compiler/iface: compress .hi fileswip/D1159

Summary:
Compress all interface files generated by the compiler with LZ4. While
having an extremely small amount of code, LZ4 is both very fast at
compression and decompression while having quite good space saving
properties.

Non-scientific size test: size of stage2 compiler .hi files:

`find ./compiler/stage2 -type f -iname '*.hi' -exec du -ch {} + | grep total$`

Without this patch: 22MB of .hi files for stage2.
With this patch: 9.2MB of .hi files for stage2.

Signed-off-by: Austin Seipp <austin@well-typed.com>

Test Plan: I ran `./validate`

Reviewers: hvr, bgamari, thomie

Subscribers: duncan

Differential Revision: https://phabricator.haskell.org/D1159

6 files changed, 2046 insertions, 19 deletions

diff --git a/compiler/cbits/lz4.c b/compiler/cbits/lz4.c
new file mode 100644
index 0000000000..b10f6aea6e
--- /dev/null
+++ b/compiler/cbits/lz4.c
@@ -0,0 +1,1520 @@
+/* NOTE(aseipp): Below is a copy of LZ4, version r131
+** Marked as @generated for code review.
+*/
+
+/*
+   LZ4 - Fast LZ compression algorithm
+   Copyright (C) 2011-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - LZ4 source repository : https://github.com/Cyan4973/lz4
+   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+
+/**************************************
+*  Tuning parameters
+**************************************/
+/*
+ * HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0:default, fastest), or in memory heap (1:requires malloc()).
+ */
+#define HEAPMODE 0
+
+/*
+ * ACCELERATION_DEFAULT :
+ * Select "acceleration" for LZ4_compress_fast() when parameter value <= 0
+ */
+#define ACCELERATION_DEFAULT 1
+
+
+/**************************************
+*  CPU Feature Detection
+**************************************/
+/*
+ * LZ4_FORCE_SW_BITCOUNT
+ * Define this parameter if your target system or compiler does not support hardware bit count
+ */
+#if defined(_MSC_VER) && defined(_WIN32_WCE)   /* Visual Studio for Windows CE does not support Hardware bit count */
+#  define LZ4_FORCE_SW_BITCOUNT
+#endif
+
+
+/**************************************
+*  Includes
+**************************************/
+#include "lz4.h"
+
+
+/**************************************
+*  Compiler Options
+**************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4293)        /* disable: C4293: too large shift (32-bits) */
+#else
+#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)   /* C99 */
+#    if defined(__GNUC__) || defined(__clang__)
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif   /* __STDC_VERSION__ */
+#endif  /* _MSC_VER */
+
+/* LZ4_GCC_VERSION is defined into lz4.h */
+#if (LZ4_GCC_VERSION >= 302) || (__INTEL_COMPILER >= 800) || defined(__clang__)
+#  define expect(expr,value)    (__builtin_expect ((expr),(value)) )
+#else
+#  define expect(expr,value)    (expr)
+#endif
+
+#define likely(expr)     expect((expr) != 0, 1)
+#define unlikely(expr)   expect((expr) != 0, 0)
+
+
+/**************************************
+*  Memory routines
+**************************************/
+#include <stdlib.h>   /* malloc, calloc, free */
+#define ALLOCATOR(n,s) calloc(n,s)
+#define FREEMEM        free
+#include <string.h>   /* memset, memcpy */
+#define MEM_INIT       memset
+
+
+/**************************************
+*  Basic Types
+**************************************/
+#if defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)   /* C99 */
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+#endif
+
+
+/**************************************
+*  Reading and writing into memory
+**************************************/
+#define STEPSIZE sizeof(size_t)
+
+static unsigned LZ4_64bits(void) { return sizeof(void*)==8; }
+
+static unsigned LZ4_isLittleEndian(void)
+{
+    const union { U32 i; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+
+static U16 LZ4_read16(const void* memPtr)
+{
+    U16 val16;
+    memcpy(&val16, memPtr, 2);
+    return val16;
+}
+
+static U16 LZ4_readLE16(const void* memPtr)
+{
+    if (LZ4_isLittleEndian())
+    {
+        return LZ4_read16(memPtr);
+    }
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)((U16)p[0] + (p[1]<<8));
+    }
+}
+
+static void LZ4_writeLE16(void* memPtr, U16 value)
+{
+    if (LZ4_isLittleEndian())
+    {
+        memcpy(memPtr, &value, 2);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE) value;
+        p[1] = (BYTE)(value>>8);
+    }
+}
+
+static U32 LZ4_read32(const void* memPtr)
+{
+    U32 val32;
+    memcpy(&val32, memPtr, 4);
+    return val32;
+}
+
+static U64 LZ4_read64(const void* memPtr)
+{
+    U64 val64;
+    memcpy(&val64, memPtr, 8);
+    return val64;
+}
+
+static size_t LZ4_read_ARCH(const void* p)
+{
+    if (LZ4_64bits())
+        return (size_t)LZ4_read64(p);
+    else
+        return (size_t)LZ4_read32(p);
+}
+
+
+static void LZ4_copy4(void* dstPtr, const void* srcPtr) { memcpy(dstPtr, srcPtr, 4); }
+
+static void LZ4_copy8(void* dstPtr, const void* srcPtr) { memcpy(dstPtr, srcPtr, 8); }
+
+/* customized version of memcpy, which may overwrite up to 7 bytes beyond dstEnd */
+static void LZ4_wildCopy(void* dstPtr, const void* srcPtr, void* dstEnd)
+{
+    BYTE* d = (BYTE*)dstPtr;
+    const BYTE* s = (const BYTE*)srcPtr;
+    BYTE* e = (BYTE*)dstEnd;
+    do { LZ4_copy8(d,s); d+=8; s+=8; } while (d<e);
+}
+
+
+/**************************************
+*  Common Constants
+**************************************/
+#define MINMATCH 4
+
+#define COPYLENGTH 8
+#define LASTLITERALS 5
+#define MFLIMIT (COPYLENGTH+MINMATCH)
+static const int LZ4_minLength = (MFLIMIT+1);
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define MAXD_LOG 16
+#define MAX_DISTANCE ((1 << MAXD_LOG) - 1)
+
+#define ML_BITS  4
+#define ML_MASK  ((1U<<ML_BITS)-1)
+#define RUN_BITS (8-ML_BITS)
+#define RUN_MASK ((1U<<RUN_BITS)-1)
+
+
+/**************************************
+*  Common Utils
+**************************************/
+#define LZ4_STATIC_ASSERT(c)    { enum { LZ4_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/**************************************
+*  Common functions
+**************************************/
+static unsigned LZ4_NbCommonBytes (register size_t val)
+{
+    if (LZ4_isLittleEndian())
+    {
+        if (LZ4_64bits())
+        {
+#       if defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanForward64( &r, (U64)val );
+            return (int)(r>>3);
+#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            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) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r;
+            _BitScanForward( &r, (U32)val );
+            return (int)(r>>3);
+#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            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 (LZ4_64bits())
+        {
+#       if defined(_MSC_VER) && defined(_WIN64) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanReverse64( &r, val );
+            return (unsigned)(r>>3);
+#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            return (__builtin_clzll((U64)val) >> 3);
+#       else
+            unsigned r;
+            if (!(val>>32)) { r=4; } else { r=0; val>>=32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        }
+        else /* 32 bits */
+        {
+#       if defined(_MSC_VER) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            unsigned long r = 0;
+            _BitScanReverse( &r, (unsigned long)val );
+            return (unsigned)(r>>3);
+#       elif (defined(__clang__) || (LZ4_GCC_VERSION >= 304)) && !defined(LZ4_FORCE_SW_BITCOUNT)
+            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
+        }
+    }
+}
+
+static unsigned LZ4_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* pInLimit)
+{
+    const BYTE* const pStart = pIn;
+
+    while (likely(pIn<pInLimit-(STEPSIZE-1)))
+    {
+        size_t diff = LZ4_read_ARCH(pMatch) ^ LZ4_read_ARCH(pIn);
+        if (!diff) { pIn+=STEPSIZE; pMatch+=STEPSIZE; continue; }
+        pIn += LZ4_NbCommonBytes(diff);
+        return (unsigned)(pIn - pStart);
+    }
+
+    if (LZ4_64bits()) if ((pIn<(pInLimit-3)) && (LZ4_read32(pMatch) == LZ4_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (LZ4_read16(pMatch) == LZ4_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (unsigned)(pIn - pStart);
+}
+
+
+#ifndef LZ4_COMMONDEFS_ONLY
+/**************************************
+*  Local Constants
+**************************************/
+#define LZ4_HASHLOG   (LZ4_MEMORY_USAGE-2)
+#define HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
+#define HASH_SIZE_U32 (1 << LZ4_HASHLOG)       /* required as macro for static allocation */
+
+static const int LZ4_64Klimit = ((64 KB) + (MFLIMIT-1));
+static const U32 LZ4_skipTrigger = 6;  /* Increase this value ==> compression run slower on incompressible data */
+
+
+/**************************************
+*  Local Structures and types
+**************************************/
+typedef struct {
+    U32 hashTable[HASH_SIZE_U32];
+    U32 currentOffset;
+    U32 initCheck;
+    const BYTE* dictionary;
+    BYTE* bufferStart;   /* obsolete, used for slideInputBuffer */
+    U32 dictSize;
+} LZ4_stream_t_internal;
+
+typedef enum { notLimited = 0, limitedOutput = 1 } limitedOutput_directive;
+typedef enum { byPtr, byU32, byU16 } tableType_t;
+
+typedef enum { noDict = 0, withPrefix64k, usingExtDict } dict_directive;
+typedef enum { noDictIssue = 0, dictSmall } dictIssue_directive;
+
+typedef enum { endOnOutputSize = 0, endOnInputSize = 1 } endCondition_directive;
+typedef enum { full = 0, partial = 1 } earlyEnd_directive;
+
+
+/**************************************
+*  Local Utils
+**************************************/
+int LZ4_versionNumber (void) { return LZ4_VERSION_NUMBER; }
+int LZ4_compressBound(int isize)  { return LZ4_COMPRESSBOUND(isize); }
+int LZ4_sizeofState() { return LZ4_STREAMSIZE; }
+
+
+
+/********************************
+*  Compression functions
+********************************/
+
+static U32 LZ4_hashSequence(U32 sequence, tableType_t const tableType)
+{
+    if (tableType == byU16)
+        return (((sequence) * 2654435761U) >> ((MINMATCH*8)-(LZ4_HASHLOG+1)));
+    else
+        return (((sequence) * 2654435761U) >> ((MINMATCH*8)-LZ4_HASHLOG));
+}
+
+static const U64 prime5bytes = 889523592379ULL;
+static U32 LZ4_hashSequence64(size_t sequence, tableType_t const tableType)
+{
+    const U32 hashLog = (tableType == byU16) ? LZ4_HASHLOG+1 : LZ4_HASHLOG;
+    const U32 hashMask = (1<<hashLog) - 1;
+    return ((sequence * prime5bytes) >> (40 - hashLog)) & hashMask;
+}
+
+static U32 LZ4_hashSequenceT(size_t sequence, tableType_t const tableType)
+{
+    if (LZ4_64bits())
+        return LZ4_hashSequence64(sequence, tableType);
+    return LZ4_hashSequence((U32)sequence, tableType);
+}
+
+static U32 LZ4_hashPosition(const void* p, tableType_t tableType) { return LZ4_hashSequenceT(LZ4_read_ARCH(p), tableType); }
+
+static void LZ4_putPositionOnHash(const BYTE* p, U32 h, void* tableBase, tableType_t const tableType, const BYTE* srcBase)
+{
+    switch (tableType)
+    {
+    case byPtr: { const BYTE** hashTable = (const BYTE**)tableBase; hashTable[h] = p; return; }
+    case byU32: { U32* hashTable = (U32*) tableBase; hashTable[h] = (U32)(p-srcBase); return; }
+    case byU16: { U16* hashTable = (U16*) tableBase; hashTable[h] = (U16)(p-srcBase); return; }
+    }
+}
+
+static void LZ4_putPosition(const BYTE* p, void* tableBase, tableType_t tableType, const BYTE* srcBase)
+{
+    U32 h = LZ4_hashPosition(p, tableType);
+    LZ4_putPositionOnHash(p, h, tableBase, tableType, srcBase);
+}
+
+static const BYTE* LZ4_getPositionOnHash(U32 h, void* tableBase, tableType_t tableType, const BYTE* srcBase)
