zstd-ruby 1.4.4.0 → 1.4.5.0

data/ext/zstdruby/libzstd/compress/zstd_compress_superblock.h

@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPRESS_ADVANCED_H
+#define ZSTD_COMPRESS_ADVANCED_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+
+#include "../zstd.h" /* ZSTD_CCtx */
+
+/*-*************************************
+*  Target Compressed Block Size
+***************************************/
+
+/* ZSTD_compressSuperBlock() :
+ * Used to compress a super block when targetCBlockSize is being used.
+ * The given block will be compressed into multiple sub blocks that are around targetCBlockSize. */
+size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               void const* src, size_t srcSize,
+                               unsigned lastBlock);
+
+#endif /* ZSTD_COMPRESS_ADVANCED_H */

data/ext/zstdruby/libzstd/compress/zstd_cwksp.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,7 +14,7 @@
 /*-*************************************
 *  Dependencies
 ***************************************/
-#include "zstd_internal.h"
+#include "../common/zstd_internal.h"
 
 #if defined (__cplusplus)
 extern "C" {
@@ -24,16 +24,6 @@ extern "C" {
 *  Constants
 ***************************************/
 
-/* define "workspace is too large" as this number of times larger than needed */
-#define ZSTD_WORKSPACETOOLARGE_FACTOR 3
-
-/* when workspace is continuously too large
- * during at least this number of times,
- * context's memory usage is considered wasteful,
- * because it's sized to handle a worst case scenario which rarely happens.
- * In which case, resize it down to free some memory */
-#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128
-
 /* Since the workspace is effectively its own little malloc implementation /
  * arena, when we run under ASAN, we should similarly insert redzones between
  * each internal element of the workspace, so ASAN will catch overruns that
@@ -468,7 +458,7 @@ MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size) {
 MEM_STATIC size_t ZSTD_cwksp_create(ZSTD_cwksp* ws, size_t size, ZSTD_customMem customMem) {
     void* workspace = ZSTD_malloc(size, customMem);
     DEBUGLOG(4, "cwksp: creating new workspace with %zd bytes", size);
-    RETURN_ERROR_IF(workspace == NULL, memory_allocation);
+    RETURN_ERROR_IF(workspace == NULL, memory_allocation, "NULL pointer!");
     ZSTD_cwksp_init(ws, workspace, size);
     return 0;
 }

data/ext/zstdruby/libzstd/compress/zstd_double_fast.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -63,10 +63,8 @@ size_t ZSTD_compressBlock_doubleFast_generic(
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
-    const U32 lowestValid = ms->window.dictLimit;
-    const U32 maxDistance = 1U << cParams->windowLog;
     /* presumes that, if there is a dictionary, it must be using Attach mode */
-    const U32 prefixLowestIndex = (endIndex - lowestValid > maxDistance) ? endIndex - maxDistance : lowestValid;
+    const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
     const BYTE* const prefixLowest = base + prefixLowestIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
@@ -96,7 +94,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
                                      dictCParams->hashLog : hBitsL;
     const U32 dictHBitsS           = dictMode == ZSTD_dictMatchState ?
                                      dictCParams->chainLog : hBitsS;
-    const U32 dictAndPrefixLength  = (U32)(ip - prefixLowest + dictEnd - dictStart);
+    const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
 
     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_generic");
 
@@ -104,13 +102,15 @@ size_t ZSTD_compressBlock_doubleFast_generic(
 
     /* if a dictionary is attached, it must be within window range */
     if (dictMode == ZSTD_dictMatchState) {
-        assert(lowestValid + maxDistance >= endIndex);
+        assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
     }
 
     /* init */
     ip += (dictAndPrefixLength == 0);
     if (dictMode == ZSTD_noDict) {
-        U32 const maxRep = (U32)(ip - prefixLowest);
+        U32 const current = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
+        U32 const maxRep = current - windowLow;
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
@@ -198,6 +198,9 @@ size_t ZSTD_compressBlock_doubleFast_generic(
         }   }
 
         ip += ((ip-anchor) >> kSearchStrength) + 1;
+#if defined(__aarch64__)
+        PREFETCH_L1(ip+256);
+#endif
         continue;
 
