ZSTD_splitBlock_4k() uses externally provided workspace

ideally, this workspace would be provided from the ZSTD_CCtx* state

lib/compress/zstd_compress.c

@@ -4489,8 +4489,11 @@ static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms,
 
 #include "zstd_preSplit.h"
 
-static size_t ZSTD_optimalBlockSize(const void* src, size_t srcSize, size_t blockSizeMax, ZSTD_strategy strat, S64 savings)
+
+static size_t ZSTD_optimalBlockSize(ZSTD_CCtx* cctx, const void* src, size_t srcSize, size_t blockSizeMax, ZSTD_strategy strat, S64 savings)
 {
+    S64 workspace[ZSTD_SLIPBLOCK_WORKSPACESIZE / 8];
+    (void)cctx;
     /* note: we currenly only split full blocks (128 KB)
      * and when there is more than 128 KB input remaining
      */
@@ -4499,7 +4502,7 @@ static size_t ZSTD_optimalBlockSize(const void* src, size_t srcSize, size_t bloc
     /* dynamic splitting has a cpu cost for analysis,
      * due to that cost it's only used for btlazy2+ strategies */
     if (strat >= ZSTD_btlazy2)
-        return ZSTD_splitBlock_4k(src, srcSize, blockSizeMax);
+        return ZSTD_splitBlock_4k(src, srcSize, blockSizeMax, workspace, sizeof(workspace));
     /* blind split strategy
      * no cpu cost, but can over-split homegeneous data.
      * heuristic, tested as being "generally better".
@@ -4537,7 +4540,7 @@ static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
     while (remaining) {
         ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
         U32 const lastBlock = lastFrameChunk & (blockSizeMax >= remaining);
-        size_t const blockSize = ZSTD_optimalBlockSize(ip, remaining, blockSizeMax, cctx->appliedParams.cParams.strategy, savings);
+        size_t const blockSize = ZSTD_optimalBlockSize(cctx, ip, remaining, blockSizeMax, cctx->appliedParams.cParams.strategy, savings);
         assert(blockSize <= remaining);
 
         /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding

lib/compress/zstd_preSplit.c

@@ -10,6 +10,7 @@
 
 #include "../common/mem.h" /* S64 */
 #include "../common/zstd_deps.h" /* ZSTD_memset */
+#include "../common/zstd_internal.h" /* ZSTD_STATIC_ASSERT */
 #include "zstd_preSplit.h"
 
 
@@ -30,22 +31,19 @@ static unsigned hash2(const void *p)
 }
 
 
-/* ==================================== */
-/* Global array -> for testing only !!! */
-/* ==================================== */
 typedef struct {
   int events[HASHTABLESIZE];
   S64 nbEvents;
 } FingerPrint;
-static FingerPrint pastEvents;
+typedef struct {
-static FingerPrint newEvents;
+    FingerPrint pastEvents;
+    FingerPrint newEvents;
+} FPStats;
 
-static void initStats(void)
+static void initStats(FPStats* fpstats)
 {
-    ZSTD_memset(&pastEvents, 0, sizeof(pastEvents));
+    ZSTD_memset(fpstats, 0, sizeof(FPStats));
-    ZSTD_memset(&newEvents, 0, sizeof(newEvents));
 }
-/* ==================================== */
 
 static void addToFingerprint(FingerPrint* fp, const void* src, size_t s)
 {
@@ -103,14 +101,14 @@ static void mergeEvents(FingerPrint* acc, const FingerPrint* newfp)
     acc->nbEvents += newfp->nbEvents;
 }
 
-static void flushEvents(void)
+static void flushEvents(FPStats* fpstats)
 {
     size_t n;
     for (n = 0; n < HASHTABLESIZE; n++) {
-        pastEvents.events[n] = newEvents.events[n];
+        fpstats->pastEvents.events[n] = fpstats->newEvents.events[n];
     }
-    pastEvents.nbEvents = newEvents.nbEvents;
+    fpstats->pastEvents.nbEvents = fpstats->newEvents.nbEvents;
-    ZSTD_memset(&newEvents, 0, sizeof(newEvents));
+    ZSTD_memset(&fpstats->newEvents, 0, sizeof(fpstats->newEvents));
 }
 
 static void removeEvents(FingerPrint* acc, const FingerPrint* slice)
@@ -125,23 +123,30 @@ static void removeEvents(FingerPrint* acc, const FingerPrint* slice)
 
 #define CHUNKSIZE (8 << 10)
 /* Note: technically, we use CHUNKSIZE, so that's 8 KB */
-size_t ZSTD_splitBlock_4k(const void* src, size_t srcSize, size_t blockSizeMax)
+size_t ZSTD_splitBlock_4k(const void* src, size_t srcSize,
+                        size_t blockSizeMax,
+                        void* workspace, size_t wkspSize)
 {
+    FPStats* const fpstats = (FPStats*)workspace;
     const char* p = (const char*)src;
     int penalty = THRESHOLD_PENALTY;
     size_t pos = 0;
     if (srcSize <= blockSizeMax) return srcSize;
     assert(blockSizeMax == (128 << 10));
+    assert(workspace != NULL);
+    assert((size_t)workspace % 8 == 0);
+    ZSTD_STATIC_ASSERT(ZSTD_SLIPBLOCK_WORKSPACESIZE == sizeof(FPStats));
+    assert(wkspSize >= sizeof(FPStats)); (void)wkspSize;
 
-    initStats();
+    initStats(fpstats);
     for (pos = 0; pos < blockSizeMax;) {
         assert(pos <= blockSizeMax - CHUNKSIZE);
-        recordFingerprint(&newEvents, p + pos, CHUNKSIZE);
+        recordFingerprint(&fpstats->newEvents, p + pos, CHUNKSIZE);
-        if (compareFingerprints(&pastEvents, &newEvents, penalty)) {
+        if (compareFingerprints(&fpstats->pastEvents, &fpstats->newEvents, penalty)) {
             return pos;
         } else {
-            mergeEvents(&pastEvents, &newEvents);
+            mergeEvents(&fpstats->pastEvents, &fpstats->newEvents);
-            ZSTD_memset(&newEvents, 0, sizeof(newEvents));
+            ZSTD_memset(&fpstats->newEvents, 0, sizeof(fpstats->newEvents));
             penalty = penalty - 1 + (penalty == 0);
         }
         pos += CHUNKSIZE;

lib/compress/zstd_preSplit.h

@@ -17,7 +17,13 @@
 extern "C" {
 #endif
 
-size_t ZSTD_splitBlock_4k(const void* src, size_t srcSize, size_t blockSizeMax);
+#define ZSTD_SLIPBLOCK_WORKSPACESIZE 8208
+
+/* note:
+ * @workspace must be aligned on 8-bytes boundaries
+ * @wkspSize must be at least >= ZSTD_SLIPBLOCK_WORKSPACESIZE
+ */
+size_t ZSTD_splitBlock_4k(const void* src, size_t srcSize, size_t blockSizeMax, void* workspace, size_t wkspSize);
 
 #if defined (__cplusplus)
 }