+{
+    if (tableType == byPtr) { const BYTE** hashTable = (const BYTE**) tableBase; return hashTable[h]; }
+    if (tableType == byU32) { U32* hashTable = (U32*) tableBase; return hashTable[h] + srcBase; }
+    { U16* hashTable = (U16*) tableBase; return hashTable[h] + srcBase; }   /* default, to ensure a return */
+}
+
+static const BYTE* LZ4_getPosition(const BYTE* p, void* tableBase, tableType_t tableType, const BYTE* srcBase)
+{
+    U32 h = LZ4_hashPosition(p, tableType);
+    return LZ4_getPositionOnHash(h, tableBase, tableType, srcBase);
+}
+
+FORCE_INLINE int LZ4_compress_generic(
+                 void* const ctx,
+                 const char* const source,
+                 char* const dest,
+                 const int inputSize,
+                 const int maxOutputSize,
+                 const limitedOutput_directive outputLimited,
+                 const tableType_t tableType,
+                 const dict_directive dict,
+                 const dictIssue_directive dictIssue,
+                 const U32 acceleration)
+{
+    LZ4_stream_t_internal* const dictPtr = (LZ4_stream_t_internal*)ctx;
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* base;
+    const BYTE* lowLimit;
+    const BYTE* const lowRefLimit = ip - dictPtr->dictSize;
+    const BYTE* const dictionary = dictPtr->dictionary;
+    const BYTE* const dictEnd = dictionary + dictPtr->dictSize;
+    const size_t dictDelta = dictEnd - (const BYTE*)source;
+    const BYTE* anchor = (const BYTE*) source;
+    const BYTE* const iend = ip + inputSize;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+
+    BYTE* op = (BYTE*) dest;
+    BYTE* const olimit = op + maxOutputSize;
+
+    U32 forwardH;
+    size_t refDelta=0;
+
+    /* Init conditions */
+    if ((U32)inputSize > (U32)LZ4_MAX_INPUT_SIZE) return 0;   /* Unsupported input size, too large (or negative) */
+    switch(dict)
+    {
+    case noDict:
+    default:
+        base = (const BYTE*)source;
+        lowLimit = (const BYTE*)source;
+        break;
+    case withPrefix64k:
+        base = (const BYTE*)source - dictPtr->currentOffset;
+        lowLimit = (const BYTE*)source - dictPtr->dictSize;
+        break;
+    case usingExtDict:
+        base = (const BYTE*)source - dictPtr->currentOffset;
+        lowLimit = (const BYTE*)source;
+        break;
+    }
+    if ((tableType == byU16) && (inputSize>=LZ4_64Klimit)) return 0;   /* Size too large (not within 64K limit) */
+    if (inputSize<LZ4_minLength) goto _last_literals;                  /* Input too small, no compression (all literals) */
+
+    /* First Byte */
+    LZ4_putPosition(ip, ctx, tableType, base);
+    ip++; forwardH = LZ4_hashPosition(ip, tableType);
+
+    /* Main Loop */
+    for ( ; ; )
+    {
+        const BYTE* match;
+        BYTE* token;
+        {
+            const BYTE* forwardIp = ip;
+            unsigned step = 1;
+            unsigned searchMatchNb = acceleration << LZ4_skipTrigger;
+
+            /* Find a match */
+            do {
+                U32 h = forwardH;
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimit)) goto _last_literals;
+
+                match = LZ4_getPositionOnHash(h, ctx, tableType, base);
+                if (dict==usingExtDict)
+                {
+                    if (match<(const BYTE*)source)
+                    {
+                        refDelta = dictDelta;
+                        lowLimit = dictionary;
+                    }
+                    else
+                    {
+                        refDelta = 0;
+                        lowLimit = (const BYTE*)source;
+                    }
+                }
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putPositionOnHash(ip, h, ctx, tableType, base);
+
+            } while ( ((dictIssue==dictSmall) ? (match < lowRefLimit) : 0)
+                || ((tableType==byU16) ? 0 : (match + MAX_DISTANCE < ip))
+                || (LZ4_read32(match+refDelta) != LZ4_read32(ip)) );
+        }
+
+        /* Catch up */
+        while ((ip>anchor) && (match+refDelta > lowLimit) && (unlikely(ip[-1]==match[refDelta-1]))) { ip--; match--; }
+
+        {
+            /* Encode Literal length */
+            unsigned litLength = (unsigned)(ip - anchor);
+            token = op++;
+            if ((outputLimited) && (unlikely(op + litLength + (2 + 1 + LASTLITERALS) + (litLength/255) > olimit)))
+                return 0;   /* Check output limit */
+            if (litLength>=RUN_MASK)
+            {
+                int len = (int)litLength-RUN_MASK;
+                *token=(RUN_MASK<<ML_BITS);
+                for(; len >= 255 ; len-=255) *op++ = 255;
+                *op++ = (BYTE)len;
+            }
+            else *token = (BYTE)(litLength<<ML_BITS);
+
+            /* Copy Literals */
+            LZ4_wildCopy(op, anchor, op+litLength);
+            op+=litLength;
+        }
+
+_next_match:
+        /* Encode Offset */
+        LZ4_writeLE16(op, (U16)(ip-match)); op+=2;
+
+        /* Encode MatchLength */
+        {
+            unsigned matchLength;
+
+            if ((dict==usingExtDict) && (lowLimit==dictionary))
+            {
+                const BYTE* limit;
+                match += refDelta;
+                limit = ip + (dictEnd-match);
+                if (limit > matchlimit) limit = matchlimit;
+                matchLength = LZ4_count(ip+MINMATCH, match+MINMATCH, limit);
+                ip += MINMATCH + matchLength;
+                if (ip==limit)
+                {
+                    unsigned more = LZ4_count(ip, (const BYTE*)source, matchlimit);
+                    matchLength += more;
+                    ip += more;
+                }
+            }
+            else
+            {
+                matchLength = LZ4_count(ip+MINMATCH, match+MINMATCH, matchlimit);
+                ip += MINMATCH + matchLength;
+            }
+
+            if ((outputLimited) && (unlikely(op + (1 + LASTLITERALS) + (matchLength>>8) > olimit)))
+                return 0;    /* Check output limit */
+            if (matchLength>=ML_MASK)
+            {
+                *token += ML_MASK;
+                matchLength -= ML_MASK;
+                for (; matchLength >= 510 ; matchLength-=510) { *op++ = 255; *op++ = 255; }
+                if (matchLength >= 255) { matchLength-=255; *op++ = 255; }
+                *op++ = (BYTE)matchLength;
+            }
+            else *token += (BYTE)(matchLength);
+        }
+
+        anchor = ip;
+
+        /* Test end of chunk */
+        if (ip > mflimit) break;
+
+        /* Fill table */
+        LZ4_putPosition(ip-2, ctx, tableType, base);
+
+        /* Test next position */
+        match = LZ4_getPosition(ip, ctx, tableType, base);
+        if (dict==usingExtDict)
+        {
+            if (match<(const BYTE*)source)
+            {
+                refDelta = dictDelta;
+                lowLimit = dictionary;
+            }
+            else
+            {
+                refDelta = 0;
+                lowLimit = (const BYTE*)source;
+            }
+        }
+        LZ4_putPosition(ip, ctx, tableType, base);
+        if ( ((dictIssue==dictSmall) ? (match>=lowRefLimit) : 1)
+            && (match+MAX_DISTANCE>=ip)
+            && (LZ4_read32(match+refDelta)==LZ4_read32(ip)) )
+        { token=op++; *token=0; goto _next_match; }
+
+        /* Prepare next loop */
+        forwardH = LZ4_hashPosition(++ip, tableType);
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {
+        const size_t lastRun = (size_t)(iend - anchor);
+        if ((outputLimited) && ((op - (BYTE*)dest) + lastRun + 1 + ((lastRun+255-RUN_MASK)/255) > (U32)maxOutputSize))
+            return 0;   /* Check output limit */
+        if (lastRun >= RUN_MASK)
+        {
+            size_t accumulator = lastRun - RUN_MASK;
+            *op++ = RUN_MASK << ML_BITS;
+            for(; accumulator >= 255 ; accumulator-=255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        }
+        else
+        {
+            *op++ = (BYTE)(lastRun<<ML_BITS);
+        }
+        memcpy(op, anchor, lastRun);
+        op += lastRun;
+    }
+
+    /* End */
+    return (int) (((char*)op)-dest);
+}
+
+
+int LZ4_compress_fast_extState(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+    LZ4_resetStream((LZ4_stream_t*)state);
+    if (acceleration < 1) acceleration = ACCELERATION_DEFAULT;
+
+    if (maxOutputSize >= LZ4_compressBound(inputSize))
+    {
+        if (inputSize < LZ4_64Klimit)
+            return LZ4_compress_generic(state, source, dest, inputSize, 0, notLimited, byU16,                        noDict, noDictIssue, acceleration);
+        else
+            return LZ4_compress_generic(state, source, dest, inputSize, 0, notLimited, LZ4_64bits() ? byU32 : byPtr, noDict, noDictIssue, acceleration);
+    }
+    else
+    {
+        if (inputSize < LZ4_64Klimit)
+            return LZ4_compress_generic(state, source, dest, inputSize, maxOutputSize, limitedOutput, byU16,                        noDict, noDictIssue, acceleration);
+        else
+            return LZ4_compress_generic(state, source, dest, inputSize, maxOutputSize, limitedOutput, LZ4_64bits() ? byU32 : byPtr, noDict, noDictIssue, acceleration);
+    }
+}
+
+
+int LZ4_compress_fast(const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+#if (HEAPMODE)
+    void* ctxPtr = ALLOCATOR(1, sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+#else
+    LZ4_stream_t ctx;
+    void* ctxPtr = &ctx;
+#endif
+
+    int result = LZ4_compress_fast_extState(ctxPtr, source, dest, inputSize, maxOutputSize, acceleration);
+
+#if (HEAPMODE)
+    FREEMEM(ctxPtr);
+#endif
+    return result;
+}
+
+
+int LZ4_compress_default(const char* source, char* dest, int inputSize, int maxOutputSize)
+{
+    return LZ4_compress_fast(source, dest, inputSize, maxOutputSize, 1);
+}
+
+
+/* hidden debug function */
+/* strangely enough, gcc generates faster code when this function is uncommented, even if unused */
+int LZ4_compress_fast_force(const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+    LZ4_stream_t ctx;
+
+    LZ4_resetStream(&ctx);
+
+    if (inputSize < LZ4_64Klimit)
+        return LZ4_compress_generic(&ctx, source, dest, inputSize, maxOutputSize, limitedOutput, byU16,                        noDict, noDictIssue, acceleration);
+    else
+        return LZ4_compress_generic(&ctx, source, dest, inputSize, maxOutputSize, limitedOutput, LZ4_64bits() ? byU32 : byPtr, noDict, noDictIssue, acceleration);
+}
+
+
+/********************************
+*  destSize variant
+********************************/
+
+static int LZ4_compress_destSize_generic(
+                       void* const ctx,
+                 const char* const src,
+                       char* const dst,
+                       int*  const srcSizePtr,
+                 const int targetDstSize,
+                 const tableType_t tableType)
+{
+    const BYTE* ip = (const BYTE*) src;
+    const BYTE* base = (const BYTE*) src;
+    const BYTE* lowLimit = (const BYTE*) src;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + *srcSizePtr;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+
+    BYTE* op = (BYTE*) dst;
+    BYTE* const oend = op + targetDstSize;
+    BYTE* const oMaxLit = op + targetDstSize - 2 /* offset */ - 8 /* because 8+MINMATCH==MFLIMIT */ - 1 /* token */;
+    BYTE* const oMaxMatch = op + targetDstSize - (LASTLITERALS + 1 /* token */);
+    BYTE* const oMaxSeq = oMaxLit - 1 /* token */;
+
+    U32 forwardH;
+
+
+    /* Init conditions */
+    if (targetDstSize < 1) return 0;                                     /* Impossible to store anything */
+    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;            /* Unsupported input size, too large (or negative) */
