xref: /freebsd/sys/contrib/zstd/lib/dictBuilder/fastcover.c (revision 2b9c00cb6bd9392645dc8afca59cf57c42df4e2d)
10f743729SConrad Meyer /*-*************************************
20f743729SConrad Meyer *  Dependencies
30f743729SConrad Meyer ***************************************/
40f743729SConrad Meyer #include <stdio.h>  /* fprintf */
50f743729SConrad Meyer #include <stdlib.h> /* malloc, free, qsort */
60f743729SConrad Meyer #include <string.h> /* memset */
70f743729SConrad Meyer #include <time.h>   /* clock */
80f743729SConrad Meyer 
90f743729SConrad Meyer #include "mem.h" /* read */
100f743729SConrad Meyer #include "pool.h"
110f743729SConrad Meyer #include "threading.h"
120f743729SConrad Meyer #include "cover.h"
130f743729SConrad Meyer #include "zstd_internal.h" /* includes zstd.h */
140f743729SConrad Meyer #ifndef ZDICT_STATIC_LINKING_ONLY
150f743729SConrad Meyer #define ZDICT_STATIC_LINKING_ONLY
160f743729SConrad Meyer #endif
170f743729SConrad Meyer #include "zdict.h"
180f743729SConrad Meyer 
190f743729SConrad Meyer 
200f743729SConrad Meyer /*-*************************************
210f743729SConrad Meyer *  Constants
220f743729SConrad Meyer ***************************************/
23a0483764SConrad Meyer #define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
240f743729SConrad Meyer #define FASTCOVER_MAX_F 31
250f743729SConrad Meyer #define FASTCOVER_MAX_ACCEL 10
260f743729SConrad Meyer #define DEFAULT_SPLITPOINT 0.75
270f743729SConrad Meyer #define DEFAULT_F 20
280f743729SConrad Meyer #define DEFAULT_ACCEL 1
290f743729SConrad Meyer 
300f743729SConrad Meyer 
310f743729SConrad Meyer /*-*************************************
320f743729SConrad Meyer *  Console display
330f743729SConrad Meyer ***************************************/
340f743729SConrad Meyer static int g_displayLevel = 2;
350f743729SConrad Meyer #define DISPLAY(...)                                                           \
360f743729SConrad Meyer   {                                                                            \
370f743729SConrad Meyer     fprintf(stderr, __VA_ARGS__);                                              \
380f743729SConrad Meyer     fflush(stderr);                                                            \
390f743729SConrad Meyer   }
400f743729SConrad Meyer #define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
410f743729SConrad Meyer   if (displayLevel >= l) {                                                     \
420f743729SConrad Meyer     DISPLAY(__VA_ARGS__);                                                      \
430f743729SConrad Meyer   } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
440f743729SConrad Meyer #define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
450f743729SConrad Meyer 
460f743729SConrad Meyer #define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
470f743729SConrad Meyer   if (displayLevel >= l) {                                                     \
480f743729SConrad Meyer     if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
490f743729SConrad Meyer       g_time = clock();                                                        \
500f743729SConrad Meyer       DISPLAY(__VA_ARGS__);                                                    \
510f743729SConrad Meyer     }                                                                          \
520f743729SConrad Meyer   }
530f743729SConrad Meyer #define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
540f743729SConrad Meyer static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
550f743729SConrad Meyer static clock_t g_time = 0;
560f743729SConrad Meyer 
570f743729SConrad Meyer 
580f743729SConrad Meyer /*-*************************************
590f743729SConrad Meyer * Hash Functions
600f743729SConrad Meyer ***************************************/
610f743729SConrad Meyer static const U64 prime6bytes = 227718039650203ULL;
620f743729SConrad Meyer static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
630f743729SConrad Meyer static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
640f743729SConrad Meyer 
650f743729SConrad Meyer static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
660f743729SConrad Meyer static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
670f743729SConrad Meyer static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
680f743729SConrad Meyer 
690f743729SConrad Meyer 
700f743729SConrad Meyer /**
710f743729SConrad Meyer  * Hash the d-byte value pointed to by p and mod 2^f
720f743729SConrad Meyer  */
730f743729SConrad Meyer static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 h, unsigned d) {
740f743729SConrad Meyer   if (d == 6) {
750f743729SConrad Meyer     return ZSTD_hash6Ptr(p, h) & ((1 << h) - 1);
760f743729SConrad Meyer   }
770f743729SConrad Meyer   return ZSTD_hash8Ptr(p, h) & ((1 << h) - 1);
780f743729SConrad Meyer }
790f743729SConrad Meyer 
800f743729SConrad Meyer 
810f743729SConrad Meyer /*-*************************************
820f743729SConrad Meyer * Acceleration
830f743729SConrad Meyer ***************************************/
840f743729SConrad Meyer typedef struct {
850f743729SConrad Meyer   unsigned finalize;    /* Percentage of training samples used for ZDICT_finalizeDictionary */
860f743729SConrad Meyer   unsigned skip;        /* Number of dmer skipped between each dmer counted in computeFrequency */
870f743729SConrad Meyer } FASTCOVER_accel_t;
880f743729SConrad Meyer 
890f743729SConrad Meyer 
900f743729SConrad Meyer static const FASTCOVER_accel_t FASTCOVER_defaultAccelParameters[FASTCOVER_MAX_ACCEL+1] = {
910f743729SConrad Meyer   { 100, 0 },   /* accel = 0, should not happen because accel = 0 defaults to accel = 1 */
920f743729SConrad Meyer   { 100, 0 },   /* accel = 1 */
930f743729SConrad Meyer   { 50, 1 },   /* accel = 2 */
940f743729SConrad Meyer   { 34, 2 },   /* accel = 3 */
950f743729SConrad Meyer   { 25, 3 },   /* accel = 4 */
960f743729SConrad Meyer   { 20, 4 },   /* accel = 5 */
970f743729SConrad Meyer   { 17, 5 },   /* accel = 6 */
980f743729SConrad Meyer   { 14, 6 },   /* accel = 7 */
990f743729SConrad Meyer   { 13, 7 },   /* accel = 8 */
1000f743729SConrad Meyer   { 11, 8 },   /* accel = 9 */
1010f743729SConrad Meyer   { 10, 9 },   /* accel = 10 */
1020f743729SConrad Meyer };
1030f743729SConrad Meyer 
1040f743729SConrad Meyer 
1050f743729SConrad Meyer /*-*************************************
1060f743729SConrad Meyer * Context
1070f743729SConrad Meyer ***************************************/
1080f743729SConrad Meyer typedef struct {
1090f743729SConrad Meyer   const BYTE *samples;
1100f743729SConrad Meyer   size_t *offsets;
1110f743729SConrad Meyer   const size_t *samplesSizes;
1120f743729SConrad Meyer   size_t nbSamples;
1130f743729SConrad Meyer   size_t nbTrainSamples;
1140f743729SConrad Meyer   size_t nbTestSamples;
1150f743729SConrad Meyer   size_t nbDmers;
1160f743729SConrad Meyer   U32 *freqs;
1170f743729SConrad Meyer   unsigned d;
1180f743729SConrad Meyer   unsigned f;
1190f743729SConrad Meyer   FASTCOVER_accel_t accelParams;
1200f743729SConrad Meyer } FASTCOVER_ctx_t;
1210f743729SConrad Meyer 
1220f743729SConrad Meyer 
1230f743729SConrad Meyer /*-*************************************
1240f743729SConrad Meyer *  Helper functions
1250f743729SConrad Meyer ***************************************/
1260f743729SConrad Meyer /**
1270f743729SConrad Meyer  * Selects the best segment in an epoch.
