Merge remote-tracking branch 'upstream/dev' into dev

NEWS

@@ -1,6 +1,9 @@
 v1.3.1
+perf: substantially decreased memory usage in Multi-threading mode, thanks to reports by Tino Reichardt
+perf: Multi-threading supports up to 256 threads. Cap at 256 when more are requested (#760)
 build: fix Visual compilation for non x86/x64 targets, reported by Greg Slazinski (#718)
-API exp : breaking change : ZSTD_getframeHeader() 
+API exp : breaking change : ZSTD_getframeHeader() provides more information
+API exp : breaking change : pinned down values of error codes
 
 v1.3.0
 cli : new : `--list` command, by Paul Cruz

build/cmake/programs/CMakeLists.txt

@@ -31,6 +31,9 @@ ENDIF (MSVC)
 
 ADD_EXECUTABLE(zstd ${PROGRAMS_DIR}/zstdcli.c ${PROGRAMS_DIR}/fileio.c ${PROGRAMS_DIR}/bench.c ${PROGRAMS_DIR}/datagen.c ${PROGRAMS_DIR}/dibio.c ${PlatformDependResources})
 TARGET_LINK_LIBRARIES(zstd libzstd_static)
+IF (CMAKE_SYSTEM_NAME MATCHES "(Solaris|SunOS)")
+    TARGET_LINK_LIBRARIES(zstd rt)
+ENDIF (CMAKE_SYSTEM_NAME MATCHES "(Solaris|SunOS)")
 INSTALL(TARGETS zstd RUNTIME DESTINATION "bin")
 
 IF (UNIX)

contrib/linux-kernel/0000-cover-letter.patch

@@ -1,7 +1,7 @@
-From 8bc9a0ae5c86a6d02d9a5274b9965ddac0e8d330 Mon Sep 17 00:00:00 2001
+From 0cd63464d182bb9708f8b25f7da3dc8e5ec6b4fa Mon Sep 17 00:00:00 2001
 From: Nick Terrell <terrelln@fb.com>
-Date: Wed, 28 Jun 2017 22:00:00 -0700
+Date: Thu, 20 Jul 2017 13:18:30 -0700
-Subject: [PATCH v2 0/4] Add xxhash and zstd modules
+Subject: [PATCH v3 0/4] Add xxhash and zstd modules
 
 Hi all,
 
@@ -24,6 +24,13 @@ v1 -> v2:
   HUF_compressWeights(), HUF_readDTableX2(), and HUF_readDTableX4() (2/4)
 - No zstd function uses more than 400 B of stack space (2/4)
 
+v2 -> v3:
+- Work around gcc-7 bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81388
+  (2/4)
+- Fix bug in dictionary compression from upstream commit cc1522351f (2/4)
+- Port upstream BtrFS commits e1ddce71d6, 389a6cfc2a, and 6acafd1eff (3/4)
+- Change default compression level for BtrFS to 3 (3/4)
+
 Nick Terrell (4):
   lib: Add xxhash module
   lib: Add zstd modules
@@ -40,7 +47,7 @@ Nick Terrell (4):
  fs/btrfs/props.c           |    6 +
  fs/btrfs/super.c           |   12 +-
  fs/btrfs/sysfs.c           |    2 +
- fs/btrfs/zstd.c            |  433 ++++++
+ fs/btrfs/zstd.c            |  435 ++++++
  fs/squashfs/Kconfig        |   14 +
  fs/squashfs/Makefile       |    1 +
  fs/squashfs/decompressor.c |    7 +
@@ -63,13 +70,13 @@ Nick Terrell (4):
  lib/zstd/fse_compress.c    |  795 ++++++++++
  lib/zstd/fse_decompress.c  |  332 +++++
  lib/zstd/huf.h             |  212 +++
- lib/zstd/huf_compress.c    |  771 ++++++++++
+ lib/zstd/huf_compress.c    |  770 ++++++++++
  lib/zstd/huf_decompress.c  |  960 ++++++++++++
  lib/zstd/mem.h             |  151 ++
  lib/zstd/zstd_common.c     |   75 +
- lib/zstd/zstd_internal.h   |  269 ++++
+ lib/zstd/zstd_internal.h   |  250 ++++
  lib/zstd/zstd_opt.h        | 1014 +++++++++++++
- 39 files changed, 14400 insertions(+), 12 deletions(-)
+ 39 files changed, 14382 insertions(+), 12 deletions(-)
  create mode 100644 fs/btrfs/zstd.c
  create mode 100644 fs/squashfs/zstd_wrapper.c
  create mode 100644 include/linux/xxhash.h

contrib/linux-kernel/0001-lib-Add-xxhash-module.patch

@@ -1,7 +1,7 @@
-From 5ac909c415ab4a18fd90794793c96e450795e8c6 Mon Sep 17 00:00:00 2001
+From fc7f26acbabda35f1c61dfc357dbb207dc8ed23d Mon Sep 17 00:00:00 2001
 From: Nick Terrell <terrelln@fb.com>
-Date: Wed, 21 Jun 2017 17:37:36 -0700
+Date: Mon, 17 Jul 2017 17:07:18 -0700
-Subject: [PATCH v2 1/4] lib: Add xxhash module
+Subject: [PATCH v3 1/4] lib: Add xxhash module
 
 Adds xxhash kernel module with xxh32 and xxh64 hashes. xxhash is an
 extremely fast non-cryptographic hash algorithm for checksumming.
@@ -327,10 +327,10 @@ index 0000000..9e1f42c
 +
 +#endif /* XXHASH_H */
 diff --git a/lib/Kconfig b/lib/Kconfig
-index 0c8b78a..b6009d7 100644
+index 6762529..5e7541f 100644
 --- a/lib/Kconfig
 +++ b/lib/Kconfig
-@@ -184,6 +184,9 @@ config CRC8
+@@ -192,6 +192,9 @@ config CRC8
  	  when they need to do cyclic redundancy check according CRC8
  	  algorithm. Module will be called crc8.
 
@@ -341,10 +341,10 @@ index 0c8b78a..b6009d7 100644
  	bool
  	depends on AUDIT && !AUDIT_ARCH
 diff --git a/lib/Makefile b/lib/Makefile
-index 0166fbc..1338226 100644
+index 40c1837..d06b68a 100644
 --- a/lib/Makefile
 +++ b/lib/Makefile
-@@ -102,6 +102,7 @@ obj-$(CONFIG_CRC32_SELFTEST)	+= crc32test.o
+@@ -102,6 +102,7 @@ obj-$(CONFIG_CRC4)	+= crc4.o
  obj-$(CONFIG_CRC7)	+= crc7.o
  obj-$(CONFIG_LIBCRC32C)	+= libcrc32c.o
  obj-$(CONFIG_CRC8)	+= crc8.o

contrib/linux-kernel/0002-lib-Add-zstd-modules.patch

@@ -1,7 +1,7 @@
-From d2626127c6d6e60e940dd9a3ed58323bdcdc4930 Mon Sep 17 00:00:00 2001
+From 686a6149b98250d66b5951e3ae05e79063e9de98 Mon Sep 17 00:00:00 2001
 From: Nick Terrell <terrelln@fb.com>
-Date: Tue, 16 May 2017 14:55:36 -0700
+Date: Mon, 17 Jul 2017 17:08:19 -0700
-Subject: [PATCH v2 2/4] lib: Add zstd modules
+Subject: [PATCH v3 2/4] lib: Add zstd modules
 
 Add zstd compression and decompression kernel modules.
 zstd offers a wide varity of compression speed and quality trade-offs.
@@ -110,6 +110,10 @@ v1 -> v2:
   HUF_compressWeights(), HUF_readDTableX2(), and HUF_readDTableX4()
 - No function uses more than 400 B of stack space
 
+v2 -> v3:
+- Work around gcc-7 bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81388
+- Fix bug in dictionary compression from upstream commit cc1522351f
+
  include/linux/zstd.h      | 1157 +++++++++++++++
  lib/Kconfig               |    8 +
  lib/Makefile              |    2 +
@@ -123,13 +127,13 @@ v1 -> v2:
  lib/zstd/fse_compress.c   |  795 +++++++++++
  lib/zstd/fse_decompress.c |  332 +++++
  lib/zstd/huf.h            |  212 +++
- lib/zstd/huf_compress.c   |  771 ++++++++++
+ lib/zstd/huf_compress.c   |  770 ++++++++++
  lib/zstd/huf_decompress.c |  960 +++++++++++++
  lib/zstd/mem.h            |  151 ++
  lib/zstd/zstd_common.c    |   75 +
- lib/zstd/zstd_internal.h  |  269 ++++
+ lib/zstd/zstd_internal.h  |  250 ++++
  lib/zstd/zstd_opt.h       | 1014 +++++++++++++
- 19 files changed, 13014 insertions(+)
+ 19 files changed, 12994 insertions(+)
  create mode 100644 include/linux/zstd.h
  create mode 100644 lib/zstd/Makefile
  create mode 100644 lib/zstd/bitstream.h
@@ -1312,10 +1316,10 @@ index 0000000..249575e
 +
 +#endif  /* ZSTD_H */
 diff --git a/lib/Kconfig b/lib/Kconfig
-index b6009d7..f00ddab 100644
+index 5e7541f..0d49ed0 100644
 --- a/lib/Kconfig
 +++ b/lib/Kconfig
-@@ -241,6 +241,14 @@ config LZ4HC_COMPRESS
+@@ -249,6 +249,14 @@ config LZ4HC_COMPRESS
  config LZ4_DECOMPRESS
  	tristate
 
@@ -1331,7 +1335,7 @@ index b6009d7..f00ddab 100644
 
  #
 diff --git a/lib/Makefile b/lib/Makefile
-index 1338226..4fcef16 100644
+index d06b68a..d5c8a4f 100644
 --- a/lib/Makefile
 +++ b/lib/Makefile
 @@ -116,6 +116,8 @@ obj-$(CONFIG_LZO_DECOMPRESS) += lzo/
@@ -10012,10 +10016,10 @@ index 0000000..2143da2
 +#endif /* HUF_H_298734234 */
 diff --git a/lib/zstd/huf_compress.c b/lib/zstd/huf_compress.c
 new file mode 100644
-index 0000000..0361f38
+index 0000000..40055a7
 --- /dev/null
 +++ b/lib/zstd/huf_compress.c
-@@ -0,0 +1,771 @@
+@@ -0,0 +1,770 @@
 +/*
 + * Huffman encoder, part of New Generation Entropy library
 + * Copyright (C) 2013-2016, Yann Collet.
@@ -10543,7 +10547,7 @@ index 0000000..0361f38
 +
 +size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
 +
-+#define HUF_FLUSHBITS(s) (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
++#define HUF_FLUSHBITS(s)  BIT_flushBits(s)
 +
 +#define HUF_FLUSHBITS_1(stream)                                            \
 +	if (sizeof((stream)->bitContainer) * 8 < HUF_TABLELOG_MAX * 2 + 7) \
@@ -10560,7 +10564,6 @@ index 0000000..0361f38
 +	BYTE *const oend = ostart + dstSize;
 +	BYTE *op = ostart;
 +	size_t n;
-+	const unsigned fast = (dstSize >= HUF_BLOCKBOUND(srcSize));
 +	BIT_CStream_t bitC;
 +
 +	/* init */
@@ -11993,10 +11996,10 @@ index 0000000..a282624
 +}
 diff --git a/lib/zstd/zstd_internal.h b/lib/zstd/zstd_internal.h
 new file mode 100644
-index 0000000..6748719
+index 0000000..f0ba474
 --- /dev/null
 +++ b/lib/zstd/zstd_internal.h
-@@ -0,0 +1,269 @@
+@@ -0,0 +1,250 @@
 +/**
 + * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
 + * All rights reserved.
@@ -12125,35 +12128,16 @@ index 0000000..6748719
 +/*-*******************************************
 +*  Shared functions to include for inlining
 +*********************************************/
-+static void ZSTD_copy8(void *dst, const void *src) { memcpy(dst, src, 8); }
++static void ZSTD_copy8(void *dst, const void *src) {
-+#define COPY8(d, s)               \
++	memcpy(dst, src, 8);
-+	{                         \
-+		ZSTD_copy8(d, s); \
-+		d += 8;           \
-+		s += 8;           \
 +}
-+
 +/*! ZSTD_wildcopy() :
 +*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
 +#define WILDCOPY_OVERLENGTH 8
 +ZSTD_STATIC void ZSTD_wildcopy(void *dst, const void *src, ptrdiff_t length)
 +{
-+	const BYTE *ip = (const BYTE *)src;
++	if (length > 0)
-+	BYTE *op = (BYTE *)dst;
++		memcpy(dst, src, length);
-+	BYTE *const oend = op + length;
-+	do
-+		COPY8(op, ip)
-+	while (op < oend);
-+}
-+
-+ZSTD_STATIC void ZSTD_wildcopy_e(void *dst, const void *src, void *dstEnd) /* should be faster for decoding, but strangely, not verified on all platform */
-+{
-+	const BYTE *ip = (const BYTE *)src;
-+	BYTE *op = (BYTE *)dst;
-+	BYTE *const oend = (BYTE *)dstEnd;
-+	do
-+		COPY8(op, ip)
-+	while (op < oend);
 +}
 +
 +/*-*******************************************

contrib/linux-kernel/0003-btrfs-Add-zstd-support.patch

@@ -1,7 +1,7 @@
-From 599f8f2aaace3df939cb145368574a52268d82d0 Mon Sep 17 00:00:00 2001
+From b0ef8fc63c9ca251ceca632f53aa1de8f1f17772 Mon Sep 17 00:00:00 2001
 From: Nick Terrell <terrelln@fb.com>
-Date: Wed, 21 Jun 2017 17:31:39 -0700
+Date: Mon, 17 Jul 2017 17:08:39 -0700
-Subject: [PATCH v2 3/4] btrfs: Add zstd support
+Subject: [PATCH v3 3/4] btrfs: Add zstd support
 
 Add zstd compression and decompression support to BtrFS. zstd at its
 fastest level compresses almost as well as zlib, while offering much
@@ -63,6 +63,10 @@ zstd source repository: https://github.com/facebook/zstd
 
 Signed-off-by: Nick Terrell <terrelln@fb.com>
 ---
+v2 -> v3:
+- Port upstream BtrFS commits e1ddce71d6, 389a6cfc2a, and 6acafd1eff
+- Change default compression level for BtrFS to 3
+
  fs/btrfs/Kconfig           |   2 +
  fs/btrfs/Makefile          |   2 +-
  fs/btrfs/compression.c     |   1 +
@@ -73,9 +77,9 @@ Signed-off-by: Nick Terrell <terrelln@fb.com>
  fs/btrfs/props.c           |   6 +
  fs/btrfs/super.c           |  12 +-
  fs/btrfs/sysfs.c           |   2 +
- fs/btrfs/zstd.c            | 433 +++++++++++++++++++++++++++++++++++++++++++++
+ fs/btrfs/zstd.c            | 435 +++++++++++++++++++++++++++++++++++++++++++++
  include/uapi/linux/btrfs.h |   8 +-
- 12 files changed, 469 insertions(+), 12 deletions(-)
+ 12 files changed, 471 insertions(+), 12 deletions(-)
  create mode 100644 fs/btrfs/zstd.c
 
 diff --git a/fs/btrfs/Kconfig b/fs/btrfs/Kconfig
@@ -105,10 +109,10 @@ index 128ce17..962a95a 100644
  	   reada.o backref.o ulist.o qgroup.o send.o dev-replace.o raid56.o \
  	   uuid-tree.o props.o hash.o free-space-tree.o
 diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
-index 10e6b28..3beb0d0 100644
+index d2ef9ac..4ff42d1 100644
 --- a/fs/btrfs/compression.c
 +++ b/fs/btrfs/compression.c
-@@ -761,6 +761,7 @@ static struct {
+@@ -704,6 +704,7 @@ static struct {
  static const struct btrfs_compress_op * const btrfs_compress_op[] = {
  	&btrfs_zlib_compress,
  	&btrfs_lzo_compress,
@@ -117,10 +121,10 @@ index 10e6b28..3beb0d0 100644
 
  void __init btrfs_init_compress(void)
 diff --git a/fs/btrfs/compression.h b/fs/btrfs/compression.h
-index 39ec43a..d99fc21 100644
+index 87f6d33..2269e00 100644
 --- a/fs/btrfs/compression.h
 +++ b/fs/btrfs/compression.h
-@@ -60,8 +60,9 @@ enum btrfs_compression_type {
+@@ -99,8 +99,9 @@ enum btrfs_compression_type {
  	BTRFS_COMPRESS_NONE  = 0,
  	BTRFS_COMPRESS_ZLIB  = 1,
  	BTRFS_COMPRESS_LZO   = 2,
@@ -132,7 +136,7 @@ index 39ec43a..d99fc21 100644
  };
 
  struct btrfs_compress_op {
-@@ -92,5 +93,6 @@ struct btrfs_compress_op {
+@@ -128,5 +129,6 @@ struct btrfs_compress_op {
 
  extern const struct btrfs_compress_op btrfs_zlib_compress;
  extern const struct btrfs_compress_op btrfs_lzo_compress;
@@ -140,10 +144,10 @@ index 39ec43a..d99fc21 100644
 
  #endif
 diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
-index 4f8f75d..61dd3dd 100644
+index 3f3eb7b..845d77c 100644
 --- a/fs/btrfs/ctree.h
 +++ b/fs/btrfs/ctree.h
-@@ -271,6 +271,7 @@ struct btrfs_super_block {
+@@ -270,6 +270,7 @@ struct btrfs_super_block {
  	 BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS |		\
  	 BTRFS_FEATURE_INCOMPAT_BIG_METADATA |		\
  	 BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO |		\
@@ -152,10 +156,10 @@ index 4f8f75d..61dd3dd 100644
  	 BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF |		\
  	 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA |	\
 diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
-index 5f678dc..49c0e91 100644
+index 080e2eb..04632f4 100644
 --- a/fs/btrfs/disk-io.c
 +++ b/fs/btrfs/disk-io.c
-@@ -2831,6 +2831,8 @@ int open_ctree(struct super_block *sb,
+@@ -2828,6 +2828,8 @@ int open_ctree(struct super_block *sb,
  	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
  	if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
  		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
@@ -165,7 +169,7 @@ index 5f678dc..49c0e91 100644
  	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
  		btrfs_info(fs_info, "has skinny extents");
 diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
-index e176375..f732cfd 100644
+index fa1b78c..b9963d9 100644
 --- a/fs/btrfs/ioctl.c
 +++ b/fs/btrfs/ioctl.c
 @@ -327,8 +327,10 @@ static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
@@ -180,7 +184,7 @@ index e176375..f732cfd 100644
  		ret = btrfs_set_prop(inode, "btrfs.compression",
  				     comp, strlen(comp), 0);
  		if (ret)
-@@ -1463,6 +1465,8 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
+@@ -1466,6 +1468,8 @@ int btrfs_defrag_file(struct inode *inode, struct file *file,
 
  	if (range->compress_type == BTRFS_COMPRESS_LZO) {
  		btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
@@ -190,10 +194,10 @@ index e176375..f732cfd 100644
 
  	ret = defrag_count;
 diff --git a/fs/btrfs/props.c b/fs/btrfs/props.c
-index d6cb155..162105f 100644
+index 4b23ae5..20631e9 100644
 --- a/fs/btrfs/props.c
 +++ b/fs/btrfs/props.c
-@@ -383,6 +383,8 @@ static int prop_compression_validate(const char *value, size_t len)
+@@ -390,6 +390,8 @@ static int prop_compression_validate(const char *value, size_t len)
  		return 0;
  	else if (!strncmp("zlib", value, len))
  		return 0;
@@ -202,7 +206,7 @@ index d6cb155..162105f 100644
 
  	return -EINVAL;
  }
-@@ -405,6 +407,8 @@ static int prop_compression_apply(struct inode *inode,
+@@ -412,6 +414,8 @@ static int prop_compression_apply(struct inode *inode,
  		type = BTRFS_COMPRESS_LZO;
  	else if (!strncmp("zlib", value, len))
  		type = BTRFS_COMPRESS_ZLIB;
@@ -211,7 +215,7 @@ index d6cb155..162105f 100644
  	else
  		return -EINVAL;
 
-@@ -422,6 +426,8 @@ static const char *prop_compression_extract(struct inode *inode)
+@@ -429,6 +433,8 @@ static const char *prop_compression_extract(struct inode *inode)
  		return "zlib";
  	case BTRFS_COMPRESS_LZO:
  		return "lzo";
@@ -221,7 +225,7 @@ index d6cb155..162105f 100644
 
  	return NULL;
 diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
-index 4f1cdd5..4f792d5 100644
+index 12540b6..c370dea 100644
 --- a/fs/btrfs/super.c
 +++ b/fs/btrfs/super.c
 @@ -513,6 +513,14 @@ int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
@@ -239,7 +243,7 @@ index 4f1cdd5..4f792d5 100644
  			} else if (strncmp(args[0].from, "no", 2) == 0) {
  				compress_type = "no";
  				btrfs_clear_opt(info->mount_opt, COMPRESS);
-@@ -1240,8 +1248,10 @@ static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
+@@ -1227,8 +1235,10 @@ static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
  	if (btrfs_test_opt(info, COMPRESS)) {
  		if (info->compress_type == BTRFS_COMPRESS_ZLIB)
  			compress_type = "zlib";
@@ -252,7 +256,7 @@ index 4f1cdd5..4f792d5 100644
  			seq_printf(seq, ",compress-force=%s", compress_type);
  		else
 diff --git a/fs/btrfs/sysfs.c b/fs/btrfs/sysfs.c
-index 1f157fb..b0dec90 100644
+index c2d5f35..2b6d37c 100644
 --- a/fs/btrfs/sysfs.c
 +++ b/fs/btrfs/sysfs.c
 @@ -200,6 +200,7 @@ BTRFS_FEAT_ATTR_INCOMPAT(mixed_backref, MIXED_BACKREF);
@@ -273,10 +277,10 @@ index 1f157fb..b0dec90 100644
  	BTRFS_FEAT_ATTR_PTR(raid56),
 diff --git a/fs/btrfs/zstd.c b/fs/btrfs/zstd.c
 new file mode 100644
-index 0000000..838741b
+index 0000000..1822068
 --- /dev/null
 +++ b/fs/btrfs/zstd.c
-@@ -0,0 +1,433 @@
+@@ -0,0 +1,435 @@
 +/*
 + * Copyright (c) 2016-present, Facebook, Inc.
 + * All rights reserved.
@@ -308,10 +312,11 @@ index 0000000..838741b
 +
 +#define ZSTD_BTRFS_MAX_WINDOWLOG 17
 +#define ZSTD_BTRFS_MAX_INPUT (1 << ZSTD_BTRFS_MAX_WINDOWLOG)
++#define ZSTD_BTRFS_DEFAULT_LEVEL 3
 +
 +static ZSTD_parameters zstd_get_btrfs_parameters(size_t src_len)
 +{
-+	ZSTD_parameters params = ZSTD_getParams(1, src_len, 0);
++	ZSTD_parameters params = ZSTD_getParams(ZSTD_BTRFS_DEFAULT_LEVEL, src_len, 0);
 +
 +	if (params.cParams.windowLog > ZSTD_BTRFS_MAX_WINDOWLOG)
 +		params.cParams.windowLog = ZSTD_BTRFS_MAX_WINDOWLOG;
@@ -330,7 +335,7 @@ index 0000000..838741b
 +{
 +	struct workspace *workspace = list_entry(ws, struct workspace, list);
 +
-+	vfree(workspace->mem);
++	kvfree(workspace->mem);
 +	kfree(workspace->buf);
 +	kfree(workspace);
 +}
@@ -341,15 +346,15 @@ index 0000000..838741b
 +			zstd_get_btrfs_parameters(ZSTD_BTRFS_MAX_INPUT);
 +	struct workspace *workspace;
 +
-+	workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
++	workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
 +	if (!workspace)
 +		return ERR_PTR(-ENOMEM);
 +
 +	workspace->size = max_t(size_t,
 +			ZSTD_CStreamWorkspaceBound(params.cParams),
 +			ZSTD_DStreamWorkspaceBound(ZSTD_BTRFS_MAX_INPUT));
-+	workspace->mem = vmalloc(workspace->size);
++	workspace->mem = kvmalloc(workspace->size, GFP_KERNEL);
-+	workspace->buf = kmalloc(PAGE_SIZE, GFP_NOFS);
++	workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 +	if (!workspace->mem || !workspace->buf)
 +		goto fail;
 +
@@ -541,12 +546,13 @@ index 0000000..838741b
 +	return ret;
 +}
 +
-+static int zstd_decompress_bio(struct list_head *ws, struct page **pages_in,
++static int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
-+		u64 disk_start,
-+		struct bio *orig_bio,
-+		size_t srclen)
 +{
 +	struct workspace *workspace = list_entry(ws, struct workspace, list);
++	struct page **pages_in = cb->compressed_pages;
++	u64 disk_start = cb->start;
++	struct bio *orig_bio = cb->orig_bio;
++	size_t srclen = cb->compressed_len;
 +	ZSTD_DStream *stream;
 +	int ret = 0;
 +	unsigned long page_in_index = 0;
@@ -711,7 +717,7 @@ index 0000000..838741b
 +	.decompress = zstd_decompress,
 +};
 diff --git a/include/uapi/linux/btrfs.h b/include/uapi/linux/btrfs.h
-index a456e53..992c150 100644
+index 9aa74f3..378230c 100644
 --- a/include/uapi/linux/btrfs.h
 +++ b/include/uapi/linux/btrfs.h
 @@ -255,13 +255,7 @@ struct btrfs_ioctl_fs_info_args {

contrib/linux-kernel/0004-squashfs-Add-zstd-support.patch

@@ -1,7 +1,7 @@
-From 5ff6a64abaea7b7f11d37cb0fdf08642316a3a90 Mon Sep 17 00:00:00 2001
+From 0cd63464d182bb9708f8b25f7da3dc8e5ec6b4fa Mon Sep 17 00:00:00 2001
 From: Nick Terrell <terrelln@fb.com>
-Date: Mon, 12 Jun 2017 12:18:23 -0700
+Date: Mon, 17 Jul 2017 17:08:59 -0700
-Subject: [PATCH v2 4/4] squashfs: Add zstd support
+Subject: [PATCH v3 4/4] squashfs: Add zstd support
 
 Add zstd compression and decompression support to SquashFS. zstd is a
 great fit for SquashFS because it can compress at ratios approaching xz,

contrib/linux-kernel/fs/btrfs/zstd.c

@@ -29,10 +29,11 @@
 
 #define ZSTD_BTRFS_MAX_WINDOWLOG 17
 #define ZSTD_BTRFS_MAX_INPUT (1 << ZSTD_BTRFS_MAX_WINDOWLOG)
+#define ZSTD_BTRFS_DEFAULT_LEVEL 3
 
 static ZSTD_parameters zstd_get_btrfs_parameters(size_t src_len)
 {
-	ZSTD_parameters params = ZSTD_getParams(1, src_len, 0);
+	ZSTD_parameters params = ZSTD_getParams(ZSTD_BTRFS_DEFAULT_LEVEL, src_len, 0);
 
 	if (params.cParams.windowLog > ZSTD_BTRFS_MAX_WINDOWLOG)
 		params.cParams.windowLog = ZSTD_BTRFS_MAX_WINDOWLOG;
@@ -51,7 +52,7 @@ static void zstd_free_workspace(struct list_head *ws)
 {
 	struct workspace *workspace = list_entry(ws, struct workspace, list);
 
-	vfree(workspace->mem);
+	kvfree(workspace->mem);
 	kfree(workspace->buf);
 	kfree(workspace);
 }
@@ -62,15 +63,15 @@ static struct list_head *zstd_alloc_workspace(void)
 			zstd_get_btrfs_parameters(ZSTD_BTRFS_MAX_INPUT);
 	struct workspace *workspace;
 
-	workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
+	workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
 	if (!workspace)
 		return ERR_PTR(-ENOMEM);
 
 	workspace->size = max_t(size_t,
 			ZSTD_CStreamWorkspaceBound(params.cParams),
 			ZSTD_DStreamWorkspaceBound(ZSTD_BTRFS_MAX_INPUT));
-	workspace->mem = vmalloc(workspace->size);
+	workspace->mem = kvmalloc(workspace->size, GFP_KERNEL);
-	workspace->buf = kmalloc(PAGE_SIZE, GFP_NOFS);
+	workspace->buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 	if (!workspace->mem || !workspace->buf)
 		goto fail;
 
@@ -262,12 +263,13 @@ out:
 	return ret;
 }
 
-static int zstd_decompress_bio(struct list_head *ws, struct page **pages_in,
+static int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
-		u64 disk_start,
-		struct bio *orig_bio,
-		size_t srclen)
 {
 	struct workspace *workspace = list_entry(ws, struct workspace, list);
+	struct page **pages_in = cb->compressed_pages;
+	u64 disk_start = cb->start;
+	struct bio *orig_bio = cb->orig_bio;
+	size_t srclen = cb->compressed_len;
 	ZSTD_DStream *stream;
 	int ret = 0;
 	unsigned long page_in_index = 0;

contrib/linux-kernel/lib/zstd/huf_compress.c

@@ -525,7 +525,7 @@ static void HUF_encodeSymbol(BIT_CStream_t *bitCPtr, U32 symbol, const HUF_CElt
 
 size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
 
-#define HUF_FLUSHBITS(s) (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+#define HUF_FLUSHBITS(s)  BIT_flushBits(s)
 
 #define HUF_FLUSHBITS_1(stream)                                            \
 	if (sizeof((stream)->bitContainer) * 8 < HUF_TABLELOG_MAX * 2 + 7) \
@@ -542,7 +542,6 @@ size_t HUF_compress1X_usingCTable(void *dst, size_t dstSize, const void *src, si
 	BYTE *const oend = ostart + dstSize;
 	BYTE *op = ostart;
 	size_t n;
-	const unsigned fast = (dstSize >= HUF_BLOCKBOUND(srcSize));
 	BIT_CStream_t bitC;
 