+    if ((tableType == byU16) && (*srcSizePtr>=LZ4_64Klimit)) return 0;   /* Size too large (not within 64K limit) */
+    if (*srcSizePtr<LZ4_minLength) goto _last_literals;                  /* Input too small, no compression (all literals) */
+
+    /* First Byte */
+    *srcSizePtr = 0;
+    LZ4_putPosition(ip, ctx, tableType, base);
+    ip++; forwardH = LZ4_hashPosition(ip, tableType);
+
+    /* Main Loop */
+    for ( ; ; )
+    {
+        const BYTE* match;
+        BYTE* token;
+        {
+            const BYTE* forwardIp = ip;
+            unsigned step = 1;
+            unsigned searchMatchNb = 1 << LZ4_skipTrigger;
+
+            /* Find a match */
+            do {
+                U32 h = forwardH;
+                ip = forwardIp;
+                forwardIp += step;
+                step = (searchMatchNb++ >> LZ4_skipTrigger);
+
+                if (unlikely(forwardIp > mflimit))
+                    goto _last_literals;
+
+                match = LZ4_getPositionOnHash(h, ctx, tableType, base);
+                forwardH = LZ4_hashPosition(forwardIp, tableType);
+                LZ4_putPositionOnHash(ip, h, ctx, tableType, base);
+
+            } while ( ((tableType==byU16) ? 0 : (match + MAX_DISTANCE < ip))
+                || (LZ4_read32(match) != LZ4_read32(ip)) );
+        }
+
+        /* Catch up */
+        while ((ip>anchor) && (match > lowLimit) && (unlikely(ip[-1]==match[-1]))) { ip--; match--; }
+
+        {
+            /* Encode Literal length */
+            unsigned litLength = (unsigned)(ip - anchor);
+            token = op++;
+            if (op + ((litLength+240)/255) + litLength > oMaxLit)
+            {
+                /* Not enough space for a last match */
+                op--;
+                goto _last_literals;
+            }
+            if (litLength>=RUN_MASK)
+            {
+                unsigned len = litLength - RUN_MASK;
+                *token=(RUN_MASK<<ML_BITS);
+                for(; len >= 255 ; len-=255) *op++ = 255;
+                *op++ = (BYTE)len;
+            }
+            else *token = (BYTE)(litLength<<ML_BITS);
+
+            /* Copy Literals */
+            LZ4_wildCopy(op, anchor, op+litLength);
+            op += litLength;
+        }
+
+_next_match:
+        /* Encode Offset */
+        LZ4_writeLE16(op, (U16)(ip-match)); op+=2;
+
+        /* Encode MatchLength */
+        {
+            size_t matchLength;
+
+            matchLength = LZ4_count(ip+MINMATCH, match+MINMATCH, matchlimit);
+
+            if (op + ((matchLength+240)/255) > oMaxMatch)
+            {
+                /* Match description too long : reduce it */
+                matchLength = (15-1) + (oMaxMatch-op) * 255;
+            }
+            //printf("offset %5i, matchLength%5i \n", (int)(ip-match), matchLength + MINMATCH);
+            ip += MINMATCH + matchLength;
+
+            if (matchLength>=ML_MASK)
+            {
+                *token += ML_MASK;
+                matchLength -= ML_MASK;
+                while (matchLength >= 255) { matchLength-=255; *op++ = 255; }
+                *op++ = (BYTE)matchLength;
+            }
+            else *token += (BYTE)(matchLength);
+        }
+
+        anchor = ip;
+
+        /* Test end of block */
+        if (ip > mflimit) break;
+        if (op > oMaxSeq) break;
+
+        /* Fill table */
+        LZ4_putPosition(ip-2, ctx, tableType, base);
+
+        /* Test next position */
+        match = LZ4_getPosition(ip, ctx, tableType, base);
+        LZ4_putPosition(ip, ctx, tableType, base);
+        if ( (match+MAX_DISTANCE>=ip)
+            && (LZ4_read32(match)==LZ4_read32(ip)) )
+        { token=op++; *token=0; goto _next_match; }
+
+        /* Prepare next loop */
+        forwardH = LZ4_hashPosition(++ip, tableType);
+    }
+
+_last_literals:
+    /* Encode Last Literals */
+    {
+        size_t lastRunSize = (size_t)(iend - anchor);
+        if (op + 1 /* token */ + ((lastRunSize+240)/255) /* litLength */ + lastRunSize /* literals */ > oend)
+        {
+            /* adapt lastRunSize to fill 'dst' */
+            lastRunSize  = (oend-op) - 1;
+            lastRunSize -= (lastRunSize+240)/255;
+        }
+        ip = anchor + lastRunSize;
+
+        if (lastRunSize >= RUN_MASK)
+        {
+            size_t accumulator = lastRunSize - RUN_MASK;
+            *op++ = RUN_MASK << ML_BITS;
+            for(; accumulator >= 255 ; accumulator-=255) *op++ = 255;
+            *op++ = (BYTE) accumulator;
+        }
+        else
+        {
+            *op++ = (BYTE)(lastRunSize<<ML_BITS);
+        }
+        memcpy(op, anchor, lastRunSize);
+        op += lastRunSize;
+    }
+
+    /* End */
+    *srcSizePtr = (int) (((const char*)ip)-src);
+    return (int) (((char*)op)-dst);
+}
+
+
+static int LZ4_compress_destSize_extState (void* state, const char* src, char* dst, int* srcSizePtr, int targetDstSize)
+{
+    LZ4_resetStream((LZ4_stream_t*)state);
+
+    if (targetDstSize >= LZ4_compressBound(*srcSizePtr))   /* compression success is guaranteed */
+    {
+        return LZ4_compress_fast_extState(state, src, dst, *srcSizePtr, targetDstSize, 1);
+    }
+    else
+    {
+        if (*srcSizePtr < LZ4_64Klimit)
+            return LZ4_compress_destSize_generic(state, src, dst, srcSizePtr, targetDstSize, byU16);
+        else
+            return LZ4_compress_destSize_generic(state, src, dst, srcSizePtr, targetDstSize, LZ4_64bits() ? byU32 : byPtr);
+    }
+}
+
+
+int LZ4_compress_destSize(const char* src, char* dst, int* srcSizePtr, int targetDstSize)
+{
+#if (HEAPMODE)
+    void* ctx = ALLOCATOR(1, sizeof(LZ4_stream_t));   /* malloc-calloc always properly aligned */
+#else
+    LZ4_stream_t ctxBody;
+    void* ctx = &ctxBody;
+#endif
+
+    int result = LZ4_compress_destSize_extState(ctx, src, dst, srcSizePtr, targetDstSize);
+
+#if (HEAPMODE)
+    FREEMEM(ctx);
+#endif
+    return result;
+}
+
+
+
+/********************************
+*  Streaming functions
+********************************/
+
+LZ4_stream_t* LZ4_createStream(void)
+{
+    LZ4_stream_t* lz4s = (LZ4_stream_t*)ALLOCATOR(8, LZ4_STREAMSIZE_U64);
+    LZ4_STATIC_ASSERT(LZ4_STREAMSIZE >= sizeof(LZ4_stream_t_internal));    /* A compilation error here means LZ4_STREAMSIZE is not large enough */
+    LZ4_resetStream(lz4s);
+    return lz4s;
+}
+
+void LZ4_resetStream (LZ4_stream_t* LZ4_stream)
+{
+    MEM_INIT(LZ4_stream, 0, sizeof(LZ4_stream_t));
+}
+
+int LZ4_freeStream (LZ4_stream_t* LZ4_stream)
+{
+    FREEMEM(LZ4_stream);
+    return (0);
+}
+
+
+#define HASH_UNIT sizeof(size_t)
+int LZ4_loadDict (LZ4_stream_t* LZ4_dict, const char* dictionary, int dictSize)
+{
+    LZ4_stream_t_internal* dict = (LZ4_stream_t_internal*) LZ4_dict;
+    const BYTE* p = (const BYTE*)dictionary;
+    const BYTE* const dictEnd = p + dictSize;
+    const BYTE* base;
+
+    if ((dict->initCheck) || (dict->currentOffset > 1 GB))  /* Uninitialized structure, or reuse overflow */
+        LZ4_resetStream(LZ4_dict);
+
+    if (dictSize < (int)HASH_UNIT)
+    {
+        dict->dictionary = NULL;
+        dict->dictSize = 0;
+        return 0;
+    }
+
+    if ((dictEnd - p) > 64 KB) p = dictEnd - 64 KB;
+    dict->currentOffset += 64 KB;
+    base = p - dict->currentOffset;
+    dict->dictionary = p;
+    dict->dictSize = (U32)(dictEnd - p);
+    dict->currentOffset += dict->dictSize;
+
+    while (p <= dictEnd-HASH_UNIT)
+    {
+        LZ4_putPosition(p, dict->hashTable, byU32, base);
+        p+=3;
+    }
+
+    return dict->dictSize;
+}
+
+
+static void LZ4_renormDictT(LZ4_stream_t_internal* LZ4_dict, const BYTE* src)
+{
+    if ((LZ4_dict->currentOffset > 0x80000000) ||
+        ((size_t)LZ4_dict->currentOffset > (size_t)src))   /* address space overflow */
+    {
+        /* rescale hash table */
+        U32 delta = LZ4_dict->currentOffset - 64 KB;
+        const BYTE* dictEnd = LZ4_dict->dictionary + LZ4_dict->dictSize;
+        int i;
+        for (i=0; i<HASH_SIZE_U32; i++)
+        {
+            if (LZ4_dict->hashTable[i] < delta) LZ4_dict->hashTable[i]=0;
+            else LZ4_dict->hashTable[i] -= delta;
+        }
+        LZ4_dict->currentOffset = 64 KB;
+        if (LZ4_dict->dictSize > 64 KB) LZ4_dict->dictSize = 64 KB;
+        LZ4_dict->dictionary = dictEnd - LZ4_dict->dictSize;
+    }
+}
+
+
+int LZ4_compress_fast_continue (LZ4_stream_t* LZ4_stream, const char* source, char* dest, int inputSize, int maxOutputSize, int acceleration)
+{
+    LZ4_stream_t_internal* streamPtr = (LZ4_stream_t_internal*)LZ4_stream;
+    const BYTE* const dictEnd = streamPtr->dictionary + streamPtr->dictSize;
+
+    const BYTE* smallest = (const BYTE*) source;
+    if (streamPtr->initCheck) return 0;   /* Uninitialized structure detected */
+    if ((streamPtr->dictSize>0) && (smallest>dictEnd)) smallest = dictEnd;
+    LZ4_renormDictT(streamPtr, smallest);
+    if (acceleration < 1) acceleration = ACCELERATION_DEFAULT;
+
+    /* Check overlapping input/dictionary space */
+    {
+        const BYTE* sourceEnd = (const BYTE*) source + inputSize;
+        if ((sourceEnd > streamPtr->dictionary) && (sourceEnd < dictEnd))
+        {
+            streamPtr->dictSize = (U32)(dictEnd - sourceEnd);
+            if (streamPtr->dictSize > 64 KB) streamPtr->dictSize = 64 KB;
+            if (streamPtr->dictSize < 4) streamPtr->dictSize = 0;
+            streamPtr->dictionary = dictEnd - streamPtr->dictSize;
+        }
+    }
+
+    /* prefix mode : source data follows dictionary */
+    if (dictEnd == (const BYTE*)source)
+    {
+        int result;
+        if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset))
+            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, withPrefix64k, dictSmall, acceleration);
+        else
+            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, withPrefix64k, noDictIssue, acceleration);
+        streamPtr->dictSize += (U32)inputSize;
+        streamPtr->currentOffset += (U32)inputSize;
+        return result;
+    }
+
+    /* external dictionary mode */
+    {
+        int result;
+        if ((streamPtr->dictSize < 64 KB) && (streamPtr->dictSize < streamPtr->currentOffset))
+            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, usingExtDict, dictSmall, acceleration);
+        else
+            result = LZ4_compress_generic(LZ4_stream, source, dest, inputSize, maxOutputSize, limitedOutput, byU32, usingExtDict, noDictIssue, acceleration);
+        streamPtr->dictionary = (const BYTE*)source;
+        streamPtr->dictSize = (U32)inputSize;
+        streamPtr->currentOffset += (U32)inputSize;
+        return result;
+    }
+}
+
+
+/* Hidden debug function, to force external dictionary mode */
+int LZ4_compress_forceExtDict (LZ4_stream_t* LZ4_dict, const char* source, char* dest, int inputSize)
+{
+    LZ4_stream_t_internal* streamPtr = (LZ4_stream_t_internal*)LZ4_dict;
+    int result;
+    const BYTE* const dictEnd = streamPtr->dictionary + streamPtr->dictSize;
+
+    const BYTE* smallest = dictEnd;
+    if (smallest > (const BYTE*) source) smallest = (const BYTE*) source;
+    LZ4_renormDictT((LZ4_stream_t_internal*)LZ4_dict, smallest);
+
+    result = LZ4_compress_generic(LZ4_dict, source, dest, inputSize, 0, notLimited, byU32, usingExtDict, noDictIssue, 1);
+
+    streamPtr->dictionary = (const BYTE*)source;
+    streamPtr->dictSize = (U32)inputSize;
+    streamPtr->currentOffset += (U32)inputSize;
+
+    return result;
+}
+
+
+int LZ4_saveDict (LZ4_stream_t* LZ4_dict, char* safeBuffer, int dictSize)
+{
+    LZ4_stream_t_internal* dict = (LZ4_stream_t_internal*) LZ4_dict;
+    const BYTE* previousDictEnd = dict->dictionary + dict->dictSize;
+
+    if ((U32)dictSize > 64 KB) dictSize = 64 KB;   /* useless to define a dictionary > 64 KB */
