DISPATCH_CASE(TAMARAMA_NFA, Tamarama, dbnt_func); \
DISPATCH_CASE(MCSHENG_NFA_8, McSheng8, dbnt_func); \
DISPATCH_CASE(MCSHENG_NFA_16, McSheng16, dbnt_func); \
+ DISPATCH_CASE(SHENG_NFA_32, Sheng32, dbnt_func); \
default: \
assert(0); \
}
const char *NFATraits<MCSHENG_NFA_16>::name = "Shengy McShengFace 16";
#endif
+template<> struct NFATraits<SHENG_NFA_32> {
+ UNUSED static const char *name;
+ static const NFACategory category = NFA_OTHER;
+ static const u32 stateAlign = 1;
+ static const bool fast = true;
+ static const nfa_dispatch_fn has_accel;
+ static const nfa_dispatch_fn has_repeats;
+ static const nfa_dispatch_fn has_repeats_other_than_firsts;
+};
+const nfa_dispatch_fn NFATraits<SHENG_NFA_32>::has_accel = has_accel_sheng;
+const nfa_dispatch_fn NFATraits<SHENG_NFA_32>::has_repeats = dispatch_false;
+const nfa_dispatch_fn NFATraits<SHENG_NFA_32>::has_repeats_other_than_firsts = dispatch_false;
+#if defined(DUMP_SUPPORT)
+const char *NFATraits<SHENG_NFA_32>::name = "Sheng 32";
+#endif
+
} // namespace
#if defined(DUMP_SUPPORT)
DISPATCH_CASE(TAMARAMA_NFA, Tamarama, dbnt_func); \
DISPATCH_CASE(MCSHENG_NFA_8, McSheng8, dbnt_func); \
DISPATCH_CASE(MCSHENG_NFA_16, McSheng16, dbnt_func); \
+ DISPATCH_CASE(SHENG_NFA_32, Sheng32, dbnt_func); \
default: \
assert(0); \
}
TAMARAMA_NFA, /**< magic nfa container */
MCSHENG_NFA_8, /**< magic pseudo nfa */
MCSHENG_NFA_16, /**< magic pseudo nfa */
+ SHENG_NFA_32, /**< magic pseudo nfa */
/** \brief bogus NFA - not used */
INVALID_NFA
};
}
/** \brief True if the given type (from NFA::type) is a Sheng DFA. */
+static really_inline int isSheng16Type(u8 t) {
+ return t == SHENG_NFA;
+}
+
+#if defined(HAVE_AVX512VBMI)
+/** \brief True if the given type (from NFA::type) is a Sheng32 DFA. */
+static really_inline int isSheng32Type(u8 t) {
+ return t == SHENG_NFA_32;
+}
+#endif
+
+/** \brief True if the given type (from NFA::type) is a Sheng/Sheng32 DFA. */
static really_inline int isShengType(u8 t) {
+#if defined(HAVE_AVX512VBMI)
+ return t == SHENG_NFA || t == SHENG_NFA_32;
+#else
return t == SHENG_NFA;
+#endif
}
/**
return MO_CONTINUE_MATCHING; /* continue execution */
}
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+const struct sheng32 *get_sheng32(const struct NFA *n) {
+ return (const struct sheng32 *)getImplNfa(n);
+}
+
+static really_inline
+const struct sstate_aux *get_aux32(const struct sheng32 *sh, u8 id) {
+ u32 offset = sh->aux_offset - sizeof(struct NFA) +
+ (id & SHENG32_STATE_MASK) * sizeof(struct sstate_aux);
+ DEBUG_PRINTF("Getting aux for state %u at offset %llu\n",
+ id & SHENG32_STATE_MASK, (u64a)offset + sizeof(struct NFA));
+ return (const struct sstate_aux *)((const char *) sh + offset);
+}
+
+static really_inline
+const union AccelAux *get_accel32(const struct sheng32 *sh, u8 id) {
+ const struct sstate_aux *saux = get_aux32(sh, id);
+ DEBUG_PRINTF("Getting accel aux at offset %u\n", saux->accel);
+ const union AccelAux *aux = (const union AccelAux *)
+ ((const char *)sh + saux->accel - sizeof(struct NFA));
+ return aux;
+}
+
+static really_inline
+const struct report_list *get_rl32(const struct sheng32 *sh,
+ const struct sstate_aux *aux) {
+ DEBUG_PRINTF("Getting report list at offset %u\n", aux->accept);
+ return (const struct report_list *)
+ ((const char *)sh + aux->accept - sizeof(struct NFA));
+}
+
+static really_inline
+const struct report_list *get_eod_rl32(const struct sheng32 *sh,
+ const struct sstate_aux *aux) {
+ DEBUG_PRINTF("Getting EOD report list at offset %u\n", aux->accept);
+ return (const struct report_list *)
+ ((const char *)sh + aux->accept_eod - sizeof(struct NFA));
+}
+
+static really_inline
+char sheng32HasAccept(const struct sheng32 *sh, const struct sstate_aux *aux,
+ ReportID report) {
+ assert(sh && aux);
+
+ const struct report_list *rl = get_rl32(sh, aux);
+ assert(ISALIGNED_N(rl, 4));
+
+ DEBUG_PRINTF("report list has %u entries\n", rl->count);
+
+ for (u32 i = 0; i < rl->count; i++) {
+ if (rl->report[i] == report) {
+ DEBUG_PRINTF("reporting %u\n", rl->report[i]);
+ return 1;
+ }
+ }
+
+ return 0;
+}
+
+static really_inline
+char fireReports32(const struct sheng32 *sh, NfaCallback cb, void *ctxt,
+ const u8 state, u64a loc, u8 *const cached_accept_state,
+ ReportID *const cached_accept_id, char eod) {
+ DEBUG_PRINTF("reporting matches @ %llu\n", loc);
+
+ if (!eod && state == *cached_accept_state) {
+ DEBUG_PRINTF("reporting %u\n", *cached_accept_id);
+ if (cb(0, loc, *cached_accept_id, ctxt) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING; /* termination requested */
+ }
+
+ return MO_CONTINUE_MATCHING; /* continue execution */
+ }
+ const struct sstate_aux *aux = get_aux32(sh, state);
+ const struct report_list *rl = eod ? get_eod_rl32(sh, aux) :
+ get_rl32(sh, aux);
+ assert(ISALIGNED(rl));
+
+ DEBUG_PRINTF("report list has %u entries\n", rl->count);
+ u32 count = rl->count;
+
+ if (!eod && count == 1) {
+ *cached_accept_state = state;
+ *cached_accept_id = rl->report[0];
+
+ DEBUG_PRINTF("reporting %u\n", rl->report[0]);
+ if (cb(0, loc, rl->report[0], ctxt) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING; /* termination requested */
+ }
+
+ return MO_CONTINUE_MATCHING; /* continue execution */
+ }
+
+ for (u32 i = 0; i < count; i++) {
+ DEBUG_PRINTF("reporting %u\n", rl->report[i]);
+ if (cb(0, loc, rl->report[i], ctxt) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING; /* termination requested */
+ }
+ }
+ return MO_CONTINUE_MATCHING; /* continue execution */
+}
+#endif // end of HAVE_AVX512VBMI
+
/* include Sheng function definitions */
#include "sheng_defs.h"
*(u8 *)dest = *(const u8 *)src;
return 0;
}
+
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+char runSheng32Cb(const struct sheng32 *sh, NfaCallback cb, void *ctxt,
+ u64a offset, u8 *const cached_accept_state,
+ ReportID *const cached_accept_id, const u8 *cur_buf,
+ const u8 *start, const u8 *end, u8 can_die,
+ u8 has_accel, u8 single, const u8 **scanned, u8 *state) {
+ DEBUG_PRINTF("Scanning %llu bytes (offset %llu) in callback mode\n",
+ (u64a)(end - start), offset);
+ DEBUG_PRINTF("start: %lli end: %lli\n", (s64a)(start - cur_buf),
+ (s64a)(end - cur_buf));
+ DEBUG_PRINTF("can die: %u has accel: %u single: %u\n", !!can_die,
+ !!has_accel, !!single);
+ int rv;
+ /* scan and report all matches */
+ if (can_die) {
+ if (has_accel) {
+ rv = sheng32_4_coda(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ } else {
+ rv = sheng32_4_cod(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ }
+ if (rv == MO_HALT_MATCHING) {
+ return MO_DEAD;
+ }
+ rv = sheng32_cod(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ *scanned, end, scanned);
+ } else {
+ if (has_accel) {
+ rv = sheng32_4_coa(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ } else {
+ rv = sheng32_4_co(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ }
+ if (rv == MO_HALT_MATCHING) {
+ return MO_DEAD;
+ }
+ rv = sheng32_co(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ *scanned, end, scanned);
+ }
+ if (rv == MO_HALT_MATCHING) {
+ return MO_DEAD;
+ }
+ return MO_ALIVE;
+}
+
+static really_inline
+void runSheng32Nm(const struct sheng32 *sh, NfaCallback cb, void *ctxt,
