150f743729SConrad Meyer     COVER_segment_t segment = FASTCOVER_selectSegment(
4160f743729SConrad Meyer         ctx, freqs, epochBegin, epochEnd, parameters, segmentFreqs);
4170f743729SConrad Meyer 
4182b9c00cbSConrad Meyer     /* If the segment covers no dmers, then we are out of content.
4192b9c00cbSConrad Meyer      * There may be new content in other epochs, for continue for some time.
4202b9c00cbSConrad Meyer      */
4210f743729SConrad Meyer     if (segment.score == 0) {
4222b9c00cbSConrad Meyer       if (++zeroScoreRun >= maxZeroScoreRun) {
4230f743729SConrad Meyer           break;
4240f743729SConrad Meyer       }
4252b9c00cbSConrad Meyer       continue;
4262b9c00cbSConrad Meyer     }
4272b9c00cbSConrad Meyer     zeroScoreRun = 0;
4280f743729SConrad Meyer 
4290f743729SConrad Meyer     /* Trim the segment if necessary and if it is too small then we are done */
4300f743729SConrad Meyer     segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
4310f743729SConrad Meyer     if (segmentSize < parameters.d) {
4320f743729SConrad Meyer       break;
4330f743729SConrad Meyer     }
4340f743729SConrad Meyer 
4350f743729SConrad Meyer     /* We fill the dictionary from the back to allow the best segments to be
4360f743729SConrad Meyer      * referenced with the smallest offsets.
4370f743729SConrad Meyer      */
4380f743729SConrad Meyer     tail -= segmentSize;
4390f743729SConrad Meyer     memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
4400f743729SConrad Meyer     DISPLAYUPDATE(
4410f743729SConrad Meyer         2, "\r%u%%       ",
442a0483764SConrad Meyer         (unsigned)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
4430f743729SConrad Meyer   }
4440f743729SConrad Meyer   DISPLAYLEVEL(2, "\r%79s\r", "");
4450f743729SConrad Meyer   return tail;
4460f743729SConrad Meyer }
4470f743729SConrad Meyer 
4480f743729SConrad Meyer /**
4490f743729SConrad Meyer  * Parameters for FASTCOVER_tryParameters().
4500f743729SConrad Meyer  */
4510f743729SConrad Meyer typedef struct FASTCOVER_tryParameters_data_s {
4520f743729SConrad Meyer     const FASTCOVER_ctx_t* ctx;
4530f743729SConrad Meyer     COVER_best_t* best;
4540f743729SConrad Meyer     size_t dictBufferCapacity;
4550f743729SConrad Meyer     ZDICT_cover_params_t parameters;
4560f743729SConrad Meyer } FASTCOVER_tryParameters_data_t;
4570f743729SConrad Meyer 
4580f743729SConrad Meyer 
4590f743729SConrad Meyer /**
4600f743729SConrad Meyer  * Tries a set of parameters and updates the COVER_best_t with the results.
4610f743729SConrad Meyer  * This function is thread safe if zstd is compiled with multithreaded support.
4620f743729SConrad Meyer  * It takes its parameters as an *OWNING* opaque pointer to support threading.
4630f743729SConrad Meyer  */
4640f743729SConrad Meyer static void FASTCOVER_tryParameters(void *opaque)
4650f743729SConrad Meyer {
4660f743729SConrad Meyer   /* Save parameters as local variables */
4670f743729SConrad Meyer   FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t *)opaque;
4680f743729SConrad Meyer   const FASTCOVER_ctx_t *const ctx = data->ctx;
4690f743729SConrad Meyer   const ZDICT_cover_params_t parameters = data->parameters;
4700f743729SConrad Meyer   size_t dictBufferCapacity = data->dictBufferCapacity;
4710f743729SConrad Meyer   size_t totalCompressedSize = ERROR(GENERIC);
4720f743729SConrad Meyer   /* Initialize array to keep track of frequency of dmer within activeSegment */
4730f743729SConrad Meyer   U16* segmentFreqs = (U16 *)calloc(((U64)1 << ctx->f), sizeof(U16));
4740f743729SConrad Meyer   /* Allocate space for hash table, dict, and freqs */
4750f743729SConrad Meyer   BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
4764d3f1eafSConrad Meyer   COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
4770f743729SConrad Meyer   U32 *freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
4780f743729SConrad Meyer   if (!segmentFreqs || !dict || !freqs) {
4790f743729SConrad Meyer     DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
4800f743729SConrad Meyer     goto _cleanup;
4810f743729SConrad Meyer   }
4820f743729SConrad Meyer   /* Copy the frequencies because we need to modify them */
4830f743729SConrad Meyer   memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
4840f743729SConrad Meyer   /* Build the dictionary */
4850f743729SConrad Meyer   { const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