 _search_next_long:
@@ -271,7 +274,7 @@ _match_stored:
                     U32 const repIndex2 = current2 - offset_2;
                     const BYTE* repMatch2 = dictMode == ZSTD_dictMatchState
                         && repIndex2 < prefixLowestIndex ?
-                            dictBase - dictIndexDelta + repIndex2 :
+                            dictBase + repIndex2 - dictIndexDelta :
                             base + repIndex2;
                     if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
                        && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {

data/ext/zstdruby/libzstd/compress/zstd_double_fast.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -15,7 +15,7 @@
 extern "C" {
 #endif
 
-#include "mem.h"      /* U32 */
+#include "../common/mem.h"      /* U32 */
 #include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
 
 void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,

data/ext/zstdruby/libzstd/compress/zstd_fast.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -61,9 +61,7 @@ ZSTD_compressBlock_fast_generic(
     const BYTE* ip1;
     const BYTE* anchor = istart;
     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
-    const U32   maxDistance = 1U << cParams->windowLog;
-    const U32   validStartIndex = ms->window.dictLimit;
-    const U32   prefixStartIndex = (endIndex - validStartIndex > maxDistance) ? endIndex - maxDistance : validStartIndex;
+    const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
     const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
@@ -74,12 +72,21 @@ ZSTD_compressBlock_fast_generic(
     DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
     ip0 += (ip0 == prefixStart);
     ip1 = ip0 + 1;
-    {   U32 const maxRep = (U32)(ip0 - prefixStart);
+    {   U32 const current = (U32)(ip0 - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
+        U32 const maxRep = current - windowLow;
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
 
     /* Main Search Loop */
+#ifdef __INTEL_COMPILER
+    /* From intel 'The vector pragma indicates that the loop should be
+     * vectorized if it is legal to do so'. Can be used together with
+     * #pragma ivdep (but have opted to exclude that because intel
+     * warns against using it).*/
+    #pragma vector always
+#endif
     while (ip1 < ilimit) {   /* < instead of <=, because check at ip0+2 */
         size_t mLength;
         BYTE const* ip2 = ip0 + 2;
@@ -91,19 +98,25 @@ ZSTD_compressBlock_fast_generic(
         U32 const current1 = (U32)(ip1-base);
         U32 const matchIndex0 = hashTable[h0];
         U32 const matchIndex1 = hashTable[h1];
-        BYTE const* repMatch = ip2-offset_1;
+        BYTE const* repMatch = ip2 - offset_1;
         const BYTE* match0 = base + matchIndex0;
         const BYTE* match1 = base + matchIndex1;
         U32 offcode;
+
+#if defined(__aarch64__)
+        PREFETCH_L1(ip0+256);
+#endif
+
         hashTable[h0] = current0;   /* update hash table */
         hashTable[h1] = current1;   /* update hash table */
 
         assert(ip0 + 1 == ip1);
 
         if ((offset_1 > 0) & (MEM_read32(repMatch) == MEM_read32(ip2))) {
-            mLength = ip2[-1] == repMatch[-1] ? 1 : 0;
+            mLength = (ip2[-1] == repMatch[-1]) ? 1 : 0;
             ip0 = ip2 - mLength;
             match0 = repMatch - mLength;
+            mLength += 4;
             offcode = 0;
             goto _match;
         }
@@ -128,19 +141,18 @@ _offset: /* Requires: ip0, match0 */
         offset_2 = offset_1;
         offset_1 = (U32)(ip0-match0);
         offcode = offset_1 + ZSTD_REP_MOVE;
-        mLength = 0;
+        mLength = 4;
         /* Count the backwards match length */
         while (((ip0>anchor) & (match0>prefixStart))
              && (ip0[-1] == match0[-1])) { ip0--; match0--; mLength++; } /* catch up */
 
 _match: /* Requires: ip0, match0, offcode */
         /* Count the forward length */
-        mLength += ZSTD_count(ip0+mLength+4, match0+mLength+4, iend) + 4;
+        mLength += ZSTD_count(ip0+mLength, match0+mLength, iend);
         ZSTD_storeSeq(seqStore, (size_t)(ip0-anchor), anchor, iend, offcode, mLength-MINMATCH);
         /* match found */
         ip0 += mLength;
         anchor = ip0;
-        ip1 = ip0 + 1;
 
         if (ip0 <= ilimit) {
             /* Fill Table */
@@ -148,19 +160,18 @@ _match: /* Requires: ip0, match0, offcode */
             hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
             hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
 