 	/* init */

contrib/linux-kernel/lib/zstd/zstd_internal.h

@@ -126,35 +126,16 @@ static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
 /*-*******************************************
 *  Shared functions to include for inlining
 *********************************************/
-static void ZSTD_copy8(void *dst, const void *src) { memcpy(dst, src, 8); }
+static void ZSTD_copy8(void *dst, const void *src) {
-#define COPY8(d, s)               \
+	memcpy(dst, src, 8);
-	{                         \
-		ZSTD_copy8(d, s); \
-		d += 8;           \
-		s += 8;           \
 }
-
 /*! ZSTD_wildcopy() :
 *   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
 #define WILDCOPY_OVERLENGTH 8
 ZSTD_STATIC void ZSTD_wildcopy(void *dst, const void *src, ptrdiff_t length)
 {
-	const BYTE *ip = (const BYTE *)src;
+	if (length > 0)
-	BYTE *op = (BYTE *)dst;
+		memcpy(dst, src, length);
-	BYTE *const oend = op + length;
-	do
-		COPY8(op, ip)
-	while (op < oend);
-}
-
-ZSTD_STATIC void ZSTD_wildcopy_e(void *dst, const void *src, void *dstEnd) /* should be faster for decoding, but strangely, not verified on all platform */
-{
-	const BYTE *ip = (const BYTE *)src;
-	BYTE *op = (BYTE *)dst;
-	BYTE *const oend = (BYTE *)dstEnd;
-	do
-		COPY8(op, ip)
-	while (op < oend);
 }
 
 /*-*******************************************

contrib/long_distance_matching/Makefile

@@ -0,0 +1,37 @@
+# ################################################################
+# Copyright (c) 2016-present, Facebook, Inc.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree. An additional grant
+# of patent rights can be found in the PATENTS file in the same directory.
+# ################################################################
+
+# This Makefile presumes libzstd is installed, using `sudo make install`
+
+CPPFLAGS+= -I../../lib/common
+CFLAGS  ?= -O3
+DEBUGFLAGS = -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
+            -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
+            -Wstrict-prototypes -Wundef -Wpointer-arith -Wformat-security \
+            -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
+            -Wredundant-decls
+CFLAGS  += $(DEBUGFLAGS) $(MOREFLAGS)
+FLAGS    = $(CPPFLAGS) $(CFLAGS)
+
+LDFLAGS += -lzstd
+
+.PHONY: default all clean
+
+default: all
+
+all: ldm 
+	
+ldm: ldm_common.c ldm.c main.c
+	$(CC) $(CPPFLAGS) $(CFLAGS) $^ $(LDFLAGS) -o $@
+
+clean:
+	@rm -f core *.o tmp* result* *.ldm *.ldm.dec \
+	ldm
+	@echo Cleaning completed
+

contrib/long_distance_matching/README.md

@@ -0,0 +1,102 @@
+This is a compression algorithm focused on finding long distance matches.
+
+It is based upon lz4 and uses nearly the same block format (github.com/lz4/lz4/blob/dev/doc/lz4_Block_format.md). The number of bytes to encode the offset is four instead of two in lz4 to reflect the longer distance matching. The block format is described in `ldm.h`.
+
+### Build
+
+Run `make`.
+
+### Compressing a file
+
+`ldm <filename>`
+
+Decompression and verification can be enabled by defining `DECOMPRESS_AND_VERIFY` in `main.c`.
+The output file names are as follows:
+- `<filename>.ldm` : compressed file
+- `<filename>.ldm.dec` : decompressed file
+
+### Parameters
+
+There are various parameters that can be tuned. These parameters can be tuned in `ldm.h` or, alternatively if `ldm_params.h` is included, in `ldm_params.h` (for easier configuration).
+
+The parameters are as follows and must all be defined:
+- `LDM_MEMORY_USAGE` : the memory usage of the underlying hash table in bytes.
+- `HASH_BUCKET_SIZE_LOG` : the log size of each bucket in the hash table (used in collision resolution).
+- `LDM_LAG` : the lag (in bytes) in inserting entries into the hash table.
+- `LDM_WINDOW_SIZE_LOG` : the log maximum window size when searching for matches.
+- `LDM_MIN_MATCH_LENGTH` : the minimum match length.
+- `INSERT_BY_TAG` : insert entries into the hash table as a function of the hash. This increases speed by reducing the number of hash table lookups and match comparisons. Certain hashes will never be inserted.
+- `USE_CHECKSUM`  : store a checksum with the hash table entries for faster comparison. This halves the number of entries the hash table can contain.
+
+The optional parameter `HASH_ONLY_EVERY_LOG` is the log inverse frequency of insertion into the hash table. That is, an entry is inserted approximately every `1 << HASH_ONLY_EVERY_LOG` times. If this parameter is not defined, the value is computed as a function of the window size and memory usage to approximate an even coverage of the window.
+
+
+### Benchmark
+
+Below is a comparison of various compression methods on a tar of four versions of llvm (versions `3.9.0`, `3.9.1`, `4.0.0`, `4.0.1`) with a total size of `727900160` B.
+
+| Method | Size | Ratio |
+|:---|---:|---:|
+|lrzip -p 32 -n -w 1 | `369968714` | `1.97`|
+|ldm | `209391361` | `3.48`|
+|lz4 | `189954338` | `3.83`|
+|lrzip -p 32 -l -w 1 | `163940343` | `4.44`|
+|zstd -1 | `126080293` | `5.77`|
+|lrzip -p 32 -n | `124821009` | `5.83`|
+|lrzip -p 32 -n -w 1 & zstd -1 | `120317909` | `6.05`|
+|zstd -3 -o | `115290952` | `6.31`|
+|lrzip -p 32 -g -L 9 -w 1 | `107168979` | `6.79`|
+|zstd -6 -o | `102772098` | `7.08`|
+|zstd -T16 -9 | `98040470` | `7.42`|
+|lrzip -p 32 -n -w 1 & zstd -T32 -19 | `88050289` | `8.27`|
+|zstd -T32 -19 | `83626098` | `8.70`|
+|lrzip -p 32 -n & zstd -1 | `36335117` | `20.03`|
+|ldm & zstd -6 | `32856232` | `22.15`|
+|lrzip -p 32 -g -L 9 | `32243594` | `22.58`|
+|lrzip -p 32 -n & zstd -6 | `30954572` | `23.52`|
+|lrzip -p 32 -n & zstd -T32 -19 | `26472064` | `27.50`|
+
+The method marked `ldm` was run with the following parameters:
+
+| Parameter | Value |
+|:---|---:|
+| `LDM_MEMORY_USAGE`    |   `23`|
+|`HASH_BUCKET_SIZE_LOG` |    `3`|
+|`LDM_LAG`              |    `0`|
+|`LDM_WINDOW_SIZE_LOG`  |   `28`|
+|`LDM_MIN_MATCH_LENGTH`|   `64`|
+|`INSERT_BY_TAG`        |    `1`|
+|`USE_CHECKSUM`         |    `1`|
+
+The compression speed was `220.5 MB/s`.
+
+### Parameter selection
+
+Below is a brief discussion of the effects of the parameters on the speed and compression ratio.
+
+#### Speed
+
+A large bottleneck in terms of speed is finding the matches and comparing to see if they are greater than the minimum match length. Generally:
+- The fewer matches found (or the lower the percentage of the literals matched), the slower the algorithm will behave.
+- Increasing `HASH_ONLY_EVERY_LOG` results in fewer inserts and, if `INSERT_BY_TAG` is set, fewer lookups in the table. This has a large effect on speed, as well as compression ratio.
+- If `HASH_ONLY_EVERY_LOG` is not set, its value is calculated based on `LDM_WINDOW_SIZE_LOG` and `LDM_MEMORY_USAGE`. Increasing `LDM_WINDOW_SIZE_LOG` has the effect of increasing `HASH_ONLY_EVERY_LOG` and increasing `LDM_MEMORY_USAGE` decreases `HASH_ONLY_EVERY_LOG`.
+- `USE_CHECKSUM` generally improves speed with hash table lookups.
+
+#### Compression ratio
+
+The compression ratio is highly correlated with the coverage of matches. As a long distance matcher, the algorithm was designed to "optimize" for long distance matches outside the zstd compression window. The compression ratio after recompressing the output of the long-distance matcher with zstd was a more important signal in development than the raw compression ratio itself.
+
+Generally, increasing `LDM_MEMORY_USAGE` will improve the compression ratio. However when using the default computed value of `HASH_ONLY_EVERY_LOG`, this increases the frequency of insertion and lookup in the table and thus may result in a decrease in speed. 
+
+Below is a table showing the speed and compression ratio when compressing the llvm tar (as described above) using different settings for `LDM_MEMORY_USAGE`. The other parameters were the same as used in the benchmark above.
+
+| `LDM_MEMORY_USAGE` | Ratio | Speed (MB/s) | Ratio after zstd -6  |
+|---:| ---: | ---: | ---: |
+| `18` | `1.85` | `232.4` | `10.92` |
+| `21` | `2.79` | `233.9` | `15.92` |
+| `23` | `3.48` | `220.5` | `18.29` |
+| `25` | `4.56` | `140.8` | `19.21` |
+
+### Compression statistics
+
+Compression statistics (and the configuration) can be enabled/disabled via `COMPUTE_STATS` and `OUTPUT_CONFIGURATION` in `ldm.h`.

contrib/long_distance_matching/ldm.c

@@ -0,0 +1,857 @@
+#include <limits.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "ldm.h"
+
+#define LDM_HASHTABLESIZE (1 << (LDM_MEMORY_USAGE))
+#define LDM_HASHTABLESIZE_U32 ((LDM_HASHTABLESIZE) >> 2)
+#define LDM_HASHTABLESIZE_U64 ((LDM_HASHTABLESIZE) >> 3)
+
+#if USE_CHECKSUM
+  #define LDM_HASH_ENTRY_SIZE_LOG 3
+#else
+  #define LDM_HASH_ENTRY_SIZE_LOG 2
+#endif
+
+// Entries are inserted into the table HASH_ONLY_EVERY + 1 times "on average".
+#ifndef HASH_ONLY_EVERY_LOG
+  #define HASH_ONLY_EVERY_LOG (LDM_WINDOW_SIZE_LOG-((LDM_MEMORY_USAGE)-(LDM_HASH_ENTRY_SIZE_LOG)))
+#endif
+
+#define HASH_ONLY_EVERY ((1 << (HASH_ONLY_EVERY_LOG)) - 1)
+
+#define HASH_BUCKET_SIZE (1 << (HASH_BUCKET_SIZE_LOG))
+#define NUM_HASH_BUCKETS_LOG ((LDM_MEMORY_USAGE)-(LDM_HASH_ENTRY_SIZE_LOG)-(HASH_BUCKET_SIZE_LOG))
+
+#define HASH_CHAR_OFFSET 10
+
+// Take the first match in the hash bucket only.
+//#define ZSTD_SKIP
+
+static const U64 prime8bytes = 11400714785074694791ULL;
+
+// Type of the small hash used to index into the hash table.
+typedef U32 hash_t;
+
+#if USE_CHECKSUM
+typedef struct LDM_hashEntry {
+  U32 offset;
+  U32 checksum;
+} LDM_hashEntry;
+#else
+typedef struct LDM_hashEntry {
+  U32 offset;
+} LDM_hashEntry;
+#endif
+
+struct LDM_compressStats {
+  U32 windowSizeLog, hashTableSizeLog;
+  U32 numMatches;
+  U64 totalMatchLength;
+  U64 totalLiteralLength;
+  U64 totalOffset;
+
+  U32 matchLengthHistogram[32];
+
+  U32 minOffset, maxOffset;
+  U32 offsetHistogram[32];
+};
+
+typedef struct LDM_hashTable LDM_hashTable;
+
+struct LDM_CCtx {
+  size_t isize;             /* Input size */
+  size_t maxOSize;          /* Maximum output size */
+
+  const BYTE *ibase;        /* Base of input */
+  const BYTE *ip;           /* Current input position */
+  const BYTE *iend;         /* End of input */
+
+  // Maximum input position such that hashing at the position does not exceed
+  // end of input.
+  const BYTE *ihashLimit;
+
+  // Maximum input position such that finding a match of at least the minimum
+  // match length does not exceed end of input.
+  const BYTE *imatchLimit;
+
+  const BYTE *obase;        /* Base of output */
+  BYTE *op;                 /* Output */
+
+  const BYTE *anchor;       /* Anchor to start of current (match) block */
+
+  LDM_compressStats stats;            /* Compression statistics */
+
+  LDM_hashTable *hashTable;
+
+  const BYTE *lastPosHashed;          /* Last position hashed */
+  U64 lastHash;
+
+  const BYTE *nextIp;                 // TODO: this is redundant (ip + step)
+  const BYTE *nextPosHashed;
+  U64 nextHash;
+
+  unsigned step;                      // ip step, should be 1.
+
+  const BYTE *lagIp;
+  U64 lagHash;
+};
+
+struct LDM_hashTable {
+  U32 numBuckets;          // The number of buckets.
+  U32 numEntries;          // numBuckets * HASH_BUCKET_SIZE.
+
+  LDM_hashEntry *entries;
+  BYTE *bucketOffsets;     // A pointer (per bucket) to the next insert position.
+};
+
+static void HASH_destroyTable(LDM_hashTable *table) {
+  free(table->entries);
+  free(table->bucketOffsets);
+  free(table);
+}
+
+/**
+ * Create a hash table that can contain size elements.
+ * The number of buckets is determined by size >> HASH_BUCKET_SIZE_LOG.
+ *
+ * Returns NULL if table creation failed.
+ */
+static LDM_hashTable *HASH_createTable(U32 size) {
+  LDM_hashTable *table = malloc(sizeof(LDM_hashTable));
+  if (!table) return NULL;
+
+  table->numBuckets = size >> HASH_BUCKET_SIZE_LOG;
+  table->numEntries = size;
+  table->entries = calloc(size, sizeof(LDM_hashEntry));
+  table->bucketOffsets = calloc(size >> HASH_BUCKET_SIZE_LOG, sizeof(BYTE));
+
+  if (!table->entries || !table->bucketOffsets) {
+    HASH_destroyTable(table);
+    return NULL;
+  }
+
+  return table;
+}
+
+static LDM_hashEntry *getBucket(const LDM_hashTable *table, const hash_t hash) {
+  return table->entries + (hash << HASH_BUCKET_SIZE_LOG);
+}
+
+static unsigned ZSTD_NbCommonBytes (register size_t val) {
+  if (MEM_isLittleEndian()) {
+    if (MEM_64bits()) {
+#    if defined(_MSC_VER) && defined(_WIN64)
+      unsigned long r = 0;
+      _BitScanForward64( &r, (U64)val );
+      return (unsigned)(r>>3);
+#     elif defined(__GNUC__) && (__GNUC__ >= 3)
+      return (__builtin_ctzll((U64)val) >> 3);
+#     else
+      static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
+                                               0, 3, 1, 3, 1, 4, 2, 7,
+                                               0, 2, 3, 6, 1, 5, 3, 5,
+                                               1, 3, 4, 4, 2, 5, 6, 7,
+                                               7, 0, 1, 2, 3, 3, 4, 6,
+                                               2, 6, 5, 5, 3, 4, 5, 6,
+                                               7, 1, 2, 4, 6, 4, 4, 5,
+                                               7, 2, 6, 5, 7, 6, 7, 7 };
+      return DeBruijnBytePos[
+          ((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
+#     endif
+  } else { /* 32 bits */
+#     if defined(_MSC_VER)
+      unsigned long r=0;
+      _BitScanForward( &r, (U32)val );
+      return (unsigned)(r>>3);
+#     elif defined(__GNUC__) && (__GNUC__ >= 3)
+      return (__builtin_ctz((U32)val) >> 3);
+#     else
+      static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
+                                               3, 2, 2, 1, 3, 2, 0, 1,
+                                               3, 3, 1, 2, 2, 2, 2, 0,
+                                               3, 1, 2, 0, 1, 0, 1, 1 };
+      return DeBruijnBytePos[
+          ((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
+#     endif
+    }
+  } else {  /* Big Endian CPU */
+    if (MEM_64bits()) {
+#     if defined(_MSC_VER) && defined(_WIN64)
+      unsigned long r = 0;
+      _BitScanReverse64( &r, val );
+      return (unsigned)(r>>3);
+#     elif defined(__GNUC__) && (__GNUC__ >= 3)
+      return (__builtin_clzll(val) >> 3);
+#     else
+      unsigned r;
+      /* calculate this way due to compiler complaining in 32-bits mode */
+      const unsigned n32 = sizeof(size_t)*4;
+      if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
+      if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+      r += (!val);
+      return r;
+#       endif
+    } else { /* 32 bits */
+#     if defined(_MSC_VER)
+      unsigned long r = 0;
+      _BitScanReverse( &r, (unsigned long)val );
+      return (unsigned)(r>>3);
+#     elif defined(__GNUC__) && (__GNUC__ >= 3)
+      return (__builtin_clz((U32)val) >> 3);
+#     else
+      unsigned r;
+      if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
+      r += (!val);
+      return r;
+#     endif
+    }
+  }
+}
+
+// From lib/compress/zstd_compress.c
+static size_t ZSTD_count(const BYTE *pIn, const BYTE *pMatch,
+                         const BYTE *const pInLimit) {
+    const BYTE * const pStart = pIn;
+    const BYTE * const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
+
+    while (pIn < pInLoopLimit) {
+        size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+        if (!diff) {
+          pIn += sizeof(size_t);
+          pMatch += sizeof(size_t);
+          continue;
+        }
+        pIn += ZSTD_NbCommonBytes(diff);
+        return (size_t)(pIn - pStart);
+    }
+
+    if (MEM_64bits()) {
+      if ((pIn < (pInLimit - 3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) {
+        pIn += 4;
+        pMatch += 4;
+      }
+    }
+    if ((pIn < (pInLimit - 1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) {
+      pIn += 2;
+      pMatch += 2;
+    }
+    if ((pIn < pInLimit) && (*pMatch == *pIn)) {
+      pIn++;
+    }
+    return (size_t)(pIn - pStart);
+}
+
+/**
+ * Count number of bytes that match backwards before pIn and pMatch.
+ *
+ * We count only bytes where pMatch > pBase and pIn > pAnchor.
+ */
+static size_t countBackwardsMatch(const BYTE *pIn, const BYTE *pAnchor,
+                                  const BYTE *pMatch, const BYTE *pBase) {
+  size_t matchLength = 0;
+  while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
+    pIn--;
+    pMatch--;
+    matchLength++;
+  }
+  return matchLength;
+}
+
+/**
+ * Returns a pointer to the entry in the hash table matching the hash and
+ * checksum with the "longest match length" as defined below. The forward and
+ * backward match lengths are written to *pForwardMatchLength and
+ * *pBackwardMatchLength.
+ *
+ * The match length is defined based on cctx->ip and the entry's offset.
+ * The forward match is computed from cctx->ip and entry->offset + cctx->ibase.
+ * The backward match is computed backwards from cctx->ip and
+ * cctx->ibase only if the forward match is longer than LDM_MIN_MATCH_LENGTH.
+ */
+static LDM_hashEntry *HASH_getBestEntry(const LDM_CCtx *cctx,
+                                        const hash_t hash,
+                                        const U32 checksum,
+                                        U64 *pForwardMatchLength,
+                                        U64 *pBackwardMatchLength) {
+  LDM_hashTable *table = cctx->hashTable;
+  LDM_hashEntry *bucket = getBucket(table, hash);
+  LDM_hashEntry *cur;
+  LDM_hashEntry *bestEntry = NULL;
+  U64 bestMatchLength = 0;
+#if !(USE_CHECKSUM)
+  (void)checksum;
+#endif
+  for (cur = bucket; cur < bucket + HASH_BUCKET_SIZE; ++cur) {
+    const BYTE *pMatch = cur->offset + cctx->ibase;
+
+    // Check checksum for faster check.
+#if USE_CHECKSUM
+    if (cur->checksum == checksum &&
+        cctx->ip - pMatch <= LDM_WINDOW_SIZE) {
+#else
+    if (cctx->ip - pMatch <= LDM_WINDOW_SIZE) {
+#endif
+      U64 forwardMatchLength = ZSTD_count(cctx->ip, pMatch, cctx->iend);
+      U64 backwardMatchLength, totalMatchLength;
+
+      // Only take matches where the forward match length is large enough
+      // for speed.
+      if (forwardMatchLength < LDM_MIN_MATCH_LENGTH) {
+        continue;
+      }
+
+      backwardMatchLength =
+          countBackwardsMatch(cctx->ip, cctx->anchor,
+                              cur->offset + cctx->ibase,
+                              cctx->ibase);
+
+      totalMatchLength = forwardMatchLength + backwardMatchLength;
+
+      if (totalMatchLength >= bestMatchLength) {
+        bestMatchLength = totalMatchLength;
+        *pForwardMatchLength = forwardMatchLength;
+        *pBackwardMatchLength = backwardMatchLength;
+
+        bestEntry = cur;
+#ifdef ZSTD_SKIP
+        return cur;
+#endif
+      }
+    }
+  }
+  if (bestEntry != NULL) {
+    return bestEntry;
+  }
+  return NULL;
+}
+
+/**
+ * Insert an entry into the hash table. The table uses a "circular buffer",
+ * with the oldest entry overwritten.
+ */
+static void HASH_insert(LDM_hashTable *table,
+                        const hash_t hash, const LDM_hashEntry entry) {
+  *(getBucket(table, hash) + table->bucketOffsets[hash]) = entry;
+  table->bucketOffsets[hash]++;
+  table->bucketOffsets[hash] &= HASH_BUCKET_SIZE - 1;
+}
+
+static void HASH_outputTableOccupancy(const LDM_hashTable *table) {
+  U32 ctr = 0;
+  LDM_hashEntry *cur = table->entries;
+  LDM_hashEntry *end = table->entries + (table->numBuckets * HASH_BUCKET_SIZE);
+  for (; cur < end; ++cur) {
+    if (cur->offset == 0) {
+      ctr++;
+    }
+  }
+
+  // The number of buckets is repeated as a check for now.
+  printf("Num buckets, bucket size: %d (2^%d), %d\n",
+         table->numBuckets, NUM_HASH_BUCKETS_LOG, HASH_BUCKET_SIZE);
+  printf("Hash table size, empty slots, %% empty: %u, %u, %.3f\n",
+         table->numEntries, ctr,
+         100.0 * (double)(ctr) / table->numEntries);
+}
+
+// TODO: This can be done more efficiently, for example by using builtin
+// functions (but it is not that important as it is only used for computing
+// stats).
+static int intLog2(U64 x) {
+  int ret = 0;
+  while (x >>= 1) {
+    ret++;
+  }
+  return ret;
+}
+
+void LDM_printCompressStats(const LDM_compressStats *stats) {
+  printf("=====================\n");
+  printf("Compression statistics\n");
+  printf("Window size, hash table size (bytes): 2^%u, 2^%u\n",
+          stats->windowSizeLog, stats->hashTableSizeLog);
+  printf("num matches, total match length, %% matched: %u, %llu, %.3f\n",
+          stats->numMatches,
+          stats->totalMatchLength,
+          100.0 * (double)stats->totalMatchLength /
+              (double)(stats->totalMatchLength + stats->totalLiteralLength));
+  printf("avg match length: %.1f\n", ((double)stats->totalMatchLength) /
+                                         (double)stats->numMatches);
+  printf("avg literal length, total literalLength: %.1f, %llu\n",
+         ((double)stats->totalLiteralLength) / (double)stats->numMatches,
+         stats->totalLiteralLength);
+  printf("avg offset length: %.1f\n",
+         ((double)stats->totalOffset) / (double)stats->numMatches);
+  printf("min offset, max offset: %u, %u\n",
+         stats->minOffset, stats->maxOffset);
+
+  printf("\n");
+  printf("offset histogram | match length histogram\n");
+  printf("offset/ML, num matches, %% of matches | num matches, %% of matches\n");
+
+  {
+    int i;
+    int logMaxOffset = intLog2(stats->maxOffset);
+    for (i = 0; i <= logMaxOffset; i++) {
+      printf("2^%*d: %10u    %6.3f%% |2^%*d:  %10u    %6.3f \n",
+             2, i,
+             stats->offsetHistogram[i],
+             100.0 * (double) stats->offsetHistogram[i] /
+                     (double) stats->numMatches,
+             2, i,
+             stats->matchLengthHistogram[i],
+             100.0 * (double) stats->matchLengthHistogram[i] /
+                     (double) stats->numMatches);
+    }
+  }
+  printf("\n");
+  printf("=====================\n");
+}
+
+/**
+ * Return the upper (most significant) NUM_HASH_BUCKETS_LOG bits.
+ */
+static hash_t getSmallHash(U64 hash) {
+  return hash >> (64 - NUM_HASH_BUCKETS_LOG);
+}
+
+/**
+ * Return the 32 bits after the upper NUM_HASH_BUCKETS_LOG bits.
+ */
+static U32 getChecksum(U64 hash) {
+  return (hash >> (64 - 32 - NUM_HASH_BUCKETS_LOG)) & 0xFFFFFFFF;
+}
+
+#if INSERT_BY_TAG
+static U32 lowerBitsFromHfHash(U64 hash) {
+  // The number of bits used so far is NUM_HASH_BUCKETS_LOG + 32.
+  // So there are 32 - NUM_HASH_BUCKETS_LOG bits left.
+  // Occasional hashing requires HASH_ONLY_EVERY_LOG bits.
+  // So if 32 - LDMHASHLOG < HASH_ONLY_EVERY_LOG, just return lower bits
+  // allowing for reuse of bits.
+  if (32 - NUM_HASH_BUCKETS_LOG < HASH_ONLY_EVERY_LOG) {
+    return hash & HASH_ONLY_EVERY;
+  } else {
+    // Otherwise shift by
+    // (32 - NUM_HASH_BUCKETS_LOG - HASH_ONLY_EVERY_LOG) bits first.
+    return (hash >> (32 - NUM_HASH_BUCKETS_LOG - HASH_ONLY_EVERY_LOG)) &
+           HASH_ONLY_EVERY;
+  }
+}
+#endif
+
+/**
+ * Get a 64-bit hash using the first len bytes from buf.
+ *
+ * Giving bytes s = s_1, s_2, ... s_k, the hash is defined to be
+ * H(s) = s_1*(a^(k-1)) + s_2*(a^(k-2)) + ... + s_k*(a^0)
+ *
+ * where the constant a is defined to be prime8bytes.
+ *
+ * The implementation adds an offset to each byte, so
+ * H(s) = (s_1 + HASH_CHAR_OFFSET)*(a^(k-1)) + ...
+ */
+static U64 getHash(const BYTE *buf, U32 len) {
+  U64 ret = 0;
+  U32 i;
+  for (i = 0; i < len; i++) {
+    ret *= prime8bytes;
+    ret += buf[i] + HASH_CHAR_OFFSET;
+  }
+  return ret;
+
+}
+
+static U64 ipow(U64 base, U64 exp) {
+  U64 ret = 1;
+  while (exp) {
+    if (exp & 1) {
+      ret *= base;
+    }
+    exp >>= 1;
+    base *= base;
+  }
+  return ret;
+}
+
+static U64 updateHash(U64 hash, U32 len,
+                      BYTE toRemove, BYTE toAdd) {
+  // TODO: this relies on compiler optimization.
+  // The exponential can be calculated explicitly as len is constant.
+  hash -= ((toRemove + HASH_CHAR_OFFSET) *
+          ipow(prime8bytes, len - 1));
+  hash *= prime8bytes;
+  hash += toAdd + HASH_CHAR_OFFSET;
+  return hash;
+}
+
+/**
+ * Update cctx->nextHash and cctx->nextPosHashed
+ * based on cctx->lastHash and cctx->lastPosHashed.
+ *
+ * This uses a rolling hash and requires that the last position hashed
+ * corresponds to cctx->nextIp - step.
+ */
+static void setNextHash(LDM_CCtx *cctx) {
+  cctx->nextHash = updateHash(
+      cctx->lastHash, LDM_HASH_LENGTH,
+      cctx->lastPosHashed[0],
+      cctx->lastPosHashed[LDM_HASH_LENGTH]);
+  cctx->nextPosHashed = cctx->nextIp;
+
+#if LDM_LAG
+  if (cctx->ip - cctx->ibase > LDM_LAG) {
+    cctx->lagHash = updateHash(
+      cctx->lagHash, LDM_HASH_LENGTH,
+      cctx->lagIp[0], cctx->lagIp[LDM_HASH_LENGTH]);
+    cctx->lagIp++;
+  }
+#endif
+}
+
+static void putHashOfCurrentPositionFromHash(LDM_CCtx *cctx, U64 hash) {
+  // Hash only every HASH_ONLY_EVERY times, based on cctx->ip.
+  // Note: this works only when cctx->step is 1.
+#if LDM_LAG
+  if (cctx -> lagIp - cctx->ibase > 0) {
+#if INSERT_BY_TAG
+    U32 hashEveryMask = lowerBitsFromHfHash(cctx->lagHash);
+    if (hashEveryMask == HASH_ONLY_EVERY) {
+#else
+    if (((cctx->ip - cctx->ibase) & HASH_ONLY_EVERY) == HASH_ONLY_EVERY) {
+#endif
+      U32 smallHash = getSmallHash(cctx->lagHash);
+
+#   if USE_CHECKSUM
+      U32 checksum = getChecksum(cctx->lagHash);
+      const LDM_hashEntry entry = { cctx->lagIp - cctx->ibase, checksum };
+#   else
+      const LDM_hashEntry entry = { cctx->lagIp - cctx->ibase };
+#   endif
+
+      HASH_insert(cctx->hashTable, smallHash, entry);
+    }
+  } else {
+#endif // LDM_LAG
+#if INSERT_BY_TAG
+    U32 hashEveryMask = lowerBitsFromHfHash(hash);
+    if (hashEveryMask == HASH_ONLY_EVERY) {
+#else
+    if (((cctx->ip - cctx->ibase) & HASH_ONLY_EVERY) == HASH_ONLY_EVERY) {
+#endif
+      U32 smallHash = getSmallHash(hash);
+
+#if USE_CHECKSUM
+      U32 checksum = getChecksum(hash);
+      const LDM_hashEntry entry = { cctx->ip - cctx->ibase, checksum };
+#else
+      const LDM_hashEntry entry = { cctx->ip - cctx->ibase };
+#endif
+      HASH_insert(cctx->hashTable, smallHash, entry);
+    }
+#if LDM_LAG
+  }
+#endif
+
+  cctx->lastPosHashed = cctx->ip;
+  cctx->lastHash = hash;
+}
+
+/**
+ * Copy over the cctx->lastHash, and cctx->lastPosHashed
+ * fields from the "next" fields.
+ *
+ * This requires that cctx->ip == cctx->nextPosHashed.
+ */
+static void LDM_updateLastHashFromNextHash(LDM_CCtx *cctx) {
+  putHashOfCurrentPositionFromHash(cctx, cctx->nextHash);
+}
+
+/**
+ * Insert hash of the current position into the hash table.
+ */
+static void LDM_putHashOfCurrentPosition(LDM_CCtx *cctx) {
+  U64 hash = getHash(cctx->ip, LDM_HASH_LENGTH);
+
+  putHashOfCurrentPositionFromHash(cctx, hash);
+}
+
+size_t LDM_initializeCCtx(LDM_CCtx *cctx,
+                          const void *src, size_t srcSize,
+                          void *dst, size_t maxDstSize) {
+  cctx->isize = srcSize;
+  cctx->maxOSize = maxDstSize;
+
+  cctx->ibase = (const BYTE *)src;
+  cctx->ip = cctx->ibase;
+  cctx->iend = cctx->ibase + srcSize;
+
+  cctx->ihashLimit = cctx->iend - LDM_HASH_LENGTH;
+  cctx->imatchLimit = cctx->iend - LDM_MIN_MATCH_LENGTH;
+
+  cctx->obase = (BYTE *)dst;
+  cctx->op = (BYTE *)dst;
+
+  cctx->anchor = cctx->ibase;
+
+  memset(&(cctx->stats), 0, sizeof(cctx->stats));
+#if USE_CHECKSUM
+  cctx->hashTable = HASH_createTable(LDM_HASHTABLESIZE_U64);
+#else
+  cctx->hashTable = HASH_createTable(LDM_HASHTABLESIZE_U32);
+#endif
+
+  if (!cctx->hashTable) return 1;
+
+  cctx->stats.minOffset = UINT_MAX;
+  cctx->stats.windowSizeLog = LDM_WINDOW_SIZE_LOG;
+  cctx->stats.hashTableSizeLog = LDM_MEMORY_USAGE;
+
+  cctx->lastPosHashed = NULL;
+
+  cctx->step = 1;   // Fixed to be 1 for now. Changing may break things.
+  cctx->nextIp = cctx->ip + cctx->step;
+  cctx->nextPosHashed = 0;
+
+  return 0;
+}
+
+void LDM_destroyCCtx(LDM_CCtx *cctx) {
+  HASH_destroyTable(cctx->hashTable);
+}
+
+/**
+ * Finds the "best" match.
+ *
+ * Returns 0 if successful and 1 otherwise (i.e. no match can be found
+ * in the remaining input that is long enough).
+ *
+ * forwardMatchLength contains the forward length of the match.
+ */
+static int LDM_findBestMatch(LDM_CCtx *cctx, const BYTE **match,
+                             U64 *forwardMatchLength, U64 *backwardMatchLength) {
+
+  LDM_hashEntry *entry = NULL;
+  cctx->nextIp = cctx->ip + cctx->step;
+
+  while (entry == NULL) {
+    U64 hash;
+    hash_t smallHash;
+    U32 checksum;
+#if INSERT_BY_TAG
+    U32 hashEveryMask;
+#endif
+    setNextHash(cctx);
+
+    hash = cctx->nextHash;
+    smallHash = getSmallHash(hash);
+    checksum = getChecksum(hash);
+#if INSERT_BY_TAG
+    hashEveryMask = lowerBitsFromHfHash(hash);
+#endif
+
+    cctx->ip = cctx->nextIp;
+    cctx->nextIp += cctx->step;
+
+    if (cctx->ip > cctx->imatchLimit) {
+      return 1;
+    }
+#if INSERT_BY_TAG
+    if (hashEveryMask == HASH_ONLY_EVERY) {
+
+      entry = HASH_getBestEntry(cctx, smallHash, checksum,
+                                forwardMatchLength, backwardMatchLength);
+    }
+#else
+    entry = HASH_getBestEntry(cctx, smallHash, checksum,
+                              forwardMatchLength, backwardMatchLength);
+#endif
+
+    if (entry != NULL) {
+      *match = entry->offset + cctx->ibase;
+    }
+
+    putHashOfCurrentPositionFromHash(cctx, hash);
+
+  }
+  setNextHash(cctx);
+  return 0;
+}
+
+void LDM_encodeLiteralLengthAndLiterals(
+    LDM_CCtx *cctx, BYTE *pToken, const U64 literalLength) {
+  /* Encode the literal length. */
+  if (literalLength >= RUN_MASK) {
+    U64 len = (U64)literalLength - RUN_MASK;
+    *pToken = (RUN_MASK << ML_BITS);
+    for (; len >= 255; len -= 255) {
+      *(cctx->op)++ = 255;
+    }
+    *(cctx->op)++ = (BYTE)len;
+  } else {
+    *pToken = (BYTE)(literalLength << ML_BITS);
+  }
+
+  /* Encode the literals. */
+  memcpy(cctx->op, cctx->anchor, literalLength);
+  cctx->op += literalLength;
+}
+
+void LDM_outputBlock(LDM_CCtx *cctx,
+                     const U64 literalLength,
+                     const U32 offset,
+                     const U64 matchLength) {
+  BYTE *pToken = cctx->op++;
+
+  /* Encode the literal length and literals. */
+  LDM_encodeLiteralLengthAndLiterals(cctx, pToken, literalLength);
+
+  /* Encode the offset. */
+  MEM_write32(cctx->op, offset);
+  cctx->op += LDM_OFFSET_SIZE;
+
+  /* Encode the match length. */
+  if (matchLength >= ML_MASK) {
+    U64 matchLengthRemaining = matchLength;
+    *pToken += ML_MASK;
+    matchLengthRemaining -= ML_MASK;
+    MEM_write32(cctx->op, 0xFFFFFFFF);
+    while (matchLengthRemaining >= 4*0xFF) {
+      cctx->op += 4;
+      MEM_write32(cctx->op, 0xffffffff);
+      matchLengthRemaining -= 4*0xFF;
+    }
+    cctx->op += matchLengthRemaining / 255;
+    *(cctx->op)++ = (BYTE)(matchLengthRemaining % 255);
+  } else {
+    *pToken += (BYTE)(matchLength);
+  }
+}
+
+// TODO: maxDstSize is unused. This function may seg fault when writing
+// beyond the size of dst, as it does not check maxDstSize. Writing to
+// a buffer and performing checks is a possible solution.
+//
+// This is based upon lz4.
+size_t LDM_compress(const void *src, size_t srcSize,
+                    void *dst, size_t maxDstSize) {
+  LDM_CCtx cctx;
+  const BYTE *match = NULL;
+  U64 forwardMatchLength = 0;
+  U64 backwardsMatchLength = 0;
+
+  if (LDM_initializeCCtx(&cctx, src, srcSize, dst, maxDstSize)) {
+    // Initialization failed.
+    return 0;
+  }
+
+#ifdef OUTPUT_CONFIGURATION
+  LDM_outputConfiguration();
+#endif
+
+  /* Hash the first position and put it into the hash table. */
+  LDM_putHashOfCurrentPosition(&cctx);
+
+  cctx.lagIp = cctx.ip;
+  cctx.lagHash = cctx.lastHash;
+
+  /**
+   * Find a match.
+   * If no more matches can be found (i.e. the length of the remaining input
+   * is less than the minimum match length), then stop searching for matches
+   * and encode the final literals.
+   */
+  while (!LDM_findBestMatch(&cctx, &match, &forwardMatchLength,
+                           &backwardsMatchLength)) {
+
+#ifdef COMPUTE_STATS
+    cctx.stats.numMatches++;
+#endif
+
+     cctx.ip -= backwardsMatchLength;
+     match -= backwardsMatchLength;
+
+    /**
+     * Write current block (literals, literal length, match offset, match
+     * length) and update pointers and hashes.
+     */
+    {
+      const U64 literalLength = cctx.ip - cctx.anchor;
+      const U32 offset = cctx.ip - match;
+      const U64 matchLength = forwardMatchLength +
+                              backwardsMatchLength -
+                              LDM_MIN_MATCH_LENGTH;
+
+      LDM_outputBlock(&cctx, literalLength, offset, matchLength);
+
+#ifdef COMPUTE_STATS
+      cctx.stats.totalLiteralLength += literalLength;
+      cctx.stats.totalOffset += offset;
+      cctx.stats.totalMatchLength += matchLength + LDM_MIN_MATCH_LENGTH;
+      cctx.stats.minOffset =
+          offset < cctx.stats.minOffset ? offset : cctx.stats.minOffset;
+      cctx.stats.maxOffset =
+          offset > cctx.stats.maxOffset ? offset : cctx.stats.maxOffset;
+      cctx.stats.offsetHistogram[(U32)intLog2(offset)]++;
+      cctx.stats.matchLengthHistogram[
+          (U32)intLog2(matchLength + LDM_MIN_MATCH_LENGTH)]++;
+#endif
+
+      // Move ip to end of block, inserting hashes at each position.
+      cctx.nextIp = cctx.ip + cctx.step;
+      while (cctx.ip < cctx.anchor + LDM_MIN_MATCH_LENGTH +
+                       matchLength + literalLength) {
+        if (cctx.ip > cctx.lastPosHashed) {
+          // TODO: Simplify.
+          LDM_updateLastHashFromNextHash(&cctx);
+          setNextHash(&cctx);
+        }
+        cctx.ip++;
+        cctx.nextIp++;
+      }
+    }
+
+    // Set start of next block to current input pointer.
+    cctx.anchor = cctx.ip;
+    LDM_updateLastHashFromNextHash(&cctx);
+  }
+
+  /* Encode the last literals (no more matches). */
+  {
+    const U64 lastRun = cctx.iend - cctx.anchor;
+    BYTE *pToken = cctx.op++;
+    LDM_encodeLiteralLengthAndLiterals(&cctx, pToken, lastRun);
+  }
+
+#ifdef COMPUTE_STATS
+  LDM_printCompressStats(&cctx.stats);
+  HASH_outputTableOccupancy(cctx.hashTable);
+#endif
+
+  {
+    const size_t ret = cctx.op - cctx.obase;
+    LDM_destroyCCtx(&cctx);
+    return ret;
+  }
+}
+
+void LDM_outputConfiguration(void) {
+  printf("=====================\n");
+  printf("Configuration\n");
+  printf("LDM_WINDOW_SIZE_LOG: %d\n", LDM_WINDOW_SIZE_LOG);
+  printf("LDM_MIN_MATCH_LENGTH, LDM_HASH_LENGTH: %d, %d\n",
+         LDM_MIN_MATCH_LENGTH, LDM_HASH_LENGTH);
+  printf("LDM_MEMORY_USAGE: %d\n", LDM_MEMORY_USAGE);
+  printf("HASH_ONLY_EVERY_LOG: %d\n", HASH_ONLY_EVERY_LOG);
+  printf("HASH_BUCKET_SIZE_LOG: %d\n", HASH_BUCKET_SIZE_LOG);
+  printf("LDM_LAG: %d\n", LDM_LAG);
+  printf("USE_CHECKSUM: %d\n", USE_CHECKSUM);
+  printf("INSERT_BY_TAG: %d\n", INSERT_BY_TAG);
+  printf("HASH_CHAR_OFFSET: %d\n", HASH_CHAR_OFFSET);
+  printf("=====================\n");
+}
+

