Reduce size of dctx by reutilizing dst buffer (#2751)

* Reduce size of dctx by reutilizing dst buffer

Co-authored-by: Binh Vo <binhvo@fb.com>

lib/README.md

@@ -155,6 +155,12 @@ The file structure is designed to make this selection manually achievable for an
 - The build macro `ZSTD_NO_INTRINSICS` can be defined to disable all explicit intrinsics.
   Compiler builtins are still used.
 
+- The build macro `ZSTD_LITBUFFEREXTRASIZE` can be set to control the amount of extra memory used
+  during decompression to store literals.  This defaults to 64kB.  Reducing it can reduce the
+  memory footprint required for decompression by increasing the portion of the literal buffer that
+  is stored in the unwritten portion of the dst buffer, at the cost of performance impact for
+  decompression.
+
 
 #### Windows : using MinGW+MSYS to create DLL
 

lib/common/zstd_internal.h

@@ -181,7 +181,7 @@ static void ZSTD_copy16(void* dst, const void* src) {
 #if defined(ZSTD_ARCH_ARM_NEON)
     vst1q_u8((uint8_t*)dst, vld1q_u8((const uint8_t*)src));
 #else
-    ZSTD_memcpy(dst, src, 16);
+    ZSTD_memmove(dst, src, 16);
 #endif
 }
 #define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
@@ -210,8 +210,6 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
     BYTE* op = (BYTE*)dst;
     BYTE* const oend = op + length;
 
-    assert(diff >= 8 || (ovtype == ZSTD_no_overlap && diff <= -WILDCOPY_VECLEN));
-
     if (ovtype == ZSTD_overlap_src_before_dst && diff < WILDCOPY_VECLEN) {
         /* Handle short offset copies. */
         do {

lib/decompress/zstd_decompress.c

@@ -916,7 +916,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         switch(blockProperties.blockType)
         {
         case bt_compressed:
-            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1);
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1, not_streaming);
             break;
         case bt_raw :
             decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
@@ -1229,7 +1229,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             {
             case bt_compressed:
                 DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
-                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1);
+                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1, is_streaming);
                 dctx->expected = 0;  /* Streaming not supported */
                 break;
             case bt_raw :
@@ -1824,7 +1824,8 @@ size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
 size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
 {
     size_t const blockSize = (size_t) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
-    unsigned long long const neededRBSize = windowSize + blockSize + (WILDCOPY_OVERLENGTH * 2);
+    /* space is needed to store the litbuffer after the output of a given block without stomping the extDict of a previous run, as well as to cover both windows against wildcopy*/
+    unsigned long long const neededRBSize = windowSize + blockSize + ZSTD_BLOCKSIZE_MAX + (WILDCOPY_OVERLENGTH * 2);
     unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
     size_t const minRBSize = (size_t) neededSize;
     RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,

lib/decompress/zstd_decompress_block.h

@@ -33,6 +33,12 @@
  */
 
 
+ /* Streaming state is used to inform allocation of the literal buffer */
+typedef enum {
+    not_streaming = 0,
+    is_streaming = 1
+} streaming_operation;
+
 /* ZSTD_decompressBlock_internal() :
  * decompress block, starting at `src`,
  * into destination buffer `dst`.
@@ -41,7 +47,7 @@
  */
 size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
                                void* dst, size_t dstCapacity,
-                         const void* src, size_t srcSize, const int frame);
+                         const void* src, size_t srcSize, const int frame, const streaming_operation streaming);
 
 /* ZSTD_buildFSETable() :
  * generate FSE decoding table for one symbol (ll, ml or off)

lib/decompress/zstd_decompress_internal.h

@@ -106,6 +106,16 @@ typedef struct {
     size_t ddictPtrCount;
 } ZSTD_DDictHashSet;
 
+#ifndef ZSTD_LITBUFFEREXTRASIZE
+#define ZSTD_LITBUFFEREXTRASIZE    (1 << 16) /* extra buffer reduces amount of dst required to store litBuffer */
+#endif
+
+typedef enum {
+    ZSTD_not_in_dst = 0,  /* Stored entirely within litExtraBuffer */
+    ZSTD_in_dst = 1,           /* Stored entirely within dst (in memory after current output write) */
+    ZSTD_split = 2            /* Split between litExtraBuffer and dst */
+} ZSTD_litLocation_e;
+
 struct ZSTD_DCtx_s
 {
     const ZSTD_seqSymbol* LLTptr;
@@ -171,7 +181,10 @@ struct ZSTD_DCtx_s
     ZSTD_outBuffer expectedOutBuffer;
 
     /* workspace */
-    BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
+    BYTE* litBuffer;
+    const BYTE* litBufferEnd;
+    ZSTD_litLocation_e litBufferLocation;
+    BYTE litExtraBuffer[ZSTD_LITBUFFEREXTRASIZE + WILDCOPY_OVERLENGTH]; /* literal buffer can be split between storage within dst and within this scratch buffer */
     BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
 
     size_t oversizedDuration;

tests/fullbench.c

@@ -123,11 +123,11 @@ static size_t local_ZSTD_decompress(const void* src, size_t srcSize,
 static ZSTD_DCtx* g_zdc = NULL;
 
 #ifndef ZSTD_DLL_IMPORT
-extern size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* ctx, const void* src, size_t srcSize);
+extern size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* ctx, const void* src, size_t srcSize, void* dst, size_t dstCapacity);
 static size_t local_ZSTD_decodeLiteralsBlock(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2)
 {
     (void)src; (void)srcSize; (void)dst; (void)dstSize;
-    return ZSTD_decodeLiteralsBlock(g_zdc, buff2, g_cSize);
+    return ZSTD_decodeLiteralsBlock(g_zdc, buff2, g_cSize, dst, dstSize);
 }
 
 static size_t local_ZSTD_decodeSeqHeaders(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2)
@@ -577,7 +577,7 @@ static int benchMem(unsigned benchNb,
             ip += ZSTD_blockHeaderSize;    /* skip block header */
             ZSTD_decompressBegin(g_zdc);
             CONTROL(iend > ip);
-            ip += ZSTD_decodeLiteralsBlock(g_zdc, ip, (size_t)(iend-ip));   /* skip literal segment */
+            ip += ZSTD_decodeLiteralsBlock(g_zdc, ip, (size_t)(iend-ip), dstBuff, dstBuffSize);   /* skip literal segment */
             g_cSize = (size_t)(iend-ip);
             memcpy(dstBuff2, ip, g_cSize);   /* copy rest of block (it starts by SeqHeader) */
             srcSize = srcSize > 128 KB ? 128 KB : srcSize;   /* speed relative to block */