Skip to content

Commit

Permalink
[HUF] Improve Huffman encoding speed
Browse files Browse the repository at this point in the history
Improve Huffman encoding speed by 20% for gcc and 10% for clang.

| Compiler |     Benchmark     | Config  |   Dataset   | Ratio | Speed MB/s (dev) | Speed MB/s (huf-cspeed) | Speed MB/s (huf-cspeed - dev) |
|----------|-------------------|---------|-------------|-------|------------------|-------------------------|-------------------------------|
| gcc      | compress          | level_1 | enwik7      | 2.43  | 253.70           | 258.72                  | 2.0%                          |
| gcc      | compress          | level_1 | silesia     | 2.88  | 341.90           | 348.15                  | 1.8%                          |
| gcc      | compress_literals | level_1 | enwik7      | 1.49  | 761.83           | 912.76                  | 19.8%                         |
| gcc      | compress_literals | level_1 | silesia     | 1.28  | 754.83           | 902.37                  | 19.5%                         |
| gcc      | compress_literals | level_7 | enwik7      | 1.29  | 502.81           | 552.79                  | 9.9%                          |
| gcc      | compress_literals | level_7 | silesia     | 1.11  | 675.97           | 776.44                  | 14.9%                         |
| clang    | compress          | level_1 | enwik7      | 2.43  | 277.54           | 280.98                  | 1.2%                          |
| clang    | compress          | level_1 | silesia     | 2.88  | 369.98           | 375.46                  | 1.5%                          |
| clang    | compress_literals | level_1 | enwik7      | 1.49  | 828.83           | 918.41                  | 10.8%                         |
| clang    | compress_literals | level_1 | silesia     | 1.28  | 815.81           | 905.41                  | 11.0%                         |
| clang    | compress_literals | level_7 | enwik7      | 1.29  | 533.13           | 553.30                  | 3.8%                          |
| clang    | compress_literals | level_7 | silesia     | 1.11  | 714.52           | 775.38                  | 8.5%                          |
  • Loading branch information
terrelln committed Jul 27, 2021
1 parent b3e372c commit 46f2710
Show file tree
Hide file tree
Showing 5 changed files with 381 additions and 87 deletions.
27 changes: 11 additions & 16 deletions lib/common/huf.h
Original file line number Diff line number Diff line change
Expand Up @@ -89,9 +89,9 @@ HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity,

/** HUF_compress4X_wksp() :
* Same as HUF_compress2(), but uses externally allocated `workSpace`.
* `workspace` must have minimum alignment of 4, and be at least as large as HUF_WORKSPACE_SIZE */
#define HUF_WORKSPACE_SIZE ((6 << 10) + 256)
#define HUF_WORKSPACE_SIZE_U32 (HUF_WORKSPACE_SIZE / sizeof(U32))
* `workspace` must be at least as large as HUF_WORKSPACE_SIZE */
#define HUF_WORKSPACE_SIZE ((8 << 10) + 256)
#define HUF_WORKSPACE_SIZE_U64 (HUF_WORKSPACE_SIZE / sizeof(U64))
HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
const void* src, size_t srcSize,
unsigned maxSymbolValue, unsigned tableLog,
Expand Down Expand Up @@ -136,15 +136,11 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,

/* static allocation of HUF's Compression Table */
/* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
struct HUF_CElt_s {
U16 val;
BYTE nbBits;
}; /* typedef'd to HUF_CElt */
typedef struct HUF_CElt_s HUF_CElt; /* consider it an incomplete type */
#define HUF_CTABLE_SIZE_U32(maxSymbolValue) ((maxSymbolValue)+1) /* Use tables of U32, for proper alignment */
#define HUF_CTABLE_SIZE(maxSymbolValue) (HUF_CTABLE_SIZE_U32(maxSymbolValue) * sizeof(U32))
typedef size_t HUF_CElt; /* consider it an incomplete type */
#define HUF_CTABLE_SIZE_ST(maxSymbolValue) ((maxSymbolValue)+2) /* Use tables of size_t, for proper alignment */
#define HUF_CTABLE_SIZE(maxSymbolValue) (HUF_CTABLE_SIZE_ST(maxSymbolValue) * sizeof(size_t))
#define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
HUF_CElt name[HUF_CTABLE_SIZE_U32(maxSymbolValue)] /* no final ; */
HUF_CElt name[HUF_CTABLE_SIZE_ST(maxSymbolValue)] /* no final ; */

/* static allocation of HUF's DTable */
typedef U32 HUF_DTable;
Expand Down Expand Up @@ -250,11 +246,10 @@ size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
* Loading a CTable saved with HUF_writeCTable() */
size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);

/** HUF_getNbBits() :
/** HUF_getNbBitsFromCTable() :
* Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
* Note 1 : is not inlined, as HUF_CElt definition is private
* Note 2 : const void* used, so that it can provide a statically allocated table as argument (which uses type U32) */
U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue);
* Note 1 : is not inlined, as HUF_CElt definition is private */
U32 HUF_getNbBitsFromCTable(const HUF_CElt* symbolTable, U32 symbolValue);

/*
* HUF_decompress() does the following:
Expand Down Expand Up @@ -306,7 +301,7 @@ size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* c
/* ====================== */

size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize); /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize); /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U64 U64 */
size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
/** HUF_compress1X_repeat() :
* Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
Expand Down
Loading

0 comments on commit 46f2710

Please sign in to comment.