1280f743729SConrad Meyer  * Segments of are scored according to the function:
1290f743729SConrad Meyer  *
1300f743729SConrad Meyer  * Let F(d) be the frequency of all dmers with hash value d.
1310f743729SConrad Meyer  * Let S_i be hash value of the dmer at position i of segment S which has length k.
1320f743729SConrad Meyer  *
1330f743729SConrad Meyer  *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
1340f743729SConrad Meyer  *
135*2b9c00cbSConrad Meyer  * Once the dmer with hash value d is in the dictionary we set F(d) = 0.
1360f743729SConrad Meyer  */
1370f743729SConrad Meyer static COVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx,
1380f743729SConrad Meyer                                               U32 *freqs, U32 begin, U32 end,
1390f743729SConrad Meyer                                               ZDICT_cover_params_t parameters,
1400f743729SConrad Meyer                                               U16* segmentFreqs) {
1410f743729SConrad Meyer   /* Constants */
1420f743729SConrad Meyer   const U32 k = parameters.k;
1430f743729SConrad Meyer   const U32 d = parameters.d;
1440f743729SConrad Meyer   const U32 f = ctx->f;
1450f743729SConrad Meyer   const U32 dmersInK = k - d + 1;
1460f743729SConrad Meyer 
1470f743729SConrad Meyer   /* Try each segment (activeSegment) and save the best (bestSegment) */
1480f743729SConrad Meyer   COVER_segment_t bestSegment = {0, 0, 0};
1490f743729SConrad Meyer   COVER_segment_t activeSegment;
1500f743729SConrad Meyer 
1510f743729SConrad Meyer   /* Reset the activeDmers in the segment */
1520f743729SConrad Meyer   /* The activeSegment starts at the beginning of the epoch. */
1530f743729SConrad Meyer   activeSegment.begin = begin;
1540f743729SConrad Meyer   activeSegment.end = begin;
1550f743729SConrad Meyer   activeSegment.score = 0;
1560f743729SConrad Meyer 
1570f743729SConrad Meyer   /* Slide the activeSegment through the whole epoch.
1580f743729SConrad Meyer    * Save the best segment in bestSegment.
1590f743729SConrad Meyer    */
1600f743729SConrad Meyer   while (activeSegment.end < end) {
1610f743729SConrad Meyer     /* Get hash value of current dmer */
162a0483764SConrad Meyer     const size_t idx = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.end, f, d);
1630f743729SConrad Meyer 
164*2b9c00cbSConrad Meyer     /* Add frequency of this index to score if this is the first occurrence of index in active segment */
165a0483764SConrad Meyer     if (segmentFreqs[idx] == 0) {
166a0483764SConrad Meyer       activeSegment.score += freqs[idx];
1670f743729SConrad Meyer     }
1680f743729SConrad Meyer     /* Increment end of segment and segmentFreqs*/
1690f743729SConrad Meyer     activeSegment.end += 1;
170a0483764SConrad Meyer     segmentFreqs[idx] += 1;
1710f743729SConrad Meyer     /* If the window is now too large, drop the first position */
1720f743729SConrad Meyer     if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
1730f743729SConrad Meyer       /* Get hash value of the dmer to be eliminated from active segment */
1740f743729SConrad Meyer       const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
1750f743729SConrad Meyer       segmentFreqs[delIndex] -= 1;
1760f743729SConrad Meyer       /* Subtract frequency of this index from score if this is the last occurrence of this index in active segment */
1770f743729SConrad Meyer       if (segmentFreqs[delIndex] == 0) {
1780f743729SConrad Meyer         activeSegment.score -= freqs[delIndex];
1790f743729SConrad Meyer       }
1800f743729SConrad Meyer       /* Increment start of segment */
1810f743729SConrad Meyer       activeSegment.begin += 1;
1820f743729SConrad Meyer     }
1830f743729SConrad Meyer 
1840f743729SConrad Meyer     /* If this segment is the best so far save it */
1850f743729SConrad Meyer     if (activeSegment.score > bestSegment.score) {
1860f743729SConrad Meyer       bestSegment = activeSegment;
1870f743729SConrad Meyer     }
1880f743729SConrad Meyer   }
1890f743729SConrad Meyer 
1900f743729SConrad Meyer   /* Zero out rest of segmentFreqs array */
1910f743729SConrad Meyer   while (activeSegment.begin < end) {
1920f743729SConrad Meyer     const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
1930f743729SConrad Meyer     segmentFreqs[delIndex] -= 1;
1940f743729SConrad Meyer     activeSegment.begin += 1;
1950f743729SConrad Meyer   }
1960f743729SConrad Meyer 
1970f743729SConrad Meyer   {