+    if ((U32)dictSize > dict->dictSize) dictSize = dict->dictSize;
+
+    memmove(safeBuffer, previousDictEnd - dictSize, dictSize);
+
+    dict->dictionary = (const BYTE*)safeBuffer;
+    dict->dictSize = (U32)dictSize;
+
+    return dictSize;
+}
+
+
+
+/*******************************
+*  Decompression functions
+*******************************/
+/*
+ * This generic decompression function cover all use cases.
+ * It shall be instantiated several times, using different sets of directives
+ * Note that it is essential this generic function is really inlined,
+ * in order to remove useless branches during compilation optimization.
+ */
+FORCE_INLINE int LZ4_decompress_generic(
+                 const char* const source,
+                 char* const dest,
+                 int inputSize,
+                 int outputSize,         /* If endOnInput==endOnInputSize, this value is the max size of Output Buffer. */
+
+                 int endOnInput,         /* endOnOutputSize, endOnInputSize */
+                 int partialDecoding,    /* full, partial */
+                 int targetOutputSize,   /* only used if partialDecoding==partial */
+                 int dict,               /* noDict, withPrefix64k, usingExtDict */
+                 const BYTE* const lowPrefix,  /* == dest if dict == noDict */
+                 const BYTE* const dictStart,  /* only if dict==usingExtDict */
+                 const size_t dictSize         /* note : = 0 if noDict */
+                 )
+{
+    /* Local Variables */
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* const iend = ip + inputSize;
+
+    BYTE* op = (BYTE*) dest;
+    BYTE* const oend = op + outputSize;
+    BYTE* cpy;
+    BYTE* oexit = op + targetOutputSize;
+    const BYTE* const lowLimit = lowPrefix - dictSize;
+
+    const BYTE* const dictEnd = (const BYTE*)dictStart + dictSize;
+    const size_t dec32table[] = {4, 1, 2, 1, 4, 4, 4, 4};
+    const size_t dec64table[] = {0, 0, 0, (size_t)-1, 0, 1, 2, 3};
+
+    const int safeDecode = (endOnInput==endOnInputSize);
+    const int checkOffset = ((safeDecode) && (dictSize < (int)(64 KB)));
+
+
+    /* Special cases */
+    if ((partialDecoding) && (oexit> oend-MFLIMIT)) oexit = oend-MFLIMIT;                         /* targetOutputSize too high => decode everything */
+    if ((endOnInput) && (unlikely(outputSize==0))) return ((inputSize==1) && (*ip==0)) ? 0 : -1;  /* Empty output buffer */
+    if ((!endOnInput) && (unlikely(outputSize==0))) return (*ip==0?1:-1);
+
+
+    /* Main Loop */
+    while (1)
+    {
+        unsigned token;
+        size_t length;
+        const BYTE* match;
+
+        /* get literal length */
+        token = *ip++;
+        if ((length=(token>>ML_BITS)) == RUN_MASK)
+        {
+            unsigned s;
+            do
+            {
+                s = *ip++;
+                length += s;
+            }
+            while (likely((endOnInput)?ip<iend-RUN_MASK:1) && (s==255));
+            if ((safeDecode) && unlikely((size_t)(op+length)<(size_t)(op))) goto _output_error;   /* overflow detection */
+            if ((safeDecode) && unlikely((size_t)(ip+length)<(size_t)(ip))) goto _output_error;   /* overflow detection */
+        }
+
+        /* copy literals */
+        cpy = op+length;
+        if (((endOnInput) && ((cpy>(partialDecoding?oexit:oend-MFLIMIT)) || (ip+length>iend-(2+1+LASTLITERALS))) )
+            || ((!endOnInput) && (cpy>oend-COPYLENGTH)))
+        {
+            if (partialDecoding)
+            {
+                if (cpy > oend) goto _output_error;                           /* Error : write attempt beyond end of output buffer */
+                if ((endOnInput) && (ip+length > iend)) goto _output_error;   /* Error : read attempt beyond end of input buffer */
+            }
+            else
+            {
+                if ((!endOnInput) && (cpy != oend)) goto _output_error;       /* Error : block decoding must stop exactly there */
+                if ((endOnInput) && ((ip+length != iend) || (cpy > oend))) goto _output_error;   /* Error : input must be consumed */
+            }
+            memcpy(op, ip, length);
+            ip += length;
+            op += length;
+            break;     /* Necessarily EOF, due to parsing restrictions */
+        }
+        LZ4_wildCopy(op, ip, cpy);
+        ip += length; op = cpy;
+
+        /* get offset */
+        match = cpy - LZ4_readLE16(ip); ip+=2;
+        if ((checkOffset) && (unlikely(match < lowLimit))) goto _output_error;   /* Error : offset outside destination buffer */
+
+        /* get matchlength */
+        length = token & ML_MASK;
+        if (length == ML_MASK)
+        {
+            unsigned s;
+            do
+            {
+                if ((endOnInput) && (ip > iend-LASTLITERALS)) goto _output_error;
+                s = *ip++;
+                length += s;
+            } while (s==255);
+            if ((safeDecode) && unlikely((size_t)(op+length)<(size_t)op)) goto _output_error;   /* overflow detection */
+        }
+        length += MINMATCH;
+
+        /* check external dictionary */
+        if ((dict==usingExtDict) && (match < lowPrefix))
+        {
+            if (unlikely(op+length > oend-LASTLITERALS)) goto _output_error;   /* doesn't respect parsing restriction */
+
+            if (length <= (size_t)(lowPrefix-match))
+            {
+                /* match can be copied as a single segment from external dictionary */
+                match = dictEnd - (lowPrefix-match);
+                memmove(op, match, length); op += length;
+            }
+            else
+            {
+                /* match encompass external dictionary and current segment */
+                size_t copySize = (size_t)(lowPrefix-match);
+                memcpy(op, dictEnd - copySize, copySize);
+                op += copySize;
+                copySize = length - copySize;
+                if (copySize > (size_t)(op-lowPrefix))   /* overlap within current segment */
+                {
+                    BYTE* const endOfMatch = op + copySize;
+                    const BYTE* copyFrom = lowPrefix;
+                    while (op < endOfMatch) *op++ = *copyFrom++;
+                }
+                else
+                {
+                    memcpy(op, lowPrefix, copySize);
+                    op += copySize;
+                }
+            }
+            continue;
+        }
+
+        /* copy repeated sequence */
+        cpy = op + length;
+        if (unlikely((op-match)<8))
+        {
+            const size_t dec64 = dec64table[op-match];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[op-match];
+            LZ4_copy4(op+4, match);
+            op += 8; match -= dec64;
+        } else { LZ4_copy8(op, match); op+=8; match+=8; }
+
+        if (unlikely(cpy>oend-12))
+        {
+            if (cpy > oend-LASTLITERALS) goto _output_error;    /* Error : last LASTLITERALS bytes must be literals */
+            if (op < oend-8)
+            {
+                LZ4_wildCopy(op, match, oend-8);
+                match += (oend-8) - op;
+                op = oend-8;
+            }
+            while (op<cpy) *op++ = *match++;
+        }
+        else
+            LZ4_wildCopy(op, match, cpy);
+        op=cpy;   /* correction */
+    }
+
+    /* end of decoding */
+    if (endOnInput)
+       return (int) (((char*)op)-dest);     /* Nb of output bytes decoded */
+    else
+       return (int) (((const char*)ip)-source);   /* Nb of input bytes read */
+
+    /* Overflow error detected */
+_output_error:
+    return (int) (-(((const char*)ip)-source))-1;
+}
+
+
+int LZ4_decompress_safe(const char* source, char* dest, int compressedSize, int maxDecompressedSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxDecompressedSize, endOnInputSize, full, 0, noDict, (BYTE*)dest, NULL, 0);
+}
+
+int LZ4_decompress_safe_partial(const char* source, char* dest, int compressedSize, int targetOutputSize, int maxDecompressedSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxDecompressedSize, endOnInputSize, partial, targetOutputSize, noDict, (BYTE*)dest, NULL, 0);
+}
+
+int LZ4_decompress_fast(const char* source, char* dest, int originalSize)
+{
+    return LZ4_decompress_generic(source, dest, 0, originalSize, endOnOutputSize, full, 0, withPrefix64k, (BYTE*)(dest - 64 KB), NULL, 64 KB);
+}
+
+
+/* streaming decompression functions */
+
+typedef struct
+{
+    const BYTE* externalDict;
+    size_t extDictSize;
+    const BYTE* prefixEnd;
+    size_t prefixSize;
+} LZ4_streamDecode_t_internal;
+
+/*
+ * If you prefer dynamic allocation methods,
+ * LZ4_createStreamDecode()
+ * provides a pointer (void*) towards an initialized LZ4_streamDecode_t structure.
+ */
+LZ4_streamDecode_t* LZ4_createStreamDecode(void)
+{
+    LZ4_streamDecode_t* lz4s = (LZ4_streamDecode_t*) ALLOCATOR(1, sizeof(LZ4_streamDecode_t));
+    return lz4s;
+}
+
+int LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream)
+{
+    FREEMEM(LZ4_stream);
+    return 0;
+}
+
+/*
+ * LZ4_setStreamDecode
+ * Use this function to instruct where to find the dictionary
+ * This function is not necessary if previous data is still available where it was decoded.
+ * Loading a size of 0 is allowed (same effect as no dictionary).
+ * Return : 1 if OK, 0 if error
+ */
+int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = (LZ4_streamDecode_t_internal*) LZ4_streamDecode;
+    lz4sd->prefixSize = (size_t) dictSize;
+    lz4sd->prefixEnd = (const BYTE*) dictionary + dictSize;
+    lz4sd->externalDict = NULL;
+    lz4sd->extDictSize  = 0;
+    return 1;
+}
+
+/*
+*_continue() :
+    These decoding functions allow decompression of multiple blocks in "streaming" mode.
+    Previously decoded blocks must still be available at the memory position where they were decoded.
+    If it's not possible, save the relevant part of decoded data into a safe buffer,
+    and indicate where it stands using LZ4_setStreamDecode()
+*/
+int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = (LZ4_streamDecode_t_internal*) LZ4_streamDecode;
+    int result;
+
+    if (lz4sd->prefixEnd == (BYTE*)dest)
+    {
+        result = LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                        endOnInputSize, full, 0,
+                                        usingExtDict, lz4sd->prefixEnd - lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += result;
+        lz4sd->prefixEnd  += result;
+    }
+    else
+    {
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = lz4sd->prefixEnd - lz4sd->extDictSize;
+        result = LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize,
+                                        endOnInputSize, full, 0,
+                                        usingExtDict, (BYTE*)dest, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = result;
+        lz4sd->prefixEnd  = (BYTE*)dest + result;
+    }
+
+    return result;
+}
+
+int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int originalSize)
+{
+    LZ4_streamDecode_t_internal* lz4sd = (LZ4_streamDecode_t_internal*) LZ4_streamDecode;
+    int result;
+
+    if (lz4sd->prefixEnd == (BYTE*)dest)
+    {
+        result = LZ4_decompress_generic(source, dest, 0, originalSize,
+                                        endOnOutputSize, full, 0,
+                                        usingExtDict, lz4sd->prefixEnd - lz4sd->prefixSize, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize += originalSize;
+        lz4sd->prefixEnd  += originalSize;
+    }
+    else
+    {
+        lz4sd->extDictSize = lz4sd->prefixSize;
+        lz4sd->externalDict = (BYTE*)dest - lz4sd->extDictSize;
+        result = LZ4_decompress_generic(source, dest, 0, originalSize,