4860f743729SConrad Meyer                                                     parameters, segmentFreqs);
4874d3f1eafSConrad Meyer 
4880f743729SConrad Meyer     const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
4894d3f1eafSConrad Meyer     selection = COVER_selectDict(dict + tail, dictBufferCapacity - tail,
4904d3f1eafSConrad Meyer          ctx->samples, ctx->samplesSizes, nbFinalizeSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
4914d3f1eafSConrad Meyer          totalCompressedSize);
4924d3f1eafSConrad Meyer 
4934d3f1eafSConrad Meyer     if (COVER_dictSelectionIsError(selection)) {
4944d3f1eafSConrad Meyer       DISPLAYLEVEL(1, "Failed to select dictionary\n");
4950f743729SConrad Meyer       goto _cleanup;
4960f743729SConrad Meyer     }
4970f743729SConrad Meyer   }
4980f743729SConrad Meyer _cleanup:
4994d3f1eafSConrad Meyer   free(dict);
5004d3f1eafSConrad Meyer   COVER_best_finish(data->best, parameters, selection);
5010f743729SConrad Meyer   free(data);
5020f743729SConrad Meyer   free(segmentFreqs);
5034d3f1eafSConrad Meyer   COVER_dictSelectionFree(selection);
5040f743729SConrad Meyer   free(freqs);
5050f743729SConrad Meyer }
5060f743729SConrad Meyer 
5070f743729SConrad Meyer 
5080f743729SConrad Meyer static void
5090f743729SConrad Meyer FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
5100f743729SConrad Meyer                                ZDICT_cover_params_t* coverParams)
5110f743729SConrad Meyer {
5120f743729SConrad Meyer     coverParams->k = fastCoverParams.k;
5130f743729SConrad Meyer     coverParams->d = fastCoverParams.d;
5140f743729SConrad Meyer     coverParams->steps = fastCoverParams.steps;
5150f743729SConrad Meyer     coverParams->nbThreads = fastCoverParams.nbThreads;
5160f743729SConrad Meyer     coverParams->splitPoint = fastCoverParams.splitPoint;
5170f743729SConrad Meyer     coverParams->zParams = fastCoverParams.zParams;
5184d3f1eafSConrad Meyer     coverParams->shrinkDict = fastCoverParams.shrinkDict;
5190f743729SConrad Meyer }
5200f743729SConrad Meyer 
5210f743729SConrad Meyer 
5220f743729SConrad Meyer static void
5230f743729SConrad Meyer FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
5240f743729SConrad Meyer                                    ZDICT_fastCover_params_t* fastCoverParams,
5250f743729SConrad Meyer                                    unsigned f, unsigned accel)
5260f743729SConrad Meyer {
5270f743729SConrad Meyer     fastCoverParams->k = coverParams.k;
5280f743729SConrad Meyer     fastCoverParams->d = coverParams.d;
5290f743729SConrad Meyer     fastCoverParams->steps = coverParams.steps;
5300f743729SConrad Meyer     fastCoverParams->nbThreads = coverParams.nbThreads;
5310f743729SConrad Meyer     fastCoverParams->splitPoint = coverParams.splitPoint;
5320f743729SConrad Meyer     fastCoverParams->f = f;
5330f743729SConrad Meyer     fastCoverParams->accel = accel;
5340f743729SConrad Meyer     fastCoverParams->zParams = coverParams.zParams;
5354d3f1eafSConrad Meyer     fastCoverParams->shrinkDict = coverParams.shrinkDict;
5360f743729SConrad Meyer }
5370f743729SConrad Meyer 
5380f743729SConrad Meyer 
5390f743729SConrad Meyer ZDICTLIB_API size_t
5400f743729SConrad Meyer ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
5410f743729SConrad Meyer                                 const void* samplesBuffer,
5420f743729SConrad Meyer                                 const size_t* samplesSizes, unsigned nbSamples,
5430f743729SConrad Meyer                                 ZDICT_fastCover_params_t parameters)
5440f743729SConrad Meyer {
5450f743729SConrad Meyer     BYTE* const dict = (BYTE*)dictBuffer;
5460f743729SConrad Meyer     FASTCOVER_ctx_t ctx;
5470f743729SConrad Meyer     ZDICT_cover_params_t coverParams;
5480f743729SConrad Meyer     FASTCOVER_accel_t accelParams;
5490f743729SConrad Meyer     /* Initialize global data */
5500f743729SConrad Meyer     g_displayLevel = parameters.zParams.notificationLevel;
5510f743729SConrad Meyer     /* Assign splitPoint and f if not provided */
5520f743729SConrad Meyer     parameters.splitPoint = 1.0;
5530f743729SConrad Meyer     parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
5540f743729SConrad Meyer     parameters.accel = parameters.accel == 0 ? DEFAULT_ACCEL : parameters.accel;
5550f743729SConrad Meyer     /* Convert to cover parameter */
5560f743729SConrad Meyer     memset(&coverParams, 0 , sizeof(coverParams));