1980f743729SConrad Meyer     /*  Zero the frequency of hash value of each dmer covered by the chosen segment. */
1990f743729SConrad Meyer     U32 pos;
2000f743729SConrad Meyer     for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
2010f743729SConrad Meyer       const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, f, d);
2020f743729SConrad Meyer       freqs[i] = 0;
2030f743729SConrad Meyer     }
2040f743729SConrad Meyer   }
2050f743729SConrad Meyer 
2060f743729SConrad Meyer   return bestSegment;
2070f743729SConrad Meyer }
2080f743729SConrad Meyer 
2090f743729SConrad Meyer 
2100f743729SConrad Meyer static int FASTCOVER_checkParameters(ZDICT_cover_params_t parameters,
2110f743729SConrad Meyer                                      size_t maxDictSize, unsigned f,
2120f743729SConrad Meyer                                      unsigned accel) {
2130f743729SConrad Meyer   /* k, d, and f are required parameters */
2140f743729SConrad Meyer   if (parameters.d == 0 || parameters.k == 0) {
2150f743729SConrad Meyer     return 0;
2160f743729SConrad Meyer   }
2170f743729SConrad Meyer   /* d has to be 6 or 8 */
2180f743729SConrad Meyer   if (parameters.d != 6 && parameters.d != 8) {
2190f743729SConrad Meyer     return 0;
2200f743729SConrad Meyer   }
2210f743729SConrad Meyer   /* k <= maxDictSize */
2220f743729SConrad Meyer   if (parameters.k > maxDictSize) {
2230f743729SConrad Meyer     return 0;
2240f743729SConrad Meyer   }
2250f743729SConrad Meyer   /* d <= k */
2260f743729SConrad Meyer   if (parameters.d > parameters.k) {
2270f743729SConrad Meyer     return 0;
2280f743729SConrad Meyer   }
2290f743729SConrad Meyer   /* 0 < f <= FASTCOVER_MAX_F*/
2300f743729SConrad Meyer   if (f > FASTCOVER_MAX_F || f == 0) {
2310f743729SConrad Meyer     return 0;
2320f743729SConrad Meyer   }
2330f743729SConrad Meyer   /* 0 < splitPoint <= 1 */
2340f743729SConrad Meyer   if (parameters.splitPoint <= 0 || parameters.splitPoint > 1) {
2350f743729SConrad Meyer     return 0;
2360f743729SConrad Meyer   }
2370f743729SConrad Meyer   /* 0 < accel <= 10 */
2380f743729SConrad Meyer   if (accel > 10 || accel == 0) {
2390f743729SConrad Meyer     return 0;
2400f743729SConrad Meyer   }
2410f743729SConrad Meyer   return 1;
2420f743729SConrad Meyer }
2430f743729SConrad Meyer 
2440f743729SConrad Meyer 
2450f743729SConrad Meyer /**
2460f743729SConrad Meyer  * Clean up a context initialized with `FASTCOVER_ctx_init()`.
2470f743729SConrad Meyer  */
2480f743729SConrad Meyer static void
2490f743729SConrad Meyer FASTCOVER_ctx_destroy(FASTCOVER_ctx_t* ctx)
2500f743729SConrad Meyer {
2510f743729SConrad Meyer     if (!ctx) return;
2520f743729SConrad Meyer 
2530f743729SConrad Meyer     free(ctx->freqs);
2540f743729SConrad Meyer     ctx->freqs = NULL;
2550f743729SConrad Meyer 
2560f743729SConrad Meyer     free(ctx->offsets);
2570f743729SConrad Meyer     ctx->offsets = NULL;
2580f743729SConrad Meyer }
2590f743729SConrad Meyer 
2600f743729SConrad Meyer 
2610f743729SConrad Meyer /**
2620f743729SConrad Meyer  * Calculate for frequency of hash value of each dmer in ctx->samples
2630f743729SConrad Meyer  */
2640f743729SConrad Meyer static void
2650f743729SConrad Meyer FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
2660f743729SConrad Meyer {
2670f743729SConrad Meyer     const unsigned f = ctx->f;
2680f743729SConrad Meyer     const unsigned d = ctx->d;
2690f743729SConrad Meyer     const unsigned skip = ctx->accelParams.skip;
2700f743729SConrad Meyer     const unsigned readLength = MAX(d, 8);
2710f743729SConrad Meyer     size_t i;
2720f743729SConrad Meyer     assert(ctx->nbTrainSamples >= 5);
2730f743729SConrad Meyer     assert(ctx->nbTrainSamples <= ctx->nbSamples);
2740f743729SConrad Meyer     for (i = 0; i < ctx->nbTrainSamples; i++) {
2750f743729SConrad Meyer         size_t start = ctx->offsets[i];  /* start of current dmer */
2760f743729SConrad Meyer         size_t const currSampleEnd = ctx->offsets[i+1];
2770f743729SConrad Meyer         while (start + readLength <= currSampleEnd) {
2780f743729SConrad Meyer             const size_t dmerIndex = FASTCOVER_hashPtrToIndex(ctx->samples + start, f, d);
2790f743729SConrad Meyer             freqs[dmerIndex]++;
2800f743729SConrad Meyer             start = start + skip + 1;
2810f743729SConrad Meyer         }
2820f743729SConrad Meyer     }
2830f743729SConrad Meyer }
2840f743729SConrad Meyer 
2850f743729SConrad Meyer 
2860f743729SConrad Meyer /**
2870f743729SConrad Meyer  * Prepare a context for dictionary building.