+                                        endOnOutputSize, full, 0,
+                                        usingExtDict, (BYTE*)dest, lz4sd->externalDict, lz4sd->extDictSize);
+        if (result <= 0) return result;
+        lz4sd->prefixSize = originalSize;
+        lz4sd->prefixEnd  = (BYTE*)dest + originalSize;
+    }
+
+    return result;
+}
+
+
+/*
+Advanced decoding functions :
+*_usingDict() :
+    These decoding functions work the same as "_continue" ones,
+    the dictionary must be explicitly provided within parameters
+*/
+
+FORCE_INLINE int LZ4_decompress_usingDict_generic(const char* source, char* dest, int compressedSize, int maxOutputSize, int safe, const char* dictStart, int dictSize)
+{
+    if (dictSize==0)
+        return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, noDict, (BYTE*)dest, NULL, 0);
+    if (dictStart+dictSize == dest)
+    {
+        if (dictSize >= (int)(64 KB - 1))
+            return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, withPrefix64k, (BYTE*)dest-64 KB, NULL, 0);
+        return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, noDict, (BYTE*)dest-dictSize, NULL, 0);
+    }
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, safe, full, 0, usingExtDict, (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+int LZ4_decompress_safe_usingDict(const char* source, char* dest, int compressedSize, int maxOutputSize, const char* dictStart, int dictSize)
+{
+    return LZ4_decompress_usingDict_generic(source, dest, compressedSize, maxOutputSize, 1, dictStart, dictSize);
+}
+
+int LZ4_decompress_fast_usingDict(const char* source, char* dest, int originalSize, const char* dictStart, int dictSize)
+{
+    return LZ4_decompress_usingDict_generic(source, dest, 0, originalSize, 0, dictStart, dictSize);
+}
+
+/* debug function */
+int LZ4_decompress_safe_forceExtDict(const char* source, char* dest, int compressedSize, int maxOutputSize, const char* dictStart, int dictSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, endOnInputSize, full, 0, usingExtDict, (BYTE*)dest, (const BYTE*)dictStart, dictSize);
+}
+
+
+/***************************************************
+*  Obsolete Functions
+***************************************************/
+/* obsolete compression functions */
+int LZ4_compress_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize) { return LZ4_compress_default(source, dest, inputSize, maxOutputSize); }
+int LZ4_compress(const char* source, char* dest, int inputSize) { return LZ4_compress_default(source, dest, inputSize, LZ4_compressBound(inputSize)); }
+int LZ4_compress_limitedOutput_withState (void* state, const char* src, char* dst, int srcSize, int dstSize) { return LZ4_compress_fast_extState(state, src, dst, srcSize, dstSize, 1); }
+int LZ4_compress_withState (void* state, const char* src, char* dst, int srcSize) { return LZ4_compress_fast_extState(state, src, dst, srcSize, LZ4_compressBound(srcSize), 1); }
+int LZ4_compress_limitedOutput_continue (LZ4_stream_t* LZ4_stream, const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_fast_continue(LZ4_stream, src, dst, srcSize, maxDstSize, 1); }
+int LZ4_compress_continue (LZ4_stream_t* LZ4_stream, const char* source, char* dest, int inputSize) { return LZ4_compress_fast_continue(LZ4_stream, source, dest, inputSize, LZ4_compressBound(inputSize), 1); }
+
+/*
+These function names are deprecated and should no longer be used.
+They are only provided here for compatibility with older user programs.
+- LZ4_uncompress is totally equivalent to LZ4_decompress_fast
+- LZ4_uncompress_unknownOutputSize is totally equivalent to LZ4_decompress_safe
+*/
+int LZ4_uncompress (const char* source, char* dest, int outputSize) { return LZ4_decompress_fast(source, dest, outputSize); }
+int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize) { return LZ4_decompress_safe(source, dest, isize, maxOutputSize); }
+
+
+/* Obsolete Streaming functions */
+
+int LZ4_sizeofStreamState() { return LZ4_STREAMSIZE; }
+
+static void LZ4_init(LZ4_stream_t_internal* lz4ds, BYTE* base)
+{
+    MEM_INIT(lz4ds, 0, LZ4_STREAMSIZE);
+    lz4ds->bufferStart = base;
+}
+
+int LZ4_resetStreamState(void* state, char* inputBuffer)
+{
+    if ((((size_t)state) & 3) != 0) return 1;   /* Error : pointer is not aligned on 4-bytes boundary */
+    LZ4_init((LZ4_stream_t_internal*)state, (BYTE*)inputBuffer);
+    return 0;
+}
+
+void* LZ4_create (char* inputBuffer)
+{
+    void* lz4ds = ALLOCATOR(8, LZ4_STREAMSIZE_U64);
+    LZ4_init ((LZ4_stream_t_internal*)lz4ds, (BYTE*)inputBuffer);
+    return lz4ds;
+}
+
+char* LZ4_slideInputBuffer (void* LZ4_Data)
+{
+    LZ4_stream_t_internal* ctx = (LZ4_stream_t_internal*)LZ4_Data;
+    int dictSize = LZ4_saveDict((LZ4_stream_t*)LZ4_Data, (char*)ctx->bufferStart, 64 KB);
+    return (char*)(ctx->bufferStart + dictSize);
+}
+
+/* Obsolete streaming decompression functions */
+
+int LZ4_decompress_safe_withPrefix64k(const char* source, char* dest, int compressedSize, int maxOutputSize)
+{
+    return LZ4_decompress_generic(source, dest, compressedSize, maxOutputSize, endOnInputSize, full, 0, withPrefix64k, (BYTE*)dest - 64 KB, NULL, 64 KB);
+}
+
+int LZ4_decompress_fast_withPrefix64k(const char* source, char* dest, int originalSize)
+{
+    return LZ4_decompress_generic(source, dest, 0, originalSize, endOnOutputSize, full, 0, withPrefix64k, (BYTE*)dest - 64 KB, NULL, 64 KB);
+}
+
+#endif   /* LZ4_COMMONDEFS_ONLY */
+
diff --git a/compiler/cbits/lz4.h b/compiler/cbits/lz4.h
new file mode 100644
index 0000000000..126221e6bd
--- /dev/null
+++ b/compiler/cbits/lz4.h
@@ -0,0 +1,364 @@
+/* NOTE(aseipp): Below is a copy of LZ4, version r131
+** Marked as @generated for code review.
+*/
+
+/*
+   LZ4 - Fast LZ compression algorithm
+   Header File
+   Copyright (C) 2011-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - LZ4 source repository : https://github.com/Cyan4973/lz4
+   - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+#pragma once
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*
+ * lz4.h provides block compression functions, and gives full buffer control to programmer.
+ * If you need to generate inter-operable compressed data (respecting LZ4 frame specification),
+ * and can let the library handle its own memory, please use lz4frame.h instead.
+*/
+
+/**************************************
+*  Version
+**************************************/
+#define LZ4_VERSION_MAJOR    1    /* for breaking interface changes  */
+#define LZ4_VERSION_MINOR    7    /* for new (non-breaking) interface capabilities */
+#define LZ4_VERSION_RELEASE  1    /* for tweaks, bug-fixes, or development */
+#define LZ4_VERSION_NUMBER (LZ4_VERSION_MAJOR *100*100 + LZ4_VERSION_MINOR *100 + LZ4_VERSION_RELEASE)
+int LZ4_versionNumber (void);
+
+/**************************************
+*  Tuning parameter
+**************************************/
+/*
+ * LZ4_MEMORY_USAGE :
+ * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+ * Increasing memory usage improves compression ratio
+ * Reduced memory usage can improve speed, due to cache effect
+ * Default value is 14, for 16KB, which nicely fits into Intel x86 L1 cache
+ */
+#define LZ4_MEMORY_USAGE 14
+
+
+/**************************************
+*  Simple Functions
+**************************************/
+
+int LZ4_compress_default(const char* source, char* dest, int sourceSize, int maxDestSize);
+int LZ4_decompress_safe (const char* source, char* dest, int compressedSize, int maxDecompressedSize);
+
+/*
+LZ4_compress_default() :
+    Compresses 'sourceSize' bytes from buffer 'source'
+    into already allocated 'dest' buffer of size 'maxDestSize'.
+    Compression is guaranteed to succeed if 'maxDestSize' >= LZ4_compressBound(sourceSize).
+    It also runs faster, so it's a recommended setting.
+    If the function cannot compress 'source' into a more limited 'dest' budget,
+    compression stops *immediately*, and the function result is zero.
+    As a consequence, 'dest' content is not valid.
+    This function never writes outside 'dest' buffer, nor read outside 'source' buffer.
+        sourceSize  : Max supported value is LZ4_MAX_INPUT_VALUE
+        maxDestSize : full or partial size of buffer 'dest' (which must be already allocated)
+        return : the number of bytes written into buffer 'dest' (necessarily <= maxOutputSize)
+              or 0 if compression fails
+
+LZ4_decompress_safe() :
+    compressedSize : is the precise full size of the compressed block.
+    maxDecompressedSize : is the size of destination buffer, which must be already allocated.
+    return : the number of bytes decompressed into destination buffer (necessarily <= maxDecompressedSize)
+             If destination buffer is not large enough, decoding will stop and output an error code (<0).
+             If the source stream is detected malformed, the function will stop decoding and return a negative result.
+             This function is protected against buffer overflow exploits, including malicious data packets.
+             It never writes outside output buffer, nor reads outside input buffer.
+*/
+
+
+/**************************************
+*  Advanced Functions
+**************************************/
+#define LZ4_MAX_INPUT_SIZE        0x7E000000   /* 2 113 929 216 bytes */
+#define LZ4_COMPRESSBOUND(isize)  ((unsigned)(isize) > (unsigned)LZ4_MAX_INPUT_SIZE ? 0 : (isize) + ((isize)/255) + 16)
+
+/*
+LZ4_compressBound() :
+    Provides the maximum size that LZ4 compression may output in a "worst case" scenario (input data not compressible)
+    This function is primarily useful for memory allocation purposes (destination buffer size).
+    Macro LZ4_COMPRESSBOUND() is also provided for compilation-time evaluation (stack memory allocation for example).
+    Note that LZ4_compress_default() compress faster when dest buffer size is >= LZ4_compressBound(srcSize)
+        inputSize  : max supported value is LZ4_MAX_INPUT_SIZE
+        return : maximum output size in a "worst case" scenario
+              or 0, if input size is too large ( > LZ4_MAX_INPUT_SIZE)
+*/
+int LZ4_compressBound(int inputSize);
+
+/*
+LZ4_compress_fast() :
+    Same as LZ4_compress_default(), but allows to select an "acceleration" factor.
+    The larger the acceleration value, the faster the algorithm, but also the lesser the compression.
+    It's a trade-off. It can be fine tuned, with each successive value providing roughly +~3% to speed.
+    An acceleration value of "1" is the same as regular LZ4_compress_default()
+    Values <= 0 will be replaced by ACCELERATION_DEFAULT (see lz4.c), which is 1.
+*/
+int LZ4_compress_fast (const char* source, char* dest, int sourceSize, int maxDestSize, int acceleration);
+
+
+/*
+LZ4_compress_fast_extState() :
+    Same compression function, just using an externally allocated memory space to store compression state.
+    Use LZ4_sizeofState() to know how much memory must be allocated,
+    and allocate it on 8-bytes boundaries (using malloc() typically).
+    Then, provide it as 'void* state' to compression function.
+*/
+int LZ4_sizeofState(void);
+int LZ4_compress_fast_extState (void* state, const char* source, char* dest, int inputSize, int maxDestSize, int acceleration);
+
+
+/*
+LZ4_compress_destSize() :
+    Reverse the logic, by compressing as much data as possible from 'source' buffer
+    into already allocated buffer 'dest' of size 'targetDestSize'.
+    This function either compresses the entire 'source' content into 'dest' if it's large enough,
+    or fill 'dest' buffer completely with as much data as possible from 'source'.
+        *sourceSizePtr : will be modified to indicate how many bytes where read from 'source' to fill 'dest'.