+ u64a offset, u8 *const cached_accept_state,
+ ReportID *const cached_accept_id, const u8 *cur_buf,
+ const u8 *start, const u8 *end, u8 can_die, u8 has_accel,
+ u8 single, const u8 **scanned, u8 *state) {
+ DEBUG_PRINTF("Scanning %llu bytes (offset %llu) in nomatch mode\n",
+ (u64a)(end - start), offset);
+ DEBUG_PRINTF("start: %lli end: %lli\n", (s64a)(start - cur_buf),
+ (s64a)(end - cur_buf));
+ DEBUG_PRINTF("can die: %u has accel: %u single: %u\n", !!can_die,
+ !!has_accel, !!single);
+ /* just scan the buffer */
+ if (can_die) {
+ if (has_accel) {
+ sheng32_4_nmda(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ } else {
+ sheng32_4_nmd(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ }
+ sheng32_nmd(state, cb, ctxt, sh, cached_accept_state, cached_accept_id,
+ single, offset, cur_buf, *scanned, end, scanned);
+ } else {
+ sheng32_4_nm(state, cb, ctxt, sh, cached_accept_state, cached_accept_id,
+ single, offset, cur_buf, start, end, scanned);
+ sheng32_nm(state, cb, ctxt, sh, cached_accept_state, cached_accept_id,
+ single, offset, cur_buf, *scanned, end, scanned);
+ }
+}
+
+static really_inline
+char runSheng32Sam(const struct sheng32 *sh, NfaCallback cb, void *ctxt,
+ u64a offset, u8 *const cached_accept_state,
+ ReportID *const cached_accept_id, const u8 *cur_buf,
+ const u8 *start, const u8 *end, u8 can_die, u8 has_accel,
+ u8 single, const u8 **scanned, u8 *state) {
+ DEBUG_PRINTF("Scanning %llu bytes (offset %llu) in stop at match mode\n",
+ (u64a)(end - start), offset);
+ DEBUG_PRINTF("start: %lli end: %lli\n", (s64a)(start - cur_buf),
+ (s64a)(end - cur_buf));
+ DEBUG_PRINTF("can die: %u has accel: %u single: %u\n", !!can_die,
+ !!has_accel, !!single);
+ int rv;
+ /* scan until first match */
+ if (can_die) {
+ if (has_accel) {
+ rv = sheng32_4_samda(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ } else {
+ rv = sheng32_4_samd(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ }
+ if (rv == MO_HALT_MATCHING) {
+ return MO_DEAD;
+ }
+ /* if we stopped before we expected, we found a match */
+ if (rv == MO_MATCHES_PENDING) {
+ return MO_MATCHES_PENDING;
+ }
+
+ rv = sheng32_samd(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ *scanned, end, scanned);
+ } else {
+ if (has_accel) {
+ rv = sheng32_4_sama(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ } else {
+ rv = sheng32_4_sam(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ start, end, scanned);
+ }
+ if (rv == MO_HALT_MATCHING) {
+ return MO_DEAD;
+ }
+ /* if we stopped before we expected, we found a match */
+ if (rv == MO_MATCHES_PENDING) {
+ return MO_MATCHES_PENDING;
+ }
+
+ rv = sheng32_sam(state, cb, ctxt, sh, cached_accept_state,
+ cached_accept_id, single, offset, cur_buf,
+ *scanned, end, scanned);
+ }
+ if (rv == MO_HALT_MATCHING) {
+ return MO_DEAD;
+ }
+ /* if we stopped before we expected, we found a match */
+ if (rv == MO_MATCHES_PENDING) {
+ return MO_MATCHES_PENDING;
+ }
+ return MO_ALIVE;
+}
+
+static never_inline
+char runSheng32(const struct sheng32 *sh, struct mq *q, s64a b_end,
+ enum MatchMode mode) {
+ u8 state = *(u8 *)q->state;
+ u8 can_die = sh->flags & SHENG_FLAG_CAN_DIE;
+ u8 has_accel = sh->flags & SHENG_FLAG_HAS_ACCEL;
+ u8 single = sh->flags & SHENG_FLAG_SINGLE_REPORT;
+
+ u8 cached_accept_state = 0;
+ ReportID cached_accept_id = 0;
+
+ DEBUG_PRINTF("starting Sheng32 execution in state %u\n",
+ state & SHENG32_STATE_MASK);
+
+ if (q->report_current) {
+ DEBUG_PRINTF("reporting current pending matches\n");
+ assert(sh);
+
+ q->report_current = 0;
+
+ int rv;
+ if (single) {
+ rv = fireSingleReport(q->cb, q->context, sh->report,
+ q_cur_offset(q));
+ } else {
+ rv = fireReports32(sh, q->cb, q->context, state, q_cur_offset(q),
+ &cached_accept_state, &cached_accept_id, 0);
+ }
+ if (rv == MO_HALT_MATCHING) {
+ DEBUG_PRINTF("exiting in state %u\n", state & SHENG32_STATE_MASK);
+ return MO_DEAD;
+ }
+
+ DEBUG_PRINTF("proceeding with matching\n");
+ }
+
+ assert(q_cur_type(q) == MQE_START);
+ s64a start = q_cur_loc(q);
+
+ DEBUG_PRINTF("offset: %lli, location: %lli, mode: %s\n", q->offset, start,
+ mode == CALLBACK_OUTPUT ? "CALLBACK OUTPUT" :
+ mode == NO_MATCHES ? "NO MATCHES" :
+ mode == STOP_AT_MATCH ? "STOP AT MATCH" : "???");
+
+ DEBUG_PRINTF("processing event @ %lli: %s\n", q->offset + q_cur_loc(q),
+ q_cur_type(q) == MQE_START ? "START" :
+ q_cur_type(q) == MQE_TOP ? "TOP" :
+ q_cur_type(q) == MQE_END ? "END" : "???");
+
+ const u8* cur_buf;
+ if (start < 0) {
+ DEBUG_PRINTF("negative location, scanning history\n");
+ DEBUG_PRINTF("min location: %zd\n", -q->hlength);
+ cur_buf = q->history + q->hlength;
+ } else {
+ DEBUG_PRINTF("positive location, scanning buffer\n");
+ DEBUG_PRINTF("max location: %lli\n", b_end);
+ cur_buf = q->buffer;
+ }
+
+ /* if we our queue event is past our end */
+ if (mode != NO_MATCHES && q_cur_loc(q) > b_end) {
+ DEBUG_PRINTF("current location past buffer end\n");
+ DEBUG_PRINTF("setting q location to %llu\n", b_end);
+ DEBUG_PRINTF("exiting in state %u\n", state & SHENG32_STATE_MASK);
+ q->items[q->cur].location = b_end;
+ return MO_ALIVE;
+ }
+
+ q->cur++;
+
+ s64a cur_start = start;
+
+ while (1) {
+ DEBUG_PRINTF("processing event @ %lli: %s\n", q->offset + q_cur_loc(q),
+ q_cur_type(q) == MQE_START ? "START" :
+ q_cur_type(q) == MQE_TOP ? "TOP" :
+ q_cur_type(q) == MQE_END ? "END" : "???");
+ s64a end = q_cur_loc(q);
+ if (mode != NO_MATCHES) {
+ end = MIN(end, b_end);
+ }
+ assert(end <= (s64a) q->length);
+ s64a cur_end = end;
+
+ /* we may cross the border between history and current buffer */
+ if (cur_start < 0) {
+ cur_end = MIN(0, cur_end);
+ }
+
+ DEBUG_PRINTF("start: %lli end: %lli\n", start, end);
+
+ /* don't scan zero length buffer */
+ if (cur_start != cur_end) {
+ const u8 * scanned = cur_buf;
+ char rv;
+
+ if (mode == NO_MATCHES) {
+ runSheng32Nm(sh, q->cb, q->context, q->offset,
+ &cached_accept_state, &cached_accept_id, cur_buf,
+ cur_buf + cur_start, cur_buf + cur_end, can_die,
+ has_accel, single, &scanned, &state);
+ } else if (mode == CALLBACK_OUTPUT) {
+ rv = runSheng32Cb(sh, q->cb, q->context, q->offset,
+ &cached_accept_state, &cached_accept_id,
+ cur_buf, cur_buf + cur_start, cur_buf + cur_end,
+ can_die, has_accel, single, &scanned, &state);
+ if (rv == MO_DEAD) {
+ DEBUG_PRINTF("exiting in state %u\n",
+ state & SHENG32_STATE_MASK);
+ return MO_DEAD;
+ }
+ } else if (mode == STOP_AT_MATCH) {
+ rv = runSheng32Sam(sh, q->cb, q->context, q->offset,
+ &cached_accept_state, &cached_accept_id,
+ cur_buf, cur_buf + cur_start,
+ cur_buf + cur_end, can_die, has_accel, single,
+ &scanned, &state);
+ if (rv == MO_DEAD) {
+ DEBUG_PRINTF("exiting in state %u\n",
+ state & SHENG32_STATE_MASK);
+ return rv;
+ } else if (rv == MO_MATCHES_PENDING) {
+ assert(q->cur);
+ DEBUG_PRINTF("found a match, setting q location to %zd\n",
+ scanned - cur_buf + 1);
+ q->cur--;
+ q->items[q->cur].type = MQE_START;
+ q->items[q->cur].location =
+ scanned - cur_buf + 1; /* due to exiting early */
+ *(u8 *)q->state = state;
+ DEBUG_PRINTF("exiting in state %u\n",
+ state & SHENG32_STATE_MASK);
+ return rv;
+ }
+ } else {
+ assert(!"invalid scanning mode!");
+ }