contrib/long_distance_matching/ldm.h

@@ -0,0 +1,197 @@
+#ifndef LDM_H
+#define LDM_H
+
+#include "mem.h"    // from /lib/common/mem.h
+
+//#include "ldm_params.h"
+
+// =============================================================================
+// Modify the parameters in ldm_params.h if "ldm_params.h" is included.
+// Otherwise, modify the parameters here.
+// =============================================================================
+
+#ifndef LDM_PARAMS_H
+  // Defines the size of the hash table.
+  // Note that this is not the number of buckets.
+  // Currently this should be less than WINDOW_SIZE_LOG + 4.
+  #define LDM_MEMORY_USAGE 23
+
+  // The number of entries in a hash bucket.
+  #define HASH_BUCKET_SIZE_LOG 3 // The maximum is 4 for now.
+
+  // Defines the lag in inserting elements into the hash table.
+  #define LDM_LAG 0
+
+  // The maximum window size when searching for matches.
+  // The maximum value is 30
+  #define LDM_WINDOW_SIZE_LOG 28
+
+  // The minimum match length.
+  // This should be a multiple of four.
+  #define LDM_MIN_MATCH_LENGTH 64
+
+  // If INSERT_BY_TAG, insert entries into the hash table as a function of the
+  // hash. Certain hashes will not be inserted.
+  //
+  // Otherwise, insert as a function of the position.
+  #define INSERT_BY_TAG 1
+
+  // Store a checksum with the hash table entries for faster comparison.
+  // This halves the number of entries the hash table can contain.
+  #define USE_CHECKSUM 1
+#endif
+
+// Output compression statistics.
+#define COMPUTE_STATS
+
+// Output the configuration.
+#define OUTPUT_CONFIGURATION
+
+// If defined, forces the probability of insertion to be approximately
+// one per (1 << HASH_ONLY_EVERY_LOG). If not defined, the probability will be
+// calculated based on the memory usage and window size for "even" insertion
+// throughout the window.
+
+// #define HASH_ONLY_EVERY_LOG 8
+
+// =============================================================================
+
+// The number of bytes storing the compressed and decompressed size
+// in the header.
+#define LDM_COMPRESSED_SIZE 8
+#define LDM_DECOMPRESSED_SIZE 8
+#define LDM_HEADER_SIZE ((LDM_COMPRESSED_SIZE)+(LDM_DECOMPRESSED_SIZE))
+
+#define ML_BITS 4
+#define ML_MASK ((1U<<ML_BITS)-1)
+#define RUN_BITS (8-ML_BITS)
+#define RUN_MASK ((1U<<RUN_BITS)-1)
+
+// The number of bytes storing the offset.
+#define LDM_OFFSET_SIZE 4
+
+#define LDM_WINDOW_SIZE (1 << (LDM_WINDOW_SIZE_LOG))
+
+// TODO: Match lengths that are too small do not use the hash table efficiently.
+// There should be a minimum hash length given the hash table size.
+#define LDM_HASH_LENGTH LDM_MIN_MATCH_LENGTH
+
+typedef struct LDM_compressStats LDM_compressStats;
+typedef struct LDM_CCtx LDM_CCtx;
+typedef struct LDM_DCtx LDM_DCtx;
+
+/**
+ *  Compresses src into dst.
+ *  Returns the compressed size if successful, 0 otherwise.
+ *
+ *  NB: This currently ignores maxDstSize and assumes enough space is available.
+ *
+ *  Block format (see lz4 documentation for more information):
+ *  github.com/lz4/lz4/blob/dev/doc/lz4_Block_format.md
+ *
+ *  A block is composed of sequences. Each sequence begins with a token, which
+ *  is a one-byte value separated into two 4-bit fields.
+ *
+ *  The first field uses the four high bits of the token and encodes the literal
+ *  length. If the field value is 0, there is no literal. If it is 15,
+ *  additional bytes are added (each ranging from 0 to 255) to the previous
+ *  value to produce a total length.
+ *
+ *  Following the token and optional length bytes are the literals.
+ *
+ *  Next are the 4 bytes representing the offset of the match (2 in lz4),
+ *  representing the position to copy the literals.
+ *
+ *  The lower four bits of the token encode the match length. With additional
+ *  bytes added similarly to the additional literal length bytes after the offset.
+ *
+ *  The last sequence is incomplete and stops right after the literals.
+ */
+size_t LDM_compress(const void *src, size_t srcSize,
+                    void *dst, size_t maxDstSize);
+
+/**
+ * Initialize the compression context.
+ *
+ * Allocates memory for the hash table.
+ *
+ * Returns 0 if successful, 1 otherwise.
+ */
+size_t LDM_initializeCCtx(LDM_CCtx *cctx,
+                          const void *src, size_t srcSize,
+                          void *dst, size_t maxDstSize);
+
+/**
+ * Frees up memory allocated in LDM_initializeCCtx().
+ */
+void LDM_destroyCCtx(LDM_CCtx *cctx);
+
+/**
+ * Prints the distribution of offsets in the hash table.
+ *
+ * The offsets are defined as the distance of the hash table entry from the
+ * current input position of the cctx.
+ */
+void LDM_outputHashTableOffsetHistogram(const LDM_CCtx *cctx);
+
+/**
+ * Outputs compression statistics to stdout.
+ */
+void LDM_printCompressStats(const LDM_compressStats *stats);
+
+/**
+ * Encode the literal length followed by the literals.
+ *
+ * The literal length is written to the upper four bits of pToken, with
+ * additional bytes written to the output as needed (see lz4).
+ *
+ * This is followed by literalLength bytes corresponding to the literals.
+ */
+void LDM_encodeLiteralLengthAndLiterals(LDM_CCtx *cctx, BYTE *pToken,
+                                        const U64 literalLength);
+
+/**
+ * Write current block (literals, literal length, match offset,
+ * match length).
+ */
+void LDM_outputBlock(LDM_CCtx *cctx,
+                     const U64 literalLength,
+                     const U32 offset,
+                     const U64 matchLength);
+
+/**
+ * Decompresses src into dst.
+ *
+ * Note: assumes src does not have a header.
+ */
+size_t LDM_decompress(const void *src, size_t srcSize,
+                      void *dst, size_t maxDstSize);
+
+/**
+ * Initialize the decompression context.
+ */
+void LDM_initializeDCtx(LDM_DCtx *dctx,
+                        const void *src, size_t compressedSize,
+                        void *dst, size_t maxDecompressedSize);
+
+/**
+ * Reads the header from src and writes the compressed size and
+ * decompressed size into compressedSize and decompressedSize respectively.
+ *
+ * NB: LDM_compress and LDM_decompress currently do not add/read headers.
+ */
+void LDM_readHeader(const void *src, U64 *compressedSize,
+                    U64 *decompressedSize);
+
+/**
+ * Write the compressed and decompressed size.
+ */
+void LDM_writeHeader(void *memPtr, U64 compressedSize,
+                     U64 decompressedSize);
+
+/**
+ * Output the configuration used.
+ */
+void LDM_outputConfiguration(void);
+
+#endif /* LDM_H */

contrib/long_distance_matching/ldm_common.c

@@ -0,0 +1,109 @@
+#include <stdio.h>
+
+#include "ldm.h"
+
+/**
+ * This function reads the header at the beginning of src and writes
+ * the compressed and decompressed size to compressedSize and
+ * decompressedSize.
+ *
+ * The header consists of 16 bytes: 8 bytes each in little-endian format
+ * of the compressed size and the decompressed size.
+ */
+void LDM_readHeader(const void *src, U64 *compressedSize,
+                    U64 *decompressedSize) {
+  const BYTE *ip = (const BYTE *)src;
+  *compressedSize = MEM_readLE64(ip);
+  *decompressedSize = MEM_readLE64(ip + 8);
+}
+
+/**
+ * Writes the 16-byte header (8-bytes each of the compressedSize and
+ * decompressedSize in little-endian format) to memPtr.
+ */
+void LDM_writeHeader(void *memPtr, U64 compressedSize,
+                     U64 decompressedSize) {
+  MEM_writeLE64(memPtr, compressedSize);
+  MEM_writeLE64((BYTE *)memPtr + 8, decompressedSize);
+}
+
+struct LDM_DCtx {
+  size_t compressedSize;
+  size_t maxDecompressedSize;
+
+  const BYTE *ibase;   /* Base of input */
+  const BYTE *ip;      /* Current input position */
+  const BYTE *iend;    /* End of source */
+
+  const BYTE *obase;   /* Base of output */
+  BYTE *op;            /* Current output position */
+  const BYTE *oend;    /* End of output */
+};
+
+void LDM_initializeDCtx(LDM_DCtx *dctx,
+                        const void *src, size_t compressedSize,
+                        void *dst, size_t maxDecompressedSize) {
+  dctx->compressedSize = compressedSize;
+  dctx->maxDecompressedSize = maxDecompressedSize;
+
+  dctx->ibase = src;
+  dctx->ip = (const BYTE *)src;
+  dctx->iend = dctx->ip + dctx->compressedSize;
+  dctx->op = dst;
+  dctx->oend = dctx->op + dctx->maxDecompressedSize;
+}
+
+size_t LDM_decompress(const void *src, size_t compressedSize,
+                      void *dst, size_t maxDecompressedSize) {
+
+  LDM_DCtx dctx;
+  LDM_initializeDCtx(&dctx, src, compressedSize, dst, maxDecompressedSize);
+
+  while (dctx.ip < dctx.iend) {
+    BYTE *cpy;
+    const BYTE *match;
+    size_t length, offset;
+
+    /* Get the literal length. */
+    const unsigned token = *(dctx.ip)++;
+    if ((length = (token >> ML_BITS)) == RUN_MASK) {
+      unsigned s;
+      do {
+        s = *(dctx.ip)++;
+        length += s;
+      } while (s == 255);
+    }
+
+    /* Copy the literals. */
+    cpy = dctx.op + length;
+    memcpy(dctx.op, dctx.ip, length);
+    dctx.ip += length;
+    dctx.op = cpy;
+
+    //TODO: dynamic offset size?
+    /* Encode the offset. */
+    offset = MEM_read32(dctx.ip);
+    dctx.ip += LDM_OFFSET_SIZE;
+    match = dctx.op - offset;
+
+    /* Get the match length. */
+    length = token & ML_MASK;
+    if (length == ML_MASK) {
+      unsigned s;
+      do {
+        s = *(dctx.ip)++;
+        length += s;
+      } while (s == 255);
+    }
+    length += LDM_MIN_MATCH_LENGTH;
+
+    /* Copy match. */
+    cpy = dctx.op + length;
+
+    // TODO: this can be made more efficient.
+    while (match < cpy - offset && dctx.op < dctx.oend) {
+      *(dctx.op)++ = *match++;
+    }
+  }
+  return dctx.op - (BYTE *)dst;
+}

contrib/long_distance_matching/ldm_params.h

@@ -0,0 +1,12 @@
+#ifndef LDM_PARAMS_H
+#define LDM_PARAMS_H
+
+#define LDM_MEMORY_USAGE 23
+#define HASH_BUCKET_SIZE_LOG 3
+#define LDM_LAG 0
+#define LDM_WINDOW_SIZE_LOG 28
+#define LDM_MIN_MATCH_LENGTH 64
+#define INSERT_BY_TAG 1
+#define USE_CHECKSUM 1
+
+#endif  // LDM_PARAMS_H

contrib/long_distance_matching/main.c

@@ -0,0 +1,269 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <zstd.h>
+
+#include <fcntl.h>
+#include "ldm.h"
+#include "zstd.h"
+
+// #define DECOMPRESS_AND_VERIFY
+
+/* Compress file given by fname and output to oname.
+ * Returns 0 if successful, error code otherwise.
+ *
+ * This adds a header from LDM_writeHeader to the beginning of the output.
+ *
+ * This might seg fault if the compressed size is > the decompress
+ * size due to the mmapping and output file size allocated to be the input size
+ * The compress function should check before writing or buffer writes.
+ */
+static int compress(const char *fname, const char *oname) {
+  int fdin, fdout;
+  struct stat statbuf;
+  char *src, *dst;
+  size_t maxCompressedSize, compressedSize;
+
+  struct timeval tv1, tv2;
+  double timeTaken;
+
+
+  /* Open the input file. */
+  if ((fdin = open(fname, O_RDONLY)) < 0) {
+    perror("Error in file opening");
+    return 1;
+  }
+
+  /* Open the output file. */
+  if ((fdout = open(oname, O_RDWR | O_CREAT | O_TRUNC, (mode_t)0600)) < 0) {
+    perror("Can't create output file");
+    return 1;
+  }
+
+  /* Find the size of the input file. */
+  if (fstat (fdin, &statbuf) < 0) {
+    perror("Fstat error");
+    return 1;
+  }
+
+  maxCompressedSize = (statbuf.st_size + LDM_HEADER_SIZE);
+
+  // Handle case where compressed size is > decompressed size.
+  // TODO: The compress function should check before writing or buffer writes.
+  maxCompressedSize += statbuf.st_size / 255;
+
+  ftruncate(fdout, maxCompressedSize);
+
+  /* mmap the input file. */
+  if ((src = mmap(0, statbuf.st_size, PROT_READ,  MAP_SHARED, fdin, 0))
+          == (caddr_t) - 1) {
+      perror("mmap error for input");
+      return 1;
+  }
+
+  /* mmap the output file. */
+  if ((dst = mmap(0, maxCompressedSize, PROT_READ | PROT_WRITE,
+                  MAP_SHARED, fdout, 0)) == (caddr_t) - 1) {
+      perror("mmap error for output");
+      return 1;
+  }
+
+  gettimeofday(&tv1, NULL);
+
+  compressedSize = LDM_HEADER_SIZE +
+      LDM_compress(src, statbuf.st_size,
+                   dst + LDM_HEADER_SIZE, maxCompressedSize);
+
+  gettimeofday(&tv2, NULL);
+
+  // Write the header.
+  LDM_writeHeader(dst, compressedSize, statbuf.st_size);
+
+  // Truncate file to compressedSize.
+  ftruncate(fdout, compressedSize);
+
+  printf("%25s : %10lu -> %10lu - %s \n", fname,
+         (size_t)statbuf.st_size, (size_t)compressedSize, oname);
+  printf("Compression ratio: %.2fx --- %.1f%%\n",
+         (double)statbuf.st_size / (double)compressedSize,
+         (double)compressedSize / (double)(statbuf.st_size) * 100.0);
+
+  timeTaken = (double) (tv2.tv_usec - tv1.tv_usec) / 1000000 +
+              (double) (tv2.tv_sec - tv1.tv_sec),
+
+  printf("Total compress time = %.3f seconds, Average scanning speed: %.3f MB/s\n",
+         timeTaken,
+         ((double)statbuf.st_size / (double) (1 << 20)) / timeTaken);
+
+  // Close files.
+  close(fdin);
+  close(fdout);
+  return 0;
+}
+
+#ifdef DECOMPRESS_AND_VERIFY
+/* Decompress file compressed using LDM_compress.
+ * The input file should have the LDM_HEADER followed by payload.
+ * Returns 0 if succesful, and an error code otherwise.
+ */
+static int decompress(const char *fname, const char *oname) {
+  int fdin, fdout;
+  struct stat statbuf;
+  char *src, *dst;
+  U64 compressedSize, decompressedSize;
+  size_t outSize;
+
+  /* Open the input file. */
+  if ((fdin = open(fname, O_RDONLY)) < 0) {
+    perror("Error in file opening");
+    return 1;
+  }
+
+  /* Open the output file. */
+  if ((fdout = open(oname, O_RDWR | O_CREAT | O_TRUNC, (mode_t)0600)) < 0) {
+    perror("Can't create output file");
+    return 1;
+  }
+
+  /* Find the size of the input file. */
+  if (fstat (fdin, &statbuf) < 0) {
+    perror("Fstat error");
+    return 1;
+  }
+
+  /* mmap the input file. */
+  if ((src = mmap(0, statbuf.st_size, PROT_READ,  MAP_SHARED, fdin, 0))
+          == (caddr_t) - 1) {
+      perror("mmap error for input");
+      return 1;
+  }
+
+  /* Read the header. */
+  LDM_readHeader(src, &compressedSize, &decompressedSize);
+
+  ftruncate(fdout, decompressedSize);
+
+  /* mmap the output file */
+  if ((dst = mmap(0, decompressedSize, PROT_READ | PROT_WRITE,
+                  MAP_SHARED, fdout, 0)) == (caddr_t) - 1) {
+      perror("mmap error for output");
+      return 1;
+  }
+
+  outSize = LDM_decompress(
+      src + LDM_HEADER_SIZE, statbuf.st_size - LDM_HEADER_SIZE,
+      dst, decompressedSize);
+  printf("Ret size out: %zu\n", outSize);
+
+  close(fdin);
+  close(fdout);
+  return 0;
+}
+
+/* Compare two files.
+ * Returns 0 iff they are the same.
+ */
+static int compare(FILE *fp0, FILE *fp1) {
+  int result = 0;
+  while (result == 0) {
+    char b0[1024];
+    char b1[1024];
+    const size_t r0 = fread(b0, 1, sizeof(b0), fp0);
+    const size_t r1 = fread(b1, 1, sizeof(b1), fp1);
+
+    result = (int)r0 - (int)r1;
+
+    if (0 == r0 || 0 == r1) break;
+
+    if (0 == result) result = memcmp(b0, b1, r0);
+  }
+  return result;
+}
+
+/* Verify the input file is the same as the decompressed file. */
+static int verify(const char *inpFilename, const char *decFilename) {
+  FILE *inpFp, *decFp;
+
+  if ((inpFp = fopen(inpFilename, "rb")) == NULL) {
+    perror("Could not open input file\n");
+    return 1;
+  }
+
+  if ((decFp = fopen(decFilename, "rb")) == NULL) {
+    perror("Could not open decompressed file\n");
+    return 1;
+  }
+
+  printf("verify : %s <-> %s\n", inpFilename, decFilename);
+  {
+    const int cmp = compare(inpFp, decFp);
+    if(0 == cmp) {
+      printf("verify : OK\n");
+    } else {
+      printf("verify : NG\n");
+      return 1;
+    }
+  }
+
+	fclose(decFp);
+	fclose(inpFp);
+  return 0;
+}
+#endif
+
+int main(int argc, const char *argv[]) {
+  const char * const exeName = argv[0];
+  char inpFilename[256] = { 0 };
+  char ldmFilename[256] = { 0 };
+  char decFilename[256] = { 0 };
+
+  if (argc < 2) {
+    printf("Wrong arguments\n");
+    printf("Usage:\n");
+    printf("%s FILE\n", exeName);
+    return 1;
+  }
+
+  snprintf(inpFilename, 256, "%s", argv[1]);
+  snprintf(ldmFilename, 256, "%s.ldm", argv[1]);
+  snprintf(decFilename, 256, "%s.ldm.dec", argv[1]);
+
+ 	printf("inp = [%s]\n", inpFilename);
+	printf("ldm = [%s]\n", ldmFilename);
+	printf("dec = [%s]\n", decFilename);
+
+  /* Compress */
+  {
+    if (compress(inpFilename, ldmFilename)) {
+        printf("Compress error\n");
+        return 1;
+    }
+  }
+
+#ifdef DECOMPRESS_AND_VERIFY
+  /* Decompress */
+  {
+    struct timeval tv1, tv2;
+    gettimeofday(&tv1, NULL);
+    if (decompress(ldmFilename, decFilename)) {
+        printf("Decompress error\n");
+        return 1;
+    }
+    gettimeofday(&tv2, NULL);
+    printf("Total decompress time = %f seconds\n",
+          (double) (tv2.tv_usec - tv1.tv_usec) / 1000000 +
+          (double) (tv2.tv_sec - tv1.tv_sec));
+  }
+  /* verify */
+  if (verify(inpFilename, decFilename)) {
+    printf("Verification error\n");
+    return 1;
+  }
+#endif
+  return 0;
+}

doc/zstd_manual.html

@@ -813,7 +813,9 @@ void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
                               * Special: value 0 means "do not change strategy". */
 