5570f743729SConrad Meyer     FASTCOVER_convertToCoverParams(parameters, &coverParams);
5580f743729SConrad Meyer     /* Checks */
5590f743729SConrad Meyer     if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
5600f743729SConrad Meyer                                    parameters.accel)) {
5610f743729SConrad Meyer       DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
5624d3f1eafSConrad Meyer       return ERROR(parameter_outOfBound);
5630f743729SConrad Meyer     }
5640f743729SConrad Meyer     if (nbSamples == 0) {
5650f743729SConrad Meyer       DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
5664d3f1eafSConrad Meyer       return ERROR(srcSize_wrong);
5670f743729SConrad Meyer     }
5680f743729SConrad Meyer     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
5690f743729SConrad Meyer       DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
5700f743729SConrad Meyer                    ZDICT_DICTSIZE_MIN);
5710f743729SConrad Meyer       return ERROR(dstSize_tooSmall);
5720f743729SConrad Meyer     }
5730f743729SConrad Meyer     /* Assign corresponding FASTCOVER_accel_t to accelParams*/
5740f743729SConrad Meyer     accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
5750f743729SConrad Meyer     /* Initialize context */
5764d3f1eafSConrad Meyer     {
5774d3f1eafSConrad Meyer       size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
5780f743729SConrad Meyer                             coverParams.d, parameters.splitPoint, parameters.f,
5794d3f1eafSConrad Meyer                             accelParams);
5804d3f1eafSConrad Meyer       if (ZSTD_isError(initVal)) {
5810f743729SConrad Meyer         DISPLAYLEVEL(1, "Failed to initialize context\n");
5824d3f1eafSConrad Meyer         return initVal;
5834d3f1eafSConrad Meyer       }
5840f743729SConrad Meyer     }
5852b9c00cbSConrad Meyer     COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, g_displayLevel);
5860f743729SConrad Meyer     /* Build the dictionary */
5870f743729SConrad Meyer     DISPLAYLEVEL(2, "Building dictionary\n");
5880f743729SConrad Meyer     {
5890f743729SConrad Meyer       /* Initialize array to keep track of frequency of dmer within activeSegment */
5900f743729SConrad Meyer       U16* segmentFreqs = (U16 *)calloc(((U64)1 << parameters.f), sizeof(U16));
5910f743729SConrad Meyer       const size_t tail = FASTCOVER_buildDictionary(&ctx, ctx.freqs, dictBuffer,
5920f743729SConrad Meyer                                                 dictBufferCapacity, coverParams, segmentFreqs);
5930f743729SConrad Meyer       const unsigned nbFinalizeSamples = (unsigned)(ctx.nbTrainSamples * ctx.accelParams.finalize / 100);
5940f743729SConrad Meyer       const size_t dictionarySize = ZDICT_finalizeDictionary(
5950f743729SConrad Meyer           dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
5960f743729SConrad Meyer           samplesBuffer, samplesSizes, nbFinalizeSamples, coverParams.zParams);
5970f743729SConrad Meyer       if (!ZSTD_isError(dictionarySize)) {
5980f743729SConrad Meyer           DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
599a0483764SConrad Meyer                       (unsigned)dictionarySize);
6000f743729SConrad Meyer       }
6010f743729SConrad Meyer       FASTCOVER_ctx_destroy(&ctx);
6020f743729SConrad Meyer       free(segmentFreqs);
6030f743729SConrad Meyer       return dictionarySize;
6040f743729SConrad Meyer     }
6050f743729SConrad Meyer }
6060f743729SConrad Meyer 
6070f743729SConrad Meyer 
6080f743729SConrad Meyer ZDICTLIB_API size_t
6090f743729SConrad Meyer ZDICT_optimizeTrainFromBuffer_fastCover(
6100f743729SConrad Meyer                     void* dictBuffer, size_t dictBufferCapacity,
6110f743729SConrad Meyer                     const void* samplesBuffer,
6120f743729SConrad Meyer                     const size_t* samplesSizes, unsigned nbSamples,
6130f743729SConrad Meyer                     ZDICT_fastCover_params_t* parameters)
6140f743729SConrad Meyer {
6150f743729SConrad Meyer     ZDICT_cover_params_t coverParams;
6160f743729SConrad Meyer     FASTCOVER_accel_t accelParams;
6170f743729SConrad Meyer     /* constants */
6180f743729SConrad Meyer     const unsigned nbThreads = parameters->nbThreads;
6190f743729SConrad Meyer     const double splitPoint =
6200f743729SConrad Meyer         parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