2880f743729SConrad Meyer  * The context is only dependent on the parameter `d` and can used multiple
2890f743729SConrad Meyer  * times.
2900f743729SConrad Meyer  * Returns 1 on success or zero on error.
2910f743729SConrad Meyer  * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
2920f743729SConrad Meyer  */
2930f743729SConrad Meyer static int
2940f743729SConrad Meyer FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
2950f743729SConrad Meyer                    const void* samplesBuffer,
2960f743729SConrad Meyer                    const size_t* samplesSizes, unsigned nbSamples,
2970f743729SConrad Meyer                    unsigned d, double splitPoint, unsigned f,
2980f743729SConrad Meyer                    FASTCOVER_accel_t accelParams)
2990f743729SConrad Meyer {
3000f743729SConrad Meyer     const BYTE* const samples = (const BYTE*)samplesBuffer;
3010f743729SConrad Meyer     const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
3020f743729SConrad Meyer     /* Split samples into testing and training sets */
3030f743729SConrad Meyer     const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
3040f743729SConrad Meyer     const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
3050f743729SConrad Meyer     const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
3060f743729SConrad Meyer     const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
3070f743729SConrad Meyer 
3080f743729SConrad Meyer     /* Checks */
3090f743729SConrad Meyer     if (totalSamplesSize < MAX(d, sizeof(U64)) ||
3100f743729SConrad Meyer         totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
3110f743729SConrad Meyer         DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
312a0483764SConrad Meyer                     (unsigned)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
3130f743729SConrad Meyer         return 0;
3140f743729SConrad Meyer     }
3150f743729SConrad Meyer 
3160f743729SConrad Meyer     /* Check if there are at least 5 training samples */
3170f743729SConrad Meyer     if (nbTrainSamples < 5) {
3180f743729SConrad Meyer         DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
3190f743729SConrad Meyer         return 0;
3200f743729SConrad Meyer     }
3210f743729SConrad Meyer 
3220f743729SConrad Meyer     /* Check if there's testing sample */
3230f743729SConrad Meyer     if (nbTestSamples < 1) {
3240f743729SConrad Meyer         DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
3250f743729SConrad Meyer         return 0;
3260f743729SConrad Meyer     }
3270f743729SConrad Meyer 
3280f743729SConrad Meyer     /* Zero the context */
3290f743729SConrad Meyer     memset(ctx, 0, sizeof(*ctx));
3300f743729SConrad Meyer     DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
331a0483764SConrad Meyer                     (unsigned)trainingSamplesSize);
3320f743729SConrad Meyer     DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
333a0483764SConrad Meyer                     (unsigned)testSamplesSize);
3340f743729SConrad Meyer 
3350f743729SConrad Meyer     ctx->samples = samples;
3360f743729SConrad Meyer     ctx->samplesSizes = samplesSizes;
3370f743729SConrad Meyer     ctx->nbSamples = nbSamples;
3380f743729SConrad Meyer     ctx->nbTrainSamples = nbTrainSamples;
3390f743729SConrad Meyer     ctx->nbTestSamples = nbTestSamples;
3400f743729SConrad Meyer     ctx->nbDmers = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
3410f743729SConrad Meyer     ctx->d = d;
3420f743729SConrad Meyer     ctx->f = f;
3430f743729SConrad Meyer     ctx->accelParams = accelParams;
3440f743729SConrad Meyer 
3450f743729SConrad Meyer     /* The offsets of each file */
3460f743729SConrad Meyer     ctx->offsets = (size_t*)calloc((nbSamples + 1), sizeof(size_t));
3470f743729SConrad Meyer     if (ctx->offsets == NULL) {
3480f743729SConrad Meyer         DISPLAYLEVEL(1, "Failed to allocate scratch buffers \n");
3490f743729SConrad Meyer         FASTCOVER_ctx_destroy(ctx);
3500f743729SConrad Meyer         return 0;
3510f743729SConrad Meyer     }
3520f743729SConrad Meyer 
3530f743729SConrad Meyer     /* Fill offsets from the samplesSizes */
3540f743729SConrad Meyer     {   U32 i;
3550f743729SConrad Meyer         ctx->offsets[0] = 0;
3560f743729SConrad Meyer         assert(nbSamples >= 5);
3570f743729SConrad Meyer         for (i = 1; i <= nbSamples; ++i) {
3580f743729SConrad Meyer             ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
3590f743729SConrad Meyer         }
3600f743729SConrad Meyer     }
3610f743729SConrad Meyer 
3620f743729SConrad Meyer     /* Initialize frequency array of size 2^f */
3630f743729SConrad Meyer     ctx->freqs = (U32*)calloc(((U64)1 << f), sizeof(U32));
3640f743729SConrad Meyer     if (ctx->freqs == NULL) {
3650f743729SConrad Meyer         DISPLAYLEVEL(1, "Failed to allocate frequency table \n");
3660f743729SConrad Meyer         FASTCOVER_ctx_destroy(ctx);
3670f743729SConrad Meyer         return 0;
3680f743729SConrad Meyer     }
3690f743729SConrad Meyer 
3700f743729SConrad Meyer     DISPLAYLEVEL(2, "Computing frequencies\n");
3710f743729SConrad Meyer     FASTCOVER_computeFrequency(ctx->freqs, ctx);
3720f743729SConrad Meyer 
3730f743729SConrad Meyer     return 1;
3740f743729SConrad Meyer }
3750f743729SConrad Meyer 
3760f743729SConrad Meyer 
3770f743729SConrad Meyer /**
3780f743729SConrad Meyer  * Given the prepared context build the dictionary.