+ assert(scanned == cur_buf + cur_end);
+
+ cur_start = cur_end;
+ }
+
+ /* if we our queue event is past our end */
+ if (mode != NO_MATCHES && q_cur_loc(q) > b_end) {
+ DEBUG_PRINTF("current location past buffer end\n");
+ DEBUG_PRINTF("setting q location to %llu\n", b_end);
+ DEBUG_PRINTF("exiting in state %u\n", state & SHENG32_STATE_MASK);
+ q->cur--;
+ q->items[q->cur].type = MQE_START;
+ q->items[q->cur].location = b_end;
+ *(u8 *)q->state = state;
+ return MO_ALIVE;
+ }
+
+ /* crossing over into actual buffer */
+ if (cur_start == 0) {
+ DEBUG_PRINTF("positive location, scanning buffer\n");
+ DEBUG_PRINTF("max offset: %lli\n", b_end);
+ cur_buf = q->buffer;
+ }
+
+ /* continue scanning the same buffer */
+ if (end != cur_end) {
+ continue;
+ }
+
+ switch (q_cur_type(q)) {
+ case MQE_END:
+ *(u8 *)q->state = state;
+ q->cur++;
+ DEBUG_PRINTF("exiting in state %u\n", state & SHENG32_STATE_MASK);
+ if (can_die) {
+ return (state & SHENG32_STATE_DEAD) ? MO_DEAD : MO_ALIVE;
+ }
+ return MO_ALIVE;
+ case MQE_TOP:
+ if (q->offset + cur_start == 0) {
+ DEBUG_PRINTF("Anchored start, going to state %u\n",
+ sh->anchored);
+ state = sh->anchored;
+ } else {
+ u8 new_state = get_aux32(sh, state)->top;
+ DEBUG_PRINTF("Top event %u->%u\n", state & SHENG32_STATE_MASK,
+ new_state & SHENG32_STATE_MASK);
+ state = new_state;
+ }
+ break;
+ default:
+ assert(!"invalid queue event");
+ break;
+ }
+ q->cur++;
+ }
+}
+
+char nfaExecSheng32_B(const struct NFA *n, u64a offset, const u8 *buffer,
+ size_t length, NfaCallback cb, void *context) {
+ DEBUG_PRINTF("smallwrite Sheng32\n");
+ assert(n->type == SHENG_NFA_32);
+ const struct sheng32 *sh = getImplNfa(n);
+ u8 state = sh->anchored;
+ u8 can_die = sh->flags & SHENG_FLAG_CAN_DIE;
+ u8 has_accel = sh->flags & SHENG_FLAG_HAS_ACCEL;
+ u8 single = sh->flags & SHENG_FLAG_SINGLE_REPORT;
+ u8 cached_accept_state = 0;
+ ReportID cached_accept_id = 0;
+
+ /* scan and report all matches */
+ int rv;
+ s64a end = length;
+ const u8 *scanned;
+
+ rv = runSheng32Cb(sh, cb, context, offset, &cached_accept_state,
+ &cached_accept_id, buffer, buffer, buffer + end, can_die,
+ has_accel, single, &scanned, &state);
+ if (rv == MO_DEAD) {
+ DEBUG_PRINTF("exiting in state %u\n",
+ state & SHENG32_STATE_MASK);
+ return MO_DEAD;
+ }
+
+ DEBUG_PRINTF("%u\n", state & SHENG32_STATE_MASK);
+
+ const struct sstate_aux *aux = get_aux32(sh, state);
+
+ if (aux->accept_eod) {
+ DEBUG_PRINTF("Reporting EOD matches\n");
+ fireReports32(sh, cb, context, state, end + offset,
+ &cached_accept_state, &cached_accept_id, 1);
+ }
+
+ return state & SHENG32_STATE_DEAD ? MO_DEAD : MO_ALIVE;
+}
+
+char nfaExecSheng32_Q(const struct NFA *n, struct mq *q, s64a end) {
+ const struct sheng32 *sh = get_sheng32(n);
+ char rv = runSheng32(sh, q, end, CALLBACK_OUTPUT);
+ return rv;
+}
+
+char nfaExecSheng32_Q2(const struct NFA *n, struct mq *q, s64a end) {
+ const struct sheng32 *sh = get_sheng32(n);
+ char rv = runSheng32(sh, q, end, STOP_AT_MATCH);
+ return rv;
+}
+
+char nfaExecSheng32_QR(const struct NFA *n, struct mq *q, ReportID report) {
+ assert(q_cur_type(q) == MQE_START);
+
+ const struct sheng32 *sh = get_sheng32(n);
+ char rv = runSheng32(sh, q, 0 /* end */, NO_MATCHES);
+
+ if (rv && nfaExecSheng32_inAccept(n, report, q)) {
+ return MO_MATCHES_PENDING;
+ }
+ return rv;
+}
+
+char nfaExecSheng32_inAccept(const struct NFA *n, ReportID report,
+ struct mq *q) {
+ assert(n && q);
+
+ const struct sheng32 *sh = get_sheng32(n);
+ u8 s = *(const u8 *)q->state;
+ DEBUG_PRINTF("checking accepts for %u\n", (u8)(s & SHENG32_STATE_MASK));
+
+ const struct sstate_aux *aux = get_aux32(sh, s);
+
+ if (!aux->accept) {
+ return 0;
+ }
+
+ return sheng32HasAccept(sh, aux, report);
+}
+
+char nfaExecSheng32_inAnyAccept(const struct NFA *n, struct mq *q) {
+ assert(n && q);
+
+ const struct sheng32 *sh = get_sheng32(n);
+ u8 s = *(const u8 *)q->state;
+ DEBUG_PRINTF("checking accepts for %u\n", (u8)(s & SHENG32_STATE_MASK));
+
+ const struct sstate_aux *aux = get_aux32(sh, s);
+ return !!aux->accept;
+}
+
+char nfaExecSheng32_testEOD(const struct NFA *nfa, const char *state,
+ UNUSED const char *streamState, u64a offset,
+ NfaCallback cb, void *ctxt) {
+ assert(nfa);
+
+ const struct sheng32 *sh = get_sheng32(nfa);
+ u8 s = *(const u8 *)state;
+ DEBUG_PRINTF("checking EOD accepts for %u\n", (u8)(s & SHENG32_STATE_MASK));
+
+ const struct sstate_aux *aux = get_aux32(sh, s);
+
+ if (!aux->accept_eod) {
+ return MO_CONTINUE_MATCHING;
+ }
+
+ return fireReports32(sh, cb, ctxt, s, offset, NULL, NULL, 1);
+}
+
+char nfaExecSheng32_reportCurrent(const struct NFA *n, struct mq *q) {
+ const struct sheng32 *sh = (const struct sheng32 *)getImplNfa(n);
+ NfaCallback cb = q->cb;
+ void *ctxt = q->context;
+ u8 s = *(u8 *)q->state;
+ const struct sstate_aux *aux = get_aux32(sh, s);
+ u64a offset = q_cur_offset(q);
+ u8 cached_state_id = 0;
+ ReportID cached_report_id = 0;
+ assert(q_cur_type(q) == MQE_START);
+
+ if (aux->accept) {
+ if (sh->flags & SHENG_FLAG_SINGLE_REPORT) {
+ fireSingleReport(cb, ctxt, sh->report, offset);
+ } else {
+ fireReports32(sh, cb, ctxt, s, offset, &cached_state_id,
+ &cached_report_id, 0);
+ }
+ }
+
+ return 0;
+}
+
+char nfaExecSheng32_initCompressedState(const struct NFA *nfa, u64a offset,
+ void *state, UNUSED u8 key) {
+ const struct sheng32 *sh = get_sheng32(nfa);
+ u8 *s = (u8 *)state;
+ *s = offset ? sh->floating: sh->anchored;
+ return !(*s & SHENG32_STATE_DEAD);
+}
+
+char nfaExecSheng32_queueInitState(const struct NFA *nfa, struct mq *q) {
+ assert(nfa->scratchStateSize == 1);
+
+ /* starting in floating state */
+ const struct sheng32 *sh = get_sheng32(nfa);
+ *(u8 *)q->state = sh->floating;
+ DEBUG_PRINTF("starting in floating state\n");
+ return 0;
+}
+
+char nfaExecSheng32_queueCompressState(UNUSED const struct NFA *nfa,
+ const struct mq *q, UNUSED s64a loc) {
+ void *dest = q->streamState;
+ const void *src = q->state;
+ assert(nfa->scratchStateSize == 1);
+ assert(nfa->streamStateSize == 1);
+ *(u8 *)dest = *(const u8 *)src;
+ return 0;
+}
+
+char nfaExecSheng32_expandState(UNUSED const struct NFA *nfa, void *dest,
+ const void *src, UNUSED u64a offset,
+ UNUSED u8 key) {
+ assert(nfa->scratchStateSize == 1);
+ assert(nfa->streamStateSize == 1);
+ *(u8 *)dest = *(const u8 *)src;
+ return 0;
+}
+#endif // end of HAVE_AVX512VBMI
char nfaExecSheng_B(const struct NFA *n, u64a offset, const u8 *buffer,
size_t length, NfaCallback cb, void *context);
+#if defined(HAVE_AVX512VBMI)
+#define nfaExecSheng32_B_Reverse NFA_API_NO_IMPL
+#define nfaExecSheng32_zombie_status NFA_API_ZOMBIE_NO_IMPL
+
+char nfaExecSheng32_Q(const struct NFA *n, struct mq *q, s64a end);
+char nfaExecSheng32_Q2(const struct NFA *n, struct mq *q, s64a end);
+char nfaExecSheng32_QR(const struct NFA *n, struct mq *q, ReportID report);
+char nfaExecSheng32_inAccept(const struct NFA *n, ReportID report,
+ struct mq *q);
+char nfaExecSheng32_inAnyAccept(const struct NFA *n, struct mq *q);
+char nfaExecSheng32_queueInitState(const struct NFA *nfa, struct mq *q);
+char nfaExecSheng32_queueCompressState(const struct NFA *nfa,
+ const struct mq *q, s64a loc);
