[linux] Opt out of row based match finder for the kernel

The row based match finder is slower without SIMD. We used to detect the
presence of SIMD to set the lower bound to 17, but that breaks
determinism. Instead, specifically opt into it for the kernel, because
it is one of the rare cases that doesn't have SIMD support.

contrib/linux-kernel/linux_zstd.h

@@ -77,6 +77,30 @@ int zstd_min_clevel(void);
  */
 int zstd_max_clevel(void);
 
+/**
+ * zstd_default_clevel() - default compression level
+ *
+ * Return: Default compression level.
+ */
+int zstd_default_clevel(void);
+
+/**
+ * struct zstd_custom_mem - custom memory allocation
+ */
+typedef ZSTD_customMem zstd_custom_mem;
+
+/**
+ * struct zstd_dict_load_method - Dictionary load method.
+ * See zstd_lib.h.
+ */
+typedef ZSTD_dictLoadMethod_e zstd_dict_load_method;
+
+/**
+ * struct zstd_dict_content_type - Dictionary context type.
+ * See zstd_lib.h.
+ */
+typedef ZSTD_dictContentType_e zstd_dict_content_type;
+
 /* ======   Parameter Selection   ====== */
 
 /**
@@ -136,6 +160,18 @@ typedef ZSTD_parameters zstd_parameters;
 zstd_parameters zstd_get_params(int level,
 	unsigned long long estimated_src_size);
 
+/**
+ * zstd_get_cparams() - returns zstd_compression_parameters for selected level
+ * @level:              The compression level
+ * @estimated_src_size: The estimated source size to compress or 0
+ *                      if unknown.
+ * @dict_size:          Dictionary size.
+ *
+ * Return:              The selected zstd_compression_parameters.
+ */
+zstd_compression_parameters zstd_get_cparams(int level,
+	unsigned long long estimated_src_size, size_t dict_size);
+
 typedef ZSTD_CCtx zstd_cctx;
 typedef ZSTD_cParameter zstd_cparameter;
 
@@ -205,6 +241,71 @@ zstd_cctx *zstd_init_cctx(void *workspace, size_t workspace_size);
 size_t zstd_compress_cctx(zstd_cctx *cctx, void *dst, size_t dst_capacity,
 	const void *src, size_t src_size, const zstd_parameters *parameters);
 
+/**
+ * zstd_create_cctx_advanced() - Create compression context
+ * @custom_mem:   Custom allocator.
+ *
+ * Return:        NULL on error, pointer to compression context otherwise.
+ */
+zstd_cctx *zstd_create_cctx_advanced(zstd_custom_mem custom_mem);
+
+/**
+ * zstd_free_cctx() - Free compression context
+ * @cdict:        Pointer to compression context.
+ *
+ * Return:        Always 0.
+ */
+size_t zstd_free_cctx(zstd_cctx* cctx);
+
+/**
+ * struct zstd_cdict - Compression dictionary.
+ * See zstd_lib.h.
+ */
+typedef ZSTD_CDict zstd_cdict;
+
+/**
+ * zstd_create_cdict_byreference() - Create compression dictionary
+ * @dict:              Pointer to dictionary buffer.
+ * @dict_size:         Size of the dictionary buffer.
+ * @dict_load_method:  Dictionary load method.
+ * @dict_content_type: Dictionary content type.
+ * @custom_mem:        Memory allocator.
+ *
+ * Note, this uses @dict by reference (ZSTD_dlm_byRef), so it should be
+ * free before zstd_cdict is destroyed.
+ *
+ * Return:             NULL on error, pointer to compression dictionary
+ *                     otherwise.
+ */
+zstd_cdict *zstd_create_cdict_byreference(const void *dict, size_t dict_size,
+					  zstd_compression_parameters cparams,
+					  zstd_custom_mem custom_mem);
+
+/**
+ * zstd_free_cdict() - Free compression dictionary
+ * @cdict:        Pointer to compression dictionary.
+ *
+ * Return:        Always 0.
+ */
+size_t zstd_free_cdict(zstd_cdict* cdict);
+
+/**
+ * zstd_compress_using_cdict() - compress src into dst using a dictionary
+ * @cctx:         The context. Must have been initialized with zstd_init_cctx().
+ * @dst:          The buffer to compress src into.
+ * @dst_capacity: The size of the destination buffer. May be any size, but
+ *                ZSTD_compressBound(srcSize) is guaranteed to be large enough.
+ * @src:          The data to compress.
+ * @src_size:     The size of the data to compress.
+ * @cdict:        The dictionary to be used.
+ *
+ * Return:        The compressed size or an error, which can be checked using
+ *                zstd_is_error().
+ */
+size_t zstd_compress_using_cdict(zstd_cctx *cctx, void *dst,
+	size_t dst_capacity, const void *src, size_t src_size,
+	const zstd_cdict *cdict);
+
 /* ======   Single-pass Decompression   ====== */
 
 typedef ZSTD_DCtx zstd_dctx;
@@ -245,6 +346,71 @@ zstd_dctx *zstd_init_dctx(void *workspace, size_t workspace_size);
 size_t zstd_decompress_dctx(zstd_dctx *dctx, void *dst, size_t dst_capacity,
 	const void *src, size_t src_size);
 