3790f743729SConrad Meyer  */
3800f743729SConrad Meyer static size_t
3810f743729SConrad Meyer FASTCOVER_buildDictionary(const FASTCOVER_ctx_t* ctx,
3820f743729SConrad Meyer                           U32* freqs,
3830f743729SConrad Meyer                           void* dictBuffer, size_t dictBufferCapacity,
3840f743729SConrad Meyer                           ZDICT_cover_params_t parameters,
3850f743729SConrad Meyer                           U16* segmentFreqs)
3860f743729SConrad Meyer {
3870f743729SConrad Meyer   BYTE *const dict = (BYTE *)dictBuffer;
3880f743729SConrad Meyer   size_t tail = dictBufferCapacity;
389*2b9c00cbSConrad Meyer   /* Divide the data into epochs. We will select one segment from each epoch. */
390*2b9c00cbSConrad Meyer   const COVER_epoch_info_t epochs = COVER_computeEpochs(
391*2b9c00cbSConrad Meyer       (U32)dictBufferCapacity, (U32)ctx->nbDmers, parameters.k, 1);
392*2b9c00cbSConrad Meyer   const size_t maxZeroScoreRun = 10;
393*2b9c00cbSConrad Meyer   size_t zeroScoreRun = 0;
3940f743729SConrad Meyer   size_t epoch;
395a0483764SConrad Meyer   DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n",
396*2b9c00cbSConrad Meyer                 (U32)epochs.num, (U32)epochs.size);
3970f743729SConrad Meyer   /* Loop through the epochs until there are no more segments or the dictionary
3980f743729SConrad Meyer    * is full.
3990f743729SConrad Meyer    */
400*2b9c00cbSConrad Meyer   for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs.num) {
401*2b9c00cbSConrad Meyer     const U32 epochBegin = (U32)(epoch * epochs.size);
402*2b9c00cbSConrad Meyer     const U32 epochEnd = epochBegin + epochs.size;
4030f743729SConrad Meyer     size_t segmentSize;
4040f743729SConrad Meyer     /* Select a segment */
4050f743729SConrad Meyer     COVER_segment_t segment = FASTCOVER_selectSegment(
4060f743729SConrad Meyer         ctx, freqs, epochBegin, epochEnd, parameters, segmentFreqs);
4070f743729SConrad Meyer 
408*2b9c00cbSConrad Meyer     /* If the segment covers no dmers, then we are out of content.
409*2b9c00cbSConrad Meyer      * There may be new content in other epochs, for continue for some time.
410*2b9c00cbSConrad Meyer      */
4110f743729SConrad Meyer     if (segment.score == 0) {
412*2b9c00cbSConrad Meyer       if (++zeroScoreRun >= maxZeroScoreRun) {
4130f743729SConrad Meyer           break;
4140f743729SConrad Meyer       }
415*2b9c00cbSConrad Meyer       continue;
416*2b9c00cbSConrad Meyer     }
417*2b9c00cbSConrad Meyer     zeroScoreRun = 0;
4180f743729SConrad Meyer 
4190f743729SConrad Meyer     /* Trim the segment if necessary and if it is too small then we are done */
4200f743729SConrad Meyer     segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
4210f743729SConrad Meyer     if (segmentSize < parameters.d) {
4220f743729SConrad Meyer       break;
4230f743729SConrad Meyer     }
4240f743729SConrad Meyer 
4250f743729SConrad Meyer     /* We fill the dictionary from the back to allow the best segments to be
4260f743729SConrad Meyer      * referenced with the smallest offsets.
4270f743729SConrad Meyer      */
4280f743729SConrad Meyer     tail -= segmentSize;
4290f743729SConrad Meyer     memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
4300f743729SConrad Meyer     DISPLAYUPDATE(
4310f743729SConrad Meyer         2, "\r%u%%       ",
432a0483764SConrad Meyer         (unsigned)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
4330f743729SConrad Meyer   }
4340f743729SConrad Meyer   DISPLAYLEVEL(2, "\r%79s\r", "");
4350f743729SConrad Meyer   return tail;
4360f743729SConrad Meyer }
4370f743729SConrad Meyer 
4380f743729SConrad Meyer 
4390f743729SConrad Meyer /**
4400f743729SConrad Meyer  * Parameters for FASTCOVER_tryParameters().
4410f743729SConrad Meyer  */
4420f743729SConrad Meyer typedef struct FASTCOVER_tryParameters_data_s {
4430f743729SConrad Meyer     const FASTCOVER_ctx_t* ctx;
4440f743729SConrad Meyer     COVER_best_t* best;
4450f743729SConrad Meyer     size_t dictBufferCapacity;
4460f743729SConrad Meyer     ZDICT_cover_params_t parameters;
4470f743729SConrad Meyer } FASTCOVER_tryParameters_data_t;
4480f743729SConrad Meyer 
4490f743729SConrad Meyer 
4500f743729SConrad Meyer /**
4510f743729SConrad Meyer  * Tries a set of parameters and updates the COVER_best_t with the results.
4520f743729SConrad Meyer  * This function is thread safe if zstd is compiled with multithreaded support.
4530f743729SConrad Meyer  * It takes its parameters as an *OWNING* opaque pointer to support threading.