     </b>/* frame parameters */<b>
-    ZSTD_p_contentSizeFlag=200, </b>/* Content size is written into frame header _whenever known_ (default:1) */<b>
+    ZSTD_p_contentSizeFlag=200, </b>/* Content size is written into frame header _whenever known_ (default:1)<b>
+                              * note that content size must be known at the beginning,
+                              * it is sent using ZSTD_CCtx_setPledgedSrcSize() */
     ZSTD_p_checksumFlag,     </b>/* A 32-bits checksum of content is written at end of frame (default:0) */<b>
     ZSTD_p_dictIDFlag,       </b>/* When applicable, dictID of dictionary is provided in frame header (default:1) */<b>
 

lib/common/bitstream.h

@@ -203,7 +203,7 @@ static const unsigned BIT_mask[] = { 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F,
 /*! BIT_initCStream() :
  *  `dstCapacity` must be > sizeof(size_t)
  *  @return : 0 if success,
-              otherwise an error code (can be tested using ERR_isError() ) */
+ *            otherwise an error code (can be tested using ERR_isError()) */
 MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
                                   void* startPtr, size_t dstCapacity)
 {
@@ -217,8 +217,8 @@ MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
 }
 
 /*! BIT_addBits() :
-    can add up to 26 bits into `bitC`.
+ *  can add up to 26 bits into `bitC`.
-    Does not check for register overflow ! */
+ *  Note : does not check for register overflow ! */
 MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
                             size_t value, unsigned nbBits)
 {
@@ -268,7 +268,7 @@ MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC)
 
 /*! BIT_closeCStream() :
  *  @return : size of CStream, in bytes,
-              or 0 if it could not fit into dstBuffer */
+ *            or 0 if it could not fit into dstBuffer */
 MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
 {
     BIT_addBitsFast(bitC, 1, 1);   /* endMark */
@@ -285,7 +285,7 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
  *  Initialize a BIT_DStream_t.
  * `bitD` : a pointer to an already allocated BIT_DStream_t structure.
  * `srcSize` must be the *exact* size of the bitStream, in bytes.
-*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
+ * @return : size of stream (== srcSize), or an errorCode if a problem is detected
  */
 MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
 {
@@ -327,7 +327,8 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
         }
         {   BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
             bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
-          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+            if (lastByte == 0) return ERROR(corruption_detected);  /* endMark not present */
+        }
         bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
     }
 
@@ -363,8 +364,7 @@ MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
  *  local register is not modified.
  *  On 32-bits, maxNbBits==24.
  *  On 64-bits, maxNbBits==56.
- *  @return : value extracted
+ * @return : value extracted */
- */
 MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
 {
 #if defined(__BMI__) && defined(__GNUC__)   /* experimental; fails if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8 */
@@ -392,8 +392,7 @@ MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
 /*! BIT_readBits() :
  *  Read (consume) next n bits from local register and update.
  *  Pay attention to not read more than nbBits contained into local register.
- *  @return : extracted value.
+ * @return : extracted value. */
- */
 MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
 {
     size_t const value = BIT_lookBits(bitD, nbBits);
@@ -415,7 +414,7 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
  *  Refill `bitD` from buffer previously set in BIT_initDStream() .
  *  This function is safe, it guarantees it will not read beyond src buffer.
  * @return : status of `BIT_DStream_t` internal register.
-              if status == BIT_DStream_unfinished, internal register is filled with >= (sizeof(bitD->bitContainer)*8 - 7) bits */
+ *           when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
 MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
 {
     if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* overflow detected, like end of stream */
@@ -446,7 +445,7 @@ MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
 }
 
 /*! BIT_endOfDStream() :
-*   @return Tells if DStream has exactly reached its end (all bits consumed).
+ * @return : 1 if DStream has _exactly_ reached its end (all bits consumed).
  */
 MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
 {

lib/common/error_private.c

@@ -20,19 +20,17 @@ const char* ERR_getErrorString(ERR_enum code)
     case PREFIX(GENERIC):  return "Error (generic)";
     case PREFIX(prefix_unknown): return "Unknown frame descriptor";
     case PREFIX(version_unsupported): return "Version not supported";
-    case PREFIX(parameter_unknown): return "Unknown parameter type";
     case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
-    case PREFIX(frameParameter_unsupportedBy32bits): return "Frame parameter unsupported in 32-bits mode";
     case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
-    case PREFIX(compressionParameter_unsupported): return "Compression parameter is not supported";
+    case PREFIX(corruption_detected): return "Corrupted block detected";
-    case PREFIX(compressionParameter_outOfBound): return "Compression parameter is out of bound";
+    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
+    case PREFIX(parameter_unsupported): return "Unsupported parameter";
+    case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
     case PREFIX(init_missing): return "Context should be init first";
     case PREFIX(memory_allocation): return "Allocation error : not enough memory";
     case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
     case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
     case PREFIX(srcSize_wrong): return "Src size is incorrect";
-    case PREFIX(corruption_detected): return "Corrupted block detected";
-    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
     case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
     case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
     case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";

lib/common/fse.h

@@ -31,13 +31,14 @@
    You can contact the author at :
    - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
 ****************************************************************** */
-#ifndef FSE_H
-#define FSE_H
 
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
+#ifndef FSE_H
+#define FSE_H
+
 
 /*-*****************************************
 *  Dependencies
@@ -297,8 +298,10 @@ FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<
 If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
 */
 
+#endif  /* FSE_H */
 
-#ifdef FSE_STATIC_LINKING_ONLY
+#if defined(FSE_STATIC_LINKING_ONLY) && !defined(FSE_H_FSE_STATIC_LINKING_ONLY)
+#define FSE_H_FSE_STATIC_LINKING_ONLY
 
 /* *** Dependency *** */
 #include "bitstream.h"
@@ -381,6 +384,11 @@ size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
 size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog);
 /**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DTABLE_SIZE_U32(maxLog)` */
 
+typedef enum {
+   FSE_repeat_none,  /**< Cannot use the previous table */
+   FSE_repeat_check, /**< Can use the previous table but it must be checked */
+   FSE_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+ } FSE_repeat;
 
 /* *****************************************
 *  FSE symbol compression API
@@ -694,5 +702,3 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
 #if defined (__cplusplus)
 }
 #endif
-
-#endif  /* FSE_H */

lib/common/huf.h

@@ -31,13 +31,13 @@
    You can contact the author at :
    - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
 ****************************************************************** */
-#ifndef HUF_H_298734234
-#define HUF_H_298734234
 
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
+#ifndef HUF_H_298734234
+#define HUF_H_298734234
 
 /* *** Dependencies *** */
 #include <stddef.h>    /* size_t */
@@ -124,6 +124,7 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity, const
 #define HUF_DECOMPRESS_WORKSPACE_SIZE (2 << 10)
 #define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
 
+#endif   /* HUF_H_298734234 */
 
 /* ******************************************************************
  *  WARNING !!
@@ -132,7 +133,8 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity, const
  *  because they are not guaranteed to remain stable in the future.
  *  Only consider them in association with static linking.
  *******************************************************************/
-#ifdef HUF_STATIC_LINKING_ONLY
+#if defined(HUF_STATIC_LINKING_ONLY) && !defined(HUF_H_HUF_STATIC_LINKING_ONLY)
+#define HUF_H_HUF_STATIC_LINKING_ONLY
 
 /* *** Dependencies *** */
 #include "mem.h"   /* U32 */
@@ -295,9 +297,6 @@ size_t HUF_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* c
 
 #endif /* HUF_STATIC_LINKING_ONLY */
 
-
 #if defined (__cplusplus)
 }
 #endif
-
-#endif   /* HUF_H_298734234 */

lib/common/pool.c

@@ -95,9 +95,9 @@ POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) {
     ctx->queue = (POOL_job*) malloc(ctx->queueSize * sizeof(POOL_job));
     ctx->queueHead = 0;
     ctx->queueTail = 0;
-    pthread_mutex_init(&ctx->queueMutex, NULL);
+    (void)pthread_mutex_init(&ctx->queueMutex, NULL);
-    pthread_cond_init(&ctx->queuePushCond, NULL);
+    (void)pthread_cond_init(&ctx->queuePushCond, NULL);
-    pthread_cond_init(&ctx->queuePopCond, NULL);
+    (void)pthread_cond_init(&ctx->queuePopCond, NULL);
     ctx->shutdown = 0;
     /* Allocate space for the thread handles */
     ctx->threads = (pthread_t*)malloc(numThreads * sizeof(pthread_t));
@@ -154,7 +154,7 @@ size_t POOL_sizeof(POOL_ctx *ctx) {
 }
 
 void POOL_add(void* ctxVoid, POOL_function function, void *opaque) {
-    POOL_ctx *ctx = (POOL_ctx *)ctxVoid;
+    POOL_ctx* const ctx = (POOL_ctx*)ctxVoid;
     if (!ctx) { return; }
 
     pthread_mutex_lock(&ctx->queueMutex);
@@ -190,7 +190,7 @@ POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) {
 }
 
 void POOL_free(POOL_ctx* ctx) {
-  if (ctx) free(ctx);
+    free(ctx);
 }
 
 void POOL_add(void* ctx, POOL_function function, void* opaque) {

lib/common/pool.h

@@ -19,11 +19,11 @@ extern "C" {
 typedef struct POOL_ctx_s POOL_ctx;
 
 /*! POOL_create() :
-    Create a thread pool with at most `numThreads` threads.
+ *  Create a thread pool with at most `numThreads` threads.
-    `numThreads` must be at least 1.
+ * `numThreads` must be at least 1.
-    The maximum number of queued jobs before blocking is `queueSize`.
+ *  The maximum number of queued jobs before blocking is `queueSize`.
-    `queueSize` must be at least 1.
+ * `queueSize` must be at least 1.
-    @return : The POOL_ctx pointer on success else NULL.
+ * @return : POOL_ctx pointer on success, else NULL.
 */
 POOL_ctx *POOL_create(size_t numThreads, size_t queueSize);
 

lib/common/threading.h

@@ -42,14 +42,14 @@ extern "C" {
 
 /* mutex */
 #define pthread_mutex_t           CRITICAL_SECTION
-#define pthread_mutex_init(a,b)   InitializeCriticalSection((a))
+#define pthread_mutex_init(a,b)   (InitializeCriticalSection((a)), 0)
 #define pthread_mutex_destroy(a)  DeleteCriticalSection((a))
 #define pthread_mutex_lock(a)     EnterCriticalSection((a))
 #define pthread_mutex_unlock(a)   LeaveCriticalSection((a))
 
 /* condition variable */
 #define pthread_cond_t             CONDITION_VARIABLE
-#define pthread_cond_init(a, b)    InitializeConditionVariable((a))
+#define pthread_cond_init(a, b)    (InitializeConditionVariable((a)), 0)
 #define pthread_cond_destroy(a)    /* No delete */
 #define pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
 #define pthread_cond_signal(a)     WakeConditionVariable((a))
@@ -80,14 +80,14 @@ int _pthread_join(pthread_t* thread, void** value_ptr);
 #else  /* ZSTD_MULTITHREAD not defined */
 /* No multithreading support */
 
-#define pthread_mutex_t int   /* #define rather than typedef, as sometimes pthread support is implicit, resulting in duplicated symbols */
+#define pthread_mutex_t int   /* #define rather than typedef, because sometimes pthread support is implicit, resulting in duplicated symbols */
-#define pthread_mutex_init(a,b)
+#define pthread_mutex_init(a,b)    ((void)a, 0)
 #define pthread_mutex_destroy(a)
 #define pthread_mutex_lock(a)
 #define pthread_mutex_unlock(a)
 
 #define pthread_cond_t int
-#define pthread_cond_init(a,b)
+#define pthread_cond_init(a,b)     ((void)a, 0)
 #define pthread_cond_destroy(a)
 #define pthread_cond_wait(a,b)
 #define pthread_cond_signal(a)

lib/common/zstd_errors.h

@@ -37,43 +37,41 @@ extern "C" {
 /*-****************************************
  *  error codes list
  *  note : this API is still considered unstable
- *         it should not be used with a dynamic library
+ *         and shall not be used with a dynamic library.
  *         only static linking is allowed
  ******************************************/
 typedef enum {
-  ZSTD_error_no_error,
+  ZSTD_error_no_error = 0,
-  ZSTD_error_GENERIC,
+  ZSTD_error_GENERIC  = 1,
-  ZSTD_error_prefix_unknown,
+  ZSTD_error_prefix_unknown                = 10,
-  ZSTD_error_version_unsupported,
+  ZSTD_error_version_unsupported           = 12,
-  ZSTD_error_parameter_unknown,
+  ZSTD_error_frameParameter_unsupported    = 14,
-  ZSTD_error_frameParameter_unsupported,
+  ZSTD_error_frameParameter_windowTooLarge = 16,
-  ZSTD_error_frameParameter_unsupportedBy32bits,
+  ZSTD_error_corruption_detected = 20,
-  ZSTD_error_frameParameter_windowTooLarge,
+  ZSTD_error_checksum_wrong      = 22,
-  ZSTD_error_compressionParameter_unsupported,
+  ZSTD_error_dictionary_corrupted      = 30,
-  ZSTD_error_compressionParameter_outOfBound,
+  ZSTD_error_dictionary_wrong          = 32,
-  ZSTD_error_init_missing,
+  ZSTD_error_dictionaryCreation_failed = 34,
-  ZSTD_error_memory_allocation,
+  ZSTD_error_parameter_unsupported   = 40,
-  ZSTD_error_stage_wrong,
+  ZSTD_error_parameter_outOfBound    = 42,
-  ZSTD_error_dstSize_tooSmall,
+  ZSTD_error_tableLog_tooLarge       = 44,
-  ZSTD_error_srcSize_wrong,
+  ZSTD_error_maxSymbolValue_tooLarge = 46,
-  ZSTD_error_corruption_detected,
+  ZSTD_error_maxSymbolValue_tooSmall = 48,
-  ZSTD_error_checksum_wrong,
+  ZSTD_error_stage_wrong       = 60,
-  ZSTD_error_tableLog_tooLarge,
+  ZSTD_error_init_missing      = 62,
-  ZSTD_error_maxSymbolValue_tooLarge,
+  ZSTD_error_memory_allocation = 64,
-  ZSTD_error_maxSymbolValue_tooSmall,
+  ZSTD_error_dstSize_tooSmall = 70,
-  ZSTD_error_dictionary_corrupted,
+  ZSTD_error_srcSize_wrong    = 72,
-  ZSTD_error_dictionary_wrong,
+  ZSTD_error_frameIndex_tooLarge = 100,
-  ZSTD_error_dictionaryCreation_failed,
+  ZSTD_error_seekableIO          = 102,
-  ZSTD_error_frameIndex_tooLarge,
+  ZSTD_error_maxCode = 120  /* never EVER use this value directly, it may change in future versions! Use ZSTD_isError() instead */
-  ZSTD_error_seekableIO,
-  ZSTD_error_maxCode
 } ZSTD_ErrorCode;
 
 /*! ZSTD_getErrorCode() :
     convert a `size_t` function result into a `ZSTD_ErrorCode` enum type,
     which can be used to compare with enum list published above */
 ZSTDERRORLIB_API ZSTD_ErrorCode ZSTD_getErrorCode(size_t functionResult);
-ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code);
+ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code);   /**< Same as ZSTD_getErrorName, but using a `ZSTD_ErrorCode` enum argument */
 
 
 #if defined (__cplusplus)

lib/common/zstd_internal.h

@@ -50,6 +50,10 @@
 #include "error_private.h"
 #define ZSTD_STATIC_LINKING_ONLY
 #include "zstd.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
 #ifndef XXH_STATIC_LINKING_ONLY
 #  define XXH_STATIC_LINKING_ONLY  /* XXH64_state_t */
 #endif
@@ -211,20 +215,6 @@ MEM_STATIC void ZSTD_wildcopy_e(void* dst, const void* src, void* dstEnd)   /* s
 *********************************************/
 typedef struct ZSTD_stats_s ZSTD_stats_t;
 
-typedef struct {
-    U32 off;
-    U32 len;
-} ZSTD_match_t;
-
-typedef struct {
-    U32 price;
-    U32 off;
-    U32 mlen;
-    U32 litlen;
-    U32 rep[ZSTD_REP_NUM];
-} ZSTD_optimal_t;
-
-
 typedef struct seqDef_s {
     U32 offset;
     U16 litLength;
@@ -242,13 +232,31 @@ typedef struct {
     BYTE* ofCode;
     U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
     U32   longLengthPos;
-    /* opt */
+    U32   rep[ZSTD_REP_NUM];
-    ZSTD_optimal_t* priceTable;
+    U32   repToConfirm[ZSTD_REP_NUM];
-    ZSTD_match_t* matchTable;
+} seqStore_t;
-    U32* matchLengthFreq;
+
-    U32* litLengthFreq;
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTD_match_t;
+
+typedef struct {
+    U32 price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_optimal_t;
+
+typedef struct {
     U32* litFreq;
+    U32* litLengthFreq;
+    U32* matchLengthFreq;
     U32* offCodeFreq;
+    ZSTD_match_t* matchTable;
+    ZSTD_optimal_t* priceTable;
+
     U32  matchLengthSum;
     U32  matchSum;
     U32  litLengthSum;
@@ -264,7 +272,19 @@ typedef struct {
     U32  cachedPrice;
     U32  cachedLitLength;
     const BYTE* cachedLiterals;
-} seqStore_t;
+} optState_t;
+
+typedef struct {
+    U32 hufCTable[HUF_CTABLE_SIZE_U32(255)];
+    FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
+    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
+    FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
+    U32 workspace[HUF_WORKSPACE_SIZE_U32];
+    HUF_repeat hufCTable_repeatMode;
+    FSE_repeat offcode_repeatMode;
+    FSE_repeat matchlength_repeatMode;
+    FSE_repeat litlength_repeatMode;
+} ZSTD_entropyCTables_t;
 
 const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);
 void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);

lib/compress/fse_compress.c

@@ -781,7 +781,7 @@ size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
 
 size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
 
-#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return f
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
 #define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
 
 /* FSE_compress_wksp() :

lib/compress/huf_compress.c

@@ -56,7 +56,7 @@
 *  Error Management
 ****************************************************************/
 #define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
-#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return f
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
 #define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
 
 
@@ -436,7 +436,7 @@ static void HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt*
 
 size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
 
-#define HUF_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+#define HUF_FLUSHBITS(s)  BIT_flushBits(s)
 
 #define HUF_FLUSHBITS_1(stream) \
     if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*2+7) HUF_FLUSHBITS(stream)
@@ -451,7 +451,6 @@ size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, si
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
     size_t n;
-    const unsigned fast = (dstSize >= HUF_BLOCKBOUND(srcSize));
     BIT_CStream_t bitC;
 
     /* init */

lib/compress/zstd_compress.c

@@ -36,13 +36,6 @@ static const U32 g_searchStrength = 8;   /* control skip over incompressible dat
 #define HASH_READ_SIZE 8
 typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
 
-/* entropy tables always have same size */
-static size_t const hufCTable_size = HUF_CTABLE_SIZE(255);
-static size_t const litlengthCTable_size = FSE_CTABLE_SIZE(LLFSELog, MaxLL);
-static size_t const offcodeCTable_size = FSE_CTABLE_SIZE(OffFSELog, MaxOff);
-static size_t const matchlengthCTable_size = FSE_CTABLE_SIZE(MLFSELog, MaxML);
-static size_t const entropyScratchSpace_size = HUF_WORKSPACE_SIZE;
-
 
 /*-*************************************
 *  Helper functions
@@ -89,8 +82,6 @@ struct ZSTD_CCtx_s {
     U32   loadedDictEnd;    /* index of end of dictionary */
     U32   forceWindow;      /* force back-references to respect limit of 1<<wLog, even for dictionary */
     ZSTD_compressionStage_e stage;
-    U32   rep[ZSTD_REP_NUM];
-    U32   repToConfirm[ZSTD_REP_NUM];
     U32   dictID;
     int   compressionLevel;
     ZSTD_parameters requestedParams;
@@ -105,16 +96,11 @@ struct ZSTD_CCtx_s {
     size_t staticSize;
 
     seqStore_t seqStore;    /* sequences storage ptrs */
+    optState_t optState;
     U32* hashTable;
     U32* hashTable3;
     U32* chainTable;
-    HUF_repeat hufCTable_repeatMode;
+    ZSTD_entropyCTables_t* entropy;
-    HUF_CElt* hufCTable;
-    U32 fseCTables_ready;
-    FSE_CTable* offcodeCTable;
-    FSE_CTable* matchlengthCTable;
-    FSE_CTable* litlengthCTable;
-    unsigned* entropyScratchSpace;
 
     /* streaming */
     char*  inBuff;
@@ -174,19 +160,9 @@ ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize)
     cctx->workSpaceSize = workspaceSize - sizeof(ZSTD_CCtx);
 
     /* entropy space (never moves) */
-    /* note : this code should be shared with resetCCtx, rather than copy/pasted */
+    if (cctx->workSpaceSize < sizeof(ZSTD_entropyCTables_t)) return NULL;
-    {   void* ptr = cctx->workSpace;
+    assert(((size_t)cctx->workSpace & 7) == 0);   /* ensure correct alignment */
-        cctx->hufCTable = (HUF_CElt*)ptr;
+    cctx->entropy = (ZSTD_entropyCTables_t*)cctx->workSpace;
-        ptr = (char*)cctx->hufCTable + hufCTable_size;
-        cctx->offcodeCTable = (FSE_CTable*) ptr;
-        ptr = (char*)ptr + offcodeCTable_size;
-        cctx->matchlengthCTable = (FSE_CTable*) ptr;
-        ptr = (char*)ptr + matchlengthCTable_size;
-        cctx->litlengthCTable = (FSE_CTable*) ptr;
-        ptr = (char*)ptr + litlengthCTable_size;
-        assert(((size_t)ptr & 3) == 0);   /* ensure correct alignment */
-        cctx->entropyScratchSpace = (unsigned*) ptr;
-    }
 
     return cctx;
 }
@@ -237,7 +213,7 @@ size_t ZSTD_setCCtxParameter(ZSTD_CCtx* cctx, ZSTD_CCtxParameter param, unsigned
     ZSTD_STATIC_ASSERT(ZSTD_dm_auto==0);
     ZSTD_STATIC_ASSERT(ZSTD_dm_rawContent==1);
     case ZSTD_p_forceRawDict : cctx->dictMode = (ZSTD_dictMode_e)(value>0); return 0;
-    default: return ERROR(parameter_unknown);
+    default: return ERROR(parameter_unsupported);
     }
 }
 
@@ -253,7 +229,7 @@ static void ZSTD_cLevelToCParams(ZSTD_CCtx* cctx)
 
 #define CLAMPCHECK(val,min,max) {            \
     if (((val)<(min)) | ((val)>(max))) {     \
-        return ERROR(compressionParameter_outOfBound);  \
+        return ERROR(parameter_outOfBound);  \
 }   }
 
 size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value)
@@ -349,7 +325,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned v
         /* restrict dictionary mode, to "rawContent" or "fullDict" only */
         ZSTD_STATIC_ASSERT((U32)ZSTD_dm_fullDict > (U32)ZSTD_dm_rawContent);
         if (value > (unsigned)ZSTD_dm_fullDict)
-            return ERROR(compressionParameter_outOfBound);
+            return ERROR(parameter_outOfBound);
         cctx->dictMode = (ZSTD_dictMode_e)value;
         return 0;
 
@@ -370,31 +346,31 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned v
         if (value==0) return 0;
         DEBUGLOG(5, " setting nbThreads : %u", value);
 #ifndef ZSTD_MULTITHREAD
-        if (value > 1) return ERROR(compressionParameter_unsupported);
+        if (value > 1) return ERROR(parameter_unsupported);
 #endif
         if ((value>1) && (cctx->nbThreads != value)) {
             if (cctx->staticSize)  /* MT not compatible with static alloc */
-                return ERROR(compressionParameter_unsupported);
+                return ERROR(parameter_unsupported);
             ZSTDMT_freeCCtx(cctx->mtctx);
             cctx->nbThreads = 1;
-            cctx->mtctx = ZSTDMT_createCCtx(value);
+            cctx->mtctx = ZSTDMT_createCCtx_advanced(value, cctx->customMem);
             if (cctx->mtctx == NULL) return ERROR(memory_allocation);
         }
         cctx->nbThreads = value;
         return 0;
 
     case ZSTD_p_jobSize:
-        if (cctx->nbThreads <= 1) return ERROR(compressionParameter_unsupported);
+        if (cctx->nbThreads <= 1) return ERROR(parameter_unsupported);
         assert(cctx->mtctx != NULL);
         return ZSTDMT_setMTCtxParameter(cctx->mtctx, ZSTDMT_p_sectionSize, value);
 
     case ZSTD_p_overlapSizeLog:
         DEBUGLOG(5, " setting overlap with nbThreads == %u", cctx->nbThreads);
-        if (cctx->nbThreads <= 1) return ERROR(compressionParameter_unsupported);
+        if (cctx->nbThreads <= 1) return ERROR(parameter_unsupported);
         assert(cctx->mtctx != NULL);
         return ZSTDMT_setMTCtxParameter(cctx->mtctx, ZSTDMT_p_overlapSectionLog, value);
 
-    default: return ERROR(parameter_unknown);
+    default: return ERROR(parameter_unsupported);
     }
 }
 
@@ -474,7 +450,8 @@ size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
     CLAMPCHECK(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
     CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
     CLAMPCHECK(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
-    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra) return ERROR(compressionParameter_unsupported);
+    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra)
+        return ERROR(parameter_unsupported);
     return 0;
 }
 
@@ -551,9 +528,7 @@ size_t ZSTD_estimateCCtxSize_advanced(ZSTD_compressionParameters cParams)
     size_t const hSize = ((size_t)1) << cParams.hashLog;
     U32    const hashLog3 = (cParams.searchLength>3) ? 0 : MIN(ZSTD_HASHLOG3_MAX, cParams.windowLog);
     size_t const h3Size = ((size_t)1) << hashLog3;
-    size_t const entropySpace = hufCTable_size + litlengthCTable_size
+    size_t const entropySpace = sizeof(ZSTD_entropyCTables_t);
-                              + offcodeCTable_size + matchlengthCTable_size
-                              + entropyScratchSpace_size;
     size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
 
     size_t const optBudget = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits))*sizeof(U32)
@@ -620,8 +595,8 @@ static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_parameters params, U64 ple
     cctx->stage = ZSTDcs_init;
     cctx->dictID = 0;
     cctx->loadedDictEnd = 0;
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) cctx->rep[i] = repStartValue[i]; }
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) cctx->seqStore.rep[i] = repStartValue[i]; }
-    cctx->seqStore.litLengthSum = 0;  /* force reset of btopt stats */
+    cctx->optState.litLengthSum = 0;  /* force reset of btopt stats */
     XXH64_reset(&cctx->xxhState, 0);
     return 0;
 }
@@ -641,8 +616,10 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
     if (crp == ZSTDcrp_continue) {
         if (ZSTD_equivalentParams(params.cParams, zc->appliedParams.cParams)) {
             DEBUGLOG(5, "ZSTD_equivalentParams()==1");
-            zc->fseCTables_ready = 0;
+            zc->entropy->hufCTable_repeatMode = HUF_repeat_none;
-            zc->hufCTable_repeatMode = HUF_repeat_none;
+            zc->entropy->offcode_repeatMode = FSE_repeat_none;
+            zc->entropy->matchlength_repeatMode = FSE_repeat_none;
+            zc->entropy->litlength_repeatMode = FSE_repeat_none;
             return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
     }   }
 
@@ -662,9 +639,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         void* ptr;
 
         /* Check if workSpace is large enough, alloc a new one if needed */
-        {   size_t const entropySpace = hufCTable_size + litlengthCTable_size
+        {   size_t const entropySpace = sizeof(ZSTD_entropyCTables_t);
-                                  + offcodeCTable_size + matchlengthCTable_size
-                                  + entropyScratchSpace_size;
             size_t const optPotentialSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits)) * sizeof(U32)
                                   + (ZSTD_OPT_NUM+1) * (sizeof(ZSTD_match_t)+sizeof(ZSTD_optimal_t));
             size_t const optSpace = ( (params.cParams.strategy == ZSTD_btopt)
@@ -689,16 +664,9 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                 ptr = zc->workSpace;
 
                 /* entropy space */
-                zc->hufCTable = (HUF_CElt*)ptr;
+                assert(((size_t)zc->workSpace & 3) == 0);   /* ensure correct alignment */
-                ptr = (char*)zc->hufCTable + hufCTable_size;  /* note : HUF_CElt* is incomplete type, size is estimated via macro */
+                assert(zc->workSpaceSize >= sizeof(ZSTD_entropyCTables_t));
-                zc->offcodeCTable = (FSE_CTable*) ptr;
+                zc->entropy = (ZSTD_entropyCTables_t*)zc->workSpace;
-                ptr = (char*)ptr + offcodeCTable_size;
-                zc->matchlengthCTable = (FSE_CTable*) ptr;
-                ptr = (char*)ptr + matchlengthCTable_size;
-                zc->litlengthCTable = (FSE_CTable*) ptr;
-                ptr = (char*)ptr + litlengthCTable_size;
-                assert(((size_t)ptr & 3) == 0);   /* ensure correct alignment */
-                zc->entropyScratchSpace = (unsigned*) ptr;
         }   }
 
         /* init params */
@@ -715,39 +683,35 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         zc->stage = ZSTDcs_init;
         zc->dictID = 0;
         zc->loadedDictEnd = 0;
-        zc->fseCTables_ready = 0;
+        zc->entropy->hufCTable_repeatMode = HUF_repeat_none;
-        zc->hufCTable_repeatMode = HUF_repeat_none;
+        zc->entropy->offcode_repeatMode = FSE_repeat_none;
+        zc->entropy->matchlength_repeatMode = FSE_repeat_none;
+        zc->entropy->litlength_repeatMode = FSE_repeat_none;
         zc->nextToUpdate = 1;
         zc->nextSrc = NULL;
         zc->base = NULL;
         zc->dictBase = NULL;
         zc->dictLimit = 0;
         zc->lowLimit = 0;
-        { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->rep[i] = repStartValue[i]; }
+        { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->seqStore.rep[i] = repStartValue[i]; }
         zc->hashLog3 = hashLog3;
-        zc->seqStore.litLengthSum = 0;
+        zc->optState.litLengthSum = 0;
 
