[libzstd] Add ZSTD_c_literalCompressionMode flag

It controls the literals compression. It is either
`auto`, `huffman`, or `uncompressed`. It defaults to
`auto`, which is the current behavior.

diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c
index 2bf8ee535cc176cfe9c295d78ea03fd04bd184bc..e18051f00ceb44b56eccd6433c18175a5a3c7bbc 100644 (file)
--- a/lib/compress/zstd_compress.c
+++ b/lib/compress/zstd_compress.c
@@ -360,6 +360,12 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         bounds.upperBound = ZSTD_dictForceCopy;       /* note : how to ensure at compile time that this is the highest value enum ? */
         return bounds;
 
+    case ZSTD_c_literalCompressionMode:
+        ZSTD_STATIC_ASSERT(ZSTD_lcm_auto < ZSTD_lcm_huffman && ZSTD_lcm_huffman < ZSTD_lcm_uncompressed);
+        bounds.lowerBound = ZSTD_lcm_auto;
+        bounds.upperBound = ZSTD_lcm_uncompressed;
+        return bounds;
+
     default:
         {   ZSTD_bounds const boundError = { ERROR(parameter_unsupported), 0, 0 };
             return boundError;
@@ -396,6 +402,7 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_c_minMatch:
     case ZSTD_c_targetLength:
     case ZSTD_c_strategy:
+    case ZSTD_c_literalCompressionMode:
         return 1;
 
     case ZSTD_c_format:
@@ -463,6 +470,9 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
     case ZSTD_c_forceAttachDict:
         return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
 
+    case ZSTD_c_literalCompressionMode:
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
     case ZSTD_c_nbWorkers:
         RETURN_ERROR_IF((value!=0) && cctx->staticSize, parameter_unsupported,
                         "MT not compatible with static alloc");
@@ -575,6 +585,13 @@ size_t ZSTD_CCtxParam_setParameter(ZSTD_CCtx_params* CCtxParams,
         return CCtxParams->attachDictPref;
     }
 
+    case ZSTD_c_literalCompressionMode : {
+        const ZSTD_literalCompressionMode_e lcm = (ZSTD_literalCompressionMode_e)value;
+        BOUNDCHECK(ZSTD_c_literalCompressionMode, lcm);
+        CCtxParams->literalCompressionMode = lcm;
+        return CCtxParams->literalCompressionMode;
+    }
+
     case ZSTD_c_nbWorkers :
 #ifndef ZSTD_MULTITHREAD
         RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
@@ -688,6 +705,9 @@ size_t ZSTD_CCtxParam_getParameter(
     case ZSTD_c_forceAttachDict :
         *value = CCtxParams->attachDictPref;
         break;
+    case ZSTD_c_literalCompressionMode :
+        *value = CCtxParams->literalCompressionMode;
+        break;
     case ZSTD_c_nbWorkers :
 #ifndef ZSTD_MULTITHREAD
         assert(CCtxParams->nbWorkers == 0);
@@ -2369,6 +2389,21 @@ static size_t ZSTD_encodeSequences(
                                         sequences, nbSeq, longOffsets);
 }
 
+static int ZSTD_disableLiteralsCompression(const ZSTD_CCtx_params* cctxParams)
+{
+    switch (cctxParams->literalCompressionMode) {
+    case ZSTD_lcm_huffman:
+        return 0;
+    case ZSTD_lcm_uncompressed:
+        return 1;
+    default:
+        assert(0 /* impossible: pre-validated */);
+        /* fall-through */
+    case ZSTD_lcm_auto:
+        return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
+    }
+}
+
 /* ZSTD_compressSequences_internal():
  * actually compresses both literals and sequences */
 MEM_STATIC size_t
@@ -2404,10 +2439,10 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
     /* Compress literals */
     {   const BYTE* const literals = seqStorePtr->litStart;
         size_t const litSize = seqStorePtr->lit - literals;
-        int const disableLiteralCompression = (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
         size_t const cSize = ZSTD_compressLiterals(
                                     &prevEntropy->huf, &nextEntropy->huf,
-                                    cctxParams->cParams.strategy, disableLiteralCompression,
+                                    cctxParams->cParams.strategy,
+                                    ZSTD_disableLiteralsCompression(cctxParams),
                                     op, dstCapacity,
                                     literals, litSize,
                                     workspace, wkspSize,

diff --git a/lib/compress/zstd_compress_internal.h b/lib/compress/zstd_compress_internal.h
index 29bca59859ea88cca1a1372ac231bc15e1b0532c..e0b54299d6a3a7407ab242c95fcd4ccf2db71b29 100644 (file)
--- a/lib/compress/zstd_compress_internal.h
+++ b/lib/compress/zstd_compress_internal.h
@@ -188,6 +188,7 @@ struct ZSTD_CCtx_params_s {
                                 * 1<<wLog, even for dictionary */
 
     ZSTD_dictAttachPref_e attachDictPref;
+    ZSTD_literalCompressionMode_e literalCompressionMode;
 
     /* Multithreading: used to pass parameters to mtctx */
     int nbWorkers;

diff --git a/lib/zstd.h b/lib/zstd.h
index 237635cbaf34d2b9c8e6c52e6fd1128ce3410384..03ba1e2dde529665c66de62e22a3219756c74f7b 100644 (file)
--- a/lib/zstd.h
+++ b/lib/zstd.h
@@ -625,6 +625,7 @@ typedef enum {
      * ZSTD_c_format
      * ZSTD_c_forceMaxWindow
      * ZSTD_c_forceAttachDict
+     * ZSTD_c_literalCompressionMode
      * 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.
@@ -632,7 +633,8 @@ typedef enum {
      ZSTD_c_experimentalParam1=500,
      ZSTD_c_experimentalParam2=10,
      ZSTD_c_experimentalParam3=1000,
-     ZSTD_c_experimentalParam4=1001
+     ZSTD_c_experimentalParam4=1001,
+     ZSTD_c_experimentalParam5=1002,
 } ZSTD_cParameter;
 
 
@@ -1064,6 +1066,15 @@ typedef enum {
     ZSTD_dictForceCopy     = 2, /* Always copy the dictionary. */
 } ZSTD_dictAttachPref_e;
 
+typedef enum {
+  ZSTD_lcm_auto = 0,          /**< Automatically determine the compression mode based on the compression level.
+                               *   Negative compression levels will be uncompressed, and positive compression
+                               *   levels will be compressed. */
+  ZSTD_lcm_huffman = 1,       /**< Always attempt Huffman compression. Uncompressed literals will still be
+                               *   emitted if Huffman compression is not profitable. */
+  ZSTD_lcm_uncompressed = 2,  /**< Always emit uncompressed literals. */
+} ZSTD_literalCompressionMode_e;
+
 
 /***************************************
 *  Frame size functions
@@ -1318,6 +1329,12 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  * See the comments on that enum for an explanation of the feature. */
 #define ZSTD_c_forceAttachDict ZSTD_c_experimentalParam4
 
+/* Controls how the literals are compressed (default is auto).
+ * The value must be of type ZSTD_literalCompressionMode_e.
+ * See ZSTD_literalCompressionMode_t enum definition for details.
+ */
+#define ZSTD_c_literalCompressionMode ZSTD_c_experimentalParam5
+
 /*! ZSTD_CCtx_getParameter() :
  *  Get the requested compression parameter value, selected by enum ZSTD_cParameter,
  *  and store it into int* value.