-            while ( ((ip0 <= ilimit) & (offset_2>0))  /* offset_2==0 means offset_2 is invalidated */
-                 && (MEM_read32(ip0) == MEM_read32(ip0 - offset_2)) ) {
-                /* store sequence */
-                size_t const rLength = ZSTD_count(ip0+4, ip0+4-offset_2, iend) + 4;
-                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
-                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
-                ip0 += rLength;
-                ip1 = ip0 + 1;
-                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, 0 /*offCode*/, rLength-MINMATCH);
-                anchor = ip0;
-                continue;   /* faster when present (confirmed on gcc-8) ... (?) */
-            }
-        }
+            if (offset_2 > 0) { /* offset_2==0 means offset_2 is invalidated */
+                while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - offset_2)) ) {
+                    /* store sequence */
+                    size_t const rLength = ZSTD_count(ip0+4, ip0+4-offset_2, iend) + 4;
+                    { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
+                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
+                    ip0 += rLength;
+                    ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, 0 /*offCode*/, rLength-MINMATCH);
+                    anchor = ip0;
+                    continue;   /* faster when present (confirmed on gcc-8) ... (?) */
+        }   }   }
+        ip1 = ip0 + 1;
     }
 
     /* save reps for next block */
@@ -387,7 +398,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     const BYTE* const ilimit = iend - 8;
     U32 offset_1=rep[0], offset_2=rep[1];
 
-    DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic");
+    DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);
 
     /* switch to "regular" variant if extDict is invalidated due to maxDistance */
     if (prefixStartIndex == dictStartIndex)
@@ -404,6 +415,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
         const BYTE* const repMatch = repBase + repIndex;
         hashTable[h] = current;   /* update hash table */
+        DEBUGLOG(7, "offset_1 = %u , current = %u", offset_1, current);
         assert(offset_1 <= current +1);   /* check repIndex */
 
         if ( (((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > dictStartIndex))

data/ext/zstdruby/libzstd/compress/zstd_fast.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -15,7 +15,7 @@
 extern "C" {
 #endif
 
-#include "mem.h"      /* U32 */
+#include "../common/mem.h"      /* U32 */
 #include "zstd_compress_internal.h"
 
 void ZSTD_fillHashTable(ZSTD_matchState_t* ms,

data/ext/zstdruby/libzstd/compress/zstd_lazy.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -660,12 +660,16 @@ ZSTD_compressBlock_lazy_generic(
     const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
                                      prefixLowestIndex - (U32)(dictEnd - dictBase) :
                                      0;
-    const U32 dictAndPrefixLength = (U32)(ip - prefixLowest + dictEnd - dictLowest);
+    const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u)", (U32)dictMode);
 
     /* init */
     ip += (dictAndPrefixLength == 0);
     if (dictMode == ZSTD_noDict) {
-        U32 const maxRep = (U32)(ip - prefixLowest);
+        U32 const current = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, ms->cParams.windowLog);
+        U32 const maxRep = current - windowLow;
         if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
     }
@@ -677,6 +681,12 @@ ZSTD_compressBlock_lazy_generic(
     }
 
     /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
     while (ip < ilimit) {
         size_t matchLength=0;
         size_t offset=0;
@@ -929,11 +939,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
     const BYTE* const ilimit = iend - 8;
     const BYTE* const base = ms->window.base;
     const U32 dictLimit = ms->window.dictLimit;
-    const U32 lowestIndex = ms->window.lowLimit;
     const BYTE* const prefixStart = base + dictLimit;
     const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const dictEnd  = dictBase + dictLimit;
-    const BYTE* const dictStart  = dictBase + lowestIndex;
+    const BYTE* const dictStart  = dictBase + ms->window.lowLimit;
+    const U32 windowLog = ms->cParams.windowLog;
 
     typedef size_t (*searchMax_f)(
                         ZSTD_matchState_t* ms,
@@ -942,10 +952,18 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
 
     U32 offset_1 = rep[0], offset_2 = rep[1];
 
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic");
+
     /* init */
     ip += (ip == prefixStart);
 
     /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
     while (ip < ilimit) {
         size_t matchLength=0;
         size_t offset=0;
@@ -953,10 +971,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
         U32 current = (U32)(ip-base);
 
