Add prefetchCDictTables CCtxParam

lib/compress/zstd_compress.c

@@ -576,6 +576,11 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         bounds.upperBound = 1;
         return bounds;
 
+    case ZSTD_c_prefetchCDictTables:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
     default:
         bounds.error = ERROR(parameter_unsupported);
         return bounds;
@@ -640,6 +645,7 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_c_useBlockSplitter:
     case ZSTD_c_useRowMatchFinder:
     case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
     default:
         return 0;
     }
@@ -695,6 +701,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
     case ZSTD_c_useBlockSplitter:
     case ZSTD_c_useRowMatchFinder:
     case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
         break;
 
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
@@ -921,6 +928,11 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         CCtxParams->deterministicRefPrefix = !!value;
         return CCtxParams->deterministicRefPrefix;
 
+    case ZSTD_c_prefetchCDictTables:
+        BOUNDCHECK(ZSTD_c_prefetchCDictTables, value);
+        CCtxParams->prefetchCDictTables = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->prefetchCDictTables;
+
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
 }
@@ -1053,6 +1065,9 @@ size_t ZSTD_CCtxParams_getParameter(
     case ZSTD_c_deterministicRefPrefix:
         *value = (int)CCtxParams->deterministicRefPrefix;
         break;
+    case ZSTD_c_prefetchCDictTables:
+        *value = (int)CCtxParams->prefetchCDictTables;
+        break;
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
     return 0;
@@ -2913,6 +2928,7 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
                                                                                     zc->appliedParams.useRowMatchFinder,
                                                                                     dictMode);
             ms->ldmSeqStore = NULL;
+            ms->prefetchCDictTables = zc->appliedParams.prefetchCDictTables == ZSTD_ps_enable;
             lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
         }
         {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;

lib/compress/zstd_compress_internal.h

@@ -235,6 +235,7 @@ struct ZSTD_matchState_t {
     const ZSTD_matchState_t* dictMatchState;
     ZSTD_compressionParameters cParams;
     const rawSeqStore_t* ldmSeqStore;
+    int prefetchCDictTables;  /* TODO document */
 };
 
 typedef struct {
@@ -331,6 +332,9 @@ struct ZSTD_CCtx_params_s {
 
     /* Internal use, for createCCtxParams() and freeCCtxParams() only */
     ZSTD_customMem customMem;
+
+    /* TODO document */
+    ZSTD_paramSwitch_e prefetchCDictTables;
 };  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
 
 #define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))

lib/compress/zstd_double_fast.c

@@ -345,6 +345,13 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
     /* if a dictionary is attached, it must be within window range */
     assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
 
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableSize = ((size_t)1) << dictCParams->hashLog;
+        size_t const chainTableSize = ((size_t)1) << dictCParams->chainLog;
+        PREFETCH_AREA(dictHashLong, hashTableSize * sizeof(U32))
+        PREFETCH_AREA(dictHashSmall, chainTableSize * sizeof(U32))
+    }
+
     /* init */
     ip += (dictAndPrefixLength == 0);
 

lib/compress/zstd_fast.c

@@ -500,6 +500,11 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
      * when translating a dict index into a local index */
     assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
 
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableSize = ((size_t)1) << dictCParams->hashLog;
+        PREFETCH_AREA(dictHashTable, hashTableSize * sizeof(U32))
+    }
+
     /* init */
     DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
     ip0 += (dictAndPrefixLength == 0);

lib/zstd.h

@@ -421,6 +421,7 @@ typedef enum {
      * ZSTD_c_validateSequences
      * ZSTD_c_useBlockSplitter
      * ZSTD_c_useRowMatchFinder
+     * ZSTD_c_prefetchCDictTables
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly;
      *        also, the enums values themselves are unstable and can still change.
@@ -439,7 +440,8 @@ typedef enum {
      ZSTD_c_experimentalParam12=1009,
      ZSTD_c_experimentalParam13=1010,
      ZSTD_c_experimentalParam14=1011,
-     ZSTD_c_experimentalParam15=1012
+     ZSTD_c_experimentalParam15=1012,
+     ZSTD_c_experimentalParam16=1013
 } ZSTD_cParameter;
 
 typedef struct {
@@ -1954,6 +1956,9 @@ ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const vo
  */
 #define ZSTD_c_deterministicRefPrefix ZSTD_c_experimentalParam15
 
+/* TODO document */
+#define ZSTD_c_prefetchCDictTables ZSTD_c_experimentalParam16
+
 /*! ZSTD_CCtx_getParameter() :
  *  Get the requested compression parameter value, selected by enum ZSTD_cParameter,
  *  and store it into int* value.