re-inserted offset evaluation

diff --git a/lib/decompress/zstd_decompress.c b/lib/decompress/zstd_decompress.c
index 5fa4e3056e07e6a527ae0ee8d57c0bd9c448f770..470aac286180e23e18ae2574b7082aaf458e2b8b 100644 (file)
--- a/lib/decompress/zstd_decompress.c
+++ b/lib/decompress/zstd_decompress.c
@@ -1248,47 +1248,49 @@ typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset
 #endif
 
 typedef size_t (*ZSTD_decompressSequences_t)(
-    ZSTD_DCtx *dctx, void *dst, size_t maxDstSize, const void *seqStart,
-    size_t seqSize, const ZSTD_longOffset_e isLongOffset);
+    ZSTD_DCtx *dctx, void *dst, size_t maxDstSize,
+    const void *seqStart, size_t seqSize, int nbSeq,
+    const ZSTD_longOffset_e isLongOffset);
 
 static size_t ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
-                                const void* seqStart, size_t seqSize,
+                                const void* seqStart, size_t seqSize, int nbSeq,
                                 const ZSTD_longOffset_e isLongOffset)
 {
 #if DYNAMIC_BMI2
     if (dctx->bmi2) {
-        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, isLongOffset);
+        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
     }
 #endif
-  return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, isLongOffset);
+  return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 
-static size_t ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
-                                    const void* seqStart, size_t seqSize,
-                                    const ZSTD_longOffset_e isLongOffset)
+static size_t ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
+                                void* dst, size_t maxDstSize,
+                                const void* seqStart, size_t seqSize, int nbSeq,
+                                const ZSTD_longOffset_e isLongOffset)
 {
 #if DYNAMIC_BMI2
     if (dctx->bmi2) {
-        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, isLongOffset);
+        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
     }
 #endif
-  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, isLongOffset);
+  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
 }
 
 static unsigned
-ZSTD_getLongOffsetsShare(const FSE_DTable* offTable)
+ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
 {
     const void* ptr = offTable;
     U32 const tableLog = ((const FSE_DTableHeader*)ptr)[0].tableLog;
-    const FSE_decode_t* table = ((const FSE_decode_t*)ptr) + 1;
+    const ZSTD_seqSymbol* table = offTable + 1;
     U32 const max = 1 << tableLog;
     U32 u, total = 0;
 
-    assert(tableLog <= OffFSELog);
     for (u=0; u<max; u++)
-        if (table[u].symbol > 23) total += 1;
+        if (table[u].nbAdditionalBits > 23) total += 1;
 
-    total <<= (OffFSELog - tableLog);
+    assert(tableLog <= OffFSELog);
+    total <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
 
     return total;
 }
@@ -1317,12 +1319,22 @@ static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
         srcSize -= litCSize;
     }
 
-    if ( frame /* windowSize exists */
-      && (dctx->fParams.windowSize > (1<<24))
-      && MEM_64bits() /* x86 benefits less from long mode than x64 */ )
-        return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, isLongOffset);
+    /* Build Decoding Tables */
+    {   int nbSeq;
+        size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
+        if (ZSTD_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        srcSize -= seqHSize;
+
+        if (dctx->fParams.windowSize > (1<<24)) {
+            U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
+            U32 const minShare = MEM_64bits() ? 5 : 13;
+            if (shareLongOffsets >= minShare)
+                return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+        }
 
-    return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, isLongOffset);
+        return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+    }
 }
 
 

diff --git a/lib/decompress/zstd_decompress_impl.h b/lib/decompress/zstd_decompress_impl.h
index 6454a7429eedfec610bcc0161f07b5494a9f3c95..18b3a296ed0abaf0ac87e87b42451222be24886d 100644 (file)
--- a/lib/decompress/zstd_decompress_impl.h
+++ b/lib/decompress/zstd_decompress_impl.h
@@ -209,7 +209,7 @@ static TARGET
 size_t FUNCTION(ZSTD_decompressSequences)(
                                ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
-                         const void* seqStart, size_t seqSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
                          const ZSTD_longOffset_e isLongOffset)
 {
     const BYTE* ip = (const BYTE*)seqStart;
@@ -222,17 +222,8 @@ size_t FUNCTION(ZSTD_decompressSequences)(
     const BYTE* const base = (const BYTE*) (dctx->base);
     const BYTE* const vBase = (const BYTE*) (dctx->vBase);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    int nbSeq;
     DEBUGLOG(5, "ZSTD_decompressSequences");
 
-    /* Build Decoding Tables */
-    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
-        DEBUGLOG(5, "ZSTD_decodeSeqHeaders: size=%u, nbSeq=%i",
-                    (U32)seqHSize, nbSeq);
-        if (ZSTD_isError(seqHSize)) return seqHSize;
-        ip += seqHSize;
-    }
-
     /* Regen sequences */
     if (nbSeq) {
         seqState_t seqState;
@@ -273,7 +264,7 @@ static TARGET
 size_t FUNCTION(ZSTD_decompressSequencesLong)(
                                ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
-                         const void* seqStart, size_t seqSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
                          const ZSTD_longOffset_e isLongOffset)
 {
     const BYTE* ip = (const BYTE*)seqStart;
@@ -286,13 +277,6 @@ size_t FUNCTION(ZSTD_decompressSequencesLong)(
     const BYTE* const prefixStart = (const BYTE*) (dctx->base);
     const BYTE* const dictStart = (const BYTE*) (dctx->vBase);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    int nbSeq;
-
-    /* Build Decoding Tables */
-    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
-        if (ZSTD_isError(seqHSize)) return seqHSize;
-        ip += seqHSize;
-    }
 
     /* Regen sequences */
     if (nbSeq) {