4540f743729SConrad Meyer  */
4550f743729SConrad Meyer static void FASTCOVER_tryParameters(void *opaque)
4560f743729SConrad Meyer {
4570f743729SConrad Meyer   /* Save parameters as local variables */
4580f743729SConrad Meyer   FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t *)opaque;
4590f743729SConrad Meyer   const FASTCOVER_ctx_t *const ctx = data->ctx;
4600f743729SConrad Meyer   const ZDICT_cover_params_t parameters = data->parameters;
4610f743729SConrad Meyer   size_t dictBufferCapacity = data->dictBufferCapacity;
4620f743729SConrad Meyer   size_t totalCompressedSize = ERROR(GENERIC);
4630f743729SConrad Meyer   /* Initialize array to keep track of frequency of dmer within activeSegment */
4640f743729SConrad Meyer   U16* segmentFreqs = (U16 *)calloc(((U64)1 << ctx->f), sizeof(U16));
4650f743729SConrad Meyer   /* Allocate space for hash table, dict, and freqs */
4660f743729SConrad Meyer   BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
4670f743729SConrad Meyer   U32 *freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
4680f743729SConrad Meyer   if (!segmentFreqs || !dict || !freqs) {
4690f743729SConrad Meyer     DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
4700f743729SConrad Meyer     goto _cleanup;
4710f743729SConrad Meyer   }
4720f743729SConrad Meyer   /* Copy the frequencies because we need to modify them */
4730f743729SConrad Meyer   memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
4740f743729SConrad Meyer   /* Build the dictionary */
4750f743729SConrad Meyer   { const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
4760f743729SConrad Meyer                                                   parameters, segmentFreqs);
4770f743729SConrad Meyer     const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
4780f743729SConrad Meyer     dictBufferCapacity = ZDICT_finalizeDictionary(
4790f743729SConrad Meyer         dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
4800f743729SConrad Meyer         ctx->samples, ctx->samplesSizes, nbFinalizeSamples, parameters.zParams);
4810f743729SConrad Meyer     if (ZDICT_isError(dictBufferCapacity)) {
4820f743729SConrad Meyer       DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
4830f743729SConrad Meyer       goto _cleanup;
4840f743729SConrad Meyer     }
4850f743729SConrad Meyer   }
4860f743729SConrad Meyer   /* Check total compressed size */
4870f743729SConrad Meyer   totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
4880f743729SConrad Meyer                                                        ctx->samples, ctx->offsets,
4890f743729SConrad Meyer                                                        ctx->nbTrainSamples, ctx->nbSamples,
4900f743729SConrad Meyer                                                        dict, dictBufferCapacity);
4910f743729SConrad Meyer _cleanup:
4920f743729SConrad Meyer   COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
4930f743729SConrad Meyer                     dictBufferCapacity);
4940f743729SConrad Meyer   free(data);
4950f743729SConrad Meyer   free(segmentFreqs);
4960f743729SConrad Meyer   free(dict);
4970f743729SConrad Meyer   free(freqs);
4980f743729SConrad Meyer }
4990f743729SConrad Meyer 
5000f743729SConrad Meyer 
5010f743729SConrad Meyer static void
5020f743729SConrad Meyer FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
5030f743729SConrad Meyer                                ZDICT_cover_params_t* coverParams)
5040f743729SConrad Meyer {
5050f743729SConrad Meyer     coverParams->k = fastCoverParams.k;
5060f743729SConrad Meyer     coverParams->d = fastCoverParams.d;
5070f743729SConrad Meyer     coverParams->steps = fastCoverParams.steps;
5080f743729SConrad Meyer     coverParams->nbThreads = fastCoverParams.nbThreads;
5090f743729SConrad Meyer     coverParams->splitPoint = fastCoverParams.splitPoint;
5100f743729SConrad Meyer     coverParams->zParams = fastCoverParams.zParams;
5110f743729SConrad Meyer }
5120f743729SConrad Meyer 
5130f743729SConrad Meyer 
5140f743729SConrad Meyer static void
5150f743729SConrad Meyer FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
5160f743729SConrad Meyer                                    ZDICT_fastCover_params_t* fastCoverParams,
5170f743729SConrad Meyer                                    unsigned f, unsigned accel)
5180f743729SConrad Meyer {
5190f743729SConrad Meyer     fastCoverParams->k = coverParams.k;
5200f743729SConrad Meyer     fastCoverParams->d = coverParams.d;
5210f743729SConrad Meyer     fastCoverParams->steps = coverParams.steps;
5220f743729SConrad Meyer     fastCoverParams->nbThreads = coverParams.nbThreads;
5230f743729SConrad Meyer     fastCoverParams->splitPoint = coverParams.splitPoint;
5240f743729SConrad Meyer     fastCoverParams->f = f;
5250f743729SConrad Meyer     fastCoverParams->accel = accel;
5260f743729SConrad Meyer     fastCoverParams->zParams = coverParams.zParams;
5270f743729SConrad Meyer }
5280f743729SConrad Meyer 
5290f743729SConrad Meyer 
5300f743729SConrad Meyer ZDICTLIB_API size_t
5310f743729SConrad Meyer ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
5320f743729SConrad Meyer                                 const void* samplesBuffer,
5330f743729SConrad Meyer                                 const size_t* samplesSizes, unsigned nbSamples,
5340f743729SConrad Meyer                                 ZDICT_fastCover_params_t parameters)
5350f743729SConrad Meyer {
5360f743729SConrad Meyer     BYTE* const dict = (BYTE*)dictBuffer;
5370f743729SConrad Meyer     FASTCOVER_ctx_t ctx;
5380f743729SConrad Meyer     ZDICT_cover_params_t coverParams;
5390f743729SConrad Meyer     FASTCOVER_accel_t accelParams;
5400f743729SConrad Meyer     /* Initialize global data */
5410f743729SConrad Meyer     g_displayLevel = parameters.zParams.notificationLevel;
5420f743729SConrad Meyer     /* Assign splitPoint and f if not provided */
5430f743729SConrad Meyer     parameters.splitPoint = 1.0;
5440f743729SConrad Meyer     parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
5450f743729SConrad Meyer     parameters.accel = parameters.accel == 0 ? DEFAULT_ACCEL : parameters.accel;
5460f743729SConrad Meyer     /* Convert to cover parameter */
5470f743729SConrad Meyer     memset(&coverParams, 0 , sizeof(coverParams));
5480f743729SConrad Meyer     FASTCOVER_convertToCoverParams(parameters, &coverParams);