6210f743729SConrad Meyer     const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
6220f743729SConrad Meyer     const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
6230f743729SConrad Meyer     const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
6240f743729SConrad Meyer     const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
6250f743729SConrad Meyer     const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
6260f743729SConrad Meyer     const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
6270f743729SConrad Meyer     const unsigned kIterations =
6280f743729SConrad Meyer         (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
6290f743729SConrad Meyer     const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
6300f743729SConrad Meyer     const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
6314d3f1eafSConrad Meyer     const unsigned shrinkDict = 0;
6320f743729SConrad Meyer     /* Local variables */
6330f743729SConrad Meyer     const int displayLevel = parameters->zParams.notificationLevel;
6340f743729SConrad Meyer     unsigned iteration = 1;
6350f743729SConrad Meyer     unsigned d;
6360f743729SConrad Meyer     unsigned k;
6370f743729SConrad Meyer     COVER_best_t best;
6380f743729SConrad Meyer     POOL_ctx *pool = NULL;
6392b9c00cbSConrad Meyer     int warned = 0;
6400f743729SConrad Meyer     /* Checks */
6410f743729SConrad Meyer     if (splitPoint <= 0 || splitPoint > 1) {
6420f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect splitPoint\n");
6434d3f1eafSConrad Meyer       return ERROR(parameter_outOfBound);
6440f743729SConrad Meyer     }
6450f743729SConrad Meyer     if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
6460f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect accel\n");
6474d3f1eafSConrad Meyer       return ERROR(parameter_outOfBound);
6480f743729SConrad Meyer     }
6490f743729SConrad Meyer     if (kMinK < kMaxD || kMaxK < kMinK) {
6500f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect k\n");
6514d3f1eafSConrad Meyer       return ERROR(parameter_outOfBound);
6520f743729SConrad Meyer     }
6530f743729SConrad Meyer     if (nbSamples == 0) {
6540f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "FASTCOVER must have at least one input file\n");
6554d3f1eafSConrad Meyer       return ERROR(srcSize_wrong);
6560f743729SConrad Meyer     }
6570f743729SConrad Meyer     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
6580f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "dictBufferCapacity must be at least %u\n",
6590f743729SConrad Meyer                    ZDICT_DICTSIZE_MIN);
6600f743729SConrad Meyer       return ERROR(dstSize_tooSmall);
6610f743729SConrad Meyer     }
6620f743729SConrad Meyer     if (nbThreads > 1) {
6630f743729SConrad Meyer       pool = POOL_create(nbThreads, 1);
6640f743729SConrad Meyer       if (!pool) {
6650f743729SConrad Meyer         return ERROR(memory_allocation);
6660f743729SConrad Meyer       }
6670f743729SConrad Meyer     }
6680f743729SConrad Meyer     /* Initialization */
6690f743729SConrad Meyer     COVER_best_init(&best);
6700f743729SConrad Meyer     memset(&coverParams, 0 , sizeof(coverParams));
6710f743729SConrad Meyer     FASTCOVER_convertToCoverParams(*parameters, &coverParams);
6720f743729SConrad Meyer     accelParams = FASTCOVER_defaultAccelParameters[accel];
6730f743729SConrad Meyer     /* Turn down global display level to clean up display at level 2 and below */
6740f743729SConrad Meyer     g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
6750f743729SConrad Meyer     /* Loop through d first because each new value needs a new context */
6760f743729SConrad Meyer     LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
6770f743729SConrad Meyer                       kIterations);
6780f743729SConrad Meyer     for (d = kMinD; d <= kMaxD; d += 2) {
6790f743729SConrad Meyer       /* Initialize the context for this value of d */
6800f743729SConrad Meyer       FASTCOVER_ctx_t ctx;
6810f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
6824d3f1eafSConrad Meyer       {
6834d3f1eafSConrad Meyer         size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams);
6844d3f1eafSConrad Meyer         if (ZSTD_isError(initVal)) {
6850f743729SConrad Meyer           LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
6860f743729SConrad Meyer           COVER_best_destroy(&best);