-        /* ensure entropy tables are close together at the beginning */
+        ptr = zc->entropy + 1;
-        assert((void*)zc->hufCTable == zc->workSpace);
-        assert((char*)zc->offcodeCTable == (char*)zc->hufCTable + hufCTable_size);
-        assert((char*)zc->matchlengthCTable == (char*)zc->offcodeCTable + offcodeCTable_size);
-        assert((char*)zc->litlengthCTable == (char*)zc->matchlengthCTable + matchlengthCTable_size);
-        assert((char*)zc->entropyScratchSpace == (char*)zc->litlengthCTable + litlengthCTable_size);
-        ptr = (char*)zc->entropyScratchSpace + entropyScratchSpace_size;
 
         /* opt parser space */
         if ((params.cParams.strategy == ZSTD_btopt) || (params.cParams.strategy == ZSTD_btultra)) {
             DEBUGLOG(5, "reserving optimal parser space");
             assert(((size_t)ptr & 3) == 0);  /* ensure ptr is properly aligned */
-            zc->seqStore.litFreq = (U32*)ptr;
+            zc->optState.litFreq = (U32*)ptr;
-            zc->seqStore.litLengthFreq = zc->seqStore.litFreq + (1<<Litbits);
+            zc->optState.litLengthFreq = zc->optState.litFreq + (1<<Litbits);
-            zc->seqStore.matchLengthFreq = zc->seqStore.litLengthFreq + (MaxLL+1);
+            zc->optState.matchLengthFreq = zc->optState.litLengthFreq + (MaxLL+1);
-            zc->seqStore.offCodeFreq = zc->seqStore.matchLengthFreq + (MaxML+1);
+            zc->optState.offCodeFreq = zc->optState.matchLengthFreq + (MaxML+1);
-            ptr = zc->seqStore.offCodeFreq + (MaxOff+1);
+            ptr = zc->optState.offCodeFreq + (MaxOff+1);
-            zc->seqStore.matchTable = (ZSTD_match_t*)ptr;
+            zc->optState.matchTable = (ZSTD_match_t*)ptr;
-            ptr = zc->seqStore.matchTable + ZSTD_OPT_NUM+1;
+            ptr = zc->optState.matchTable + ZSTD_OPT_NUM+1;
-            zc->seqStore.priceTable = (ZSTD_optimal_t*)ptr;
+            zc->optState.priceTable = (ZSTD_optimal_t*)ptr;
-            ptr = zc->seqStore.priceTable + ZSTD_OPT_NUM+1;
+            ptr = zc->optState.priceTable + ZSTD_OPT_NUM+1;
         }
 
         /* table Space */
@@ -783,7 +747,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
  *        do not use with extDict variant ! */
 void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
     int i;
-    for (i=0; i<ZSTD_REP_NUM; i++) cctx->rep[i] = 0;
+    for (i=0; i<ZSTD_REP_NUM; i++) cctx->seqStore.rep[i] = 0;
 }
 
 
@@ -830,16 +794,7 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
     dstCCtx->dictID       = srcCCtx->dictID;
 
     /* copy entropy tables */
-    dstCCtx->fseCTables_ready = srcCCtx->fseCTables_ready;
+    memcpy(dstCCtx->entropy, srcCCtx->entropy, sizeof(ZSTD_entropyCTables_t));
-    if (srcCCtx->fseCTables_ready) {
-        memcpy(dstCCtx->litlengthCTable, srcCCtx->litlengthCTable, litlengthCTable_size);
-        memcpy(dstCCtx->matchlengthCTable, srcCCtx->matchlengthCTable, matchlengthCTable_size);
-        memcpy(dstCCtx->offcodeCTable, srcCCtx->offcodeCTable, offcodeCTable_size);
-    }
-    dstCCtx->hufCTable_repeatMode = srcCCtx->hufCTable_repeatMode;
-    if (srcCCtx->hufCTable_repeatMode) {
-        memcpy(dstCCtx->hufCTable, srcCCtx->hufCTable, hufCTable_size);
-    }
 
     return 0;
 }
@@ -956,7 +911,8 @@ static size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, cons
 
 static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
 
-static size_t ZSTD_compressLiterals (ZSTD_CCtx* zc,
+static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t * entropy,
+                                     ZSTD_strategy strategy,
                                      void* dst, size_t dstCapacity,
                                const void* src, size_t srcSize)
 {
@@ -970,28 +926,28 @@ static size_t ZSTD_compressLiterals (ZSTD_CCtx* zc,
 
     /* small ? don't even attempt compression (speed opt) */
 #   define LITERAL_NOENTROPY 63
-    {   size_t const minLitSize = zc->hufCTable_repeatMode == HUF_repeat_valid ? 6 : LITERAL_NOENTROPY;
+    {   size_t const minLitSize = entropy->hufCTable_repeatMode == HUF_repeat_valid ? 6 : LITERAL_NOENTROPY;
         if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
     }
 
     if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall);   /* not enough space for compression */
-    {   HUF_repeat repeat = zc->hufCTable_repeatMode;
+    {   HUF_repeat repeat = entropy->hufCTable_repeatMode;
-        int const preferRepeat = zc->appliedParams.cParams.strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
+        int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
         if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
         cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
-                                      zc->entropyScratchSpace, entropyScratchSpace_size, zc->hufCTable, &repeat, preferRepeat)
+                                      entropy->workspace, sizeof(entropy->workspace), (HUF_CElt*)entropy->hufCTable, &repeat, preferRepeat)
                                 : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
-                                      zc->entropyScratchSpace, entropyScratchSpace_size, zc->hufCTable, &repeat, preferRepeat);
+                                      entropy->workspace, sizeof(entropy->workspace), (HUF_CElt*)entropy->hufCTable, &repeat, preferRepeat);
         if (repeat != HUF_repeat_none) { hType = set_repeat; }    /* reused the existing table */
-        else { zc->hufCTable_repeatMode = HUF_repeat_check; }       /* now have a table to reuse */
+        else { entropy->hufCTable_repeatMode = HUF_repeat_check; }       /* now have a table to reuse */
     }
 
     if ((cLitSize==0) | (cLitSize >= srcSize - minGain)) {
-        zc->hufCTable_repeatMode = HUF_repeat_none;
+        entropy->hufCTable_repeatMode = HUF_repeat_none;
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
     }
     if (cLitSize==1) {
-        zc->hufCTable_repeatMode = HUF_repeat_none;
+        entropy->hufCTable_repeatMode = HUF_repeat_none;
         return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
     }
 
@@ -1062,135 +1018,78 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
         mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
 }
 
-MEM_STATIC size_t ZSTD_compressSequences (ZSTD_CCtx* zc,
+MEM_STATIC symbolEncodingType_e ZSTD_selectEncodingType(FSE_repeat* repeatMode,
-                              void* dst, size_t dstCapacity,
+        size_t const mostFrequent, size_t nbSeq, U32 defaultNormLog)
-                              size_t srcSize)
 {
-    const int longOffsets = zc->appliedParams.cParams.windowLog > STREAM_ACCUMULATOR_MIN;
-    const seqStore_t* seqStorePtr = &(zc->seqStore);
-    U32 count[MaxSeq+1];
-    S16 norm[MaxSeq+1];
-    FSE_CTable* CTable_LitLength = zc->litlengthCTable;
-    FSE_CTable* CTable_OffsetBits = zc->offcodeCTable;
-    FSE_CTable* CTable_MatchLength = zc->matchlengthCTable;
-    U32 LLtype, Offtype, MLtype;   /* compressed, raw or rle */
-    const seqDef* const sequences = seqStorePtr->sequencesStart;
-    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
-    const BYTE* const llCodeTable = seqStorePtr->llCode;
-    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
-    BYTE* const ostart = (BYTE*)dst;
-    BYTE* const oend = ostart + dstCapacity;
-    BYTE* op = ostart;
-    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
-    BYTE* seqHead;
-    BYTE scratchBuffer[1<<MAX(MLFSELog,LLFSELog)];
-
-    /* Compress literals */
-    {   const BYTE* const literals = seqStorePtr->litStart;
-        size_t const litSize = seqStorePtr->lit - literals;
-        size_t const cSize = ZSTD_compressLiterals(zc, op, dstCapacity, literals, litSize);
-        if (ZSTD_isError(cSize)) return cSize;
-        op += cSize;
-    }
-
-    /* Sequences Header */
-    if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead */) return ERROR(dstSize_tooSmall);
-    if (nbSeq < 0x7F) *op++ = (BYTE)nbSeq;
-    else if (nbSeq < LONGNBSEQ) op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
-    else op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
-    if (nbSeq==0) goto _check_compressibility;
-
-    /* seqHead : flags for FSE encoding type */
-    seqHead = op++;
-
 #define MIN_SEQ_FOR_DYNAMIC_FSE   64
 #define MAX_SEQ_FOR_STATIC_FSE  1000
 
-    /* convert length/distances into codes */
-    ZSTD_seqToCodes(seqStorePtr);
-
-    /* CTable for Literal Lengths */
-    {   U32 max = MaxLL;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, zc->entropyScratchSpace);
     if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
-            *op++ = llCodeTable[0];
+        *repeatMode = FSE_repeat_check;
-            FSE_buildCTable_rle(CTable_LitLength, (BYTE)max);
+        return set_rle;
-            LLtype = set_rle;
+    }
-        } else if ((zc->fseCTables_ready) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+    if ((*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-            LLtype = set_repeat;
+        return set_repeat;
-        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (LL_defaultNormLog-1)))) {
+    }
-            FSE_buildCTable_wksp(CTable_LitLength, LL_defaultNorm, MaxLL, LL_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
+    if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) {
-            LLtype = set_basic;
+        *repeatMode = FSE_repeat_valid;
-        } else {
+        return set_basic;
+    }
+    *repeatMode = FSE_repeat_check;
+    return set_compressed;
+}
+
+MEM_STATIC size_t ZSTD_buildCTable(void* dst, size_t dstCapacity,
+        FSE_CTable* CTable, U32 FSELog, symbolEncodingType_e type,
+        U32* count, U32 max,
+        BYTE const* codeTable, size_t nbSeq,
+        S16 const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+        void* workspace, size_t workspaceSize)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE const* const oend = op + dstCapacity;
+
+    switch (type) {
+    case set_rle:
+        *op = codeTable[0];
+        CHECK_F(FSE_buildCTable_rle(CTable, (BYTE)max));
+        return 1;
+    case set_repeat:
+        return 0;
+    case set_basic:
+        CHECK_F(FSE_buildCTable_wksp(CTable, defaultNorm, defaultMax, defaultNormLog, workspace, workspaceSize));
+        return 0;
+    case set_compressed: {
+        S16 norm[MaxSeq + 1];
         size_t nbSeq_1 = nbSeq;
-            const U32 tableLog = FSE_optimalTableLog(LLFSELog, nbSeq, max);
+        const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
-            if (count[llCodeTable[nbSeq-1]]>1) { count[llCodeTable[nbSeq-1]]--; nbSeq_1--; }
+        if (count[codeTable[nbSeq-1]] > 1) {
-            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
+            count[codeTable[nbSeq-1]]--;
+            nbSeq_1--;
+        }
+        CHECK_F(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max));
         {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
             if (FSE_isError(NCountSize)) return NCountSize;
-              op += NCountSize; }
+            CHECK_F(FSE_buildCTable_wksp(CTable, norm, max, tableLog, workspace, workspaceSize));
-            FSE_buildCTable_wksp(CTable_LitLength, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
+            return NCountSize;
-            LLtype = set_compressed;
+        }
-    }   }
+    }
+    default: return assert(0), ERROR(GENERIC);
+    }
+}
 
-    /* CTable for Offsets */
+MEM_STATIC size_t ZSTD_encodeSequences(void* dst, size_t dstCapacity,
-    {   U32 max = MaxOff;
+    FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, zc->entropyScratchSpace);
+    FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
-        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
+    FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
-            *op++ = ofCodeTable[0];
+    seqDef const* sequences, size_t nbSeq, int longOffsets)
-            FSE_buildCTable_rle(CTable_OffsetBits, (BYTE)max);
+{
-            Offtype = set_rle;
+    BIT_CStream_t blockStream;
-        } else if ((zc->fseCTables_ready) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-            Offtype = set_repeat;
-        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (OF_defaultNormLog-1)))) {
-            FSE_buildCTable_wksp(CTable_OffsetBits, OF_defaultNorm, MaxOff, OF_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
-            Offtype = set_basic;
-        } else {
-            size_t nbSeq_1 = nbSeq;
-            const U32 tableLog = FSE_optimalTableLog(OffFSELog, nbSeq, max);
-            if (count[ofCodeTable[nbSeq-1]]>1) { count[ofCodeTable[nbSeq-1]]--; nbSeq_1--; }
-            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
-            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
-              if (FSE_isError(NCountSize)) return NCountSize;
-              op += NCountSize; }
-            FSE_buildCTable_wksp(CTable_OffsetBits, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
-            Offtype = set_compressed;
-    }   }
-
-    /* CTable for MatchLengths */
-    {   U32 max = MaxML;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, zc->entropyScratchSpace);
-        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
-            *op++ = *mlCodeTable;
-            FSE_buildCTable_rle(CTable_MatchLength, (BYTE)max);
-            MLtype = set_rle;
-        } else if ((zc->fseCTables_ready) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-            MLtype = set_repeat;
-        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (ML_defaultNormLog-1)))) {
-            FSE_buildCTable_wksp(CTable_MatchLength, ML_defaultNorm, MaxML, ML_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
-            MLtype = set_basic;
-        } else {
-            size_t nbSeq_1 = nbSeq;
-            const U32 tableLog = FSE_optimalTableLog(MLFSELog, nbSeq, max);
-            if (count[mlCodeTable[nbSeq-1]]>1) { count[mlCodeTable[nbSeq-1]]--; nbSeq_1--; }
-            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
-            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
-              if (FSE_isError(NCountSize)) return NCountSize;
-              op += NCountSize; }
-            FSE_buildCTable_wksp(CTable_MatchLength, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
-            MLtype = set_compressed;
-    }   }
-
-    *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
-    zc->fseCTables_ready = 0;
-
-    /* Encoding Sequences */
-    {   BIT_CStream_t blockStream;
     FSE_CState_t  stateMatchLength;
     FSE_CState_t  stateOffsetBits;
     FSE_CState_t  stateLitLength;
 
-        CHECK_E(BIT_initCStream(&blockStream, op, oend-op), dstSize_tooSmall); /* not enough space remaining */
+    CHECK_E(BIT_initCStream(&blockStream, dst, dstCapacity), dstSize_tooSmall); /* not enough space remaining */
 
     /* first symbols */
     FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
@@ -1253,20 +1152,113 @@ MEM_STATIC size_t ZSTD_compressSequences (ZSTD_CCtx* zc,
 
     {   size_t const streamSize = BIT_closeCStream(&blockStream);
         if (streamSize==0) return ERROR(dstSize_tooSmall);   /* not enough space */
-            op += streamSize;
+        return streamSize;
+    }
+}
+
+MEM_STATIC size_t ZSTD_compressSequences (seqStore_t* seqStorePtr,
+                              ZSTD_entropyCTables_t* entropy,
+                              ZSTD_compressionParameters const* cParams,
+                              void* dst, size_t dstCapacity,
+                              size_t srcSize)
+{
+    const int longOffsets = cParams->windowLog > STREAM_ACCUMULATOR_MIN;
+    U32 count[MaxSeq+1];
+    FSE_CTable* CTable_LitLength = entropy->litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = entropy->offcodeCTable;
+    FSE_CTable* CTable_MatchLength = entropy->matchlengthCTable;
+    U32 LLtype, Offtype, MLtype;   /* compressed, raw or rle */
+    const seqDef* const sequences = seqStorePtr->sequencesStart;
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    BYTE* seqHead;
+
+    ZSTD_STATIC_ASSERT(sizeof(entropy->workspace) >= (1<<MAX(MLFSELog,LLFSELog)));
+
+    /* Compress literals */
+    {   const BYTE* const literals = seqStorePtr->litStart;
+        size_t const litSize = seqStorePtr->lit - literals;
+        size_t const cSize = ZSTD_compressLiterals(
+                entropy, cParams->strategy, op, dstCapacity, literals, litSize);
+        if (ZSTD_isError(cSize))
+          return cSize;
+        op += cSize;
+    }
+
+    /* Sequences Header */
+    if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead */) return ERROR(dstSize_tooSmall);
+    if (nbSeq < 0x7F) *op++ = (BYTE)nbSeq;
+    else if (nbSeq < LONGNBSEQ) op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
+    else op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
+    if (nbSeq==0) goto _check_compressibility;
+
+    /* seqHead : flags for FSE encoding type */
+    seqHead = op++;
+
+    /* convert length/distances into codes */
+    ZSTD_seqToCodes(seqStorePtr);
+    /* CTable for Literal Lengths */
+    {   U32 max = MaxLL;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, entropy->workspace);
+        LLtype = ZSTD_selectEncodingType(&entropy->litlength_repeatMode, mostFrequent, nbSeq, LL_defaultNormLog);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
+                    count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                    entropy->workspace, sizeof(entropy->workspace));
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
     }   }
+    /* CTable for Offsets */
+    {   U32 max = MaxOff;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, entropy->workspace);
+        Offtype = ZSTD_selectEncodingType(&entropy->offcode_repeatMode, mostFrequent, nbSeq, OF_defaultNormLog);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
+                    count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, MaxOff,
+                    entropy->workspace, sizeof(entropy->workspace));
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
+    }   }
+    /* CTable for MatchLengths */
+    {   U32 max = MaxML;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, entropy->workspace);
+        MLtype = ZSTD_selectEncodingType(&entropy->matchlength_repeatMode, mostFrequent, nbSeq, ML_defaultNormLog);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
+                    count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
+                    entropy->workspace, sizeof(entropy->workspace));
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
+    }   }
+
+    *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
+
+    {   size_t const streamSize = ZSTD_encodeSequences(op, oend - op,
+                CTable_MatchLength, mlCodeTable,
+                CTable_OffsetBits, ofCodeTable,
+                CTable_LitLength, llCodeTable,
+                sequences, nbSeq, longOffsets);
+        if (ZSTD_isError(streamSize)) return streamSize;
+        op += streamSize;
+    }
+
 
     /* check compressibility */
 _check_compressibility:
     {   size_t const minGain = ZSTD_minGain(srcSize);
         size_t const maxCSize = srcSize - minGain;
         if ((size_t)(op-ostart) >= maxCSize) {
-            zc->hufCTable_repeatMode = HUF_repeat_none;
+            entropy->hufCTable_repeatMode = HUF_repeat_none;
+            entropy->offcode_repeatMode = FSE_repeat_none;
+            entropy->matchlength_repeatMode = FSE_repeat_none;
+            entropy->litlength_repeatMode = FSE_repeat_none;
             return 0;
     }   }
 
     /* confirm repcodes */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->rep[i] = zc->repToConfirm[i]; }
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->rep[i] = seqStorePtr->repToConfirm[i]; }
 
     return op - ostart;
 }
@@ -1491,7 +1483,7 @@ void ZSTD_compressBlock_fast_generic(ZSTD_CCtx* cctx,
     const BYTE* const lowest = base + lowestIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
-    U32 offset_1=cctx->rep[0], offset_2=cctx->rep[1];
+    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
     U32 offsetSaved = 0;
 
     /* init */
@@ -1552,8 +1544,8 @@ void ZSTD_compressBlock_fast_generic(ZSTD_CCtx* cctx,
     }   }   }
 
     /* save reps for next block */
-    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
+    seqStorePtr->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
-    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
+    seqStorePtr->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1601,7 +1593,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    U32 offset_1=ctx->rep[0], offset_2=ctx->rep[1];
+    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
 
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
@@ -1667,7 +1659,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
     }   }   }
 
     /* save reps for next block */
-    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
+    seqStorePtr->repToConfirm[0] = offset_1; seqStorePtr->repToConfirm[1] = offset_2;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1736,7 +1728,7 @@ void ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
     const BYTE* const lowest = base + lowestIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
-    U32 offset_1=cctx->rep[0], offset_2=cctx->rep[1];
+    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
     U32 offsetSaved = 0;
 
     /* init */
@@ -1823,8 +1815,8 @@ void ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
     }   }   }
 
     /* save reps for next block */
-    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
+    seqStorePtr->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
-    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
+    seqStorePtr->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1873,7 +1865,7 @@ static void ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    U32 offset_1=ctx->rep[0], offset_2=ctx->rep[1];
+    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
 
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
@@ -1973,7 +1965,7 @@ static void ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
     }   }   }
 
     /* save reps for next block */
-    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
+    seqStorePtr->repToConfirm[0] = offset_1; seqStorePtr->repToConfirm[1] = offset_2;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -2409,7 +2401,7 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
                         size_t* offsetPtr,
                         U32 maxNbAttempts, U32 matchLengthSearch);
     searchMax_f const searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
-    U32 offset_1 = ctx->rep[0], offset_2 = ctx->rep[1], savedOffset=0;
+    U32 offset_1 = seqStorePtr->rep[0], offset_2 = seqStorePtr->rep[1], savedOffset=0;
 
     /* init */
     ip += (ip==base);
@@ -2519,8 +2511,8 @@ _storeSequence:
     }   }
 
     /* Save reps for next block */
-    ctx->repToConfirm[0] = offset_1 ? offset_1 : savedOffset;
+    seqStorePtr->repToConfirm[0] = offset_1 ? offset_1 : savedOffset;
-    ctx->repToConfirm[1] = offset_2 ? offset_2 : savedOffset;
+    seqStorePtr->repToConfirm[1] = offset_2 ? offset_2 : savedOffset;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -2578,7 +2570,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                         U32 maxNbAttempts, U32 matchLengthSearch);
     searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
 
-    U32 offset_1 = ctx->rep[0], offset_2 = ctx->rep[1];
+    U32 offset_1 = seqStorePtr->rep[0], offset_2 = seqStorePtr->rep[1];
 
     /* init */
     ctx->nextToUpdate3 = ctx->nextToUpdate;
@@ -2714,7 +2706,7 @@ _storeSequence:
     }   }
 
     /* Save reps for next block */
-    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
+    seqStorePtr->repToConfirm[0] = offset_1; seqStorePtr->repToConfirm[1] = offset_2;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -2823,7 +2815,7 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCa
     if (current > zc->nextToUpdate + 384)
         zc->nextToUpdate = current - MIN(192, (U32)(current - zc->nextToUpdate - 384));   /* limited update after finding a very long match */
     blockCompressor(zc, src, srcSize);
-    return ZSTD_compressSequences(zc, dst, dstCapacity, srcSize);
+    return ZSTD_compressSequences(&zc->seqStore, zc->entropy, &zc->appliedParams.cParams, dst, dstCapacity, srcSize);
 }
 
 
@@ -3105,13 +3097,14 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t
     const BYTE* const dictEnd = dictPtr + dictSize;
     short offcodeNCount[MaxOff+1];
     unsigned offcodeMaxValue = MaxOff;
-    BYTE scratchBuffer[1<<MAX(MLFSELog,LLFSELog)];
+
+    ZSTD_STATIC_ASSERT(sizeof(cctx->entropy->workspace) >= (1<<MAX(MLFSELog,LLFSELog)));
 
     dictPtr += 4;   /* skip magic number */
     cctx->dictID = cctx->appliedParams.fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr);
     dictPtr += 4;
 
-    {   size_t const hufHeaderSize = HUF_readCTable(cctx->hufCTable, 255, dictPtr, dictEnd-dictPtr);
+    {   size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)cctx->entropy->hufCTable, 255, dictPtr, dictEnd-dictPtr);
         if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
         dictPtr += hufHeaderSize;
     }
@@ -3121,7 +3114,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
         if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
         /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
-        CHECK_E( FSE_buildCTable_wksp(cctx->offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, scratchBuffer, sizeof(scratchBuffer)),
+        CHECK_E( FSE_buildCTable_wksp(cctx->entropy->offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, cctx->entropy->workspace, sizeof(cctx->entropy->workspace)),
                  dictionary_corrupted);
         dictPtr += offcodeHeaderSize;
     }
@@ -3133,7 +3126,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t
         if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
         /* Every match length code must have non-zero probability */
         CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
-        CHECK_E( FSE_buildCTable_wksp(cctx->matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, scratchBuffer, sizeof(scratchBuffer)),
+        CHECK_E( FSE_buildCTable_wksp(cctx->entropy->matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, cctx->entropy->workspace, sizeof(cctx->entropy->workspace)),
                  dictionary_corrupted);
         dictPtr += matchlengthHeaderSize;
     }
@@ -3145,15 +3138,15 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t
         if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
         /* Every literal length code must have non-zero probability */
         CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
-        CHECK_E( FSE_buildCTable_wksp(cctx->litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, scratchBuffer, sizeof(scratchBuffer)),
+        CHECK_E( FSE_buildCTable_wksp(cctx->entropy->litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, cctx->entropy->workspace, sizeof(cctx->entropy->workspace)),
                  dictionary_corrupted);
         dictPtr += litlengthHeaderSize;
     }
 
     if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
-    cctx->rep[0] = MEM_readLE32(dictPtr+0);
+    cctx->seqStore.rep[0] = MEM_readLE32(dictPtr+0);
-    cctx->rep[1] = MEM_readLE32(dictPtr+4);
+    cctx->seqStore.rep[1] = MEM_readLE32(dictPtr+4);
-    cctx->rep[2] = MEM_readLE32(dictPtr+8);
+    cctx->seqStore.rep[2] = MEM_readLE32(dictPtr+8);
     dictPtr += 12;
 
     {   size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
@@ -3167,12 +3160,14 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t
         /* All repCodes must be <= dictContentSize and != 0*/
         {   U32 u;
             for (u=0; u<3; u++) {
-                if (cctx->rep[u] == 0) return ERROR(dictionary_corrupted);
+                if (cctx->seqStore.rep[u] == 0) return ERROR(dictionary_corrupted);
-                if (cctx->rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
+                if (cctx->seqStore.rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
         }   }
 
-        cctx->fseCTables_ready = 1;
+        cctx->entropy->hufCTable_repeatMode = HUF_repeat_valid;
-        cctx->hufCTable_repeatMode = HUF_repeat_valid;
+        cctx->entropy->offcode_repeatMode = FSE_repeat_valid;
+        cctx->entropy->matchlength_repeatMode = FSE_repeat_valid;
+        cctx->entropy->litlength_repeatMode = FSE_repeat_valid;
         return ZSTD_loadDictionaryContent(cctx, dictPtr, dictContentSize);
     }
 }

lib/compress/zstd_opt.h

@@ -22,173 +22,173 @@
 /*-*************************************
 *  Price functions for optimal parser
 ***************************************/
-FORCE_INLINE void ZSTD_setLog2Prices(seqStore_t* ssPtr)
+FORCE_INLINE void ZSTD_setLog2Prices(optState_t* optPtr)
 {
-    ssPtr->log2matchLengthSum = ZSTD_highbit32(ssPtr->matchLengthSum+1);
+    optPtr->log2matchLengthSum = ZSTD_highbit32(optPtr->matchLengthSum+1);
-    ssPtr->log2litLengthSum = ZSTD_highbit32(ssPtr->litLengthSum+1);
+    optPtr->log2litLengthSum = ZSTD_highbit32(optPtr->litLengthSum+1);
-    ssPtr->log2litSum = ZSTD_highbit32(ssPtr->litSum+1);
+    optPtr->log2litSum = ZSTD_highbit32(optPtr->litSum+1);
-    ssPtr->log2offCodeSum = ZSTD_highbit32(ssPtr->offCodeSum+1);
+    optPtr->log2offCodeSum = ZSTD_highbit32(optPtr->offCodeSum+1);
-    ssPtr->factor = 1 + ((ssPtr->litSum>>5) / ssPtr->litLengthSum) + ((ssPtr->litSum<<1) / (ssPtr->litSum + ssPtr->matchSum));
+    optPtr->factor = 1 + ((optPtr->litSum>>5) / optPtr->litLengthSum) + ((optPtr->litSum<<1) / (optPtr->litSum + optPtr->matchSum));
 }
 
 
-MEM_STATIC void ZSTD_rescaleFreqs(seqStore_t* ssPtr, const BYTE* src, size_t srcSize)
+MEM_STATIC void ZSTD_rescaleFreqs(optState_t* optPtr, const BYTE* src, size_t srcSize)
 {
     unsigned u;
 
-    ssPtr->cachedLiterals = NULL;
+    optPtr->cachedLiterals = NULL;
-    ssPtr->cachedPrice = ssPtr->cachedLitLength = 0;
+    optPtr->cachedPrice = optPtr->cachedLitLength = 0;
-    ssPtr->staticPrices = 0;
+    optPtr->staticPrices = 0;
 
-    if (ssPtr->litLengthSum == 0) {
+    if (optPtr->litLengthSum == 0) {
-        if (srcSize <= 1024) ssPtr->staticPrices = 1;
+        if (srcSize <= 1024) optPtr->staticPrices = 1;
 
-        assert(ssPtr->litFreq!=NULL);
+        assert(optPtr->litFreq!=NULL);
         for (u=0; u<=MaxLit; u++)
-            ssPtr->litFreq[u] = 0;
+            optPtr->litFreq[u] = 0;
         for (u=0; u<srcSize; u++)
-            ssPtr->litFreq[src[u]]++;
+            optPtr->litFreq[src[u]]++;
 