5490f743729SConrad Meyer     /* Checks */
5500f743729SConrad Meyer     if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
5510f743729SConrad Meyer                                    parameters.accel)) {
5520f743729SConrad Meyer       DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
5530f743729SConrad Meyer       return ERROR(GENERIC);
5540f743729SConrad Meyer     }
5550f743729SConrad Meyer     if (nbSamples == 0) {
5560f743729SConrad Meyer       DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
5570f743729SConrad Meyer       return ERROR(GENERIC);
5580f743729SConrad Meyer     }
5590f743729SConrad Meyer     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
5600f743729SConrad Meyer       DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
5610f743729SConrad Meyer                    ZDICT_DICTSIZE_MIN);
5620f743729SConrad Meyer       return ERROR(dstSize_tooSmall);
5630f743729SConrad Meyer     }
5640f743729SConrad Meyer     /* Assign corresponding FASTCOVER_accel_t to accelParams*/
5650f743729SConrad Meyer     accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
5660f743729SConrad Meyer     /* Initialize context */
5670f743729SConrad Meyer     if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
5680f743729SConrad Meyer                             coverParams.d, parameters.splitPoint, parameters.f,
5690f743729SConrad Meyer                             accelParams)) {
5700f743729SConrad Meyer       DISPLAYLEVEL(1, "Failed to initialize context\n");
5710f743729SConrad Meyer       return ERROR(GENERIC);
5720f743729SConrad Meyer     }
573*2b9c00cbSConrad Meyer     COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, g_displayLevel);
5740f743729SConrad Meyer     /* Build the dictionary */
5750f743729SConrad Meyer     DISPLAYLEVEL(2, "Building dictionary\n");
5760f743729SConrad Meyer     {
5770f743729SConrad Meyer       /* Initialize array to keep track of frequency of dmer within activeSegment */
5780f743729SConrad Meyer       U16* segmentFreqs = (U16 *)calloc(((U64)1 << parameters.f), sizeof(U16));
5790f743729SConrad Meyer       const size_t tail = FASTCOVER_buildDictionary(&ctx, ctx.freqs, dictBuffer,
5800f743729SConrad Meyer                                                 dictBufferCapacity, coverParams, segmentFreqs);
5810f743729SConrad Meyer       const unsigned nbFinalizeSamples = (unsigned)(ctx.nbTrainSamples * ctx.accelParams.finalize / 100);
5820f743729SConrad Meyer       const size_t dictionarySize = ZDICT_finalizeDictionary(
5830f743729SConrad Meyer           dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
5840f743729SConrad Meyer           samplesBuffer, samplesSizes, nbFinalizeSamples, coverParams.zParams);
5850f743729SConrad Meyer       if (!ZSTD_isError(dictionarySize)) {
5860f743729SConrad Meyer           DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
587a0483764SConrad Meyer                       (unsigned)dictionarySize);
5880f743729SConrad Meyer       }
5890f743729SConrad Meyer       FASTCOVER_ctx_destroy(&ctx);
5900f743729SConrad Meyer       free(segmentFreqs);
5910f743729SConrad Meyer       return dictionarySize;
5920f743729SConrad Meyer     }
5930f743729SConrad Meyer }
5940f743729SConrad Meyer 
5950f743729SConrad Meyer 
5960f743729SConrad Meyer ZDICTLIB_API size_t
5970f743729SConrad Meyer ZDICT_optimizeTrainFromBuffer_fastCover(
5980f743729SConrad Meyer                     void* dictBuffer, size_t dictBufferCapacity,
5990f743729SConrad Meyer                     const void* samplesBuffer,
6000f743729SConrad Meyer                     const size_t* samplesSizes, unsigned nbSamples,
6010f743729SConrad Meyer                     ZDICT_fastCover_params_t* parameters)
6020f743729SConrad Meyer {
6030f743729SConrad Meyer     ZDICT_cover_params_t coverParams;
6040f743729SConrad Meyer     FASTCOVER_accel_t accelParams;
6050f743729SConrad Meyer     /* constants */
6060f743729SConrad Meyer     const unsigned nbThreads = parameters->nbThreads;
6070f743729SConrad Meyer     const double splitPoint =
6080f743729SConrad Meyer         parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
6090f743729SConrad Meyer     const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
6100f743729SConrad Meyer     const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
6110f743729SConrad Meyer     const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
6120f743729SConrad Meyer     const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
6130f743729SConrad Meyer     const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
6140f743729SConrad Meyer     const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
6150f743729SConrad Meyer     const unsigned kIterations =
6160f743729SConrad Meyer         (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
6170f743729SConrad Meyer     const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
6180f743729SConrad Meyer     const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
6190f743729SConrad Meyer     /* Local variables */
6200f743729SConrad Meyer     const int displayLevel = parameters->zParams.notificationLevel;
6210f743729SConrad Meyer     unsigned iteration = 1;
6220f743729SConrad Meyer     unsigned d;
6230f743729SConrad Meyer     unsigned k;
6240f743729SConrad Meyer     COVER_best_t best;
6250f743729SConrad Meyer     POOL_ctx *pool = NULL;
626*2b9c00cbSConrad Meyer     int warned = 0;
6270f743729SConrad Meyer     /* Checks */
6280f743729SConrad Meyer     if (splitPoint <= 0 || splitPoint > 1) {
6290f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect splitPoint\n");
6300f743729SConrad Meyer       return ERROR(GENERIC);
6310f743729SConrad Meyer     }
6320f743729SConrad Meyer     if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
6330f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect accel\n");
6340f743729SConrad Meyer       return ERROR(GENERIC);
6350f743729SConrad Meyer     }
6360f743729SConrad Meyer     if (kMinK < kMaxD || kMaxK < kMinK) {
6370f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect k\n");
6380f743729SConrad Meyer       return ERROR(GENERIC);
6390f743729SConrad Meyer     }
6400f743729SConrad Meyer     if (nbSamples == 0) {
6410f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "FASTCOVER must have at least one input file\n");