6870f743729SConrad Meyer           POOL_free(pool);
6884d3f1eafSConrad Meyer           return initVal;
6894d3f1eafSConrad Meyer         }
6900f743729SConrad Meyer       }
6912b9c00cbSConrad Meyer       if (!warned) {
6922b9c00cbSConrad Meyer         COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, displayLevel);
6932b9c00cbSConrad Meyer         warned = 1;
6942b9c00cbSConrad Meyer       }
6950f743729SConrad Meyer       /* Loop through k reusing the same context */
6960f743729SConrad Meyer       for (k = kMinK; k <= kMaxK; k += kStepSize) {
6970f743729SConrad Meyer         /* Prepare the arguments */
6980f743729SConrad Meyer         FASTCOVER_tryParameters_data_t *data = (FASTCOVER_tryParameters_data_t *)malloc(
6990f743729SConrad Meyer             sizeof(FASTCOVER_tryParameters_data_t));
7000f743729SConrad Meyer         LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
7010f743729SConrad Meyer         if (!data) {
7020f743729SConrad Meyer           LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
7030f743729SConrad Meyer           COVER_best_destroy(&best);
7040f743729SConrad Meyer           FASTCOVER_ctx_destroy(&ctx);
7050f743729SConrad Meyer           POOL_free(pool);
7064d3f1eafSConrad Meyer           return ERROR(memory_allocation);
7070f743729SConrad Meyer         }
7080f743729SConrad Meyer         data->ctx = &ctx;
7090f743729SConrad Meyer         data->best = &best;
7100f743729SConrad Meyer         data->dictBufferCapacity = dictBufferCapacity;
7110f743729SConrad Meyer         data->parameters = coverParams;
7120f743729SConrad Meyer         data->parameters.k = k;
7130f743729SConrad Meyer         data->parameters.d = d;
7140f743729SConrad Meyer         data->parameters.splitPoint = splitPoint;
7150f743729SConrad Meyer         data->parameters.steps = kSteps;
7164d3f1eafSConrad Meyer         data->parameters.shrinkDict = shrinkDict;
7170f743729SConrad Meyer         data->parameters.zParams.notificationLevel = g_displayLevel;
7180f743729SConrad Meyer         /* Check the parameters */
7190f743729SConrad Meyer         if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
7200f743729SConrad Meyer                                        data->ctx->f, accel)) {
7210f743729SConrad Meyer           DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
7220f743729SConrad Meyer           free(data);
7230f743729SConrad Meyer           continue;
7240f743729SConrad Meyer         }
7250f743729SConrad Meyer         /* Call the function and pass ownership of data to it */
7260f743729SConrad Meyer         COVER_best_start(&best);
7270f743729SConrad Meyer         if (pool) {
7280f743729SConrad Meyer           POOL_add(pool, &FASTCOVER_tryParameters, data);
7290f743729SConrad Meyer         } else {
7300f743729SConrad Meyer           FASTCOVER_tryParameters(data);
7310f743729SConrad Meyer         }
7320f743729SConrad Meyer         /* Print status */
7330f743729SConrad Meyer         LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
734a0483764SConrad Meyer                            (unsigned)((iteration * 100) / kIterations));
7350f743729SConrad Meyer         ++iteration;
7360f743729SConrad Meyer       }
7370f743729SConrad Meyer       COVER_best_wait(&best);
7380f743729SConrad Meyer       FASTCOVER_ctx_destroy(&ctx);
7390f743729SConrad Meyer     }
7400f743729SConrad Meyer     LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
7410f743729SConrad Meyer     /* Fill the output buffer and parameters with output of the best parameters */
7420f743729SConrad Meyer     {
7430f743729SConrad Meyer       const size_t dictSize = best.dictSize;
7440f743729SConrad Meyer       if (ZSTD_isError(best.compressedSize)) {
7450f743729SConrad Meyer         const size_t compressedSize = best.compressedSize;
7460f743729SConrad Meyer         COVER_best_destroy(&best);
7470f743729SConrad Meyer         POOL_free(pool);
7480f743729SConrad Meyer         return compressedSize;
7490f743729SConrad Meyer       }
7500f743729SConrad Meyer       FASTCOVER_convertToFastCoverParams(best.parameters, parameters, f, accel);
7510f743729SConrad Meyer       memcpy(dictBuffer, best.dict, dictSize);
7520f743729SConrad Meyer       COVER_best_destroy(&best);
7530f743729SConrad Meyer       POOL_free(pool);
7540f743729SConrad Meyer       return dictSize;
7550f743729SConrad Meyer     }
7560f743729SConrad Meyer 
7570f743729SConrad Meyer }
758