+char nfaExecSheng32_expandState(const struct NFA *nfa, void *dest,
+ const void *src, u64a offset, u8 key);
+char nfaExecSheng32_initCompressedState(const struct NFA *nfa, u64a offset,
+ void *state, u8 key);
+char nfaExecSheng32_testEOD(const struct NFA *nfa, const char *state,
+ const char *streamState, u64a offset,
+ NfaCallback callback, void *context);
+char nfaExecSheng32_reportCurrent(const struct NFA *n, struct mq *q);
+
+char nfaExecSheng32_B(const struct NFA *n, u64a offset, const u8 *buffer,
+ size_t length, NfaCallback cb, void *context);
+#else // !HAVE_AVX512VBMI
+#define nfaExecSheng32_B_Reverse NFA_API_NO_IMPL
+#define nfaExecSheng32_zombie_status NFA_API_ZOMBIE_NO_IMPL
+#define nfaExecSheng32_Q NFA_API_NO_IMPL
+#define nfaExecSheng32_Q2 NFA_API_NO_IMPL
+#define nfaExecSheng32_QR NFA_API_NO_IMPL
+#define nfaExecSheng32_inAccept NFA_API_NO_IMPL
+#define nfaExecSheng32_inAnyAccept NFA_API_NO_IMPL
+#define nfaExecSheng32_queueInitState NFA_API_NO_IMPL
+#define nfaExecSheng32_queueCompressState NFA_API_NO_IMPL
+#define nfaExecSheng32_expandState NFA_API_NO_IMPL
+#define nfaExecSheng32_initCompressedState NFA_API_NO_IMPL
+#define nfaExecSheng32_testEOD NFA_API_NO_IMPL
+#define nfaExecSheng32_reportCurrent NFA_API_NO_IMPL
+#define nfaExecSheng32_B NFA_API_NO_IMPL
+#endif // end of HAVE_AVX512VBMI
+
#endif /* SHENG_H_ */
return a & SHENG_STATE_DEAD;
}
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+u8 isDeadState32(const u8 a) {
+ return a & SHENG32_STATE_DEAD;
+}
+#endif
+
static really_inline
u8 isAcceptState(const u8 a) {
return a & SHENG_STATE_ACCEPT;
}
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+u8 isAcceptState32(const u8 a) {
+ return a & SHENG32_STATE_ACCEPT;
+}
+#endif
+
static really_inline
u8 isAccelState(const u8 a) {
return a & SHENG_STATE_ACCEL;
}
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+u8 isAccelState32(const u8 a) {
+ return a & SHENG32_STATE_ACCEL;
+}
+#endif
+
static really_inline
u8 hasInterestingStates(const u8 a, const u8 b, const u8 c, const u8 d) {
return (a | b | c | d) & (SHENG_STATE_FLAG_MASK);
}
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+u8 hasInterestingStates32(const u8 a, const u8 b, const u8 c, const u8 d) {
+ return (a | b | c | d) & (SHENG32_STATE_FLAG_MASK);
+}
+#endif
+
/* these functions should be optimized out, used by NO_MATCHES mode */
static really_inline
u8 dummyFunc4(UNUSED const u8 a, UNUSED const u8 b, UNUSED const u8 c,
#define SHENG_IMPL sheng_cod
#define DEAD_FUNC isDeadState
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_cod
+#define DEAD_FUNC32 isDeadState32
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl.h"
#undef SHENG_IMPL
#undef DEAD_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef DEAD_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* callback output, can't die */
#define SHENG_IMPL sheng_co
#define DEAD_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_co
+#define DEAD_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl.h"
#undef SHENG_IMPL
#undef DEAD_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef DEAD_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* stop at match, can die */
#define SHENG_IMPL sheng_samd
#define DEAD_FUNC isDeadState
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_samd
+#define DEAD_FUNC32 isDeadState32
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 1
#include "sheng_impl.h"
#undef SHENG_IMPL
#undef DEAD_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef DEAD_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* stop at match, can't die */
#define SHENG_IMPL sheng_sam
#define DEAD_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_sam
+#define DEAD_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 1
#include "sheng_impl.h"
#undef SHENG_IMPL
#undef DEAD_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef DEAD_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* no match, can die */
#define SHENG_IMPL sheng_nmd
#define DEAD_FUNC isDeadState
#define ACCEPT_FUNC dummyFunc
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_nmd
+#define DEAD_FUNC32 isDeadState32
+#define ACCEPT_FUNC32 dummyFunc
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl.h"
#undef SHENG_IMPL
#undef DEAD_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef DEAD_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* no match, can't die */
#define SHENG_IMPL sheng_nm
#define DEAD_FUNC dummyFunc
#define ACCEPT_FUNC dummyFunc
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_nm
+#define DEAD_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 dummyFunc
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl.h"
#undef SHENG_IMPL
#undef DEAD_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef DEAD_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/*
#define INNER_ACCEL_FUNC isAccelState
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_coda
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 isDeadState32
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 isAccelState32
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* callback output, can die, not accelerated */
#define INNER_ACCEL_FUNC dummyFunc
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_cod
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 isDeadState32
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 dummyFunc
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* callback output, can't die, accelerated */
#define INNER_ACCEL_FUNC isAccelState
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_coa
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 dummyFunc
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 isAccelState32
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* callback output, can't die, not accelerated */
#define INNER_ACCEL_FUNC dummyFunc
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_co
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 dummyFunc
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 dummyFunc
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* stop at match, can die, accelerated */
#define INNER_ACCEL_FUNC isAccelState
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_samda
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 isDeadState32
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 isAccelState32
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 1
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* stop at match, can die, not accelerated */
#define INNER_ACCEL_FUNC dummyFunc
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_samd
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 isDeadState32
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 dummyFunc
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 1
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* stop at match, can't die, accelerated */
#define INNER_ACCEL_FUNC isAccelState
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_sama
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 dummyFunc
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 isAccelState32
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 1