6420f743729SConrad Meyer       return ERROR(GENERIC);
6430f743729SConrad Meyer     }
6440f743729SConrad Meyer     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
6450f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 1, "dictBufferCapacity must be at least %u\n",
6460f743729SConrad Meyer                    ZDICT_DICTSIZE_MIN);
6470f743729SConrad Meyer       return ERROR(dstSize_tooSmall);
6480f743729SConrad Meyer     }
6490f743729SConrad Meyer     if (nbThreads > 1) {
6500f743729SConrad Meyer       pool = POOL_create(nbThreads, 1);
6510f743729SConrad Meyer       if (!pool) {
6520f743729SConrad Meyer         return ERROR(memory_allocation);
6530f743729SConrad Meyer       }
6540f743729SConrad Meyer     }
6550f743729SConrad Meyer     /* Initialization */
6560f743729SConrad Meyer     COVER_best_init(&best);
6570f743729SConrad Meyer     memset(&coverParams, 0 , sizeof(coverParams));
6580f743729SConrad Meyer     FASTCOVER_convertToCoverParams(*parameters, &coverParams);
6590f743729SConrad Meyer     accelParams = FASTCOVER_defaultAccelParameters[accel];
6600f743729SConrad Meyer     /* Turn down global display level to clean up display at level 2 and below */
6610f743729SConrad Meyer     g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
6620f743729SConrad Meyer     /* Loop through d first because each new value needs a new context */
6630f743729SConrad Meyer     LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
6640f743729SConrad Meyer                       kIterations);
6650f743729SConrad Meyer     for (d = kMinD; d <= kMaxD; d += 2) {
6660f743729SConrad Meyer       /* Initialize the context for this value of d */
6670f743729SConrad Meyer       FASTCOVER_ctx_t ctx;
6680f743729SConrad Meyer       LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
6690f743729SConrad Meyer       if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams)) {
6700f743729SConrad Meyer         LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
6710f743729SConrad Meyer         COVER_best_destroy(&best);
6720f743729SConrad Meyer         POOL_free(pool);
6730f743729SConrad Meyer         return ERROR(GENERIC);
6740f743729SConrad Meyer       }
675*2b9c00cbSConrad Meyer       if (!warned) {
676*2b9c00cbSConrad Meyer         COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, displayLevel);
677*2b9c00cbSConrad Meyer         warned = 1;
678*2b9c00cbSConrad Meyer       }
6790f743729SConrad Meyer       /* Loop through k reusing the same context */
6800f743729SConrad Meyer       for (k = kMinK; k <= kMaxK; k += kStepSize) {
6810f743729SConrad Meyer         /* Prepare the arguments */
6820f743729SConrad Meyer         FASTCOVER_tryParameters_data_t *data = (FASTCOVER_tryParameters_data_t *)malloc(
6830f743729SConrad Meyer             sizeof(FASTCOVER_tryParameters_data_t));
6840f743729SConrad Meyer         LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
6850f743729SConrad Meyer         if (!data) {
6860f743729SConrad Meyer           LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
6870f743729SConrad Meyer           COVER_best_destroy(&best);
6880f743729SConrad Meyer           FASTCOVER_ctx_destroy(&ctx);
6890f743729SConrad Meyer           POOL_free(pool);
6900f743729SConrad Meyer           return ERROR(GENERIC);
6910f743729SConrad Meyer         }
6920f743729SConrad Meyer         data->ctx = &ctx;
6930f743729SConrad Meyer         data->best = &best;
6940f743729SConrad Meyer         data->dictBufferCapacity = dictBufferCapacity;
6950f743729SConrad Meyer         data->parameters = coverParams;
6960f743729SConrad Meyer         data->parameters.k = k;
6970f743729SConrad Meyer         data->parameters.d = d;
6980f743729SConrad Meyer         data->parameters.splitPoint = splitPoint;
6990f743729SConrad Meyer         data->parameters.steps = kSteps;
7000f743729SConrad Meyer         data->parameters.zParams.notificationLevel = g_displayLevel;
7010f743729SConrad Meyer         /* Check the parameters */
7020f743729SConrad Meyer         if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
7030f743729SConrad Meyer                                        data->ctx->f, accel)) {
7040f743729SConrad Meyer           DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
7050f743729SConrad Meyer           free(data);
7060f743729SConrad Meyer           continue;
7070f743729SConrad Meyer         }
7080f743729SConrad Meyer         /* Call the function and pass ownership of data to it */
7090f743729SConrad Meyer         COVER_best_start(&best);
7100f743729SConrad Meyer         if (pool) {
7110f743729SConrad Meyer           POOL_add(pool, &FASTCOVER_tryParameters, data);
7120f743729SConrad Meyer         } else {
7130f743729SConrad Meyer           FASTCOVER_tryParameters(data);
7140f743729SConrad Meyer         }
7150f743729SConrad Meyer         /* Print status */
7160f743729SConrad Meyer         LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
717a0483764SConrad Meyer                            (unsigned)((iteration * 100) / kIterations));
7180f743729SConrad Meyer         ++iteration;
7190f743729SConrad Meyer       }
7200f743729SConrad Meyer       COVER_best_wait(&best);
7210f743729SConrad Meyer       FASTCOVER_ctx_destroy(&ctx);
7220f743729SConrad Meyer     }
7230f743729SConrad Meyer     LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
7240f743729SConrad Meyer     /* Fill the output buffer and parameters with output of the best parameters */
7250f743729SConrad Meyer     {
7260f743729SConrad Meyer       const size_t dictSize = best.dictSize;
7270f743729SConrad Meyer       if (ZSTD_isError(best.compressedSize)) {
7280f743729SConrad Meyer         const size_t compressedSize = best.compressedSize;
7290f743729SConrad Meyer         COVER_best_destroy(&best);
7300f743729SConrad Meyer         POOL_free(pool);
7310f743729SConrad Meyer         return compressedSize;
7320f743729SConrad Meyer       }
7330f743729SConrad Meyer       FASTCOVER_convertToFastCoverParams(best.parameters, parameters, f, accel);
7340f743729SConrad Meyer       memcpy(dictBuffer, best.dict, dictSize);
7350f743729SConrad Meyer       COVER_best_destroy(&best);
7360f743729SConrad Meyer       POOL_free(pool);
7370f743729SConrad Meyer       return dictSize;
7380f743729SConrad Meyer     }
7390f743729SConrad Meyer 
7400f743729SConrad Meyer }
741