+                         New value is necessarily <= old value.
+        return : Nb bytes written into 'dest' (necessarily <= targetDestSize)
+              or 0 if compression fails
+*/
+int LZ4_compress_destSize (const char* source, char* dest, int* sourceSizePtr, int targetDestSize);
+
+
+/*
+LZ4_decompress_fast() :
+    originalSize : is the original and therefore uncompressed size
+    return : the number of bytes read from the source buffer (in other words, the compressed size)
+             If the source stream is detected malformed, the function will stop decoding and return a negative result.
+             Destination buffer must be already allocated. Its size must be a minimum of 'originalSize' bytes.
+    note : This function fully respect memory boundaries for properly formed compressed data.
+           It is a bit faster than LZ4_decompress_safe().
+           However, it does not provide any protection against intentionally modified data stream (malicious input).
+           Use this function in trusted environment only (data to decode comes from a trusted source).
+*/
+int LZ4_decompress_fast (const char* source, char* dest, int originalSize);
+
+/*
+LZ4_decompress_safe_partial() :
+    This function decompress a compressed block of size 'compressedSize' at position 'source'
+    into destination buffer 'dest' of size 'maxDecompressedSize'.
+    The function tries to stop decompressing operation as soon as 'targetOutputSize' has been reached,
+    reducing decompression time.
+    return : the number of bytes decoded in the destination buffer (necessarily <= maxDecompressedSize)
+       Note : this number can be < 'targetOutputSize' should the compressed block to decode be smaller.
+             Always control how many bytes were decoded.
+             If the source stream is detected malformed, the function will stop decoding and return a negative result.
+             This function never writes outside of output buffer, and never reads outside of input buffer. It is therefore protected against malicious data packets
+*/
+int LZ4_decompress_safe_partial (const char* source, char* dest, int compressedSize, int targetOutputSize, int maxDecompressedSize);
+
+
+/***********************************************
+*  Streaming Compression Functions
+***********************************************/
+#define LZ4_STREAMSIZE_U64 ((1 << (LZ4_MEMORY_USAGE-3)) + 4)
+#define LZ4_STREAMSIZE     (LZ4_STREAMSIZE_U64 * sizeof(long long))
+/*
+ * LZ4_stream_t
+ * information structure to track an LZ4 stream.
+ * important : init this structure content before first use !
+ * note : only allocated directly the structure if you are statically linking LZ4
+ *        If you are using liblz4 as a DLL, please use below construction methods instead.
+ */
+typedef struct { long long table[LZ4_STREAMSIZE_U64]; } LZ4_stream_t;
+
+/*
+ * LZ4_resetStream
+ * Use this function to init an allocated LZ4_stream_t structure
+ */
+void LZ4_resetStream (LZ4_stream_t* streamPtr);
+
+/*
+ * LZ4_createStream will allocate and initialize an LZ4_stream_t structure
+ * LZ4_freeStream releases its memory.
+ * In the context of a DLL (liblz4), please use these methods rather than the static struct.
+ * They are more future proof, in case of a change of LZ4_stream_t size.
+ */
+LZ4_stream_t* LZ4_createStream(void);
+int           LZ4_freeStream (LZ4_stream_t* streamPtr);
+
+/*
+ * LZ4_loadDict
+ * Use this function to load a static dictionary into LZ4_stream.
+ * Any previous data will be forgotten, only 'dictionary' will remain in memory.
+ * Loading a size of 0 is allowed.
+ * Return : dictionary size, in bytes (necessarily <= 64 KB)
+ */
+int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
+
+/*
+ * LZ4_compress_fast_continue
+ * Compress buffer content 'src', using data from previously compressed blocks as dictionary to improve compression ratio.
+ * Important : Previous data blocks are assumed to still be present and unmodified !
+ * 'dst' buffer must be already allocated.
+ * If maxDstSize >= LZ4_compressBound(srcSize), compression is guaranteed to succeed, and runs faster.
+ * If not, and if compressed data cannot fit into 'dst' buffer size, compression stops, and function returns a zero.
+ */
+int LZ4_compress_fast_continue (LZ4_stream_t* streamPtr, const char* src, char* dst, int srcSize, int maxDstSize, int acceleration);
+
+/*
+ * LZ4_saveDict
+ * If previously compressed data block is not guaranteed to remain available at its memory location
+ * save it into a safer place (char* safeBuffer)
+ * Note : you don't need to call LZ4_loadDict() afterwards,
+ *        dictionary is immediately usable, you can therefore call LZ4_compress_fast_continue()
+ * Return : saved dictionary size in bytes (necessarily <= dictSize), or 0 if error
+ */
+int LZ4_saveDict (LZ4_stream_t* streamPtr, char* safeBuffer, int dictSize);
+
+
+/************************************************
+*  Streaming Decompression Functions
+************************************************/
+
+#define LZ4_STREAMDECODESIZE_U64  4
+#define LZ4_STREAMDECODESIZE     (LZ4_STREAMDECODESIZE_U64 * sizeof(unsigned long long))
+typedef struct { unsigned long long table[LZ4_STREAMDECODESIZE_U64]; } LZ4_streamDecode_t;
+/*
+ * LZ4_streamDecode_t
+ * information structure to track an LZ4 stream.
+ * init this structure content using LZ4_setStreamDecode or memset() before first use !
+ *
+ * In the context of a DLL (liblz4) please prefer usage of construction methods below.
+ * They are more future proof, in case of a change of LZ4_streamDecode_t size in the future.
+ * LZ4_createStreamDecode will allocate and initialize an LZ4_streamDecode_t structure
+ * LZ4_freeStreamDecode releases its memory.
+ */
+LZ4_streamDecode_t* LZ4_createStreamDecode(void);
+int                 LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_stream);
+
+/*
+ * LZ4_setStreamDecode
+ * Use this function to instruct where to find the dictionary.
+ * Setting a size of 0 is allowed (same effect as reset).
+ * Return : 1 if OK, 0 if error
+ */
+int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize);
+
+/*
+*_continue() :
+    These decoding functions allow decompression of multiple blocks in "streaming" mode.
+    Previously decoded blocks *must* remain available at the memory position where they were decoded (up to 64 KB)
+    In the case of a ring buffers, decoding buffer must be either :
+    - Exactly same size as encoding buffer, with same update rule (block boundaries at same positions)
+      In which case, the decoding & encoding ring buffer can have any size, including very small ones ( < 64 KB).
+    - Larger than encoding buffer, by a minimum of maxBlockSize more bytes.
+      maxBlockSize is implementation dependent. It's the maximum size you intend to compress into a single block.
+      In which case, encoding and decoding buffers do not need to be synchronized,
+      and encoding ring buffer can have any size, including small ones ( < 64 KB).
+    - _At least_ 64 KB + 8 bytes + maxBlockSize.
+      In which case, encoding and decoding buffers do not need to be synchronized,
+      and encoding ring buffer can have any size, including larger than decoding buffer.
+    Whenever these conditions are not possible, save the last 64KB of decoded data into a safe buffer,
+    and indicate where it is saved using LZ4_setStreamDecode()
+*/
+int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int compressedSize, int maxDecompressedSize);
+int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* source, char* dest, int originalSize);
+
+
+/*
+Advanced decoding functions :
+*_usingDict() :
+    These decoding functions work the same as
+    a combination of LZ4_setStreamDecode() followed by LZ4_decompress_x_continue()
+    They are stand-alone. They don't need nor update an LZ4_streamDecode_t structure.
+*/
+int LZ4_decompress_safe_usingDict (const char* source, char* dest, int compressedSize, int maxDecompressedSize, const char* dictStart, int dictSize);
+int LZ4_decompress_fast_usingDict (const char* source, char* dest, int originalSize, const char* dictStart, int dictSize);
+
+
+
+/**************************************
+*  Obsolete Functions
+**************************************/
+/* Deprecate Warnings */
+/* Should these warnings messages be a problem,
+   it is generally possible to disable them,
+   with -Wno-deprecated-declarations for gcc
+   or _CRT_SECURE_NO_WARNINGS in Visual for example.
+   You can also define LZ4_DEPRECATE_WARNING_DEFBLOCK. */
+#ifndef LZ4_DEPRECATE_WARNING_DEFBLOCK
+#  define LZ4_DEPRECATE_WARNING_DEFBLOCK
+#  define LZ4_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#  if (LZ4_GCC_VERSION >= 405) || defined(__clang__)
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif (LZ4_GCC_VERSION >= 301)
+#    define LZ4_DEPRECATED(message) __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define LZ4_DEPRECATED(message) __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement LZ4_DEPRECATED for this compiler")
+#    define LZ4_DEPRECATED(message)
+#  endif
+#endif /* LZ4_DEPRECATE_WARNING_DEFBLOCK */
+
+/* Obsolete compression functions */
+/* These functions are planned to start generate warnings by r131 approximately */
+int LZ4_compress               (const char* source, char* dest, int sourceSize);
+int LZ4_compress_limitedOutput (const char* source, char* dest, int sourceSize, int maxOutputSize);
+int LZ4_compress_withState               (void* state, const char* source, char* dest, int inputSize);
+int LZ4_compress_limitedOutput_withState (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
+int LZ4_compress_continue                (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize);
+int LZ4_compress_limitedOutput_continue  (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
+
+/* Obsolete decompression functions */
+/* These function names are completely deprecated and must no longer be used.
+   They are only provided here for compatibility with older programs.
+    - LZ4_uncompress is the same as LZ4_decompress_fast
+    - LZ4_uncompress_unknownOutputSize is the same as LZ4_decompress_safe
+   These function prototypes are now disabled; uncomment them only if you really need them.
+   It is highly recommended to stop using these prototypes and migrate to maintained ones */
+/* int LZ4_uncompress (const char* source, char* dest, int outputSize); */
+/* int LZ4_uncompress_unknownOutputSize (const char* source, char* dest, int isize, int maxOutputSize); */
+
+/* Obsolete streaming functions; use new streaming interface whenever possible */
+LZ4_DEPRECATED("use LZ4_createStream() instead") void* LZ4_create (char* inputBuffer);
+LZ4_DEPRECATED("use LZ4_createStream() instead") int   LZ4_sizeofStreamState(void);
+LZ4_DEPRECATED("use LZ4_resetStream() instead")  int   LZ4_resetStreamState(void* state, char* inputBuffer);
+LZ4_DEPRECATED("use LZ4_saveDict() instead")     char* LZ4_slideInputBuffer (void* state);
+
+/* Obsolete streaming decoding functions */
+LZ4_DEPRECATED("use LZ4_decompress_safe_usingDict() instead") int LZ4_decompress_safe_withPrefix64k (const char* src, char* dst, int compressedSize, int maxDstSize);
+LZ4_DEPRECATED("use LZ4_decompress_fast_usingDict() instead") int LZ4_decompress_fast_withPrefix64k (const char* src, char* dst, int originalSize);
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/compiler/ghc.cabal.in b/compiler/ghc.cabal.in
index 28227f32bf..eb3227a36a 100644
--- a/compiler/ghc.cabal.in
+++ b/compiler/ghc.cabal.in
@@ -122,6 +122,7 @@ Library
         parser/cutils.c
         ghci/keepCAFsForGHCi.c
         cbits/genSym.c
+        cbits/lz4.c
 