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* stop at match, can't die, not accelerated */
#define INNER_ACCEL_FUNC dummyFunc
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC isAcceptState
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_sam
+#define INTERESTING_FUNC32 hasInterestingStates32
+#define INNER_DEAD_FUNC32 dummyFunc
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 dummyFunc
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 isAcceptState32
+#endif
#define STOP_AT_MATCH 1
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* no-match have interesting func as dummy, and die/accel checks are outer */
#define INNER_ACCEL_FUNC dummyFunc
#define OUTER_ACCEL_FUNC isAccelState
#define ACCEPT_FUNC dummyFunc
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_nmda
+#define INTERESTING_FUNC32 dummyFunc4
+#define INNER_DEAD_FUNC32 dummyFunc
+#define OUTER_DEAD_FUNC32 isDeadState32
+#define INNER_ACCEL_FUNC32 dummyFunc
+#define OUTER_ACCEL_FUNC32 isAccelState32
+#define ACCEPT_FUNC32 dummyFunc
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* no match, can die, not accelerated */
#define INNER_ACCEL_FUNC dummyFunc
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC dummyFunc
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_nmd
+#define INTERESTING_FUNC32 dummyFunc4
+#define INNER_DEAD_FUNC32 dummyFunc
+#define OUTER_DEAD_FUNC32 isDeadState32
+#define INNER_ACCEL_FUNC32 dummyFunc
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 dummyFunc
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
/* there is no performance benefit in accelerating a no-match case that can't
#define INNER_ACCEL_FUNC dummyFunc
#define OUTER_ACCEL_FUNC dummyFunc
#define ACCEPT_FUNC dummyFunc
+#if defined(HAVE_AVX512VBMI)
+#define SHENG32_IMPL sheng32_4_nm
+#define INTERESTING_FUNC32 dummyFunc4
+#define INNER_DEAD_FUNC32 dummyFunc
+#define OUTER_DEAD_FUNC32 dummyFunc
+#define INNER_ACCEL_FUNC32 dummyFunc
+#define OUTER_ACCEL_FUNC32 dummyFunc
+#define ACCEPT_FUNC32 dummyFunc
+#endif
#define STOP_AT_MATCH 0
#include "sheng_impl4.h"
#undef SHENG_IMPL
#undef INNER_ACCEL_FUNC
#undef OUTER_ACCEL_FUNC
#undef ACCEPT_FUNC
+#if defined(HAVE_AVX512VBMI)
+#undef SHENG32_IMPL
+#undef INTERESTING_FUNC32
+#undef INNER_DEAD_FUNC32
+#undef OUTER_DEAD_FUNC32
+#undef INNER_ACCEL_FUNC32
+#undef OUTER_ACCEL_FUNC32
+#undef ACCEPT_FUNC32
+#endif
#undef STOP_AT_MATCH
#endif // SHENG_DEFS_H
*scan_end = cur_buf;
return MO_CONTINUE_MATCHING;
}
+
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
+ const struct sheng32 *s,
+ u8 *const cached_accept_state,
+ ReportID *const cached_accept_id,
+ u8 single, u64a base_offset, const u8 *buf, const u8 *start,
+ const u8 *end, const u8 **scan_end) {
+ DEBUG_PRINTF("Starting DFA execution in state %u\n",
+ *state & SHENG32_STATE_MASK);
+ const u8 *cur_buf = start;
+ if (DEAD_FUNC32(*state)) {
+ DEBUG_PRINTF("Dead on arrival\n");
+ *scan_end = end;
+ return MO_CONTINUE_MATCHING;
+ }
+ DEBUG_PRINTF("Scanning %lli bytes\n", (s64a)(end - start));
+
+ m512 cur_state = set64x8(*state);
+ const m512 *masks = s->succ_masks;
+
+ while (likely(cur_buf != end)) {
+ const u8 c = *cur_buf;
+ const m512 succ_mask = masks[c];
+ cur_state = vpermb512(cur_state, succ_mask);
+ const u8 tmp = movd512(cur_state);
+
+ DEBUG_PRINTF("c: %02hhx '%c'\n", c, ourisprint(c) ? c : '?');
+ DEBUG_PRINTF("s: %u (flag: %u)\n", tmp & SHENG32_STATE_MASK,
+ tmp & SHENG32_STATE_FLAG_MASK);
+
+ if (unlikely(ACCEPT_FUNC32(tmp))) {
+ DEBUG_PRINTF("Accept state %u reached\n", tmp & SHENG32_STATE_MASK);
+ u64a match_offset = base_offset + (cur_buf - buf) + 1;
+ DEBUG_PRINTF("Match @ %llu\n", match_offset);
+ if (STOP_AT_MATCH) {
+ DEBUG_PRINTF("Stopping at match @ %lli\n",
+ (u64a)(cur_buf - start));
+ *state = tmp;
+ *scan_end = cur_buf;
+ return MO_MATCHES_PENDING;
+ }
+ if (single) {
+ if (fireSingleReport(cb, ctxt, s->report, match_offset) ==
+ MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ } else {
+ if (fireReports32(s, cb, ctxt, tmp, match_offset,
+ cached_accept_state, cached_accept_id,
+ 0) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ }
+ }
+ cur_buf++;
+ }
+ *state = movd512(cur_state);
+ *scan_end = cur_buf;
+ return MO_CONTINUE_MATCHING;
+}
+#endif
*scan_end = cur_buf;
return MO_CONTINUE_MATCHING;
}
+
+#if defined(HAVE_AVX512VBMI)
+static really_inline
+char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
+ const struct sheng32 *s,
+ u8 *const cached_accept_state,
+ ReportID *const cached_accept_id,
+ u8 single, u64a base_offset, const u8 *buf, const u8 *start,
+ const u8 *end, const u8 **scan_end) {
+ DEBUG_PRINTF("Starting DFAx4 execution in state %u\n",
+ *state & SHENG32_STATE_MASK);
+ const u8 *cur_buf = start;
+ const u8 *min_accel_dist = start;
+ base_offset++;
+ DEBUG_PRINTF("Scanning %llu bytes\n", (u64a)(end - start));
+
+ if (INNER_ACCEL_FUNC32(*state) || OUTER_ACCEL_FUNC32(*state)) {
+ DEBUG_PRINTF("Accel state reached @ 0\n");
+ const union AccelAux *aaux =
+ get_accel32(s, *state & SHENG32_STATE_MASK);
+ const u8 *new_offset = run_accel(aaux, cur_buf, end);
+ if (new_offset < cur_buf + BAD_ACCEL_DIST) {
+ min_accel_dist = new_offset + BIG_ACCEL_PENALTY;
+ } else {
+ min_accel_dist = new_offset + SMALL_ACCEL_PENALTY;
+ }
+ DEBUG_PRINTF("Next accel chance: %llu\n",
+ (u64a)(min_accel_dist - start));
+ DEBUG_PRINTF("Accel scanned %zu bytes\n", new_offset - cur_buf);
+ cur_buf = new_offset;
+ DEBUG_PRINTF("New offset: %lli\n", (s64a)(cur_buf - start));
+ }
+ if (INNER_DEAD_FUNC32(*state) || OUTER_DEAD_FUNC32(*state)) {
+ DEBUG_PRINTF("Dead on arrival\n");
+ *scan_end = end;
+ return MO_CONTINUE_MATCHING;
+ }
+
+ m512 cur_state = set64x8(*state);
+ const m512 *masks = s->succ_masks;
+
+ while (likely(end - cur_buf >= 4)) {
+ const u8 *b1 = cur_buf;
+ const u8 *b2 = cur_buf + 1;
+ const u8 *b3 = cur_buf + 2;
+ const u8 *b4 = cur_buf + 3;
+ const u8 c1 = *b1;
+ const u8 c2 = *b2;
+ const u8 c3 = *b3;
+ const u8 c4 = *b4;
+
+ const m512 succ_mask1 = masks[c1];
+ cur_state = vpermb512(cur_state, succ_mask1);
+ const u8 a1 = movd512(cur_state);
+
+ const m512 succ_mask2 = masks[c2];
+ cur_state = vpermb512(cur_state, succ_mask2);
+ const u8 a2 = movd512(cur_state);
+
+ const m512 succ_mask3 = masks[c3];
+ cur_state = vpermb512(cur_state, succ_mask3);
+ const u8 a3 = movd512(cur_state);
+
+ const m512 succ_mask4 = masks[c4];
+ cur_state = vpermb512(cur_state, succ_mask4);
+ const u8 a4 = movd512(cur_state);
+
+ DEBUG_PRINTF("c: %02hhx '%c'\n", c1, ourisprint(c1) ? c1 : '?');
+ DEBUG_PRINTF("s: %u (flag: %u)\n", a1 & SHENG32_STATE_MASK,
+ a1 & SHENG32_STATE_FLAG_MASK);
+
+ DEBUG_PRINTF("c: %02hhx '%c'\n", c2, ourisprint(c2) ? c2 : '?');
+ DEBUG_PRINTF("s: %u (flag: %u)\n", a2 & SHENG32_STATE_MASK,
+ a2 & SHENG32_STATE_FLAG_MASK);
+
+ DEBUG_PRINTF("c: %02hhx '%c'\n", c3, ourisprint(c3) ? c3 : '?');
+ DEBUG_PRINTF("s: %u (flag: %u)\n", a3 & SHENG32_STATE_MASK,
+ a3 & SHENG32_STATE_FLAG_MASK);
+
+ DEBUG_PRINTF("c: %02hhx '%c'\n", c4, ourisprint(c4) ? c4 : '?');
+ DEBUG_PRINTF("s: %u (flag: %u)\n", a4 & SHENG32_STATE_MASK,
+ a4 & SHENG32_STATE_FLAG_MASK);
+
+ if (unlikely(INTERESTING_FUNC32(a1, a2, a3, a4))) {
+ if (ACCEPT_FUNC32(a1)) {