         /* check repCode */
-        {   const U32 repIndex = (U32)(current+1 - offset_1);
+        {   const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current+1, windowLog);
+            const U32 repIndex = (U32)(current+1 - offset_1);
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
-            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))   /* intentional overflow */
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))   /* intentional overflow */
             if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -983,10 +1002,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
             current++;
             /* check repCode */
             if (offset) {
+                const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current, windowLog);
                 const U32 repIndex = (U32)(current - offset_1);
                 const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                 const BYTE* const repMatch = repBase + repIndex;
-                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
                 if (MEM_read32(ip) == MEM_read32(repMatch)) {
                     /* repcode detected */
                     const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -1013,10 +1033,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
                 current++;
                 /* check repCode */
                 if (offset) {
+                    const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current, windowLog);
                     const U32 repIndex = (U32)(current - offset_1);
                     const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                     const BYTE* const repMatch = repBase + repIndex;
-                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
                     if (MEM_read32(ip) == MEM_read32(repMatch)) {
                         /* repcode detected */
                         const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -1057,10 +1078,12 @@ _storeSequence:
 
         /* check immediate repcode */
         while (ip <= ilimit) {
-            const U32 repIndex = (U32)((ip-base) - offset_2);
+            const U32 repCurrent = (U32)(ip-base);
+            const U32 windowLow = ZSTD_getLowestMatchIndex(ms, repCurrent, windowLog);
+            const U32 repIndex = repCurrent - offset_2;
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
-            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
             if (MEM_read32(ip) == MEM_read32(repMatch)) {
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;

data/ext/zstdruby/libzstd/compress/zstd_lazy.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the

data/ext/zstdruby/libzstd/compress/zstd_ldm.c

@@ -1,15 +1,16 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #include "zstd_ldm.h"
 
-#include "debug.h"
+#include "../common/debug.h"
 #include "zstd_fast.h"          /* ZSTD_fillHashTable() */
 #include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
 
@@ -223,6 +224,20 @@ static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
     return rollingHash;
 }
 
+void ZSTD_ldm_fillHashTable(
+            ldmState_t* state, const BYTE* ip,
+            const BYTE* iend, ldmParams_t const* params)
+{
+    DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
+    if ((size_t)(iend - ip) >= params->minMatchLength) {
+        U64 startingHash = ZSTD_rollingHash_compute(ip, params->minMatchLength);
+        ZSTD_ldm_fillLdmHashTable(
+            state, startingHash, ip, iend - params->minMatchLength, state->window.base,
+            params->hashLog - params->bucketSizeLog,
+            *params);
+    }
+}
+
 
 /** ZSTD_ldm_limitTableUpdate() :
  *
@@ -449,6 +464,8 @@ size_t ZSTD_ldm_generateSequences(
             U32 const correction = ZSTD_window_correctOverflow(
                 &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
             ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
+            /* invalidate dictionaries on overflow correction */
+            ldmState->loadedDictEnd = 0;
         }
         /* 2. We enforce the maximum offset allowed.
          *
@@ -457,8 +474,14 @@ size_t ZSTD_ldm_generateSequences(
          * TODO: * Test the chunk size.
          *       * Try invalidation after the sequence generation and test the
          *         the offset against maxDist directly.
+         *
+         * NOTE: Because of dictionaries + sequence splitting we MUST make sure
+         * that any offset used is valid at the END of the sequence, since it may
+         * be split into two sequences. This condition holds when using
+         * ZSTD_window_enforceMaxDist(), but if we move to checking offsets
+         * against maxDist directly, we'll have to carefully handle that case.
          */
-        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, NULL, NULL);
+        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
         /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
         newLeftoverSize = ZSTD_ldm_generateSequences_internal(
             ldmState, sequences, params, chunkStart, chunkSize);
@@ -566,14 +589,13 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
         if (sequence.offset == 0)
             break;
 
-        assert(sequence.offset <= (1U << cParams->windowLog));
         assert(ip + sequence.litLength + sequence.matchLength <= iend);
 
         /* Fill tables for block compressor */
         ZSTD_ldm_limitTableUpdate(ms, ip);
         ZSTD_ldm_fillFastTables(ms, ip);
         /* Run the block compressor */
-        DEBUGLOG(5, "calling block compressor on segment of size %u", sequence.litLength);
+        DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
         {
             size_t const newLitLength =
                 blockCompressor(ms, seqStore, rep, ip, sequence.litLength);