+/**
+ * struct zstd_ddict - Decompression dictionary.
+ * See zstd_lib.h.
+ */
+typedef ZSTD_DDict zstd_ddict;
+
+/**
+ * zstd_create_ddict_byreference() - Create decompression dictionary
+ * @dict:              Pointer to dictionary buffer.
+ * @dict_size:         Size of the dictionary buffer.
+ * @dict_load_method:  Dictionary load method.
+ * @dict_content_type: Dictionary content type.
+ * @custom_mem:        Memory allocator.
+ *
+ * Note, this uses @dict by reference (ZSTD_dlm_byRef), so it should be
+ * free before zstd_ddict is destroyed.
+ *
+ * Return:             NULL on error, pointer to decompression dictionary
+ *                     otherwise.
+ */
+zstd_ddict *zstd_create_ddict_byreference(const void *dict, size_t dict_size,
+					  zstd_custom_mem custom_mem);
+/**
+ * zstd_free_ddict() - Free decompression dictionary
+ * @dict:         Pointer to the dictionary.
+ *
+ * Return:        Always 0.
+ */
+size_t zstd_free_ddict(zstd_ddict *ddict);
+
+/**
+ * zstd_create_dctx_advanced() - Create decompression context
+ * @custom_mem:   Custom allocator.
+ *
+ * Return:        NULL on error, pointer to decompression context otherwise.
+ */
+zstd_dctx *zstd_create_dctx_advanced(zstd_custom_mem custom_mem);
+
+/**
+ * zstd_free_dctx() -- Free decompression context
+ * @dctx:         Pointer to decompression context.
+ * Return:        Always 0.
+ */
+size_t zstd_free_dctx(zstd_dctx *dctx);
+
+/**
+ * zstd_decompress_using_ddict() - decompress src into dst using a dictionary
+ * @dctx:         The decompression context.
+ * @dst:          The buffer to decompress src into.
+ * @dst_capacity: The size of the destination buffer. Must be at least as large
+ *                as the decompressed size. If the caller cannot upper bound the
+ *                decompressed size, then it's better to use the streaming API.
+ * @src:          The zstd compressed data to decompress. Multiple concatenated
+ *                frames and skippable frames are allowed.
+ * @src_size:     The exact size of the data to decompress.
+ * @ddict:        The dictionary to be used.
+ *
+ * Return:        The decompressed size or an error, which can be checked using
+ *                zstd_is_error().
+ */
+size_t zstd_decompress_using_ddict(zstd_dctx *dctx,
+	void *dst, size_t dst_capacity, const void *src, size_t src_size,
+	const zstd_ddict *ddict);
+
+
 /* ======   Streaming Buffers   ====== */
 
 /**

contrib/linux-kernel/zstd_compress_module.c

@@ -67,6 +67,12 @@ int zstd_max_clevel(void)
 }
 EXPORT_SYMBOL(zstd_max_clevel);
 
+int zstd_default_clevel(void)
+{
+	return ZSTD_defaultCLevel();
+}
+EXPORT_SYMBOL(zstd_default_clevel);
+
 size_t zstd_compress_bound(size_t src_size)
 {
 	return ZSTD_compressBound(src_size);
@@ -80,6 +86,13 @@ zstd_parameters zstd_get_params(int level,
 }
 EXPORT_SYMBOL(zstd_get_params);
 
+zstd_compression_parameters zstd_get_cparams(int level,
+	unsigned long long estimated_src_size, size_t dict_size)
+{
+	return ZSTD_getCParams(level, estimated_src_size, dict_size);
+}
+EXPORT_SYMBOL(zstd_get_cparams);
+
 size_t zstd_cctx_set_param(zstd_cctx *cctx, ZSTD_cParameter param, int value)
 {
 	return ZSTD_CCtx_setParameter(cctx, param, value);
@@ -146,6 +159,33 @@ zstd_cctx *zstd_init_cctx(void *workspace, size_t workspace_size)
 }
 EXPORT_SYMBOL(zstd_init_cctx);
 