+ u64a match_offset = base_offset + b1 - buf;
+ DEBUG_PRINTF("Accept state %u reached\n",
+ a1 & SHENG32_STATE_MASK);
+ DEBUG_PRINTF("Match @ %llu\n", match_offset);
+ if (STOP_AT_MATCH) {
+ DEBUG_PRINTF("Stopping at match @ %lli\n",
+ (s64a)(b1 - start));
+ *scan_end = b1;
+ *state = a1;
+ return MO_MATCHES_PENDING;
+ }
+ if (single) {
+ if (fireSingleReport(cb, ctxt, s->report, match_offset) ==
+ MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ } else {
+ if (fireReports32(s, cb, ctxt, a1, match_offset,
+ cached_accept_state, cached_accept_id,
+ 0) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ }
+ }
+ if (ACCEPT_FUNC32(a2)) {
+ u64a match_offset = base_offset + b2 - buf;
+ DEBUG_PRINTF("Accept state %u reached\n",
+ a2 & SHENG32_STATE_MASK);
+ DEBUG_PRINTF("Match @ %llu\n", match_offset);
+ if (STOP_AT_MATCH) {
+ DEBUG_PRINTF("Stopping at match @ %lli\n",
+ (s64a)(b2 - start));
+ *scan_end = b2;
+ *state = a2;
+ return MO_MATCHES_PENDING;
+ }
+ if (single) {
+ if (fireSingleReport(cb, ctxt, s->report, match_offset) ==
+ MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ } else {
+ if (fireReports32(s, cb, ctxt, a2, match_offset,
+ cached_accept_state, cached_accept_id,
+ 0) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ }
+ }
+ if (ACCEPT_FUNC32(a3)) {
+ u64a match_offset = base_offset + b3 - buf;
+ DEBUG_PRINTF("Accept state %u reached\n",
+ a3 & SHENG32_STATE_MASK);
+ DEBUG_PRINTF("Match @ %llu\n", match_offset);
+ if (STOP_AT_MATCH) {
+ DEBUG_PRINTF("Stopping at match @ %lli\n",
+ (s64a)(b3 - start));
+ *scan_end = b3;
+ *state = a3;
+ return MO_MATCHES_PENDING;
+ }
+ if (single) {
+ if (fireSingleReport(cb, ctxt, s->report, match_offset) ==
+ MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ } else {
+ if (fireReports32(s, cb, ctxt, a3, match_offset,
+ cached_accept_state, cached_accept_id,
+ 0) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ }
+ }
+ if (ACCEPT_FUNC32(a4)) {
+ u64a match_offset = base_offset + b4 - buf;
+ DEBUG_PRINTF("Accept state %u reached\n",
+ a4 & SHENG32_STATE_MASK);
+ DEBUG_PRINTF("Match @ %llu\n", match_offset);
+ if (STOP_AT_MATCH) {
+ DEBUG_PRINTF("Stopping at match @ %lli\n",
+ (s64a)(b4 - start));
+ *scan_end = b4;
+ *state = a4;
+ return MO_MATCHES_PENDING;
+ }
+ if (single) {
+ if (fireSingleReport(cb, ctxt, s->report, match_offset) ==
+ MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ } else {
+ if (fireReports32(s, cb, ctxt, a4, match_offset,
+ cached_accept_state, cached_accept_id,
+ 0) == MO_HALT_MATCHING) {
+ return MO_HALT_MATCHING;
+ }
+ }
+ }
+ if (INNER_DEAD_FUNC32(a4)) {
+ DEBUG_PRINTF("Dead state reached @ %lli\n", (s64a)(b4 - buf));
+ *scan_end = end;
+ *state = a4;
+ return MO_CONTINUE_MATCHING;
+ }
+ if (cur_buf > min_accel_dist && INNER_ACCEL_FUNC32(a4)) {
+ DEBUG_PRINTF("Accel state reached @ %lli\n", (s64a)(b4 - buf));
+ const union AccelAux *aaux =
+ get_accel32(s, a4 & SHENG32_STATE_MASK);
+ const u8 *new_offset = run_accel(aaux, cur_buf + 4, end);
+ if (new_offset < cur_buf + 4 + BAD_ACCEL_DIST) {
+ min_accel_dist = new_offset + BIG_ACCEL_PENALTY;
+ } else {
+ min_accel_dist = new_offset + SMALL_ACCEL_PENALTY;
+ }
+ DEBUG_PRINTF("Next accel chance: %llu\n",
+ (u64a)(min_accel_dist - start));
+ DEBUG_PRINTF("Accel scanned %llu bytes\n",
+ (u64a)(new_offset - cur_buf - 4));
+ cur_buf = new_offset;
+ DEBUG_PRINTF("New offset: %llu\n", (u64a)(cur_buf - buf));
+ continue;
+ }
+ }
+ if (OUTER_DEAD_FUNC32(a4)) {
+ DEBUG_PRINTF("Dead state reached @ %lli\n", (s64a)(cur_buf - buf));
+ *scan_end = end;
+ *state = a4;
+ return MO_CONTINUE_MATCHING;
+ };
+ if (cur_buf > min_accel_dist && OUTER_ACCEL_FUNC32(a4)) {
+ DEBUG_PRINTF("Accel state reached @ %lli\n", (s64a)(b4 - buf));
+ const union AccelAux *aaux =
+ get_accel32(s, a4 & SHENG32_STATE_MASK);
+ const u8 *new_offset = run_accel(aaux, cur_buf + 4, end);
+ if (new_offset < cur_buf + 4 + BAD_ACCEL_DIST) {
+ min_accel_dist = new_offset + BIG_ACCEL_PENALTY;
+ } else {
+ min_accel_dist = new_offset + SMALL_ACCEL_PENALTY;
+ }
+ DEBUG_PRINTF("Next accel chance: %llu\n",
+ (u64a)(min_accel_dist - start));
+ DEBUG_PRINTF("Accel scanned %llu bytes\n",
+ (u64a)(new_offset - cur_buf - 4));
+ cur_buf = new_offset;
+ DEBUG_PRINTF("New offset: %llu\n", (u64a)(cur_buf - buf));
+ continue;
+ };
+ cur_buf += 4;
+ }
+ *state = movd512(cur_state);
+ *scan_end = cur_buf;
+ return MO_CONTINUE_MATCHING;
+}
+#endif
#define SHENG_STATE_MASK 0xF
#define SHENG_STATE_FLAG_MASK 0x70
+#if defined (HAVE_AVX512VBMI)
+#define SHENG32_STATE_ACCEPT 0x20
+#define SHENG32_STATE_DEAD 0x40
+#define SHENG32_STATE_ACCEL 0x80
+#define SHENG32_STATE_MASK 0x1F
+#define SHENG32_STATE_FLAG_MASK 0xE0
+#endif
+
#define SHENG_FLAG_SINGLE_REPORT 0x1
#define SHENG_FLAG_CAN_DIE 0x2
#define SHENG_FLAG_HAS_ACCEL 0x4
ReportID report;
};
+#if defined (HAVE_AVX512VBMI)
+struct sheng32 {
+ m512 succ_masks[256];
+ u32 length;
+ u32 aux_offset;
+ u32 report_offset;
+ u32 accel_offset;
+ u8 n_states;
+ u8 anchored;
+ u8 floating;
+ u8 flags;
+ ReportID report;
+};
+#endif
+
#endif /* SHENG_INTERNAL_H_ */
}
DEBUG_PRINTF("chr %3u: %s\n", chr, o.str().c_str());
}
+
+static really_inline
+void dumpShuffleMask32(const u8 chr, const u8 *buf, unsigned sz) {
+ stringstream o;
+
+ for (unsigned i = 0; i < sz; i++) {
+ o.width(2);
+ o << (buf[i] & SHENG32_STATE_MASK) << " ";
+ }
+ DEBUG_PRINTF("chr %3u: %s\n", chr, o.str().c_str());
+}
#endif
static
}
}
+template <typename T>
static
u8 getShengState(dstate &state, dfa_info &info,
map<dstate_id_t, AccelScheme> &accelInfo) {
+ return 0;
+}
+
+template <>
+u8 getShengState<sheng>(dstate &state, dfa_info &info,
+ map<dstate_id_t, AccelScheme> &accelInfo) {
u8 s = state.impl_id;
if (!state.reports.empty()) {
s |= SHENG_STATE_ACCEPT;
return s;
}
+#if defined(HAVE_AVX512VBMI)
+template <>
+u8 getShengState<sheng32>(dstate &state, dfa_info &info,
+ map<dstate_id_t, AccelScheme> &accelInfo) {
+ u8 s = state.impl_id;
+ if (!state.reports.empty()) {
+ s |= SHENG32_STATE_ACCEPT;
+ }
+ if (info.isDead(state)) {
+ s |= SHENG32_STATE_DEAD;
+ }
+ if (accelInfo.find(info.raw_id(state.impl_id)) != accelInfo.end()) {
+ s |= SHENG32_STATE_ACCEL;
+ }
+ return s;
+}
+#endif
+
+template <typename T>
static
void fillAccelAux(struct NFA *n, dfa_info &info,
map<dstate_id_t, AccelScheme> &accelInfo) {
DEBUG_PRINTF("Filling accel aux structures\n");
- sheng *s = (sheng *)getMutableImplNfa(n);
+ T *s = (T *)getMutableImplNfa(n);
u32 offset = s->accel_offset;
for (dstate_id_t i = 0; i < info.size(); i++) {
}
}
+template <typename T>
static
void populateBasicInfo(struct NFA *n, dfa_info &info,
map<dstate_id_t, AccelScheme> &accelInfo, u32 aux_offset,
u32 report_offset, u32 accel_offset, u32 total_size,
u32 dfa_size) {
+}
+
+template <>
+void populateBasicInfo<sheng>(struct NFA *n, dfa_info &info,
+ map<dstate_id_t, AccelScheme> &accelInfo,
+ u32 aux_offset, u32 report_offset,
+ u32 accel_offset, u32 total_size,
+ u32 dfa_size) {
n->length = total_size;
n->scratchStateSize = 1;
n->streamStateSize = 1;
s->length = dfa_size;
s->flags |= info.can_die ? SHENG_FLAG_CAN_DIE : 0;
- s->anchored = getShengState(info.anchored, info, accelInfo);
