Skip to content

Commit

Permalink
Reinforced Teddy with 1-byte approach, based on "shift-or" and AVX2.
Browse files Browse the repository at this point in the history
  • Loading branch information
@fatchanghao
fatchanghao authored and Matthew Barr committed Aug 21, 2017
1 parent b09e3ac commit dbd3f66
Show file tree
Hide file tree
Showing 10 changed files with 1,069 additions and 1,232 deletions.
6 changes: 1 addition & 5 deletions src/fdr/fdr_confirm.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -78,12 +78,8 @@ struct LitInfo {
struct FDRConfirm {
CONF_TYPE andmsk;
CONF_TYPE mult;
u32 nBitsOrSoleID; // if flags is NO_CONFIRM then this is soleID
u32 flags; // sole meaning is 'non-zero means no-confirm' (that is all)
u32 nBits;
hwlm_group_t groups;
u32 soleLitSize;
u32 soleLitCmp;
u32 soleLitMsk;
};

static really_inline
Expand Down
50 changes: 3 additions & 47 deletions src/fdr/fdr_confirm_compile.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -130,7 +130,7 @@ void fillLitInfo(const vector<hwlmLiteral> &lits, vector<LitInfo> &tmpLitInfo,

static
bytecode_ptr<FDRConfirm> getFDRConfirm(const vector<hwlmLiteral> &lits,
bool make_small, bool make_confirm) {
bool make_small) {
// Every literal must fit within CONF_TYPE.
assert(all_of_in(lits, [](const hwlmLiteral &lit) {
return lit.s.size() <= sizeof(CONF_TYPE);
Expand All @@ -153,42 +153,6 @@ bytecode_ptr<FDRConfirm> getFDRConfirm(const vector<hwlmLiteral> &lits,
}

CONF_TYPE mult = (CONF_TYPE)0x0b4e0ef37bc32127ULL;
u32 flags = 0;
// we use next three variables for 'confirmless' case to speed-up
// confirmation process
u32 soleLitSize = 0;
u32 soleLitCmp = 0;
u32 soleLitMsk = 0;

if (!make_confirm) {
flags = FDRC_FLAG_NO_CONFIRM;
if (lits[0].noruns) {
// messy - need to clean this up later as flags is sorta kinda
// obsoleted
flags |= FDRC_FLAG_NOREPEAT;
}
mult = 0;
soleLitSize = lits[0].s.size() - 1;
// we can get to this point only in confirmless case;
// it means that we have only one literal per FDRConfirm (no packing),
// with no literal mask and size of literal is less or equal
// to the number of masks of Teddy engine;
// maximum number of masks for Teddy is 4, so the size of
// literal is definitely less or equal to size of u32
assert(lits[0].s.size() <= sizeof(u32));
for (u32 i = 0; i < lits[0].s.size(); i++) {
u32 shiftLoc = (sizeof(u32) - i - 1) * 8;
u8 c = lits[0].s[lits[0].s.size() - i - 1];
if (lits[0].nocase && ourisalpha(c)) {
soleLitCmp |= (u32)(c & CASE_CLEAR) << shiftLoc;
soleLitMsk |= (u32)CASE_CLEAR << shiftLoc;
}
else {
soleLitCmp |= (u32)c << shiftLoc;
soleLitMsk |= (u32)0xff << shiftLoc;
}
}
}

// we can walk the vector and assign elements from the vectors to a
// map by hash value
Expand Down Expand Up @@ -276,11 +240,7 @@ bytecode_ptr<FDRConfirm> getFDRConfirm(const vector<hwlmLiteral> &lits,

fdrc->andmsk = andmsk;
fdrc->mult = mult;
fdrc->nBitsOrSoleID = (flags & FDRC_FLAG_NO_CONFIRM) ? lits[0].id : nBits;
fdrc->flags = flags;
fdrc->soleLitSize = soleLitSize;
fdrc->soleLitCmp = soleLitCmp;
fdrc->soleLitMsk = soleLitMsk;
fdrc->nBits = nBits;

fdrc->groups = gm;

