ddidderr/zstd-rs
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

feat(compress): move block size estimation to Rust

Browse Source 
Block splitting still estimated literal and sequence section sizes through
three C helpers. Move the pure estimator into zstd_compress_stats.rs so its
histogram, HUF, and FSE cost calculations use the Rust implementations already
behind the compression ABI. Preserve the original literal headers, multi-stream
Huffman surcharge, sequence additional-bit costs, entropy-header accounting,
and fallback estimates; leave C responsible for split recursion and table
ownership.

Test Plan:
- `cargo test --manifest-path rust/Cargo.toml --no-default-features --features compression` -- 229 passed
- `cargo clippy --manifest-path rust/Cargo.toml --no-default-features --features compression` plus benches/tests -- passed before and after formatting
- `cargo +nightly fmt --manifest-path rust/Cargo.toml --all` -- passed
- `make -B -C lib -j2 lib` -- passed
- `make -C tests -j2 test-zstream` -- passed, including 84 deterministic and 15,375 randomized cases
This commit is contained in:
ddidderr
2026-07-18 07:08:27 +02:00
parent d334aaafa3
commit 511882e213
2 changed files with 341 additions and 109 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/compress/zstd_compress.c
+16 -102
View File
@@ -257,6 +257,14 @@ size_t ZSTD_rust_buildBlockEntropyStats(
int strategy, int disableLiteralCompression,
ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize);
size_t ZSTD_rust_estimateBlockSize(
const BYTE* literals, size_t litSize,
const BYTE* ofCodeTable, const BYTE* llCodeTable,
const BYTE* mlCodeTable, size_t nbSeq,
const ZSTD_entropyCTables_t* entropy,
const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
void* workspace, size_t wkspSize,
int writeLitEntropy, int writeSeqEntropy);
size_t ZSTD_rust_copyBlockSequences(
SeqCollector* seqCollector, const SeqStore_t* seqStore,
const U32 prevRepcodes[ZSTD_REP_NUM]);
@@ -2600,101 +2608,8 @@ size_t ZSTD_buildBlockEntropyStats(
workspace, wkspSize);
}
/* Returns the size estimate for the literals section (header + content) of a block */
static size_t
ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
const ZSTD_hufCTables_t* huf,
const ZSTD_hufCTablesMetadata_t* hufMetadata,
void* workspace, size_t wkspSize,
int writeEntropy)
{
unsigned* const countWksp = (unsigned*)workspace;
unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
size_t literalSectionHeaderSize = 3 + (litSize >= 1 KB) + (litSize >= 16 KB);
U32 singleStream = litSize < 256;
if (hufMetadata->hType == set_basic) return litSize;
else if (hufMetadata->hType == set_rle) return 1;
else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
if (ZSTD_isError(largest)) return litSize;
{ size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
if (!singleStream) cLitSizeEstimate += 6; /* multi-stream huffman uses 6-byte jump table */
return cLitSizeEstimate + literalSectionHeaderSize;
} }
assert(0); /* impossible */
return 0;
}
/* Returns the size estimate for the FSE-compressed symbols (of, ml, ll) of a block */
static size_t
ZSTD_estimateBlockSize_symbolType(SymbolEncodingType_e type,
const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
const FSE_CTable* fseCTable,
const U8* additionalBits,
short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
void* workspace, size_t wkspSize)
{
unsigned* const countWksp = (unsigned*)workspace;
const BYTE* ctp = codeTable;
const BYTE* const ctStart = ctp;
const BYTE* const ctEnd = ctStart + nbSeq;
size_t cSymbolTypeSizeEstimateInBits = 0;
unsigned max = maxCode;
HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize); /* can't fail */
if (type == set_basic) {
/* We selected this encoding type, so it must be valid. */
assert(max <= defaultMax);
(void)defaultMax;
cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
} else if (type == set_rle) {
cSymbolTypeSizeEstimateInBits = 0;
} else if (type == set_compressed || type == set_repeat) {
cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
}
if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) {
return nbSeq * 10;
}
while (ctp < ctEnd) {
if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
ctp++;
}
return cSymbolTypeSizeEstimateInBits >> 3;
}
/* Returns the size estimate for the sequences section (header + content) of a block */
static size_t
ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
const BYTE* llCodeTable,
const BYTE* mlCodeTable,
size_t nbSeq,
const ZSTD_fseCTables_t* fseTables,
const ZSTD_fseCTablesMetadata_t* fseMetadata,
void* workspace, size_t wkspSize,
int writeEntropy)
{
size_t sequencesSectionHeaderSize = 1 /* seqHead */ + 1 /* min seqSize size */ + (nbSeq >= 128) + (nbSeq >= LONGNBSEQ);
size_t cSeqSizeEstimate = 0;
cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, nbSeq, MaxOff,
fseTables->offcodeCTable, NULL,
OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->llType, llCodeTable, nbSeq, MaxLL,
fseTables->litlengthCTable, LL_bits,
LL_defaultNorm, LL_defaultNormLog, MaxLL,
workspace, wkspSize);
cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, nbSeq, MaxML,
fseTables->matchlengthCTable, ML_bits,
ML_defaultNorm, ML_defaultNormLog, MaxML,
workspace, wkspSize);
if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
return cSeqSizeEstimate + sequencesSectionHeaderSize;
}
/* Returns the size estimate for a given stream of literals, of, ll, ml */
/* The block-size estimator is implemented in Rust; C retains the stateful
* block-splitting recursion and the entropy-table ownership. */
static size_t
ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
const BYTE* ofCodeTable,
@@ -2706,13 +2621,12 @@ ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
void* workspace, size_t wkspSize,
int writeLitEntropy, int writeSeqEntropy)
{
size_t const literalsSize = ZSTD_estimateBlockSize_literal(literals, litSize,
&entropy->huf, &entropyMetadata->hufMetadata,
workspace, wkspSize, writeLitEntropy);
size_t const seqSize = ZSTD_estimateBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
workspace, wkspSize, writeSeqEntropy);
return seqSize + literalsSize + ZSTD_blockHeaderSize;
return ZSTD_rust_estimateBlockSize(
literals, litSize,
ofCodeTable, llCodeTable, mlCodeTable, nbSeq,
entropy, entropyMetadata,
workspace, wkspSize,
writeLitEntropy, writeSeqEntropy);
}
/* Builds entropy statistics and uses them for blocksize estimation.