- s->floating = getShengState(info.floating, info, accelInfo);
+ s->anchored = getShengState<sheng>(info.anchored, info, accelInfo);
+ s->floating = getShengState<sheng>(info.floating, info, accelInfo);
}
+#if defined(HAVE_AVX512VBMI)
+template <>
+void populateBasicInfo<sheng32>(struct NFA *n, dfa_info &info,
+ map<dstate_id_t, AccelScheme> &accelInfo,
+ u32 aux_offset, u32 report_offset,
+ u32 accel_offset, u32 total_size,
+ u32 dfa_size) {
+ n->length = total_size;
+ n->scratchStateSize = 1;
+ n->streamStateSize = 1;
+ n->nPositions = info.size();
+ n->type = SHENG_NFA_32;
+ n->flags |= info.raw.hasEodReports() ? NFA_ACCEPTS_EOD : 0;
+
+ sheng32 *s = (sheng32 *)getMutableImplNfa(n);
+ s->aux_offset = aux_offset;
+ s->report_offset = report_offset;
+ s->accel_offset = accel_offset;
+ s->n_states = info.size();
+ s->length = dfa_size;
+ s->flags |= info.can_die ? SHENG_FLAG_CAN_DIE : 0;
+
+ s->anchored = getShengState<sheng32>(info.anchored, info, accelInfo);
+ s->floating = getShengState<sheng32>(info.floating, info, accelInfo);
+}
+#endif
+
+template <typename T>
static
void fillTops(NFA *n, dfa_info &info, dstate_id_t id,
map<dstate_id_t, AccelScheme> &accelInfo) {
- sheng *s = (sheng *)getMutableImplNfa(n);
+ T *s = (T *)getMutableImplNfa(n);
u32 aux_base = s->aux_offset;
DEBUG_PRINTF("Filling tops for state %u\n", id);
DEBUG_PRINTF("Top transition for state %u: %u\n", id, top_state.impl_id);
- aux->top = getShengState(top_state, info, accelInfo);
+ aux->top = getShengState<T>(top_state, info, accelInfo);
}
+template <typename T>
static
void fillAux(NFA *n, dfa_info &info, dstate_id_t id, vector<u32> &reports,
vector<u32> &reports_eod, vector<u32> &report_offsets) {
- sheng *s = (sheng *)getMutableImplNfa(n);
+ T *s = (T *)getMutableImplNfa(n);
u32 aux_base = s->aux_offset;
auto raw_id = info.raw_id(id);
DEBUG_PRINTF("EOD report list offset: %u\n", aux->accept_eod);
}
+template <typename T>
static
void fillSingleReport(NFA *n, ReportID r_id) {
- sheng *s = (sheng *)getMutableImplNfa(n);
+ T *s = (T *)getMutableImplNfa(n);
DEBUG_PRINTF("Single report ID: %u\n", r_id);
s->report = r_id;
s->flags |= SHENG_FLAG_SINGLE_REPORT;
}
+template <typename T>
static
-void createShuffleMasks(sheng *s, dfa_info &info,
+void createShuffleMasks(T *s, dfa_info &info,
map<dstate_id_t, AccelScheme> &accelInfo) {
+}
+
+template <>
+void createShuffleMasks<sheng>(sheng *s, dfa_info &info,
+ map<dstate_id_t, AccelScheme> &accelInfo) {
for (u16 chr = 0; chr < 256; chr++) {
u8 buf[16] = {0};
for (dstate_id_t idx = 0; idx < info.size(); idx++) {
auto &succ_state = info.next(idx, chr);
- buf[idx] = getShengState(succ_state, info, accelInfo);
+ buf[idx] = getShengState<sheng>(succ_state, info, accelInfo);
}
#ifdef DEBUG
dumpShuffleMask(chr, buf, sizeof(buf));
}
}
-bool has_accel_sheng(const NFA *) {
- return true; /* consider the sheng region as accelerated */
-}
-
-bytecode_ptr<NFA> shengCompile(raw_dfa &raw, const CompileContext &cc,
- const ReportManager &rm, bool only_accel_init,
- set<dstate_id_t> *accel_states) {
- if (!cc.grey.allowSheng) {
- DEBUG_PRINTF("Sheng is not allowed!\n");
- return nullptr;
- }
-
- sheng_build_strat strat(raw, rm, only_accel_init);
- dfa_info info(strat);
-
- DEBUG_PRINTF("Trying to compile a %zu state Sheng\n", raw.states.size());
+#if defined(HAVE_AVX512VBMI)
+template <>
+void createShuffleMasks<sheng32>(sheng32 *s, dfa_info &info,
+ map<dstate_id_t, AccelScheme> &accelInfo) {
+ for (u16 chr = 0; chr < 256; chr++) {
+ u8 buf[64] = {0};
- DEBUG_PRINTF("Anchored start state id: %u, floating start state id: %u\n",
- raw.start_anchored, raw.start_floating);
+ assert(info.size() <= 32);
+ for (dstate_id_t idx = 0; idx < info.size(); idx++) {
+ auto &succ_state = info.next(idx, chr);
- DEBUG_PRINTF("This DFA %s die so effective number of states is %zu\n",
- info.can_die ? "can" : "cannot", info.size());
- if (info.size() > 16) {
- DEBUG_PRINTF("Too many states\n");
- return nullptr;
+ buf[idx] = getShengState<sheng32>(succ_state, info, accelInfo);
+ buf[32 + idx] = buf[idx];
+ }
+#ifdef DEBUG
+ dumpShuffleMask32(chr, buf, sizeof(buf));
+#endif
+ memcpy(&s->succ_masks[chr], buf, sizeof(m512));
}
+}
+#endif
+bool has_accel_sheng(const NFA *) {
+ return true; /* consider the sheng region as accelerated */
+}
+
+template <typename T>
+static
+bytecode_ptr<NFA> shengCompile_int(raw_dfa &raw, const CompileContext &cc,
+ set<dstate_id_t> *accel_states,
+ sheng_build_strat &strat,
+ dfa_info &info) {
if (!cc.streaming) { /* TODO: work out if we can do the strip in streaming
* mode with our semantics */
raw.stripExtraEodReports();
DEBUG_PRINTF("Anchored start state: %u, floating start state: %u\n",
info.anchored.impl_id, info.floating.impl_id);
- u32 nfa_size = ROUNDUP_16(sizeof(NFA) + sizeof(sheng));
+ u32 nfa_size = ROUNDUP_16(sizeof(NFA) + sizeof(T));
vector<u32> reports, eod_reports, report_offsets;
u8 isSingle = 0;
ReportID single_report = 0;
auto nfa = make_zeroed_bytecode_ptr<NFA>(total_size);
- populateBasicInfo(nfa.get(), info, accelInfo, nfa_size, reports_offset,
- accel_offset, total_size, total_size - sizeof(NFA));
+ populateBasicInfo<T>(nfa.get(), info, accelInfo, nfa_size,
+ reports_offset, accel_offset, total_size,
+ total_size - sizeof(NFA));
DEBUG_PRINTF("Setting up aux and report structures\n");
ri->fillReportLists(nfa.get(), reports_offset, report_offsets);
for (dstate_id_t idx = 0; idx < info.size(); idx++) {
- fillTops(nfa.get(), info, idx, accelInfo);
- fillAux(nfa.get(), info, idx, reports, eod_reports, report_offsets);
+ fillTops<T>(nfa.get(), info, idx, accelInfo);
+ fillAux<T>(nfa.get(), info, idx, reports, eod_reports,
+ report_offsets);
}
if (isSingle) {
- fillSingleReport(nfa.get(), single_report);
+ fillSingleReport<T>(nfa.get(), single_report);
}
- fillAccelAux(nfa.get(), info, accelInfo);
+ fillAccelAux<T>(nfa.get(), info, accelInfo);
if (accel_states) {
fillAccelOut(accelInfo, accel_states);
}
- createShuffleMasks((sheng *)getMutableImplNfa(nfa.get()), info, accelInfo);
+ createShuffleMasks<T>((T *)getMutableImplNfa(nfa.get()), info, accelInfo);
return nfa;
}
+bytecode_ptr<NFA> shengCompile(raw_dfa &raw, const CompileContext &cc,
+ const ReportManager &rm, bool only_accel_init,
+ set<dstate_id_t> *accel_states) {
+ if (!cc.grey.allowSheng) {
+ DEBUG_PRINTF("Sheng is not allowed!\n");
+ return nullptr;
+ }
+
+ sheng_build_strat strat(raw, rm, only_accel_init);
+ dfa_info info(strat);
+
+ DEBUG_PRINTF("Trying to compile a %zu state Sheng\n", raw.states.size());
+
+ DEBUG_PRINTF("Anchored start state id: %u, floating start state id: %u\n",
+ raw.start_anchored, raw.start_floating);
+
+ DEBUG_PRINTF("This DFA %s die so effective number of states is %zu\n",
+ info.can_die ? "can" : "cannot", info.size());
+ if (info.size() > 16) {
+#if defined(HAVE_AVX512VBMI)
+ if (info.size() > 32) {
+ DEBUG_PRINTF("Too many states\n");
+ return nullptr;
+ }
+ return shengCompile_int<sheng32>(raw, cc, accel_states, strat, info);
+#else
+ DEBUG_PRINTF("Too many states\n");
+ return nullptr;
+#endif
+ }
+
+ return shengCompile_int<sheng>(raw, cc, accel_states, strat, info);
+}
+
} // namespace ue2
static