Expand Down Expand Up @@ -334,12 +294,8 @@ setupFullConfs(const vector<hwlmLiteral> &lits,
const EngineDescription &eng,
map<BucketIndex, vector<LiteralIndex>> &bucketToLits,
bool make_small) {
bool makeConfirm = true;
unique_ptr<TeddyEngineDescription> teddyDescr =
getTeddyDescription(eng.getID());
if (teddyDescr) {
makeConfirm = teddyDescr->needConfirm(lits);
}

BC2CONF bc2Conf;
u32 totalConfirmSize = 0;
Expand All @@ -351,7 +307,7 @@ setupFullConfs(const vector<hwlmLiteral> &lits,
}

DEBUG_PRINTF("b %d sz %zu\n", b, vl.size());
auto fc = getFDRConfirm(vl, make_small, makeConfirm);
auto fc = getFDRConfirm(vl, make_small);
totalConfirmSize += fc.size();
bc2Conf.emplace(b, move(fc));
}
Expand Down
79 changes: 2 additions & 77 deletions src/fdr/fdr_confirm_runtime.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,11 +43,12 @@ void confWithBit(const struct FDRConfirm *fdrc, const struct FDR_Runtime_Args *a
size_t i, hwlmcb_rv_t *control, u32 *last_match,
u64a conf_key) {
assert(i < a->len);
assert(i >= a->start_offset);
assert(ISALIGNED(fdrc));

const u8 * buf = a->buf;
u32 c = CONF_HASH_CALL(conf_key, fdrc->andmsk, fdrc->mult,
fdrc->nBitsOrSoleID);
fdrc->nBits);
u32 start = getConfirmLitIndex(fdrc)[c];
if (likely(!start)) {
return;
Expand Down Expand Up @@ -94,80 +95,4 @@ void confWithBit(const struct FDRConfirm *fdrc, const struct FDR_Runtime_Args *a
} while (oldNext);
}

// 'light-weight' confirmation function which is used by 1-mask Teddy;
// in the 'confirmless' case it simply calls callback function,
// otherwise it calls 'confWithBit' function for the full confirmation procedure
static really_inline
void confWithBit1(const struct FDRConfirm *fdrc,
const struct FDR_Runtime_Args *a, size_t i,
hwlmcb_rv_t *control, u32 *last_match, u64a conf_key) {
assert(i < a->len);
assert(ISALIGNED(fdrc));

if (unlikely(fdrc->mult)) {
confWithBit(fdrc, a, i, control, last_match, conf_key);
return;
} else {
u32 id = fdrc->nBitsOrSoleID;

if ((*last_match == id) && (fdrc->flags & FDRC_FLAG_NOREPEAT)) {
return;
}
*last_match = id;
*control = a->cb(i, i, id, a->ctxt);
}
}

// This is 'light-weight' confirmation function which is used by 2-3-4-mask Teddy
// In the 'confirmless' case it makes fast 32-bit comparison,
// otherwise it calls 'confWithBit' function for the full confirmation procedure
static really_inline
void confWithBitMany(const struct FDRConfirm *fdrc,
const struct FDR_Runtime_Args *a, size_t i, CautionReason r,
hwlmcb_rv_t *control, u32 *last_match, u64a conf_key) {
assert(i < a->len);
assert(ISALIGNED(fdrc));

if (i < a->start_offset) {
return;
}

if (unlikely(fdrc->mult)) {
confWithBit(fdrc, a, i, control, last_match, conf_key);
return;
} else {
const u32 id = fdrc->nBitsOrSoleID;
const u32 len = fdrc->soleLitSize;

if ((*last_match == id) && (fdrc->flags & FDRC_FLAG_NOREPEAT)) {
return;
}

if (r == VECTORING && len > i - a->start_offset) {
if (len > i + a->len_history) {
return;
}

u32 cmp = (u32)a->buf[i] << 24;

if (len <= i) {
for (u32 j = 1; j <= len; j++) {
cmp |= (u32)a->buf[i - j] << (24 - (j * 8));
}
} else {
for (u32 j = 1; j <= i; j++) {
cmp |= (u32)a->buf[i - j] << (24 - (j * 8));
}
cmp |= (u32)(a->histBytes >> (40 + i * 8));
}

if ((fdrc->soleLitMsk & cmp) != fdrc->soleLitCmp) {
return;
}
}
*last_match = id;
*control = a->cb(i - len, i, id, a->ctxt);
}
}

#endif
Loading

0 comments on commit dbd3f66

Please sign in to comment.