Skip to content

[dfast] Use ZSTD_selectAddr() for prefix checks in dictMatchState #4866

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jeffchen006 wants to merge 1 commit into
base: dev
Choose a base branch
from
+20 −9
Open

[dfast] Use ZSTD_selectAddr() for prefix checks in dictMatchState #4866

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 20 additions & 9 deletions lib/compress/zstd_double_fast.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -359,6 +359,10 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
const U32 dictHBitsL = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
const U32 dictHBitsS = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
/* Array of ~random data, should have low probability of matching data.
* We load from here instead of from the prefix, if a prefix match index
* is invalid. Used to avoid unpredictable branches. */
const BYTE dummy[] = {0x12,0x34,0x56,0x78,0x9a,0xbc,0xde,0xf0,0xe2,0xb4};

DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");

Expand Down Expand Up @@ -413,7 +417,13 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
goto _match_stored;
}

if ((matchIndexL >= prefixLowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
/* matchIndexL >= prefixLowestIndex is an unpredictable branch while the
* prefix is small, e.g. for small inputs with an attached dictionary.
* Read from a safe address instead, and test the index only when the
* bytes match. That is rare, so both branches are well predicted.
* The result is the same as testing the index first. */
if ( (MEM_read64(ZSTD_selectAddr(matchIndexL, prefixLowestIndex, matchLong, &dummy[0])) == MEM_read64(ip))
&& (matchIndexL >= prefixLowestIndex) ) {
/* check prefix long match */
mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
offset = (U32)(ip-matchLong);
Expand All @@ -432,12 +442,12 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
goto _match_found;
} }

if (matchIndexS > prefixLowestIndex) {
/* short match candidate */
if (MEM_read32(match) == MEM_read32(ip)) {
goto _search_next_long;
}
} else if (dictTagsMatchS) {
/* short match candidate, same method as above */
if ( (MEM_read32(ZSTD_selectAddr(matchIndexS, prefixLowestIndex, match, &dummy[0])) == MEM_read32(ip))
&& (matchIndexS > prefixLowestIndex) ) {
goto _search_next_long;
}
if (dictTagsMatchS && (matchIndexS <= prefixLowestIndex)) {
/* check dictMatchState short match */
U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
match = dictBase + dictMatchIndexS;
Expand All @@ -462,8 +472,9 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
const BYTE* matchL3 = base + matchIndexL3;
hashLong[hl3] = curr + 1;

/* check prefix long +1 match */
if ((matchIndexL3 >= prefixLowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1))) {
/* check prefix long +1 match, same method as above */
if ( (MEM_read64(ZSTD_selectAddr(matchIndexL3, prefixLowestIndex, matchL3, &dummy[0])) == MEM_read64(ip+1))
&& (matchIndexL3 >= prefixLowestIndex) ) {
mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
ip++;
offset = (U32)(ip-matchL3);
Expand Down