-        ssPtr->litSum = 0;
+        optPtr->litSum = 0;
-        ssPtr->litLengthSum = MaxLL+1;
+        optPtr->litLengthSum = MaxLL+1;
-        ssPtr->matchLengthSum = MaxML+1;
+        optPtr->matchLengthSum = MaxML+1;
-        ssPtr->offCodeSum = (MaxOff+1);
+        optPtr->offCodeSum = (MaxOff+1);
-        ssPtr->matchSum = (ZSTD_LITFREQ_ADD<<Litbits);
+        optPtr->matchSum = (ZSTD_LITFREQ_ADD<<Litbits);
 
         for (u=0; u<=MaxLit; u++) {
-            ssPtr->litFreq[u] = 1 + (ssPtr->litFreq[u]>>ZSTD_FREQ_DIV);
+            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u]>>ZSTD_FREQ_DIV);
-            ssPtr->litSum += ssPtr->litFreq[u];
+            optPtr->litSum += optPtr->litFreq[u];
         }
         for (u=0; u<=MaxLL; u++)
-            ssPtr->litLengthFreq[u] = 1;
+            optPtr->litLengthFreq[u] = 1;
         for (u=0; u<=MaxML; u++)
-            ssPtr->matchLengthFreq[u] = 1;
+            optPtr->matchLengthFreq[u] = 1;
         for (u=0; u<=MaxOff; u++)
-            ssPtr->offCodeFreq[u] = 1;
+            optPtr->offCodeFreq[u] = 1;
     } else {
-        ssPtr->matchLengthSum = 0;
+        optPtr->matchLengthSum = 0;
-        ssPtr->litLengthSum = 0;
+        optPtr->litLengthSum = 0;
-        ssPtr->offCodeSum = 0;
+        optPtr->offCodeSum = 0;
-        ssPtr->matchSum = 0;
+        optPtr->matchSum = 0;
-        ssPtr->litSum = 0;
+        optPtr->litSum = 0;
 
         for (u=0; u<=MaxLit; u++) {
-            ssPtr->litFreq[u] = 1 + (ssPtr->litFreq[u]>>(ZSTD_FREQ_DIV+1));
+            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u]>>(ZSTD_FREQ_DIV+1));
-            ssPtr->litSum += ssPtr->litFreq[u];
+            optPtr->litSum += optPtr->litFreq[u];
         }
         for (u=0; u<=MaxLL; u++) {
-            ssPtr->litLengthFreq[u] = 1 + (ssPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
+            optPtr->litLengthFreq[u] = 1 + (optPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
-            ssPtr->litLengthSum += ssPtr->litLengthFreq[u];
+            optPtr->litLengthSum += optPtr->litLengthFreq[u];
         }
         for (u=0; u<=MaxML; u++) {
-            ssPtr->matchLengthFreq[u] = 1 + (ssPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
+            optPtr->matchLengthFreq[u] = 1 + (optPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
-            ssPtr->matchLengthSum += ssPtr->matchLengthFreq[u];
+            optPtr->matchLengthSum += optPtr->matchLengthFreq[u];
-            ssPtr->matchSum += ssPtr->matchLengthFreq[u] * (u + 3);
+            optPtr->matchSum += optPtr->matchLengthFreq[u] * (u + 3);
         }
-        ssPtr->matchSum *= ZSTD_LITFREQ_ADD;
+        optPtr->matchSum *= ZSTD_LITFREQ_ADD;
         for (u=0; u<=MaxOff; u++) {
-            ssPtr->offCodeFreq[u] = 1 + (ssPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
+            optPtr->offCodeFreq[u] = 1 + (optPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
-            ssPtr->offCodeSum += ssPtr->offCodeFreq[u];
+            optPtr->offCodeSum += optPtr->offCodeFreq[u];
         }
     }
 
-    ZSTD_setLog2Prices(ssPtr);
+    ZSTD_setLog2Prices(optPtr);
 }
 
 
-FORCE_INLINE U32 ZSTD_getLiteralPrice(seqStore_t* ssPtr, U32 litLength, const BYTE* literals)
+FORCE_INLINE U32 ZSTD_getLiteralPrice(optState_t* optPtr, U32 litLength, const BYTE* literals)
 {
     U32 price, u;
 
-    if (ssPtr->staticPrices)
+    if (optPtr->staticPrices)
         return ZSTD_highbit32((U32)litLength+1) + (litLength*6);
 
     if (litLength == 0)
-        return ssPtr->log2litLengthSum - ZSTD_highbit32(ssPtr->litLengthFreq[0]+1);
+        return optPtr->log2litLengthSum - ZSTD_highbit32(optPtr->litLengthFreq[0]+1);
 
     /* literals */
-    if (ssPtr->cachedLiterals == literals) {
+    if (optPtr->cachedLiterals == literals) {
-        U32 const additional = litLength - ssPtr->cachedLitLength;
+        U32 const additional = litLength - optPtr->cachedLitLength;
-        const BYTE* literals2 = ssPtr->cachedLiterals + ssPtr->cachedLitLength;
+        const BYTE* literals2 = optPtr->cachedLiterals + optPtr->cachedLitLength;
-        price = ssPtr->cachedPrice + additional * ssPtr->log2litSum;
+        price = optPtr->cachedPrice + additional * optPtr->log2litSum;
         for (u=0; u < additional; u++)
-            price -= ZSTD_highbit32(ssPtr->litFreq[literals2[u]]+1);
+            price -= ZSTD_highbit32(optPtr->litFreq[literals2[u]]+1);
-        ssPtr->cachedPrice = price;
+        optPtr->cachedPrice = price;
-        ssPtr->cachedLitLength = litLength;
+        optPtr->cachedLitLength = litLength;
     } else {
-        price = litLength * ssPtr->log2litSum;
+        price = litLength * optPtr->log2litSum;
         for (u=0; u < litLength; u++)
-            price -= ZSTD_highbit32(ssPtr->litFreq[literals[u]]+1);
+            price -= ZSTD_highbit32(optPtr->litFreq[literals[u]]+1);
 
         if (litLength >= 12) {
-            ssPtr->cachedLiterals = literals;
+            optPtr->cachedLiterals = literals;
-            ssPtr->cachedPrice = price;
+            optPtr->cachedPrice = price;
-            ssPtr->cachedLitLength = litLength;
+            optPtr->cachedLitLength = litLength;
         }
     }
 
     /* literal Length */
     {   const BYTE LL_deltaCode = 19;
         const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
-        price += LL_bits[llCode] + ssPtr->log2litLengthSum - ZSTD_highbit32(ssPtr->litLengthFreq[llCode]+1);
+        price += LL_bits[llCode] + optPtr->log2litLengthSum - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
     }
 
     return price;
 }
 
 
-FORCE_INLINE U32 ZSTD_getPrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength, const int ultra)
+FORCE_INLINE U32 ZSTD_getPrice(optState_t* optPtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength, const int ultra)
 {
     /* offset */
     U32 price;
     BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
 
-    if (seqStorePtr->staticPrices)
+    if (optPtr->staticPrices)
-        return ZSTD_getLiteralPrice(seqStorePtr, litLength, literals) + ZSTD_highbit32((U32)matchLength+1) + 16 + offCode;
+        return ZSTD_getLiteralPrice(optPtr, litLength, literals) + ZSTD_highbit32((U32)matchLength+1) + 16 + offCode;
 
-    price = offCode + seqStorePtr->log2offCodeSum - ZSTD_highbit32(seqStorePtr->offCodeFreq[offCode]+1);
+    price = offCode + optPtr->log2offCodeSum - ZSTD_highbit32(optPtr->offCodeFreq[offCode]+1);
     if (!ultra && offCode >= 20) price += (offCode-19)*2;
 
     /* match Length */
     {   const BYTE ML_deltaCode = 36;
         const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
-        price += ML_bits[mlCode] + seqStorePtr->log2matchLengthSum - ZSTD_highbit32(seqStorePtr->matchLengthFreq[mlCode]+1);
+        price += ML_bits[mlCode] + optPtr->log2matchLengthSum - ZSTD_highbit32(optPtr->matchLengthFreq[mlCode]+1);
     }
 
-    return price + ZSTD_getLiteralPrice(seqStorePtr, litLength, literals) + seqStorePtr->factor;
+    return price + ZSTD_getLiteralPrice(optPtr, litLength, literals) + optPtr->factor;
 }
 
 
-MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength)
+MEM_STATIC void ZSTD_updatePrice(optState_t* optPtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength)
 {
     U32 u;
 
     /* literals */
-    seqStorePtr->litSum += litLength*ZSTD_LITFREQ_ADD;
+    optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
     for (u=0; u < litLength; u++)
-        seqStorePtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+        optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
 
     /* literal Length */
     {   const BYTE LL_deltaCode = 19;
         const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
-        seqStorePtr->litLengthFreq[llCode]++;
+        optPtr->litLengthFreq[llCode]++;
-        seqStorePtr->litLengthSum++;
+        optPtr->litLengthSum++;
     }
 
     /* match offset */
     {   BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
-        seqStorePtr->offCodeSum++;
+        optPtr->offCodeSum++;
-        seqStorePtr->offCodeFreq[offCode]++;
+        optPtr->offCodeFreq[offCode]++;
     }
 
     /* match Length */
     {   const BYTE ML_deltaCode = 36;
         const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
-        seqStorePtr->matchLengthFreq[mlCode]++;
+        optPtr->matchLengthFreq[mlCode]++;
-        seqStorePtr->matchLengthSum++;
+        optPtr->matchLengthSum++;
     }
 
-    ZSTD_setLog2Prices(seqStorePtr);
+    ZSTD_setLog2Prices(optPtr);
 }
 
 
@@ -417,6 +417,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                                     const void* src, size_t srcSize, const int ultra)
 {
     seqStore_t* seqStorePtr = &(ctx->seqStore);
+    optState_t* optStatePtr = &(ctx->optState);
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
@@ -430,16 +431,16 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
     const U32 mls = ctx->appliedParams.cParams.searchLength;
     const U32 minMatch = (ctx->appliedParams.cParams.searchLength == 3) ? 3 : 4;
 
-    ZSTD_optimal_t* opt = seqStorePtr->priceTable;
+    ZSTD_optimal_t* opt = optStatePtr->priceTable;
-    ZSTD_match_t* matches = seqStorePtr->matchTable;
+    ZSTD_match_t* matches = optStatePtr->matchTable;
     const BYTE* inr;
     U32 offset, rep[ZSTD_REP_NUM];
 
     /* init */
     ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(seqStorePtr, (const BYTE*)src, srcSize);
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize);
     ip += (ip==prefixStart);
-    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=ctx->rep[i]; }
+    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=seqStorePtr->rep[i]; }
 
     /* Match Loop */
     while (ip < ilimit) {
@@ -462,7 +463,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                     }
                     best_off = i - (ip == anchor);
                     do {
-                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                        price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
                         if (mlen > last_pos || price < opt[mlen].price)
                             SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
                         mlen--;
@@ -487,7 +488,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
             mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
             best_mlen = matches[u].len;
             while (mlen <= best_mlen) {
-                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
                 if (mlen > last_pos || price < opt[mlen].price)
                     SET_PRICE(mlen, mlen, matches[u].off, litlen, price);   /* note : macro modifies last_pos */
                 mlen++;
@@ -507,12 +508,12 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
            if (opt[cur-1].mlen == 1) {
                 litlen = opt[cur-1].litlen + 1;
                 if (cur > litlen) {
-                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-litlen);
+                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-litlen);
                 } else
-                    price = ZSTD_getLiteralPrice(seqStorePtr, litlen, anchor);
+                    price = ZSTD_getLiteralPrice(optStatePtr, litlen, anchor);
            } else {
                 litlen = 1;
-                price = opt[cur - 1].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1);
+                price = opt[cur - 1].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-1);
            }
 
            if (cur > last_pos || price <= opt[cur].price)
@@ -554,12 +555,12 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                            if (opt[cur].mlen == 1) {
                                 litlen = opt[cur].litlen;
                                 if (cur > litlen) {
-                                    price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
+                                    price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
                                 } else
-                                    price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                                    price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
                             } else {
                                 litlen = 0;
-                                price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
+                                price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
                             }
 
                             if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
@@ -586,12 +587,12 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                     if (opt[cur].mlen == 1) {
                         litlen = opt[cur].litlen;
                         if (cur > litlen)
-                            price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
+                            price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
                         else
-                            price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                            price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
                     } else {
                         litlen = 0;
-                        price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
+                        price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
                     }
 
                     if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
@@ -645,13 +646,13 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
                 if (litLength==0) offset--;
             }
 
-            ZSTD_updatePrice(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
+            ZSTD_updatePrice(optStatePtr, litLength, anchor, offset, mlen-MINMATCH);
             ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
             anchor = ip = ip + mlen;
     }    }   /* for (cur=0; cur < last_pos; ) */
 
     /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->repToConfirm[i] = rep[i]; }
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -666,6 +667,7 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                                      const void* src, size_t srcSize, const int ultra)
 {
     seqStore_t* seqStorePtr = &(ctx->seqStore);
+    optState_t* optStatePtr = &(ctx->optState);
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
@@ -683,16 +685,16 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
     const U32 mls = ctx->appliedParams.cParams.searchLength;
     const U32 minMatch = (ctx->appliedParams.cParams.searchLength == 3) ? 3 : 4;
 
-    ZSTD_optimal_t* opt = seqStorePtr->priceTable;
+    ZSTD_optimal_t* opt = optStatePtr->priceTable;
-    ZSTD_match_t* matches = seqStorePtr->matchTable;
+    ZSTD_match_t* matches = optStatePtr->matchTable;
     const BYTE* inr;
 
     /* init */
     U32 offset, rep[ZSTD_REP_NUM];
-    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=ctx->rep[i]; }
+    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=seqStorePtr->rep[i]; }
 
     ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(seqStorePtr, (const BYTE*)src, srcSize);
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize);
     ip += (ip==prefixStart);
 
     /* Match Loop */
@@ -726,7 +728,7 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                     best_off = i - (ip==anchor);
                     litlen = opt[0].litlen;
                     do {
-                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                        price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
                         if (mlen > last_pos || price < opt[mlen].price)
                             SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
                         mlen--;
@@ -756,7 +758,7 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
             best_mlen = matches[u].len;
             litlen = opt[0].litlen;
             while (mlen <= best_mlen) {
-                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
                 if (mlen > last_pos || price < opt[mlen].price)
                     SET_PRICE(mlen, mlen, matches[u].off, litlen, price);
                 mlen++;
@@ -773,12 +775,12 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
             if (opt[cur-1].mlen == 1) {
                 litlen = opt[cur-1].litlen + 1;
                 if (cur > litlen) {
-                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-litlen);
+                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-litlen);
                 } else
-                    price = ZSTD_getLiteralPrice(seqStorePtr, litlen, anchor);
+                    price = ZSTD_getLiteralPrice(optStatePtr, litlen, anchor);
             } else {
                 litlen = 1;
-                price = opt[cur - 1].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1);
+                price = opt[cur - 1].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-1);
             }
 
             if (cur > last_pos || price <= opt[cur].price)
@@ -826,12 +828,12 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                             if (opt[cur].mlen == 1) {
                                 litlen = opt[cur].litlen;
                                 if (cur > litlen) {
-                                    price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
+                                    price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
                                 } else
-                                    price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                                    price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
                             } else {
                                 litlen = 0;
-                                price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
+                                price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
                             }
 
                             if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
@@ -858,12 +860,12 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                     if (opt[cur].mlen == 1) {
                         litlen = opt[cur].litlen;
                         if (cur > litlen)
-                            price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
+                            price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
                         else
-                            price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                            price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
                     } else {
                         litlen = 0;
-                        price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
+                        price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
                     }
 
                     if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
@@ -918,13 +920,13 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
                 if (litLength==0) offset--;
             }
 
-            ZSTD_updatePrice(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
+            ZSTD_updatePrice(optStatePtr, litLength, anchor, offset, mlen-MINMATCH);
             ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
             anchor = ip = ip + mlen;
     }    }   /* for (cur=0; cur < last_pos; ) */
 
     /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->repToConfirm[i] = rep[i]; }
 
     /* Last Literals */
     {   size_t lastLLSize = iend - anchor;

lib/compress/zstdmt_compress.c

@@ -9,7 +9,8 @@
 
 
 /* ======   Tuning parameters   ====== */
-#define ZSTDMT_NBTHREADS_MAX 128
+#define ZSTDMT_NBTHREADS_MAX 256
+#define ZSTDMT_OVERLAPLOG_DEFAULT 6
 
 
 /* ======   Compiler specifics   ====== */
@@ -73,6 +74,7 @@ static unsigned long long GetCurrentClockTimeMicroseconds(void)
 
 
 /* =====   Buffer Pool   ===== */
+/* a single Buffer Pool can be invoked from multiple threads in parallel */
 
 typedef struct buffer_s {
     void* start;
@@ -82,6 +84,8 @@ typedef struct buffer_s {
 static const buffer_t g_nullBuffer = { NULL, 0 };
 
 typedef struct ZSTDMT_bufferPool_s {
+    pthread_mutex_t poolMutex;
+    size_t bufferSize;
     unsigned totalBuffers;
     unsigned nbBuffers;
     ZSTD_customMem cMem;
@@ -90,10 +94,15 @@ typedef struct ZSTDMT_bufferPool_s {
 
 static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbThreads, ZSTD_customMem cMem)
 {
-    unsigned const maxNbBuffers = 2*nbThreads + 2;
+    unsigned const maxNbBuffers = 2*nbThreads + 3;
     ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
         sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
     if (bufPool==NULL) return NULL;
+    if (pthread_mutex_init(&bufPool->poolMutex, NULL)) {
+        ZSTD_free(bufPool, cMem);
+        return NULL;
+    }
+    bufPool->bufferSize = 64 KB;
     bufPool->totalBuffers = maxNbBuffers;
     bufPool->nbBuffers = 0;
     bufPool->cMem = cMem;
@@ -106,6 +115,7 @@ static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
     if (!bufPool) return;   /* compatibility with free on NULL */
     for (u=0; u<bufPool->totalBuffers; u++)
         ZSTD_free(bufPool->bTable[u].start, bufPool->cMem);
+    pthread_mutex_destroy(&bufPool->poolMutex);
     ZSTD_free(bufPool, bufPool->cMem);
 }
 
@@ -116,65 +126,85 @@ static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool)
                             + (bufPool->totalBuffers - 1) * sizeof(buffer_t);
     unsigned u;
     size_t totalBufferSize = 0;
+    pthread_mutex_lock(&bufPool->poolMutex);
     for (u=0; u<bufPool->totalBuffers; u++)
         totalBufferSize += bufPool->bTable[u].size;
+    pthread_mutex_unlock(&bufPool->poolMutex);
 
     return poolSize + totalBufferSize;
 }
 
-/** ZSTDMT_getBuffer() :
+static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* bufPool, size_t bSize)
- *  assumption : invocation from main thread only ! */
-static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* pool, size_t bSize)
 {
-    if (pool->nbBuffers) {   /* try to use an existing buffer */
+    bufPool->bufferSize = bSize;
-        buffer_t const buf = pool->bTable[--(pool->nbBuffers)];
-        size_t const availBufferSize = buf.size;
-        if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize))
-            /* large enough, but not too much */
-            return buf;
-        /* size conditions not respected : scratch this buffer, create new one */
-        ZSTD_free(buf.start, pool->cMem);
 }
+
+/** ZSTDMT_getBuffer() :
+ *  assumption : bufPool must be valid */
+static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const bSize = bufPool->bufferSize;
+    DEBUGLOG(5, "ZSTDMT_getBuffer");
+    pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers) {   /* try to use an existing buffer */
+        buffer_t const buf = bufPool->bTable[--(bufPool->nbBuffers)];
+        size_t const availBufferSize = buf.size;
+        if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize)) {
+            /* large enough, but not too much */
+            pthread_mutex_unlock(&bufPool->poolMutex);
+            return buf;
+        }
+        /* size conditions not respected : scratch this buffer, create new one */
+        DEBUGLOG(5, "existing buffer does not meet size conditions => freeing");
+        ZSTD_free(buf.start, bufPool->cMem);
+    }
+    pthread_mutex_unlock(&bufPool->poolMutex);
     /* create new buffer */
+    DEBUGLOG(5, "create a new buffer");
     {   buffer_t buffer;
-        void* const start = ZSTD_malloc(bSize, pool->cMem);
+        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
-        if (start==NULL) bSize = 0;
         buffer.start = start;   /* note : start can be NULL if malloc fails ! */
-        buffer.size = bSize;
+        buffer.size = (start==NULL) ? 0 : bSize;
         return buffer;
     }
 }
 
 /* store buffer for later re-use, up to pool capacity */
-static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* pool, buffer_t buf)
+static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
 {
-    if (buf.start == NULL) return;   /* release on NULL */
+    if (buf.start == NULL) return;   /* compatible with release on NULL */
-    if (pool->nbBuffers < pool->totalBuffers) {
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer");
-        pool->bTable[pool->nbBuffers++] = buf;   /* store for later re-use */
+    pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers < bufPool->totalBuffers) {
+        bufPool->bTable[bufPool->nbBuffers++] = buf;  /* stored for later use */
+        pthread_mutex_unlock(&bufPool->poolMutex);
         return;
     }
+    pthread_mutex_unlock(&bufPool->poolMutex);
     /* Reached bufferPool capacity (should not happen) */
-    ZSTD_free(buf.start, pool->cMem);
+    DEBUGLOG(5, "buffer pool capacity reached => freeing ");
+    ZSTD_free(buf.start, bufPool->cMem);
 }
 
 
 /* =====   CCtx Pool   ===== */
+/* a single CCtx Pool can be invoked from multiple threads in parallel */
 
 typedef struct {
+    pthread_mutex_t poolMutex;
     unsigned totalCCtx;
     unsigned availCCtx;
     ZSTD_customMem cMem;
     ZSTD_CCtx* cctx[1];   /* variable size */
 } ZSTDMT_CCtxPool;
 
-/* assumption : CCtxPool invocation only from main thread */
-
 /* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */
 static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
 {
     unsigned u;
     for (u=0; u<pool->totalCCtx; u++)
         ZSTD_freeCCtx(pool->cctx[u]);  /* note : compatible with free on NULL */
+    pthread_mutex_destroy(&pool->poolMutex);
     ZSTD_free(pool, pool->cMem);
 }
 
@@ -186,6 +216,10 @@ static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads,
     ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
         sizeof(ZSTDMT_CCtxPool) + (nbThreads-1)*sizeof(ZSTD_CCtx*), cMem);
     if (!cctxPool) return NULL;
+    if (pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
+        ZSTD_free(cctxPool, cMem);
+        return NULL;
+    }
     cctxPool->cMem = cMem;
     cctxPool->totalCCtx = nbThreads;
     cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */
@@ -198,50 +232,57 @@ static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads,
 /* only works during initialization phase, not during compression */
 static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
 {
-    unsigned const nbThreads = cctxPool->totalCCtx;
+    pthread_mutex_lock(&cctxPool->poolMutex);
+    {   unsigned const nbThreads = cctxPool->totalCCtx;
         size_t const poolSize = sizeof(*cctxPool)
                                 + (nbThreads-1)*sizeof(ZSTD_CCtx*);
         unsigned u;
         size_t totalCCtxSize = 0;
-    for (u=0; u<nbThreads; u++)
+        for (u=0; u<nbThreads; u++) {
             totalCCtxSize += ZSTD_sizeof_CCtx(cctxPool->cctx[u]);
-
+        }
+        pthread_mutex_unlock(&cctxPool->poolMutex);
         return poolSize + totalCCtxSize;
     }
-
-static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* pool)
-{
-    if (pool->availCCtx) {
-        pool->availCCtx--;
-        return pool->cctx[pool->availCCtx];
 }
-    return ZSTD_createCCtx();   /* note : can be NULL, when creation fails ! */
+
+static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool)
+{
+    DEBUGLOG(5, "ZSTDMT_getCCtx");
+    pthread_mutex_lock(&cctxPool->poolMutex);
+    if (cctxPool->availCCtx) {
+        cctxPool->availCCtx--;
+        {   ZSTD_CCtx* const cctx = cctxPool->cctx[cctxPool->availCCtx];
+            pthread_mutex_unlock(&cctxPool->poolMutex);
+            return cctx;
+    }   }
+    pthread_mutex_unlock(&cctxPool->poolMutex);
+    DEBUGLOG(5, "create one more CCtx");
+    return ZSTD_createCCtx_advanced(cctxPool->cMem);   /* note : can be NULL, when creation fails ! */
 }
 
 static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
 {
     if (cctx==NULL) return;   /* compatibility with release on NULL */
+    pthread_mutex_lock(&pool->poolMutex);
     if (pool->availCCtx < pool->totalCCtx)
         pool->cctx[pool->availCCtx++] = cctx;
-    else
+    else {
         /* pool overflow : should not happen, since totalCCtx==nbThreads */
+        DEBUGLOG(5, "CCtx pool overflow : free cctx");
         ZSTD_freeCCtx(cctx);
     }
+    pthread_mutex_unlock(&pool->poolMutex);
+}
 
 
 /* =====   Thread worker   ===== */
 
 typedef struct {
-    buffer_t buffer;
-    size_t filled;
-} inBuff_t;
-
-typedef struct {
-    ZSTD_CCtx* cctx;
     buffer_t src;
     const void* srcStart;
-    size_t   srcSize;
     size_t   dictSize;
+    size_t   srcSize;
     buffer_t dstBuff;
     size_t   cSize;
     size_t   dstFlushed;
@@ -253,6 +294,8 @@ typedef struct {
     pthread_cond_t* jobCompleted_cond;
     ZSTD_parameters params;
     const ZSTD_CDict* cdict;
+    ZSTDMT_CCtxPool* cctxPool;
+    ZSTDMT_bufferPool* bufPool;
     unsigned long long fullFrameSize;
 } ZSTDMT_jobDescription;
 
@@ -260,37 +303,56 @@ typedef struct {
 void ZSTDMT_compressChunk(void* jobDescription)
 {
     ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
+    ZSTD_CCtx* cctx = ZSTDMT_getCCtx(job->cctxPool);
     const void* const src = (const char*)job->srcStart + job->dictSize;
-    buffer_t const dstBuff = job->dstBuff;
+    buffer_t dstBuff = job->dstBuff;
     DEBUGLOG(5, "job (first:%u) (last:%u) : dictSize %u, srcSize %u",
                  job->firstChunk, job->lastChunk, (U32)job->dictSize, (U32)job->srcSize);
+
+    if (cctx==NULL) {
+        job->cSize = ERROR(memory_allocation);
+        goto _endJob;
+    }
+
+    if (dstBuff.start == NULL) {
+        dstBuff = ZSTDMT_getBuffer(job->bufPool);
+        if (dstBuff.start==NULL) {
+            job->cSize = ERROR(memory_allocation);
+            goto _endJob;
+        }
+        job->dstBuff = dstBuff;
+    }
+
     if (job->cdict) {  /* should only happen for first segment */
-        size_t const initError = ZSTD_compressBegin_usingCDict_advanced(job->cctx, job->cdict, job->params.fParams, job->fullFrameSize);
+        size_t const initError = ZSTD_compressBegin_usingCDict_advanced(cctx, job->cdict, job->params.fParams, job->fullFrameSize);
         DEBUGLOG(5, "using CDict");
         if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
     } else {  /* srcStart points at reloaded section */
         if (!job->firstChunk) job->params.fParams.contentSizeFlag = 0;  /* ensure no srcSize control */
-        {   size_t const dictModeError = ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceRawDict, 1);  /* Force loading dictionary in "content-only" mode (no header analysis) */
+        {   size_t const dictModeError = ZSTD_setCCtxParameter(cctx, ZSTD_p_forceRawDict, 1);  /* Force loading dictionary in "content-only" mode (no header analysis) */
-            size_t const initError = ZSTD_compressBegin_advanced(job->cctx, job->srcStart, job->dictSize, job->params, job->fullFrameSize);
+            size_t const initError = ZSTD_compressBegin_advanced(cctx, job->srcStart, job->dictSize, job->params, job->fullFrameSize);
             if (ZSTD_isError(initError) || ZSTD_isError(dictModeError)) { job->cSize = initError; goto _endJob; }
-            ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceWindow, 1);
+            ZSTD_setCCtxParameter(cctx, ZSTD_p_forceWindow, 1);
     }   }
     if (!job->firstChunk) {  /* flush and overwrite frame header when it's not first segment */
-        size_t const hSize = ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, 0);
+        size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, 0);
         if (ZSTD_isError(hSize)) { job->cSize = hSize; goto _endJob; }
-        ZSTD_invalidateRepCodes(job->cctx);
+        ZSTD_invalidateRepCodes(cctx);
     }
 
     DEBUGLOG(5, "Compressing : ");
     DEBUG_PRINTHEX(4, job->srcStart, 12);
     job->cSize = (job->lastChunk) ?
-                 ZSTD_compressEnd     (job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize) :
+                 ZSTD_compressEnd     (cctx, dstBuff.start, dstBuff.size, src, job->srcSize) :
-                 ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize);
+                 ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, job->srcSize);
     DEBUGLOG(5, "compressed %u bytes into %u bytes   (first:%u) (last:%u)",
                 (unsigned)job->srcSize, (unsigned)job->cSize, job->firstChunk, job->lastChunk);
     DEBUGLOG(5, "dstBuff.size : %u ; => %s", (U32)dstBuff.size, ZSTD_getErrorName(job->cSize));
 
 _endJob:
+    ZSTDMT_releaseCCtx(job->cctxPool, cctx);
+    ZSTDMT_releaseBuffer(job->bufPool, job->src);
+    job->src = g_nullBuffer; job->srcStart = NULL;
     PTHREAD_MUTEX_LOCK(job->jobCompleted_mutex);
     job->jobCompleted = 1;
     job->jobScanned = 0;
@@ -303,15 +365,19 @@ _endJob:
 /* =====   Multi-threaded compression   ===== */
 /* ------------------------------------------ */
 
+typedef struct {
+    buffer_t buffer;
+    size_t filled;
+} inBuff_t;
+
 struct ZSTDMT_CCtx_s {
     POOL_ctx* factory;
     ZSTDMT_jobDescription* jobs;
-    ZSTDMT_bufferPool* buffPool;
+    ZSTDMT_bufferPool* bufPool;
     ZSTDMT_CCtxPool* cctxPool;
     pthread_mutex_t jobCompleted_mutex;
     pthread_cond_t jobCompleted_cond;
     size_t targetSectionSize;
-    size_t marginSize;
     size_t inBuffSize;
     size_t dictSize;
     size_t targetDictSize;
@@ -324,7 +390,7 @@ struct ZSTDMT_CCtx_s {
     unsigned nextJobID;
     unsigned frameEnded;
     unsigned allJobsCompleted;
-    unsigned overlapRLog;
+    unsigned overlapLog;
     unsigned long long frameContentSize;
     size_t sectionSize;
     ZSTD_customMem cMem;
@@ -347,7 +413,8 @@ ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbThreads, ZSTD_customMem cMem)
     U32 nbJobs = nbThreads + 2;
     DEBUGLOG(3, "ZSTDMT_createCCtx_advanced");
 