+zstd_cctx *zstd_create_cctx_advanced(zstd_custom_mem custom_mem)
+{
+	return ZSTD_createCCtx_advanced(custom_mem);
+}
+EXPORT_SYMBOL(zstd_create_cctx_advanced);
+
+size_t zstd_free_cctx(zstd_cctx *cctx)
+{
+	return ZSTD_freeCCtx(cctx);
+}
+EXPORT_SYMBOL(zstd_free_cctx);
+
+zstd_cdict *zstd_create_cdict_byreference(const void *dict, size_t dict_size,
+					  zstd_compression_parameters cparams,
+					  zstd_custom_mem custom_mem)
+{
+	return ZSTD_createCDict_advanced(dict, dict_size, ZSTD_dlm_byRef,
+					 ZSTD_dct_auto, cparams, custom_mem);
+}
+EXPORT_SYMBOL(zstd_create_cdict_byreference);
+
+size_t zstd_free_cdict(zstd_cdict *cdict)
+{
+	return ZSTD_freeCDict(cdict);
+}
+EXPORT_SYMBOL(zstd_free_cdict);
+
 size_t zstd_compress_cctx(zstd_cctx *cctx, void *dst, size_t dst_capacity,
 	const void *src, size_t src_size, const zstd_parameters *parameters)
 {
@@ -154,6 +194,15 @@ size_t zstd_compress_cctx(zstd_cctx *cctx, void *dst, size_t dst_capacity,
 }
 EXPORT_SYMBOL(zstd_compress_cctx);
 
+size_t zstd_compress_using_cdict(zstd_cctx *cctx, void *dst,
+	size_t dst_capacity, const void *src, size_t src_size,
+	const ZSTD_CDict *cdict)
+{
+	return ZSTD_compress_usingCDict(cctx, dst, dst_capacity,
+					src, src_size, cdict);
+}
+EXPORT_SYMBOL(zstd_compress_using_cdict);
+
 size_t zstd_cstream_workspace_bound(const zstd_compression_parameters *cparams)
 {
 	return ZSTD_estimateCStreamSize_usingCParams(*cparams);

contrib/linux-kernel/zstd_decompress_module.c

@@ -44,6 +44,33 @@ size_t zstd_dctx_workspace_bound(void)
 }
 EXPORT_SYMBOL(zstd_dctx_workspace_bound);
 
+zstd_dctx *zstd_create_dctx_advanced(zstd_custom_mem custom_mem)
+{
+	return ZSTD_createDCtx_advanced(custom_mem);
+}
+EXPORT_SYMBOL(zstd_create_dctx_advanced);
+
+size_t zstd_free_dctx(zstd_dctx *dctx)
+{
+	return ZSTD_freeDCtx(dctx);
+}
+EXPORT_SYMBOL(zstd_free_dctx);
+
+zstd_ddict *zstd_create_ddict_byreference(const void *dict, size_t dict_size,
+					  zstd_custom_mem custom_mem)
+{
+	return ZSTD_createDDict_advanced(dict, dict_size, ZSTD_dlm_byRef,
+					 ZSTD_dct_auto, custom_mem);
+
+}
+EXPORT_SYMBOL(zstd_create_ddict_byreference);
+
+size_t zstd_free_ddict(zstd_ddict *ddict)
+{
+	return ZSTD_freeDDict(ddict);
+}
+EXPORT_SYMBOL(zstd_free_ddict);
+
 zstd_dctx *zstd_init_dctx(void *workspace, size_t workspace_size)
 {
 	if (workspace == NULL)
@@ -59,6 +86,15 @@ size_t zstd_decompress_dctx(zstd_dctx *dctx, void *dst, size_t dst_capacity,
 }
 EXPORT_SYMBOL(zstd_decompress_dctx);
 
+size_t zstd_decompress_using_ddict(zstd_dctx *dctx,
+	void *dst, size_t dst_capacity, const void* src, size_t src_size,
+	const zstd_ddict* ddict)
+{
+	return ZSTD_decompress_usingDDict(dctx, dst, dst_capacity, src,
+					  src_size, ddict);
+}
+EXPORT_SYMBOL(zstd_decompress_using_ddict);
+
 size_t zstd_dstream_workspace_bound(size_t max_window_size)
 {
 	return ZSTD_estimateDStreamSize(max_window_size);

lib/compress/zstd_compress.c

@@ -237,10 +237,18 @@ static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_Para
 /* Returns row matchfinder usage given an initial mode and cParams */
 static ZSTD_ParamSwitch_e ZSTD_resolveRowMatchFinderMode(ZSTD_ParamSwitch_e mode,
                                                          const ZSTD_compressionParameters* const cParams) {
+#ifdef ZSTD_LINUX_KERNEL
+    /* The Linux Kernel does not use SIMD, and 128KB is a very common size, e.g. in BtrFS.
+     * The row match finder is slower for this size without SIMD, so disable it.
+     */
+    const unsigned kWindowLogLowerBound = 17;
+#else
+    const unsigned kWindowLogLowerBound = 14;
+#endif
     if (mode != ZSTD_ps_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
     mode = ZSTD_ps_disable;
     if (!ZSTD_rowMatchFinderSupported(cParams->strategy)) return mode;
-    if (cParams->windowLog > 14) mode = ZSTD_ps_enable;
+    if (cParams->windowLog > kWindowLogLowerBound) mode = ZSTD_ps_enable;
     return mode;
 }