     hs-source-dirs:
         backpack
@@ -454,6 +455,7 @@ Library
         GraphPpr
         IOEnv
         ListSetOps
+        LZ4
         Maybes
         MonadUtils
         OrdList
diff --git a/compiler/ghc.mk b/compiler/ghc.mk
index e1634fda28..4695b4c109 100644
--- a/compiler/ghc.mk
+++ b/compiler/ghc.mk
@@ -535,6 +535,7 @@ compiler_stage2_dll0_MODULES = \
 	Lexer \
 	ListSetOps \
 	Literal \
+	LZ4 \
 	Maybes \
 	MkCore \
 	MkId \
diff --git a/compiler/utils/Binary.hs b/compiler/utils/Binary.hs
index 8f0d8e50dc..e06891da11 100644
--- a/compiler/utils/Binary.hs
+++ b/compiler/utils/Binary.hs
@@ -70,10 +70,12 @@ import FastMutInt
 import Fingerprint
 import BasicTypes
 import SrcLoc
+import qualified LZ4 as LZ4
 
 import Foreign
 import Data.Array
 import Data.ByteString (ByteString)
+import qualified Data.ByteString          as B
 import qualified Data.ByteString.Internal as BS
 import qualified Data.ByteString.Unsafe   as BS
 import Data.IORef
@@ -81,8 +83,7 @@ import Data.Char                ( ord, chr )
 import Data.Time
 import Data.Typeable
 import Data.Typeable.Internal
-import Control.Monad            ( when )
-import System.IO as IO
+import Control.Monad            ( when, liftM )
 import System.IO.Unsafe         ( unsafeInterleaveIO )
 import System.IO.Error          ( mkIOError, eofErrorType )
 import GHC.Real                 ( Ratio(..) )
@@ -184,29 +185,27 @@ isEOFBin (BinMem _ ix_r sz_r _) = do
 