-    if ((nbThreads < 1) | (nbThreads > ZSTDMT_NBTHREADS_MAX)) return NULL;
+    if (nbThreads < 1) return NULL;
+    nbThreads = MIN(nbThreads , ZSTDMT_NBTHREADS_MAX);
     if ((cMem.customAlloc!=NULL) ^ (cMem.customFree!=NULL))
         /* invalid custom allocator */
         return NULL;
@@ -358,18 +425,24 @@ ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbThreads, ZSTD_customMem cMem)
     mtctx->nbThreads = nbThreads;
     mtctx->allJobsCompleted = 1;
     mtctx->sectionSize = 0;
-    mtctx->overlapRLog = 3;
+    mtctx->overlapLog = ZSTDMT_OVERLAPLOG_DEFAULT;
     mtctx->factory = POOL_create(nbThreads, 1);
     mtctx->jobs = ZSTDMT_allocJobsTable(&nbJobs, cMem);
     mtctx->jobIDMask = nbJobs - 1;
-    mtctx->buffPool = ZSTDMT_createBufferPool(nbThreads, cMem);
+    mtctx->bufPool = ZSTDMT_createBufferPool(nbThreads, cMem);
     mtctx->cctxPool = ZSTDMT_createCCtxPool(nbThreads, cMem);
-    if (!mtctx->factory | !mtctx->jobs | !mtctx->buffPool | !mtctx->cctxPool) {
+    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool) {
+        ZSTDMT_freeCCtx(mtctx);
+        return NULL;
+    }
+    if (pthread_mutex_init(&mtctx->jobCompleted_mutex, NULL)) {
+        ZSTDMT_freeCCtx(mtctx);
+        return NULL;
+    }
+    if (pthread_cond_init(&mtctx->jobCompleted_cond, NULL)) {
         ZSTDMT_freeCCtx(mtctx);
         return NULL;
     }
-    pthread_mutex_init(&mtctx->jobCompleted_mutex, NULL);   /* Todo : check init function return */
-    pthread_cond_init(&mtctx->jobCompleted_cond, NULL);
     DEBUGLOG(3, "mt_cctx created, for %u threads", nbThreads);
     return mtctx;
 }
@@ -386,15 +459,13 @@ static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
     unsigned jobID;
     DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
     for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
-        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].dstBuff);
+        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
         mtctx->jobs[jobID].dstBuff = g_nullBuffer;
-        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].src);
+        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].src);
         mtctx->jobs[jobID].src = g_nullBuffer;
-        ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[jobID].cctx);
-        mtctx->jobs[jobID].cctx = NULL;
     }
     memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));
-    ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->inBuff.buffer);
+    ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer);
     mtctx->inBuff.buffer = g_nullBuffer;
     mtctx->allJobsCompleted = 1;
 }
@@ -404,7 +475,7 @@ size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
     if (mtctx==NULL) return 0;   /* compatible with free on NULL */
     POOL_free(mtctx->factory);
     if (!mtctx->allJobsCompleted) ZSTDMT_releaseAllJobResources(mtctx); /* stop workers first */
-    ZSTDMT_freeBufferPool(mtctx->buffPool);  /* release job resources into pools first */
+    ZSTDMT_freeBufferPool(mtctx->bufPool);  /* release job resources into pools first */
     ZSTD_free(mtctx->jobs, mtctx->cMem);
     ZSTDMT_freeCCtxPool(mtctx->cctxPool);
     ZSTD_freeCDict(mtctx->cdictLocal);
@@ -419,7 +490,7 @@ size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
     if (mtctx == NULL) return 0;   /* supports sizeof NULL */
     return sizeof(*mtctx)
             + POOL_sizeof(mtctx->factory)
-        + ZSTDMT_sizeof_bufferPool(mtctx->buffPool)
+            + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
             + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
             + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
             + ZSTD_sizeof_CDict(mtctx->cdictLocal);
@@ -434,10 +505,10 @@ size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSDTMT_parameter parameter,
         return 0;
     case ZSTDMT_p_overlapSectionLog :
         DEBUGLOG(5, "ZSTDMT_p_overlapSectionLog : %u", value);
-        mtctx->overlapRLog = (value >= 9) ? 0 : 9 - value;
+        mtctx->overlapLog = (value >= 9) ? 9 : value;
         return 0;
     default :
-        return ERROR(compressionParameter_unsupported);
+        return ERROR(parameter_unsupported);
     }
 }
 
@@ -463,8 +534,9 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
                          const void* src, size_t srcSize,
                          const ZSTD_CDict* cdict,
                                ZSTD_parameters const params,
-                           unsigned overlapRLog)
+                               unsigned overlapLog)
 {
+    unsigned const overlapRLog = (overlapLog>9) ? 0 : 9-overlapLog;
     size_t const overlapSize = (overlapRLog>=9) ? 0 : (size_t)1 << (params.cParams.windowLog - overlapRLog);
     unsigned nbChunks = computeNbChunks(srcSize, params.cParams.windowLog, mtctx->nbThreads);
     size_t const proposedChunkSize = (srcSize + (nbChunks-1)) / nbChunks;
@@ -473,6 +545,7 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
     size_t remainingSrcSize = srcSize;
     unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbChunks : (unsigned)(dstCapacity / ZSTD_compressBound(avgChunkSize));  /* presumes avgChunkSize >= 256 KB, which should be the case */
     size_t frameStartPos = 0, dstBufferPos = 0;
+    XXH64_state_t xxh64;
 
     DEBUGLOG(4, "nbChunks  : %2u   (chunkSize : %u bytes)   ", nbChunks, (U32)avgChunkSize);
     if (nbChunks==1) {   /* fallback to single-thread mode */
@@ -480,7 +553,9 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
         if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, params.fParams);
         return ZSTD_compress_advanced(cctx, dst, dstCapacity, src, srcSize, NULL, 0, params);
     }
-    assert(avgChunkSize >= 256 KB);  /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), which is useful to avoid allocating extra buffers */
+    assert(avgChunkSize >= 256 KB);  /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), which is required for compressWithinDst */
+    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgChunkSize) );
+    XXH64_reset(&xxh64, 0);
 
     if (nbChunks > mtctx->jobIDMask+1) {  /* enlarge job table */
         U32 nbJobs = nbChunks;
@@ -496,17 +571,10 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
             size_t const chunkSize = MIN(remainingSrcSize, avgChunkSize);
             size_t const dstBufferCapacity = ZSTD_compressBound(chunkSize);
             buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
-            buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : ZSTDMT_getBuffer(mtctx->buffPool, dstBufferCapacity);
+            buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer;
-            ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(mtctx->cctxPool);
             size_t dictSize = u ? overlapSize : 0;
 
-            if ((cctx==NULL) || (dstBuffer.start==NULL)) {
+            mtctx->jobs[u].src = g_nullBuffer;
-                mtctx->jobs[u].cSize = ERROR(memory_allocation);   /* job result */
-                mtctx->jobs[u].jobCompleted = 1;
-                nbChunks = u+1;   /* only wait and free u jobs, instead of initially expected nbChunks ones */
-                break;   /* let's wait for previous jobs to complete, but don't start new ones */
-            }
-
             mtctx->jobs[u].srcStart = srcStart + frameStartPos - dictSize;
             mtctx->jobs[u].dictSize = dictSize;
             mtctx->jobs[u].srcSize = chunkSize;
@@ -516,13 +584,18 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
             /* do not calculate checksum within sections, but write it in header for first section */
             if (u!=0) mtctx->jobs[u].params.fParams.checksumFlag = 0;
             mtctx->jobs[u].dstBuff = dstBuffer;
-            mtctx->jobs[u].cctx = cctx;
+            mtctx->jobs[u].cctxPool = mtctx->cctxPool;
+            mtctx->jobs[u].bufPool = mtctx->bufPool;
             mtctx->jobs[u].firstChunk = (u==0);
             mtctx->jobs[u].lastChunk = (u==nbChunks-1);
             mtctx->jobs[u].jobCompleted = 0;
             mtctx->jobs[u].jobCompleted_mutex = &mtctx->jobCompleted_mutex;
             mtctx->jobs[u].jobCompleted_cond = &mtctx->jobCompleted_cond;
 
+            if (params.fParams.checksumFlag) {
+                XXH64_update(&xxh64, srcStart + frameStartPos, chunkSize);
+            }
+
             DEBUGLOG(5, "posting job %u   (%u bytes)", u, (U32)chunkSize);
             DEBUG_PRINTHEX(6, mtctx->jobs[u].srcStart, 12);
             POOL_add(mtctx->factory, ZSTDMT_compressChunk, &mtctx->jobs[u]);
@@ -533,8 +606,8 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
     }   }
 
     /* collect result */
-    {   unsigned chunkID;
+    {   size_t error = 0, dstPos = 0;
-        size_t error = 0, dstPos = 0;
+        unsigned chunkID;
         for (chunkID=0; chunkID<nbChunks; chunkID++) {
             DEBUGLOG(5, "waiting for chunk %u ", chunkID);
             PTHREAD_MUTEX_LOCK(&mtctx->jobCompleted_mutex);
@@ -545,8 +618,6 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
             pthread_mutex_unlock(&mtctx->jobCompleted_mutex);
             DEBUGLOG(5, "ready to write chunk %u ", chunkID);
 
-            ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[chunkID].cctx);
-            mtctx->jobs[chunkID].cctx = NULL;
             mtctx->jobs[chunkID].srcStart = NULL;
             {   size_t const cSize = mtctx->jobs[chunkID].cSize;
                 if (ZSTD_isError(cSize)) error = cSize;
@@ -556,13 +627,25 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
                         memmove((char*)dst + dstPos, mtctx->jobs[chunkID].dstBuff.start, cSize);  /* may overlap when chunk compressed within dst */
                     if (chunkID >= compressWithinDst) {  /* chunk compressed into its own buffer, which must be released */
                         DEBUGLOG(5, "releasing buffer %u>=%u", chunkID, compressWithinDst);
-                        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[chunkID].dstBuff);
+                        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[chunkID].dstBuff);
                     }
                     mtctx->jobs[chunkID].dstBuff = g_nullBuffer;
                 }
                 dstPos += cSize ;
             }
-        }
+        }  /* for (chunkID=0; chunkID<nbChunks; chunkID++) */
+
+        DEBUGLOG(4, "checksumFlag : %u ", params.fParams.checksumFlag);
+        if (params.fParams.checksumFlag) {
+            U32 const checksum = (U32)XXH64_digest(&xxh64);
+            if (dstPos + 4 > dstCapacity) {
+                error = ERROR(dstSize_tooSmall);
+            } else {
+                DEBUGLOG(4, "writing checksum : %08X \n", checksum);
+                MEM_writeLE32((char*)dst + dstPos, checksum);
+                dstPos += 4;
+        }   }
+
         if (!error) DEBUGLOG(4, "compressed size : %u  ", (U32)dstPos);
         return error ? error : dstPos;
     }
@@ -574,10 +657,10 @@ size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
                      const void* src, size_t srcSize,
                            int compressionLevel)
 {
-    U32 const overlapRLog = (compressionLevel >= ZSTD_maxCLevel()) ? 0 : 3;
+    U32 const overlapLog = (compressionLevel >= ZSTD_maxCLevel()) ? 9 : ZSTDMT_OVERLAPLOG_DEFAULT;
     ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
     params.fParams.contentSizeFlag = 1;
-    return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapRLog);
+    return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
 }
 
 
@@ -642,18 +725,16 @@ size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
         zcs->cdict = cdict;
     }
 
-    zcs->targetDictSize = (zcs->overlapRLog>=9) ? 0 : (size_t)1 << (zcs->params.cParams.windowLog - zcs->overlapRLog);
+    zcs->targetDictSize = (zcs->overlapLog==0) ? 0 : (size_t)1 << (zcs->params.cParams.windowLog - (9 - zcs->overlapLog));
-    DEBUGLOG(4, "overlapRLog : %u ", zcs->overlapRLog);
+    DEBUGLOG(4, "overlapLog : %u ", zcs->overlapLog);
     DEBUGLOG(4, "overlap Size : %u KB", (U32)(zcs->targetDictSize>>10));
     zcs->targetSectionSize = zcs->sectionSize ? zcs->sectionSize : (size_t)1 << (zcs->params.cParams.windowLog + 2);
     zcs->targetSectionSize = MAX(ZSTDMT_SECTION_SIZE_MIN, zcs->targetSectionSize);
     zcs->targetSectionSize = MAX(zcs->targetDictSize, zcs->targetSectionSize);
     DEBUGLOG(4, "Section Size : %u KB", (U32)(zcs->targetSectionSize>>10));
-    zcs->marginSize = zcs->targetSectionSize >> 2;
+    zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize;
-    zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize + zcs->marginSize;
+    ZSTDMT_setBufferSize(zcs->bufPool, MAX(zcs->inBuffSize, ZSTD_compressBound(zcs->targetSectionSize)) );
-    zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);
+    zcs->inBuff.buffer = g_nullBuffer;
-    if (zcs->inBuff.buffer.start == NULL) return ERROR(memory_allocation);
-    zcs->inBuff.filled = 0;
     zcs->dictSize = 0;
     zcs->doneJobID = 0;
     zcs->nextJobID = 0;
@@ -665,7 +746,8 @@ size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
 
 size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
                              const void* dict, size_t dictSize,
-                                ZSTD_parameters params, unsigned long long pledgedSrcSize)
+                                   ZSTD_parameters params,
+                                   unsigned long long pledgedSrcSize)
 {
     DEBUGLOG(5, "ZSTDMT_initCStream_advanced");
     return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, NULL, params, pledgedSrcSize);
@@ -701,19 +783,8 @@ size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel) {
 
 static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsigned endFrame)
 {
-    size_t const dstBufferCapacity = ZSTD_compressBound(srcSize);
-    buffer_t const dstBuffer = ZSTDMT_getBuffer(zcs->buffPool, dstBufferCapacity);
-    ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(zcs->cctxPool);
     unsigned const jobID = zcs->nextJobID & zcs->jobIDMask;
 
-    if ((cctx==NULL) || (dstBuffer.start==NULL)) {
-        zcs->jobs[jobID].jobCompleted = 1;
-        zcs->nextJobID++;
-        ZSTDMT_waitForAllJobsCompleted(zcs);
-        ZSTDMT_releaseAllJobResources(zcs);
-        return ERROR(memory_allocation);
-    }
-
     DEBUGLOG(4, "preparing job %u to compress %u bytes with %u preload ",
                 zcs->nextJobID, (U32)srcSize, (U32)zcs->dictSize);
     zcs->jobs[jobID].src = zcs->inBuff.buffer;
@@ -726,8 +797,9 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsi
     if (zcs->nextJobID) zcs->jobs[jobID].params.fParams.checksumFlag = 0;
     zcs->jobs[jobID].cdict = zcs->nextJobID==0 ? zcs->cdict : NULL;
     zcs->jobs[jobID].fullFrameSize = zcs->frameContentSize;
-    zcs->jobs[jobID].dstBuff = dstBuffer;
+    zcs->jobs[jobID].dstBuff = g_nullBuffer;
-    zcs->jobs[jobID].cctx = cctx;
+    zcs->jobs[jobID].cctxPool = zcs->cctxPool;
+    zcs->jobs[jobID].bufPool = zcs->bufPool;
     zcs->jobs[jobID].firstChunk = (zcs->nextJobID==0);
     zcs->jobs[jobID].lastChunk = endFrame;
     zcs->jobs[jobID].jobCompleted = 0;
@@ -735,11 +807,13 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsi
     zcs->jobs[jobID].jobCompleted_mutex = &zcs->jobCompleted_mutex;
     zcs->jobs[jobID].jobCompleted_cond = &zcs->jobCompleted_cond;
 
+    if (zcs->params.fParams.checksumFlag)
+        XXH64_update(&zcs->xxhState, (const char*)zcs->inBuff.buffer.start + zcs->dictSize, srcSize);
+
     /* get a new buffer for next input */
     if (!endFrame) {
         size_t const newDictSize = MIN(srcSize + zcs->dictSize, zcs->targetDictSize);
-        DEBUGLOG(5, "ZSTDMT_createCompressionJob::endFrame = %u", endFrame);
+        zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->bufPool);
-        zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);
         if (zcs->inBuff.buffer.start == NULL) {   /* not enough memory to allocate next input buffer */
             zcs->jobs[jobID].jobCompleted = 1;
             zcs->nextJobID++;
@@ -747,26 +821,20 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsi
             ZSTDMT_releaseAllJobResources(zcs);
             return ERROR(memory_allocation);
         }
-        DEBUGLOG(5, "inBuff currently filled to %u", (U32)zcs->inBuff.filled);
         zcs->inBuff.filled -= srcSize + zcs->dictSize - newDictSize;
-        DEBUGLOG(5, "new job : inBuff filled to %u, with %u dict and %u src",
-                    (U32)zcs->inBuff.filled, (U32)newDictSize,
-                    (U32)(zcs->inBuff.filled - newDictSize));
         memmove(zcs->inBuff.buffer.start,
             (const char*)zcs->jobs[jobID].srcStart + zcs->dictSize + srcSize - newDictSize,
             zcs->inBuff.filled);
-        DEBUGLOG(5, "new inBuff pre-filled");
         zcs->dictSize = newDictSize;
     } else {   /* if (endFrame==1) */
-        DEBUGLOG(5, "ZSTDMT_createCompressionJob::endFrame = %u", endFrame);
         zcs->inBuff.buffer = g_nullBuffer;
         zcs->inBuff.filled = 0;
         zcs->dictSize = 0;
         zcs->frameEnded = 1;
-        if (zcs->nextJobID == 0)
+        if (zcs->nextJobID == 0) {
             /* single chunk exception : checksum is calculated directly within worker thread */
             zcs->params.fParams.checksumFlag = 0;
-    }
+    }   }
 
     DEBUGLOG(4, "posting job %u : %u bytes  (end:%u) (note : doneJob = %u=>%u)",
                 zcs->nextJobID,
@@ -804,11 +872,8 @@ static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsi
                 ZSTDMT_releaseAllJobResources(zcs);
                 return job.cSize;
             }
-            ZSTDMT_releaseCCtx(zcs->cctxPool, job.cctx);
-            zcs->jobs[wJobID].cctx = NULL;
             DEBUGLOG(5, "zcs->params.fParams.checksumFlag : %u ", zcs->params.fParams.checksumFlag);
             if (zcs->params.fParams.checksumFlag) {
-                XXH64_update(&zcs->xxhState, (const char*)job.srcStart + job.dictSize, job.srcSize);
                 if (zcs->frameEnded && (zcs->doneJobID+1 == zcs->nextJobID)) {  /* write checksum at end of last section */
                     U32 const checksum = (U32)XXH64_digest(&zcs->xxhState);
                     DEBUGLOG(5, "writing checksum : %08X \n", checksum);
@@ -816,9 +881,6 @@ static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsi
                     job.cSize += 4;
                     zcs->jobs[wJobID].cSize += 4;
             }   }
-            ZSTDMT_releaseBuffer(zcs->buffPool, job.src);
-            zcs->jobs[wJobID].srcStart = NULL;
-            zcs->jobs[wJobID].src = g_nullBuffer;
             zcs->jobs[wJobID].jobScanned = 1;
         }
         {   size_t const toWrite = MIN(job.cSize - job.dstFlushed, output->size - output->pos);
@@ -828,7 +890,7 @@ static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsi
             job.dstFlushed += toWrite;
         }
         if (job.dstFlushed == job.cSize) {   /* output buffer fully flushed => move to next one */
-            ZSTDMT_releaseBuffer(zcs->buffPool, job.dstBuff);
+            ZSTDMT_releaseBuffer(zcs->bufPool, job.dstBuff);
             zcs->jobs[wJobID].dstBuff = g_nullBuffer;
             zcs->jobs[wJobID].jobCompleted = 0;
             zcs->doneJobID++;
@@ -852,18 +914,18 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
                                      ZSTD_inBuffer* input,
                                      ZSTD_EndDirective endOp)
 {
-    size_t const newJobThreshold = mtctx->dictSize + mtctx->targetSectionSize + mtctx->marginSize;
+    size_t const newJobThreshold = mtctx->dictSize + mtctx->targetSectionSize;
     assert(output->pos <= output->size);
     assert(input->pos  <= input->size);
     if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
         /* current frame being ended. Only flush/end are allowed. Or start new frame with init */
         return ERROR(stage_wrong);
     }
-    if (mtctx->nbThreads==1) {
+    if (mtctx->nbThreads==1) {  /* delegate to single-thread (synchronous) */
         return ZSTD_compressStream_generic(mtctx->cctxPool->cctx[0], output, input, endOp);
     }
 
-    /* single-pass shortcut (note : this is blocking-mode) */
+    /* single-pass shortcut (note : this is synchronous-mode) */
     if ( (mtctx->nextJobID==0)      /* just started */
       && (mtctx->inBuff.filled==0)  /* nothing buffered */
       && (endOp==ZSTD_e_end)        /* end order */
@@ -871,24 +933,29 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
         size_t const cSize = ZSTDMT_compress_advanced(mtctx,
                 (char*)output->dst + output->pos, output->size - output->pos,
                 (const char*)input->src + input->pos, input->size - input->pos,
-                mtctx->cdict, mtctx->params, mtctx->overlapRLog);
+                mtctx->cdict, mtctx->params, mtctx->overlapLog);
         if (ZSTD_isError(cSize)) return cSize;
         input->pos = input->size;
         output->pos += cSize;
-        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->inBuff.buffer);  /* was allocated in initStream */
+        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer);  /* was allocated in initStream */
         mtctx->allJobsCompleted = 1;
         mtctx->frameEnded = 1;
         return 0;
     }
 
     /* fill input buffer */
-    if ((input->src) && (mtctx->inBuff.buffer.start)) {   /* support NULL input */
+    if (input->size > input->pos) {   /* support NULL input */
-        size_t const toLoad = MIN(input->size - input->pos, mtctx->inBuffSize - mtctx->inBuff.filled);
+        if (mtctx->inBuff.buffer.start == NULL) {
-        DEBUGLOG(2, "inBuff:%08X;  inBuffSize=%u;  ToCopy=%u", (U32)(size_t)mtctx->inBuff.buffer.start, (U32)mtctx->inBuffSize, (U32)toLoad);
+            mtctx->inBuff.buffer = ZSTDMT_getBuffer(mtctx->bufPool);
+            if (mtctx->inBuff.buffer.start == NULL) return ERROR(memory_allocation);
+            mtctx->inBuff.filled = 0;
+        }
+        {   size_t const toLoad = MIN(input->size - input->pos, mtctx->inBuffSize - mtctx->inBuff.filled);
+            DEBUGLOG(5, "inBuff:%08X;  inBuffSize=%u;  ToCopy=%u", (U32)(size_t)mtctx->inBuff.buffer.start, (U32)mtctx->inBuffSize, (U32)toLoad);
             memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, toLoad);
             input->pos += toLoad;
             mtctx->inBuff.filled += toLoad;
-    }
+    }   }
 
     if ( (mtctx->inBuff.filled >= newJobThreshold)  /* filled enough : let's compress */
       && (mtctx->nextJobID <= mtctx->doneJobID + mtctx->jobIDMask) ) {   /* avoid overwriting job round buffer */

lib/compress/zstdmt_compress.h

@@ -16,8 +16,8 @@
 
 
 /* Note : This is an internal API.
- *        Some methods are still exposed (ZSTDLIB_API), because for some time,
+ *        Some methods are still exposed (ZSTDLIB_API),
- *        it used to be the only way to invoke MT compression.
+ *        because it used to be the only way to invoke MT compression.
  *        Now, it's recommended to use ZSTD_compress_generic() instead.
  *        These methods will stop being exposed in a future version */
 
@@ -68,7 +68,7 @@ ZSTDLIB_API size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
                                      const void* src, size_t srcSize,
                                      const ZSTD_CDict* cdict,
                                            ZSTD_parameters const params,
-                                           unsigned overlapRLog);
+                                           unsigned overlapLog);
 
 ZSTDLIB_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
                                         const void* dict, size_t dictSize,   /* dict can be released after init, a local copy is preserved within zcs */

lib/decompress/zstd_decompress.c

@@ -95,7 +95,7 @@ typedef struct {
     HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
     U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
     U32 rep[ZSTD_REP_NUM];
-} ZSTD_entropyTables_t;
+} ZSTD_entropyDTables_t;
 
 struct ZSTD_DCtx_s
 {
@@ -103,7 +103,7 @@ struct ZSTD_DCtx_s
     const FSE_DTable* MLTptr;
     const FSE_DTable* OFTptr;
     const HUF_DTable* HUFptr;
-    ZSTD_entropyTables_t entropy;
+    ZSTD_entropyDTables_t entropy;
     const void* previousDstEnd;   /* detect continuity */
     const void* base;             /* start of current segment */
     const void* vBase;            /* virtual start of previous segment if it was just before current one */
@@ -1842,7 +1842,7 @@ static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dict
 /* ZSTD_loadEntropy() :
  * dict : must point at beginning of a valid zstd dictionary
  * @return : size of entropy tables read */
-static size_t ZSTD_loadEntropy(ZSTD_entropyTables_t* entropy, const void* const dict, size_t const dictSize)
+static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const dict, size_t const dictSize)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
@@ -1933,7 +1933,7 @@ struct ZSTD_DDict_s {
     void* dictBuffer;
     const void* dictContent;
     size_t dictSize;
-    ZSTD_entropyTables_t entropy;
+    ZSTD_entropyDTables_t entropy;
     U32 dictID;
     U32 entropyPresent;
     ZSTD_customMem cMem;
@@ -2227,7 +2227,7 @@ size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds,
 {
     switch(paramType)
     {
-        default : return ERROR(parameter_unknown);
+        default : return ERROR(parameter_unsupported);
         case DStream_p_maxWindowSize : zds->maxWindowSize = paramValue ? paramValue : (U32)(-1); break;
     }
     return 0;

lib/dictBuilder/cover.c

@@ -714,11 +714,9 @@ typedef struct COVER_best_s {
  * Initialize the `COVER_best_t`.
  */
 static void COVER_best_init(COVER_best_t *best) {
-  if (!best) {
+  if (best==NULL) return; /* compatible with init on NULL */
-    return;
+  (void)pthread_mutex_init(&best->mutex, NULL);
-  }
+  (void)pthread_cond_init(&best->cond, NULL);
-  pthread_mutex_init(&best->mutex, NULL);
-  pthread_cond_init(&best->cond, NULL);
   best->liveJobs = 0;
   best->dict = NULL;
   best->dictSize = 0;

lib/dictBuilder/zdict.c

@@ -695,7 +695,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
         DISPLAYLEVEL(1, "Not enough memory \n");
         goto _cleanup;
     }
-    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionary_wrong); goto _cleanup; }   /* too large dictionary */
+    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
     for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
     for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
     for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;

lib/legacy/zstd_v04.c

@@ -2776,7 +2776,7 @@ static size_t ZSTD_decodeFrameHeader_Part2(ZSTD_DCtx* zc, const void* src, size_
     size_t result;
     if (srcSize != zc->headerSize) return ERROR(srcSize_wrong);
     result = ZSTD_getFrameParams(&(zc->params), src, srcSize);
-    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupportedBy32bits);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
     return result;
 }
 

lib/legacy/zstd_v05.c

@@ -2888,7 +2888,7 @@ static size_t ZSTDv05_decodeFrameHeader_Part2(ZSTDv05_DCtx* zc, const void* src,
     if (srcSize != zc->headerSize)
         return ERROR(srcSize_wrong);
     result = ZSTDv05_getFrameParams(&(zc->params), src, srcSize);
-    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupportedBy32bits);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
     return result;
 }
 

lib/legacy/zstd_v06.c

@@ -3084,7 +3084,7 @@ size_t ZSTDv06_getFrameParams(ZSTDv06_frameParams* fparamsPtr, const void* src,
 static size_t ZSTDv06_decodeFrameHeader(ZSTDv06_DCtx* zc, const void* src, size_t srcSize)
 {
     size_t const result = ZSTDv06_getFrameParams(&(zc->fParams), src, srcSize);
-    if ((MEM_32bits()) && (zc->fParams.windowLog > 25)) return ERROR(frameParameter_unsupportedBy32bits);
+    if ((MEM_32bits()) && (zc->fParams.windowLog > 25)) return ERROR(frameParameter_unsupported);
     return result;
 }
 

lib/zstd.h

@@ -954,7 +954,9 @@ typedef enum {
                               * Special: value 0 means "do not change strategy". */
 
     /* frame parameters */
-    ZSTD_p_contentSizeFlag=200, /* Content size is written into frame header _whenever known_ (default:1) */
+    ZSTD_p_contentSizeFlag=200, /* Content size is written into frame header _whenever known_ (default:1)
+                              * note that content size must be known at the beginning,
+                              * it is sent using ZSTD_CCtx_setPledgedSrcSize() */
     ZSTD_p_checksumFlag,     /* A 32-bits checksum of content is written at end of frame (default:0) */
     ZSTD_p_dictIDFlag,       /* When applicable, dictID of dictionary is provided in frame header (default:1) */
 

programs/zstd.1

@@ -1,5 +1,5 @@
 .
-.TH "ZSTD" "1" "June 2017" "zstd 1.3.0" "User Commands"
+.TH "ZSTD" "1" "July 2017" "zstd 1.3.1" "User Commands"
 .
 .SH "NAME"
 \fBzstd\fR \- zstd, zstdmt, unzstd, zstdcat \- Compress or decompress \.zst files
@@ -105,7 +105,7 @@ unlocks high compression levels 20+ (maximum 22), using a lot more memory\. Note
 .
 .TP
 \fB\-T#\fR, \fB\-\-threads=#\fR
-Compress using \fB#\fR threads (default: 1)\. If \fB#\fR is 0, attempt to detect and use the number of physical CPU cores\. This modifier does nothing if \fBzstd\fR is compiled without multithread support\.
+Compress using \fB#\fR threads (default: 1)\. If \fB#\fR is 0, attempt to detect and use the number of physical CPU cores\. In all cases, the nb of threads is capped to ZSTDMT_NBTHREADS_MAX==256\. This modifier does nothing if \fBzstd\fR is compiled without multithread support\.
 .
 .TP
 \fB\-D file\fR

programs/zstd.1.md

@@ -108,6 +108,7 @@ the last one takes effect.
 * `-T#`, `--threads=#`:
     Compress using `#` threads (default: 1).
     If `#` is 0, attempt to detect and use the number of physical CPU cores.
+    In all cases, the nb of threads is capped to ZSTDMT_NBTHREADS_MAX==256.
     This modifier does nothing if `zstd` is compiled without multithread support.
 * `-D file`:
     use `file` as Dictionary to compress or decompress FILE(s)

programs/zstdcli.c

@@ -719,6 +719,10 @@ int main(int argCount, const char* argv[])
         goto _end;
     }
 