const sstate_aux *get_aux(const NFA *n, dstate_id_t i) {
- assert(n && isShengType(n->type));
+ assert(n && isSheng16Type(n->type));
const sheng *s = (const sheng *)getImplNfa(n);
const sstate_aux *aux_base =
return aux;
}
+#if defined(HAVE_AVX512VBMI)
+static
+const sstate_aux *get_aux32(const NFA *n, dstate_id_t i) {
+ assert(n && isSheng32Type(n->type));
+
+ const sheng32 *s = (const sheng32 *)getImplNfa(n);
+ const sstate_aux *aux_base =
+ (const sstate_aux *)((const char *)n + s->aux_offset);
+
+ const sstate_aux *aux = aux_base + i;
+
+ assert((const char *)aux < (const char *)s + s->length);
+
+ return aux;
+}
+#endif
+
static
void dumpHeader(FILE *f, const sheng *s) {
fprintf(f, "number of states: %u, DFA engine size: %u\n", s->n_states,
!!(s->flags & SHENG_FLAG_SINGLE_REPORT));
}
+#if defined(HAVE_AVX512VBMI)
+static
+void dumpHeader32(FILE *f, const sheng32 *s) {
+ fprintf(f, "number of states: %u, DFA engine size: %u\n", s->n_states,
+ s->length);
+ fprintf(f, "aux base offset: %u, reports base offset: %u, "
+ "accel offset: %u\n",
+ s->aux_offset, s->report_offset, s->accel_offset);
+ fprintf(f, "anchored start state: %u, floating start state: %u\n",
+ s->anchored & SHENG32_STATE_MASK, s->floating & SHENG32_STATE_MASK);
+ fprintf(f, "has accel: %u can die: %u single report: %u\n",
+ !!(s->flags & SHENG_FLAG_HAS_ACCEL),
+ !!(s->flags & SHENG_FLAG_CAN_DIE),
+ !!(s->flags & SHENG_FLAG_SINGLE_REPORT));
+}
+#endif
+
static
void dumpAux(FILE *f, u32 state, const sstate_aux *aux) {
fprintf(f, "state id: %u, reports offset: %u, EOD reports offset: %u, "
aux->top & SHENG_STATE_MASK);
}
+#if defined(HAVE_AVX512VBMI)
+static
+void dumpAux32(FILE *f, u32 state, const sstate_aux *aux) {
+ fprintf(f, "state id: %u, reports offset: %u, EOD reports offset: %u, "
+ "accel offset: %u, top: %u\n",
+ state, aux->accept, aux->accept_eod, aux->accel,
+ aux->top & SHENG32_STATE_MASK);
+}
+#endif
+
static
void dumpReports(FILE *f, const report_list *rl) {
fprintf(f, "reports count: %u\n", rl->count);
}
}
+#if defined(HAVE_AVX512VBMI)
+static
+void dumpMasks32(FILE *f, const sheng32 *s) {
+ //u8 flags[64];
+ //memcpy(flags, &s->flag_mask, sizeof(m512));
+ for (u32 chr = 0; chr < 256; chr++) {
+ u8 buf[64];
+ m512 succ_mask = s->succ_masks[chr];
+ memcpy(buf, &succ_mask, sizeof(m512));
+
+ fprintf(f, "%3u: ", chr);
+ for (u32 pos = 0; pos < 64; pos++) {
+ u8 c = buf[pos];
+ if (c & SHENG32_STATE_FLAG_MASK) {
+ fprintf(f, "%2u* ", c & SHENG32_STATE_MASK);
+ } else {
+ fprintf(f, "%2u ", c & SHENG32_STATE_MASK);
+ }
+ }
+ fprintf(f, "\n");
+ }
+}
+#endif
+
static
void nfaExecSheng_dumpText(const NFA *nfa, FILE *f) {
assert(nfa->type == SHENG_NFA);
fprintf(f, "\n");
}
+#if defined(HAVE_AVX512VBMI)
+static
+void nfaExecSheng32_dumpText(const NFA *nfa, FILE *f) {
+ assert(nfa->type == SHENG_NFA_32);
+ const sheng32 *s = (const sheng32 *)getImplNfa(nfa);
+
+ fprintf(f, "sheng32 DFA\n");
+ dumpHeader32(f, s);
+
+ for (u32 state = 0; state < s->n_states; state++) {
+ const sstate_aux *aux = get_aux32(nfa, state);
+ dumpAux32(f, state, aux);
+ if (aux->accept) {
+ fprintf(f, "report list:\n");
+ const report_list *rl =
+ (const report_list *)((const char *)nfa + aux->accept);
+ dumpReports(f, rl);
+ }
+ if (aux->accept_eod) {
+ fprintf(f, "EOD report list:\n");
+ const report_list *rl =
+ (const report_list *)((const char *)nfa + aux->accept_eod);
+ dumpReports(f, rl);
+ }
+ if (aux->accel) {
+ fprintf(f, "accel:\n");
+ const AccelAux *accel =
+ (const AccelAux *)((const char *)nfa + aux->accel);
+ dumpAccelInfo(f, *accel);
+ }
+ }
+
+ fprintf(f, "\n");
+
+ dumpMasks32(f, s);
+
+ fprintf(f, "\n");
+}
+#endif
+
static
void dumpDotPreambleDfa(FILE *f) {
dumpDotPreamble(f);
fprintf(f, "0 [style=invis];\n");
}
+template <typename T>
static
-void describeNode(const NFA *n, const sheng *s, u16 i, FILE *f) {
+void describeNode(const NFA *n, const T *s, u16 i, FILE *f) {
+}
+
+template <>
+void describeNode<sheng>(const NFA *n, const sheng *s, u16 i, FILE *f) {
const sstate_aux *aux = get_aux(n, i);
fprintf(f, "%u [ width = 1, fixedsize = true, fontsize = 12, "
}
}
+#if defined(HAVE_AVX512VBMI)
+template <>
+void describeNode<sheng32>(const NFA *n, const sheng32 *s, u16 i, FILE *f) {
+ const sstate_aux *aux = get_aux32(n, i);
+
+ fprintf(f, "%u [ width = 1, fixedsize = true, fontsize = 12, "
+ "label = \"%u\" ]; \n",
+ i, i);
+
+ if (aux->accept_eod) {
+ fprintf(f, "%u [ color = darkorchid ];\n", i);
+ }
+
+ if (aux->accept) {
+ fprintf(f, "%u [ shape = doublecircle ];\n", i);
+ }
+
+ if (aux->top && (aux->top & SHENG32_STATE_MASK) != i) {
+ fprintf(f, "%u -> %u [color = darkgoldenrod weight=0.1 ]\n", i,
+ aux->top & SHENG32_STATE_MASK);
+ }
+
+ if (i == (s->anchored & SHENG32_STATE_MASK)) {
+ fprintf(f, "STARTA -> %u [color = blue ]\n", i);
+ }
+
+ if (i == (s->floating & SHENG32_STATE_MASK)) {
+ fprintf(f, "STARTF -> %u [color = red ]\n", i);
+ }
+}
+#endif
+
static
void describeEdge(FILE *f, const u16 *t, u16 i) {
for (u16 s = 0; s < N_CHARS; s++) {
static
void shengGetTransitions(const NFA *n, u16 state, u16 *t) {
- assert(isShengType(n->type));
+ assert(isSheng16Type(n->type));
const sheng *s = (const sheng *)getImplNfa(n);
const sstate_aux *aux = get_aux(n, state);
t[TOP] = aux->top & SHENG_STATE_MASK;
}
+#if defined(HAVE_AVX512VBMI)
+static
+void sheng32GetTransitions(const NFA *n, u16 state, u16 *t) {
+ assert(isSheng32Type(n->type));
+ const sheng32 *s = (const sheng32 *)getImplNfa(n);
+ const sstate_aux *aux = get_aux32(n, state);
+
+ for (unsigned i = 0; i < N_CHARS; i++) {
+ u8 buf[64];
+ m512 succ_mask = s->succ_masks[i];
+
+ memcpy(buf, &succ_mask, sizeof(m512));
+
+ t[i] = buf[state] & SHENG32_STATE_MASK;
+ }
+
+ t[TOP] = aux->top & SHENG32_STATE_MASK;
+}
+#endif
+
static
void nfaExecSheng_dumpDot(const NFA *nfa, FILE *f) {
assert(nfa->type == SHENG_NFA);
dumpDotPreambleDfa(f);
for (u16 i = 1; i < s->n_states; i++) {
- describeNode(nfa, s, i, f);
+ describeNode<sheng>(nfa, s, i, f);
u16 t[ALPHABET_SIZE];
fprintf(f, "}\n");
}
+#if defined(HAVE_AVX512VBMI)
+static
+void nfaExecSheng32_dumpDot(const NFA *nfa, FILE *f) {
+ assert(nfa->type == SHENG_NFA_32);
+ const sheng32 *s = (const sheng32 *)getImplNfa(nfa);
+
+ dumpDotPreambleDfa(f);
+
+ for (u16 i = 1; i < s->n_states; i++) {
+ describeNode<sheng32>(nfa, s, i, f);
+
+ u16 t[ALPHABET_SIZE];
+
+ sheng32GetTransitions(nfa, i, t);
+
+ describeEdge(f, t, i);
+ }
+
+ fprintf(f, "}\n");
+}
+#endif
+
void nfaExecSheng_dump(const NFA *nfa, const string &base) {
assert(nfa->type == SHENG_NFA);
nfaExecSheng_dumpText(nfa, StdioFile(base + ".txt", "w"));
nfaExecSheng_dumpDot(nfa, StdioFile(base + ".dot", "w"));
}
+void nfaExecSheng32_dump(UNUSED const NFA *nfa, UNUSED const string &base) {
+#if defined(HAVE_AVX512VBMI)
+ assert(nfa->type == SHENG_NFA_32);
+ nfaExecSheng32_dumpText(nfa, StdioFile(base + ".txt", "w"));
+ nfaExecSheng32_dumpDot(nfa, StdioFile(base + ".dot", "w"));
+#endif
+}
+
} // namespace ue2
namespace ue2 {
void nfaExecSheng_dump(const struct NFA *nfa, const std::string &base);
+void nfaExecSheng32_dump(const struct NFA *nfa, const std::string &base);
} // namespace ue2
projects / thirdparty / vectorscan.git / commitdiff