 writeBinMem :: BinHandle -> FilePath -> IO ()
 writeBinMem (BinMem _ ix_r _ arr_r) fn = do
-  h <- openBinaryFile fn WriteMode
   arr <- readIORef arr_r
   ix  <- readFastMutInt ix_r
-  withForeignPtr arr $ \p -> hPutBuf h p ix
-  hClose h
+  let bs = LZ4.compress $ BS.fromForeignPtr arr 0 ix
+  case bs of
+    Nothing -> error "Binary.writeBinMem: compression failed"
+    Just x  -> B.writeFile fn x
 
 readBinMem :: FilePath -> IO BinHandle
 -- Return a BinHandle with a totally undefined State
 readBinMem filename = do
-  h <- openBinaryFile filename ReadMode
-  filesize' <- hFileSize h
-  let filesize = fromIntegral filesize'
-  arr <- mallocForeignPtrBytes (filesize*2)
-  count <- withForeignPtr arr $ \p -> hGetBuf h p filesize
-  when (count /= filesize) $
-       error ("Binary.readBinMem: only read " ++ show count ++ " bytes")
-  hClose h
-  arr_r <- newIORef arr
-  ix_r <- newFastMutInt
-  writeFastMutInt ix_r 0
-  sz_r <- newFastMutInt
-  writeFastMutInt sz_r filesize
-  return (BinMem noUserData ix_r sz_r arr_r)
+  bs <- liftM LZ4.decompress (B.readFile filename)
+  case bs of
+    Nothing -> error "Binary.readBinMem: decompression failed"
+    Just x  -> do
+      let (arr, ix, size) = BS.toForeignPtr x
+      arr_r <- newIORef arr
+      ix_r <- newFastMutInt
+      writeFastMutInt ix_r ix
+      sz_r <- newFastMutInt
+      writeFastMutInt sz_r size
+      return (BinMem noUserData ix_r sz_r arr_r)
 
 fingerprintBinMem :: BinHandle -> IO Fingerprint
 fingerprintBinMem (BinMem _ ix_r _ arr_r) = do
diff --git a/compiler/utils/LZ4.hs b/compiler/utils/LZ4.hs
new file mode 100644
index 0000000000..337b415227
--- /dev/null
+++ b/compiler/utils/LZ4.hs
@@ -0,0 +1,141 @@
+{-# LANGUAGE CPP #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+-- |
+-- Module      : LZ4
+-- Copyright   : (c) Mark Wotton, Austin Seipp 2012-2015
+-- License     : BSD3
+--
+-- Compression utilities (currently utilizing @LZ4 r127@).
+--
+module LZ4
+       ( compress            -- :: S.ByteString -> S.ByteString
+       , decompress          -- :: S.ByteString -> Maybe S.ByteString
+       ) where
+
+import Prelude hiding (max)
+import Data.Word
+import Data.Bits
+import Foreign.Ptr
+import Foreign.C
+--import Foreign.Marshal.Alloc
+import System.IO.Unsafe (unsafePerformIO)
+#if __GLASGOW_HASKELL__ < 709
+import Control.Applicative
+#endif
+--import Control.Monad
+import Data.Monoid ((<>))
+
+import qualified Data.ByteString as S
+import qualified Data.ByteString.Internal as SI
+import qualified Data.ByteString.Lazy.Builder as B
+import qualified Data.ByteString.Lazy as L
+import qualified Data.ByteString.Unsafe as U
+
+--------------------------------------------------------------------------------
+-- Compression
+
+-- | Compresses the input 'ByteString'.
+--
+-- Will return 'Nothing' if the compression fails. Otherwise, returns
+-- @Just xs@ with the compressed string.
+compress :: S.ByteString -> Maybe S.ByteString
+compress xs | S.null xs = Just S.empty
+compress xs = unsafePerformIO $
+  U.unsafeUseAsCStringLen xs $ \(cstr,len) -> do
+    let len' = fromIntegral len :: CInt
+    let max = c_LZ4_compressBound len'
+    bs <- SI.createAndTrim (fromIntegral max) $ \output ->
+            fromIntegral <$> c_LZ4_compress cstr output len'
+    case (S.null bs) of
+      True -> return Nothing
+      -- Prefix the compressed string with the uncompressed length
+      False -> return $ Just (format (fromIntegral len) bs)
+{-# INLINEABLE compress #-}
+
+-- | Compresses a block of allocated memory and passes the compressed
+-- block on to a continuation.
+{-
+compress :: Ptr a                  -- ^ Memory block
+         -> Int                    -- ^ Size of memory block (in bytes)
+         -> (Ptr a -> Int -> IO a) -- ^ Continuation given compressed memory
+         -> IO ()
+compress ptr count k = do
+  let len' = fromIntegral count :: CInt
+  let max = fromIntegral (c_LZ4_compressBound len')
+  allocaBytes max $ \output -> do
+    sz <- c_LZ4_compress ptr output len'
+    when (sz == 0) $ error "LZ4: compression failed"
+    k output (fromIntegral sz)
+  return ()
+-}
+
+--------------------------------------------------------------------------------
+-- Decompression
+
+-- | Decompress the input 'ByteString'.
+decompress :: S.ByteString -> Maybe S.ByteString
+decompress xs | S.null xs = Just S.empty
+-- Get the length of the uncompressed buffer and do our thing
+decompress xs = maybe Nothing (unsafePerformIO . go) (unformat xs)
+  where
+    go (l, str) =
+      U.unsafeUseAsCString str $ \cstr -> do
+        out <- SI.createAndTrim l $ \p -> do
+          r :: Int <- fromIntegral <$> c_LZ4_uncompress cstr p (fromIntegral l)
+          --- NOTE: r is the count of bytes c_LZ4_uncompress read from
+          --- input buffer, and NOT the count of bytes used in result
+          --- buffer
+          return $! if (r <= 0) then 0 else l
+        return $! if (S.null out) then Nothing else (Just out)
+{-# INLINEABLE decompress #-}
+
+--------------------------------------------------------------------------------
+-- Utilities
+
+-- | Pushes a Word32 and a ByteString into the format we use to correctly
+-- encode/decode.
+format :: Word32 -> S.ByteString -> S.ByteString
+format l xs = L.toStrict $ B.toLazyByteString $
+     B.word32LE l
+  <> B.word32LE (fromIntegral $ S.length xs)
+  <> B.byteString xs
+
+-- | Gets a ByteString and it's length from the compressed format.
+unformat :: S.ByteString -> Maybe (Int, S.ByteString)
+unformat xs
+  | S.length xs < 8        = Nothing -- Need at least 8 bytes
+  | bsLen /= S.length rest = Nothing -- Header doesn't match real size
+  | otherwise              = Just (fromIntegral origLen, rest)
+  where
+    origLen = fromIntegral (read32LE l0 l1 l2 l3) :: Int
+    bsLen   = fromIntegral (read32LE s0 s1 s2 s3) :: Int
+
+    [l0,l1,l2,l3] = S.unpack (S.take 4 xs)
+    [s0,s1,s2,s3] = S.unpack (S.take 4 $ S.drop 4 xs)
+    rest = S.drop 8 xs
+
+read32LE :: Word8 -> Word8 -> Word8 -> Word8 -> Word32
+read32LE x0 x1 x2 x3
+  = fi x0 + (fi x1 `shiftL` 8) + (fi x2 `shiftL` 16) + (fi x3 `shiftL` 24)
+  where fi = fromIntegral :: Word8 -> Word32
+
+--------------------------------------------------------------------------------
+-- FFI Bindings
+
+-- | Worst case compression bounds on an input string.
+foreign import ccall unsafe "LZ4_compressBound"
+  c_LZ4_compressBound :: CInt -> CInt
+
+-- | Compresses a string.
+foreign import ccall unsafe "LZ4_compress"
+  c_LZ4_compress :: Ptr a   -- ^ Source
+                 -> Ptr b   -- ^ Dest
+                 -> CInt    -- ^ Input size
+                 -> IO CInt -- ^ Result
+
+-- | Decompresses a string.
+foreign import ccall unsafe "LZ4_decompress_fast"
+  c_LZ4_uncompress :: Ptr a   -- ^ Source
+                   -> Ptr b   -- ^ Dest
+                   -> CInt    -- ^ Size of ORIGINAL INPUT
+                   -> IO CInt -- ^ Result