+#ifndef ZSTD_NODECOMPRESS
+    if (operation==zom_test) { outFileName=nulmark; FIO_setRemoveSrcFile(0); } /* test mode */
+#endif
+
     /* No input filename ==> use stdin and stdout */
     filenameIdx += !filenameIdx;   /* filenameTable[0] is stdin by default */
     if (!strcmp(filenameTable[0], stdinmark) && !outFileName) outFileName = stdoutmark;   /* when input is stdin, default output is stdout */
@@ -763,7 +767,6 @@ int main(int argCount, const char* argv[])
 #endif
     } else {  /* decompression or test */
 #ifndef ZSTD_NODECOMPRESS
-        if (operation==zom_test) { outFileName=nulmark; FIO_setRemoveSrcFile(0); } /* test mode */
         FIO_setMemLimit(memLimit);
         if (filenameIdx==1 && outFileName)
             operationResult = FIO_decompressFilename(outFileName, filenameTable[0], dictFileName);

tests/Makefile

@@ -108,12 +108,12 @@ fullbench-dll: $(PRGDIR)/datagen.c fullbench.c
 	$(MAKE) -C $(ZSTDDIR) libzstd
 	$(CC) $(FLAGS) $^ -o $@$(EXT) -DZSTD_DLL_IMPORT=1 $(ZSTDDIR)/dll/libzstd.dll
 
-fuzzer   : $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c fuzzer.c
+fuzzer : CPPFLAGS += $(MULTITHREAD_CPP)
+fuzzer : LDFLAGS += $(MULTITHREAD_LD)
+fuzzer32: CFLAGS += -m32
+fuzzer fuzzer32 : $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c fuzzer.c
 	$(CC) $(FLAGS) $^ -o $@$(EXT)
 
-fuzzer32 : $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c fuzzer.c
-	$(CC) -m32 $(FLAGS) $^ -o $@$(EXT)
-
 fuzzer-dll : LDFLAGS+= -L$(ZSTDDIR) -lzstd
 fuzzer-dll : $(ZSTDDIR)/common/xxhash.c $(PRGDIR)/datagen.c fuzzer.c
 	$(MAKE) -C $(ZSTDDIR) libzstd
@@ -192,7 +192,7 @@ else
 	$(CC) $(FLAGS) $^ -o $@$(EXT) -Wl,-rpath=$(ZSTDDIR) $(ZSTDDIR)/libzstd.so
 endif
 
-pool  : pool.c $(ZSTDDIR)/common/pool.c $(ZSTDDIR)/common/threading.c
+poolTests  : poolTests.c $(ZSTDDIR)/common/pool.c $(ZSTDDIR)/common/threading.c
 	$(CC)    $(FLAGS) $(MULTITHREAD) $^ -o $@$(EXT)
 
 namespaceTest:
@@ -213,7 +213,7 @@ clean:
         fuzzer-dll$(EXT) zstreamtest-dll$(EXT) zbufftest-dll$(EXT)\
         zstreamtest$(EXT) zstreamtest32$(EXT) \
         datagen$(EXT) paramgrill$(EXT) roundTripCrash$(EXT) longmatch$(EXT) \
-        symbols$(EXT) invalidDictionaries$(EXT) legacy$(EXT) pool$(EXT) \
+        symbols$(EXT) invalidDictionaries$(EXT) legacy$(EXT) poolTests$(EXT) \
 	decodecorpus$(EXT)
 	@echo Cleaning completed
 
@@ -375,7 +375,7 @@ test-decodecorpus-cli: decodecorpus
 	cd ..
 	@rm -rf testdir
 
-test-pool: pool
+test-pool: poolTests
-	$(QEMU_SYS) ./pool
+	$(QEMU_SYS) ./poolTests
 
 endif

tests/fuzzer.c

@@ -51,14 +51,14 @@ static const U32 nbTestsDefault = 30000;
 /*-************************************
 *  Display Macros
 **************************************/
-#define DISPLAY(...)          fprintf(stderr, __VA_ARGS__)
+#define DISPLAY(...)          fprintf(stdout, __VA_ARGS__)
 #define DISPLAYLEVEL(l, ...)  if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); }
 static U32 g_displayLevel = 2;
 
 #define DISPLAYUPDATE(l, ...) if (g_displayLevel>=l) { \
             if ((FUZ_clockSpan(g_displayClock) > g_refreshRate) || (g_displayLevel>=4)) \
             { g_displayClock = clock(); DISPLAY(__VA_ARGS__); \
-            if (g_displayLevel>=4) fflush(stderr); } }
+            if (g_displayLevel>=4) fflush(stdout); } }
 static const clock_t g_refreshRate = CLOCKS_PER_SEC / 6;
 static clock_t g_displayClock = 0;
 
@@ -97,7 +97,165 @@ static unsigned FUZ_highbit32(U32 v32)
 
 
 /*=============================================
-*   Basic Unit tests
+*   Memory Tests
+=============================================*/
+#if defined(__APPLE__) && defined(__MACH__)
+
+#include <malloc/malloc.h>    /* malloc_size */
+
+typedef struct {
+    unsigned long long totalMalloc;
+    size_t currentMalloc;
+    size_t peakMalloc;
+    unsigned nbMalloc;
+    unsigned nbFree;
+} mallocCounter_t;
+
+static const mallocCounter_t INIT_MALLOC_COUNTER = { 0, 0, 0, 0, 0 };
+
+static void* FUZ_mallocDebug(void* counter, size_t size)
+{
+    mallocCounter_t* const mcPtr = (mallocCounter_t*)counter;
+    void* const ptr = malloc(size);
+    if (ptr==NULL) return NULL;
+    DISPLAYLEVEL(4, "allocating %u KB => effectively %u KB \n",
+        (U32)(size >> 10), (U32)(malloc_size(ptr) >> 10));  /* OS-X specific */
+    mcPtr->totalMalloc += size;
+    mcPtr->currentMalloc += size;
+    if (mcPtr->currentMalloc > mcPtr->peakMalloc)
+        mcPtr->peakMalloc = mcPtr->currentMalloc;
+    mcPtr->nbMalloc += 1;
+    return ptr;
+}
+
+static void FUZ_freeDebug(void* counter, void* address)
+{
+    mallocCounter_t* const mcPtr = (mallocCounter_t*)counter;
+    DISPLAYLEVEL(4, "freeing %u KB \n", (U32)(malloc_size(address) >> 10));
+    mcPtr->nbFree += 1;
+    mcPtr->currentMalloc -= malloc_size(address);  /* OS-X specific */
+    free(address);
+}
+
+static void FUZ_displayMallocStats(mallocCounter_t count)
+{
+    DISPLAYLEVEL(3, "peak:%6u KB,  nbMallocs:%2u, total:%6u KB \n",
+        (U32)(count.peakMalloc >> 10),
+        count.nbMalloc,
+        (U32)(count.totalMalloc >> 10));
+}
+
+#define CHECK_Z(f) {                               \
+    size_t const err = f;                          \
+    if (ZSTD_isError(err)) {                       \
+        DISPLAY("Error => %s : %s ",               \
+                #f, ZSTD_getErrorName(err));       \
+        exit(1);                                   \
+}   }
+
+static int FUZ_mallocTests(unsigned seed, double compressibility, unsigned part)
+{
+    size_t const inSize = 64 MB + 16 MB + 4 MB + 1 MB + 256 KB + 64 KB; /* 85.3 MB */
+    size_t const outSize = ZSTD_compressBound(inSize);
+    void* const inBuffer = malloc(inSize);
+    void* const outBuffer = malloc(outSize);
+
+    /* test only played in verbose mode, as they are long */
+    if (g_displayLevel<3) return 0;
+
+    /* Create compressible noise */
+    if (!inBuffer || !outBuffer) {
+        DISPLAY("Not enough memory, aborting\n");
+        exit(1);
+    }
+    RDG_genBuffer(inBuffer, inSize, compressibility, 0. /*auto*/, seed);
+
+    /* simple compression tests */
+    if (part <= 1)
+    {   int compressionLevel;
+        for (compressionLevel=1; compressionLevel<=6; compressionLevel++) {
+            mallocCounter_t malcount = INIT_MALLOC_COUNTER;
+            ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount };
+            ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(cMem);
+            CHECK_Z( ZSTD_compressCCtx(cctx, outBuffer, outSize, inBuffer, inSize, compressionLevel) );
+            ZSTD_freeCCtx(cctx);
+            DISPLAYLEVEL(3, "compressCCtx level %i : ", compressionLevel);
+            FUZ_displayMallocStats(malcount);
+    }   }
+
+    /* streaming compression tests */
+    if (part <= 2)
+    {   int compressionLevel;
+        for (compressionLevel=1; compressionLevel<=6; compressionLevel++) {
+            mallocCounter_t malcount = INIT_MALLOC_COUNTER;
+            ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount };
+            ZSTD_CCtx* const cstream = ZSTD_createCStream_advanced(cMem);
+            ZSTD_outBuffer out = { outBuffer, outSize, 0 };
+            ZSTD_inBuffer in = { inBuffer, inSize, 0 };
+            CHECK_Z( ZSTD_initCStream(cstream, compressionLevel) );
+            CHECK_Z( ZSTD_compressStream(cstream, &out, &in) );
+            CHECK_Z( ZSTD_endStream(cstream, &out) );
+            ZSTD_freeCStream(cstream);
+            DISPLAYLEVEL(3, "compressStream level %i : ", compressionLevel);
+            FUZ_displayMallocStats(malcount);
+    }   }
+
+    /* advanced MT API test */
+    if (part <= 3)
+    {   U32 nbThreads;
+        for (nbThreads=1; nbThreads<=4; nbThreads++) {
+            int compressionLevel;
+            for (compressionLevel=1; compressionLevel<=6; compressionLevel++) {
+                mallocCounter_t malcount = INIT_MALLOC_COUNTER;
+                ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount };
+                ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(cMem);
+                ZSTD_outBuffer out = { outBuffer, outSize, 0 };
+                ZSTD_inBuffer in = { inBuffer, inSize, 0 };
+                CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_compressionLevel, (U32)compressionLevel) );
+                CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_nbThreads, nbThreads) );
+                while ( ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_end) ) {}
+                ZSTD_freeCCtx(cctx);
+                DISPLAYLEVEL(3, "compress_generic,-T%u,end level %i : ",
+                                nbThreads, compressionLevel);
+                FUZ_displayMallocStats(malcount);
+    }   }   }
+
+    /* advanced MT streaming API test */
+    if (part <= 4)
+    {   U32 nbThreads;
+        for (nbThreads=1; nbThreads<=4; nbThreads++) {
+            int compressionLevel;
+            for (compressionLevel=1; compressionLevel<=6; compressionLevel++) {
+                mallocCounter_t malcount = INIT_MALLOC_COUNTER;
+                ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount };
+                ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(cMem);
+                ZSTD_outBuffer out = { outBuffer, outSize, 0 };
+                ZSTD_inBuffer in = { inBuffer, inSize, 0 };
+                CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_compressionLevel, (U32)compressionLevel) );
+                CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_nbThreads, nbThreads) );
+                CHECK_Z( ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_continue) );
+                while ( ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_end) ) {}
+                ZSTD_freeCCtx(cctx);
+                DISPLAYLEVEL(3, "compress_generic,-T%u,continue level %i : ",
+                                nbThreads, compressionLevel);
+                FUZ_displayMallocStats(malcount);
+    }   }   }
+
+    return 0;
+}
+
+#else
+
+static int FUZ_mallocTests(unsigned seed, double compressibility, unsigned part)
+{
+    (void)seed; (void)compressibility; (void)part;
+    return 0;
+}
+
+#endif
+
+/*=============================================
+*   Unit tests
 =============================================*/
 
 #define CHECK_V(var, fn)  size_t const var = fn; if (ZSTD_isError(var)) goto _output_error
@@ -108,7 +266,8 @@ static int basicUnitTests(U32 seed, double compressibility)
 {
     size_t const CNBuffSize = 5 MB;
     void* const CNBuffer = malloc(CNBuffSize);
-    void* const compressedBuffer = malloc(ZSTD_compressBound(CNBuffSize));
+    size_t const compressedBufferSize = ZSTD_compressBound(CNBuffSize);
+    void* const compressedBuffer = malloc(compressedBufferSize);
     void* const decodedBuffer = malloc(CNBuffSize);
     ZSTD_DCtx* dctx = ZSTD_createDCtx();
     int testResult = 0;
@@ -136,11 +295,21 @@ static int basicUnitTests(U32 seed, double compressibility)
 
 
     DISPLAYLEVEL(4, "test%3i : compress %u bytes : ", testNb++, (U32)CNBuffSize);
-    CHECKPLUS(r, ZSTD_compress(compressedBuffer, ZSTD_compressBound(CNBuffSize),
+    {   ZSTD_CCtx* cctx = ZSTD_createCCtx();
+        if (cctx==NULL) goto _output_error;
+        CHECKPLUS(r, ZSTD_compressCCtx(cctx,
+                            compressedBuffer, compressedBufferSize,
                             CNBuffer, CNBuffSize, 1),
                   cSize=r );
         DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n", (U32)cSize, (double)cSize/CNBuffSize*100);
 
+        DISPLAYLEVEL(4, "test%3i : size of cctx for level 1 : ", testNb++);
+        {   size_t const cctxSize = ZSTD_sizeof_CCtx(cctx);
+            DISPLAYLEVEL(4, "%u bytes \n", (U32)cctxSize);
+        }
+        ZSTD_freeCCtx(cctx);
+    }
+
 
     DISPLAYLEVEL(4, "test%3i : ZSTD_getFrameContentSize test : ", testNb++);
     {   unsigned long long const rSize = ZSTD_getFrameContentSize(compressedBuffer, cSize);
@@ -216,7 +385,7 @@ static int basicUnitTests(U32 seed, double compressibility)
 
             DISPLAYLEVEL(4, "test%3i : simple compression test with static CCtx : ", testNb++);
             CHECKPLUS(r, ZSTD_compressCCtx(staticCCtx,
-                            compressedBuffer, ZSTD_compressBound(CNBuffSize),
+                            compressedBuffer, compressedBufferSize,
                             CNBuffer, CNBuffSize, STATIC_CCTX_LEVEL),
                       cSize=r );
             DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n",
@@ -285,7 +454,7 @@ static int basicUnitTests(U32 seed, double compressibility)
 
         DISPLAYLEVEL(4, "test%3i : compress %u bytes with 2 threads : ", testNb++, (U32)CNBuffSize);
         CHECKPLUS(r, ZSTDMT_compressCCtx(mtctx,
-                                compressedBuffer, ZSTD_compressBound(CNBuffSize),
+                                compressedBuffer, compressedBufferSize,
                                 CNBuffer, CNBuffSize,
                                 1),
                   cSize=r );
@@ -311,6 +480,23 @@ static int basicUnitTests(U32 seed, double compressibility)
         }   }
         DISPLAYLEVEL(4, "OK \n");
 
+        DISPLAYLEVEL(4, "test%3i : compress -T2 with checksum : ", testNb++);
+        {   ZSTD_parameters params = ZSTD_getParams(1, CNBuffSize, 0);
+            params.fParams.checksumFlag = 1;
+            params.fParams.contentSizeFlag = 1;
+            CHECKPLUS(r, ZSTDMT_compress_advanced(mtctx,
+                                    compressedBuffer, compressedBufferSize,
+                                    CNBuffer, CNBuffSize,
+                                    NULL, params, 3 /*overlapRLog*/),
+                      cSize=r );
+        }
+        DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n", (U32)cSize, (double)cSize/CNBuffSize*100);
+
+        DISPLAYLEVEL(4, "test%3i : decompress %u bytes : ", testNb++, (U32)CNBuffSize);
+        { size_t const r = ZSTD_decompress(decodedBuffer, CNBuffSize, compressedBuffer, cSize);
+          if (r != CNBuffSize) goto _output_error; }
+        DISPLAYLEVEL(4, "OK \n");
+
         ZSTDMT_freeCCtx(mtctx);
     }
 
@@ -372,7 +558,7 @@ static int basicUnitTests(U32 seed, double compressibility)
 
         DISPLAYLEVEL(4, "test%3i : compress with flat dictionary : ", testNb++);
         cSize = 0;
-        CHECKPLUS(r, ZSTD_compressEnd(ctxOrig, compressedBuffer, ZSTD_compressBound(CNBuffSize),
+        CHECKPLUS(r, ZSTD_compressEnd(ctxOrig, compressedBuffer, compressedBufferSize,
                                            (const char*)CNBuffer + dictSize, CNBuffSize - dictSize),
                   cSize += r);
         DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n", (U32)cSize, (double)cSize/CNBuffSize*100);
@@ -388,7 +574,7 @@ static int basicUnitTests(U32 seed, double compressibility)
         DISPLAYLEVEL(4, "test%3i : compress with duplicated context : ", testNb++);
         {   size_t const cSizeOrig = cSize;
             cSize = 0;
-            CHECKPLUS(r, ZSTD_compressEnd(ctxDuplicated, compressedBuffer, ZSTD_compressBound(CNBuffSize),
+            CHECKPLUS(r, ZSTD_compressEnd(ctxDuplicated, compressedBuffer, compressedBufferSize,
                                                (const char*)CNBuffer + dictSize, CNBuffSize - dictSize),
                       cSize += r);
             if (cSize != cSizeOrig) goto _output_error;   /* should be identical ==> same size */
@@ -473,7 +659,7 @@ static int basicUnitTests(U32 seed, double compressibility)
         DISPLAYLEVEL(4, "OK : %u \n", dictID);
 
         DISPLAYLEVEL(4, "test%3i : compress with dictionary : ", testNb++);
-        cSize = ZSTD_compress_usingDict(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize),
+        cSize = ZSTD_compress_usingDict(cctx, compressedBuffer, compressedBufferSize,
                                         CNBuffer, CNBuffSize,
                                         dictBuffer, dictSize, 4);
         if (ZSTD_isError(cSize)) goto _output_error;
@@ -511,7 +697,7 @@ static int basicUnitTests(U32 seed, double compressibility)
                                             1 /* byReference */, ZSTD_dm_auto,
                                             cParams, ZSTD_defaultCMem);
             DISPLAYLEVEL(4, "(size : %u) : ", (U32)ZSTD_sizeof_CDict(cdict));
-            cSize = ZSTD_compress_usingCDict(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize),
+            cSize = ZSTD_compress_usingCDict(cctx, compressedBuffer, compressedBufferSize,
                                                  CNBuffer, CNBuffSize, cdict);
             ZSTD_freeCDict(cdict);
             if (ZSTD_isError(cSize)) goto _output_error;
@@ -546,7 +732,7 @@ static int basicUnitTests(U32 seed, double compressibility)
                     goto _output_error;
                 }
                 cSize = ZSTD_compress_usingCDict(cctx,
-                                compressedBuffer, ZSTD_compressBound(CNBuffSize),
+                                compressedBuffer, compressedBufferSize,
                                 CNBuffer, CNBuffSize, cdict);
                 if (ZSTD_isError(cSize)) {
                     DISPLAY("ZSTD_compress_usingCDict failed ");
@@ -560,7 +746,7 @@ static int basicUnitTests(U32 seed, double compressibility)
         {   ZSTD_frameParameters const fParams = { 0 /* frameSize */, 1 /* checksum */, 1 /* noDictID*/ };
             ZSTD_compressionParameters const cParams = ZSTD_getCParams(1, CNBuffSize, dictSize);
             ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dictBuffer, dictSize, 1 /*byRef*/, ZSTD_dm_auto, cParams, ZSTD_defaultCMem);
-            cSize = ZSTD_compress_usingCDict_advanced(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize),
+            cSize = ZSTD_compress_usingCDict_advanced(cctx, compressedBuffer, compressedBufferSize,
                                                  CNBuffer, CNBuffSize, cdict, fParams);
             ZSTD_freeCDict(cdict);
             if (ZSTD_isError(cSize)) goto _output_error;
@@ -584,7 +770,7 @@ static int basicUnitTests(U32 seed, double compressibility)
         DISPLAYLEVEL(4, "test%3i : ZSTD_compress_advanced, no dictID : ", testNb++);
         {   ZSTD_parameters p = ZSTD_getParams(3, CNBuffSize, dictSize);
             p.fParams.noDictIDFlag = 1;
-            cSize = ZSTD_compress_advanced(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize),
+            cSize = ZSTD_compress_advanced(cctx, compressedBuffer, compressedBufferSize,
                                            CNBuffer, CNBuffSize,
                                            dictBuffer, dictSize, p);
             if (ZSTD_isError(cSize)) goto _output_error;
@@ -821,6 +1007,42 @@ static int basicUnitTests(U32 seed, double compressibility)
       if (r != _3BYTESTESTLENGTH) goto _output_error; }
     DISPLAYLEVEL(4, "OK \n");
 
+    DISPLAYLEVEL(4, "test%3i : incompressible data and ill suited dictionary : ", testNb++);
+    RDG_genBuffer(CNBuffer, CNBuffSize, 0.0, 0.1, seed);
+    {   /* Train a dictionary on low characters */
+        size_t dictSize = 16 KB;
+        void* const dictBuffer = malloc(dictSize);
+        size_t const totalSampleSize = 1 MB;
+        size_t const sampleUnitSize = 8 KB;
+        U32 const nbSamples = (U32)(totalSampleSize / sampleUnitSize);
+        size_t* const samplesSizes = (size_t*) malloc(nbSamples * sizeof(size_t));
+        if (!dictBuffer || !samplesSizes) goto _output_error;
+        { U32 u; for (u=0; u<nbSamples; u++) samplesSizes[u] = sampleUnitSize; }
+        dictSize = ZDICT_trainFromBuffer(dictBuffer, dictSize, CNBuffer, samplesSizes, nbSamples);
+        if (ZDICT_isError(dictSize)) goto _output_error;
+        /* Reverse the characters to make the dictionary ill suited */
+        {   U32 u;
+            for (u = 0; u < CNBuffSize; ++u) {
+              ((BYTE*)CNBuffer)[u] = 255 - ((BYTE*)CNBuffer)[u];
+            }
+        }
+        {   /* Compress the data */
+            size_t const inputSize = 500;
+            size_t const outputSize = ZSTD_compressBound(inputSize);
+            void* const outputBuffer = malloc(outputSize);
+            ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+            if (!outputBuffer || !cctx) goto _output_error;
+            CHECK(ZSTD_compress_usingDict(cctx, outputBuffer, outputSize, CNBuffer, inputSize, dictBuffer, dictSize, 1));
+            free(outputBuffer);
+            ZSTD_freeCCtx(cctx);
+        }
+
+        free(dictBuffer);
+        free(samplesSizes);
+    }
+    DISPLAYLEVEL(4, "OK \n");
+
+
     /* findFrameCompressedSize on skippable frames */
     DISPLAYLEVEL(4, "test%3i : frame compressed size of skippable frame : ", testNb++);
     {   const char* frame = "\x50\x2a\x4d\x18\x05\x0\x0\0abcde";
@@ -892,6 +1114,7 @@ static size_t FUZ_randomLength(U32* seed, U32 maxLog)
         goto _output_error;                                   \
 }   }
 
+#undef CHECK_Z
 #define CHECK_Z(f) {                                          \
     size_t const err = f;                                     \
     if (ZSTD_isError(err)) {                                  \
@@ -1223,6 +1446,19 @@ static unsigned readU32FromChar(const char** stringPtr)
     return result;
 }
 
+/** longCommandWArg() :
+ *  check if *stringPtr is the same as longCommand.
+ *  If yes, @return 1 and advances *stringPtr to the position which immediately follows longCommand.
+ *  @return 0 and doesn't modify *stringPtr otherwise.
+ */
+static unsigned longCommandWArg(const char** stringPtr, const char* longCommand)
+{
+    size_t const comSize = strlen(longCommand);
+    int const result = !strncmp(*stringPtr, longCommand, comSize);
+    if (result) *stringPtr += comSize;
+    return result;
+}
+
 int main(int argc, const char** argv)
 {
     U32 seed = 0;
@@ -1235,6 +1471,7 @@ int main(int argc, const char** argv)
     U32 mainPause = 0;
     U32 maxDuration = 0;
     int bigTests = 1;
+    U32 memTestsOnly = 0;
     const char* const programName = argv[0];
 
     /* Check command line */
@@ -1245,6 +1482,9 @@ int main(int argc, const char** argv)
         /* Handle commands. Aggregated commands are allowed */
         if (argument[0]=='-') {
 
+            if (longCommandWArg(&argument, "--memtest=")) { memTestsOnly = readU32FromChar(&argument); continue; }
+
+            if (!strcmp(argument, "--memtest")) { memTestsOnly=1; continue; }
             if (!strcmp(argument, "--no-big-tests")) { bigTests=0; continue; }
 
             argument++;
@@ -1316,6 +1556,11 @@ int main(int argc, const char** argv)
     DISPLAY("Seed = %u\n", seed);
     if (proba!=FUZ_compressibility_default) DISPLAY("Compressibility : %u%%\n", proba);
 
+    if (memTestsOnly) {
+        g_displayLevel = MAX(3, g_displayLevel);
+        return FUZ_mallocTests(seed, ((double)proba) / 100, memTestsOnly);
+    }
+
     if (nbTests < testNb) nbTests = testNb;
 
     if (testNb==0)

tests/playTests.sh

@@ -7,17 +7,17 @@ die() {
 
 roundTripTest() {
     if [ -n "$3" ]; then
-        local_c="$3"
+        cLevel="$3"
-        local_p="$2"
+        proba="$2"
     else
-        local_c="$2"
+        cLevel="$2"
-        local_p=""
+        proba=""
     fi
 
     rm -f tmp1 tmp2
-    $ECHO "roundTripTest: ./datagen $1 $local_p | $ZSTD -v$local_c | $ZSTD -d"
+    $ECHO "roundTripTest: ./datagen $1 $proba | $ZSTD -v$cLevel | $ZSTD -d"
-    ./datagen $1 $local_p | $MD5SUM > tmp1
+    ./datagen $1 $proba | $MD5SUM > tmp1
-    ./datagen $1 $local_p | $ZSTD --ultra -v$local_c | $ZSTD -d  | $MD5SUM > tmp2
+    ./datagen $1 $proba | $ZSTD --ultra -v$cLevel | $ZSTD -d  | $MD5SUM > tmp2
     $DIFF -q tmp1 tmp2
 }
 
@@ -383,6 +383,7 @@ $ZSTD -t --rm tmp1.zst
 test -f tmp1.zst   # check file is still present
 split -b16384 tmp1.zst tmpSplit.
 $ZSTD -t tmpSplit.* && die "bad file not detected !"
+./datagen | $ZSTD -c | $ZSTD -t
 
 
 $ECHO "\n**** benchmark mode tests **** "
@@ -625,16 +626,15 @@ roundTripTest -g35000000 -P75 10
 roundTripTest -g35000000 -P75 11
 roundTripTest -g35000000 -P75 12
 
-roundTripTest -g18000000 -P80 13
+roundTripTest -g18000013 -P80 13
-roundTripTest -g18000000 -P80 14
+roundTripTest -g18000014 -P80 14
-roundTripTest -g18000000 -P80 15
+roundTripTest -g18000015 -P80 15
-roundTripTest -g18000000 -P80 16
+roundTripTest -g18000016 -P80 16
-roundTripTest -g18000000 -P80 17
+roundTripTest -g18000017 -P80 17
+roundTripTest -g18000018 -P94 18
+roundTripTest -g18000019 -P94 19
 
-roundTripTest -g50000000 -P94 18
+roundTripTest -g68000020 -P99 20
-roundTripTest -g50000000 -P94 19
-
-roundTripTest -g99000000 -P99 20
 roundTripTest -g6000000000 -P99 1
 
 fileRoundTripTest -g4193M -P99 1
@@ -644,7 +644,8 @@ then
     $ECHO "\n**** zstdmt long round-trip tests **** "
     roundTripTest -g99000000 -P99 "20 -T2"
     roundTripTest -g6000000000 -P99 "1 -T2"
-    fileRoundTripTest -g4193M -P98 " -T0"
+    roundTripTest -g1500000000 -P97 "1 -T999"
+    fileRoundTripTest -g4195M -P98 " -T0"
 else
     $ECHO "\n**** no multithreading, skipping zstdmt tests **** "
 fi

tests/zstreamtest.c

@@ -95,19 +95,6 @@ unsigned int FUZ_rand(unsigned int* seedPtr)
     return rand32 >> 5;
 }
 
-static void* allocFunction(void* opaque, size_t size)
-{
-    void* address = malloc(size);
-    (void)opaque;
-    return address;
-}
-
-static void freeFunction(void* opaque, void* address)
-{
-    (void)opaque;
-    free(address);
-}
-
 
 /*======================================================
 *   Basic Unit tests
@@ -1390,13 +1377,12 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double
         /* multi-segments compression test */
         XXH64_reset(&xxhState, 0);
         {   ZSTD_outBuffer outBuff = { cBuffer, cBufferSize, 0 } ;
-            U32 n;
+            for (cSize=0, totalTestSize=0 ; (totalTestSize < maxTestSize) ; ) {
-            for (n=0, cSize=0, totalTestSize=0 ; totalTestSize < maxTestSize ; n++) {
                 /* compress random chunks into randomly sized dst buffers */
                 size_t const randomSrcSize = FUZ_randomLength(&lseed, maxSampleLog);
                 size_t const srcSize = MIN(maxTestSize-totalTestSize, randomSrcSize);
                 size_t const srcStart = FUZ_rand(&lseed) % (srcBufferSize - srcSize);
-                size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog);
+                size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog+1);
                 size_t const dstBuffSize = MIN(cBufferSize - cSize, randomDstSize);
                 ZSTD_EndDirective const flush = (FUZ_rand(&lseed) & 15) ? ZSTD_e_continue : ZSTD_e_flush;
                 ZSTD_inBuffer inBuff = { srcBuffer+srcStart, srcSize, 0 };
@@ -1415,7 +1401,7 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double
             {   size_t remainingToFlush = (size_t)(-1);
                 while (remainingToFlush) {
                     ZSTD_inBuffer inBuff = { NULL, 0, 0 };
-                    size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog);
+                    size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog+1);
                     size_t const adjustedDstSize = MIN(cBufferSize - cSize, randomDstSize);
                     outBuff.size = outBuff.pos + adjustedDstSize;
                     DISPLAYLEVEL(5, "End-flush into dst buffer of size %u \n", (U32)adjustedDstSize);
@@ -1543,7 +1529,6 @@ int main(int argc, const char** argv)
     int bigTests = (sizeof(size_t) == 8);
     e_api selected_api = simple_api;
     const char* const programName = argv[0];
-    ZSTD_customMem const customMem = { allocFunction, freeFunction, NULL };
     ZSTD_customMem const customNULL = ZSTD_defaultCMem;
 
     /* Check command line */
@@ -1657,10 +1642,7 @@ int main(int argc, const char** argv)
 
     if (testNb==0) {
         result = basicUnitTests(0, ((double)proba) / 100, customNULL);  /* constant seed for predictability */
-        if (!result) {
+    }
-            DISPLAYLEVEL(3, "Unit tests using customMem :\n")
-            result = basicUnitTests(0, ((double)proba) / 100, customMem);  /* use custom memory allocation functions */
-    }   }
 
     if (!result) {
         switch(selected_api)