mirror of
https://github.com/VectorCamp/vectorscan.git
synced 2025-06-28 16:41:01 +03:00
938c026256
256b wide SVE vectors allow some simplification of truffle. Up to 40% speedup on graviton3. Going from 12500 MB/s to 17000 MB/s onhe microbenchmark. SVE2 also offer this capability for 128b vector with a speedup around 25% compared to normal SVE Add unit tests and benchmark for this wide variant Signed-off-by: Yoan Picchi <yoan.picchi@arm.com>
463 lines
14 KiB
C++
463 lines
14 KiB
C++
/*
|
|
* Copyright (c) 2015-2017, Intel Corporation
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions are met:
|
|
*
|
|
* * Redistributions of source code must retain the above copyright notice,
|
|
* this list of conditions and the following disclaimer.
|
|
* * Redistributions in binary form must reproduce the above copyright
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
* documentation and/or other materials provided with the distribution.
|
|
* * Neither the name of Intel Corporation nor the names of its contributors
|
|
* may be used to endorse or promote products derived from this software
|
|
* without specific prior written permission.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
|
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
|
|
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
|
|
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
|
|
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
|
|
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
|
|
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
|
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
|
* POSSIBILITY OF SUCH DAMAGE.
|
|
*/
|
|
|
|
#include "config.h"
|
|
|
|
#include "mcclellandump.h"
|
|
|
|
#include "accel.h"
|
|
#include "accel_dump.h"
|
|
#include "nfa_dump_internal.h"
|
|
#include "nfa_internal.h"
|
|
#include "mcclellan_internal.h"
|
|
#include "rdfa.h"
|
|
#include "ue2common.h"
|
|
#include "util/charreach.h"
|
|
#include "util/dump_charclass.h"
|
|
#include "util/dump_util.h"
|
|
#include "util/unaligned.h"
|
|
|
|
#include <cctype>
|
|
#include <cstdio>
|
|
#include <cstdlib>
|
|
#include <cstring>
|
|
#include <map>
|
|
#include <set>
|
|
#include <vector>
|
|
|
|
#ifndef DUMP_SUPPORT
|
|
#error No dump support!
|
|
#endif
|
|
|
|
using namespace std;
|
|
|
|
namespace ue2 {
|
|
|
|
const mstate_aux *getAux(const NFA *n, dstate_id_t i) {
|
|
assert(n && isDfaType(n->type));
|
|
|
|
const mcclellan *m = reinterpret_cast<const mcclellan *>(getImplNfa(n));
|
|
const mstate_aux *aux_base
|
|
= reinterpret_cast<const mstate_aux *>(reinterpret_cast<const char *>(n) + m->aux_offset);
|
|
|
|
const mstate_aux *aux = aux_base + i;
|
|
|
|
assert(reinterpret_cast<const char *>(aux) < reinterpret_cast<const char *>(n) + m->length);
|
|
return aux;
|
|
}
|
|
|
|
static
|
|
void mcclellanGetTransitions(const NFA *n, u16 s, u16 *t) {
|
|
assert(isMcClellanType(n->type));
|
|
const mcclellan *m = reinterpret_cast<const mcclellan *>(getImplNfa(n));
|
|
const mstate_aux *aux = getAux(n, s);
|
|
const u32 as = m->alphaShift;
|
|
|
|
if (n->type == MCCLELLAN_NFA_8) {
|
|
const u8 *succ_table = reinterpret_cast<const u8 *>(reinterpret_cast<const char *>(m)
|
|
+ sizeof(mcclellan));
|
|
for (u16 c = 0; c < N_CHARS; c++) {
|
|
t[c] = succ_table[((u32)s << as) + m->remap[c]];
|
|
}
|
|
} else {
|
|
u16 base_s = s;
|
|
const char *winfo_base = reinterpret_cast<const char *>(n) + m->sherman_offset;
|
|
const char *state_base
|
|
= winfo_base + SHERMAN_FIXED_SIZE * (s - m->sherman_limit);
|
|
|
|
if (s >= m->sherman_limit) {
|
|
base_s = unaligned_load_u16(state_base + SHERMAN_DADDY_OFFSET);
|
|
}
|
|
|
|
const u16 *succ_table = reinterpret_cast<const u16 *>(reinterpret_cast<const char *>(m)
|
|
+ sizeof(mcclellan));
|
|
for (u16 c = 0; c < N_CHARS; c++) {
|
|
const u8 *addr = reinterpret_cast<const u8*>(succ_table + (((u32)base_s << as)
|
|
+ m->remap[c]));
|
|
t[c] = unaligned_load_u16(addr);
|
|
t[c] &= STATE_MASK;
|
|
}
|
|
|
|
if (s >= m->sherman_limit) {
|
|
UNUSED char type = *(state_base + SHERMAN_TYPE_OFFSET);
|
|
assert(type == SHERMAN_STATE);
|
|
u8 len = *(reinterpret_cast<const u8 *>(SHERMAN_LEN_OFFSET + state_base));
|
|
const char *chars = state_base + SHERMAN_CHARS_OFFSET;
|
|
const u16 *states = reinterpret_cast<const u16 *>(state_base
|
|
+ SHERMAN_STATES_OFFSET(len));
|
|
|
|
for (u8 i = 0; i < len; i++) {
|
|
for (u16 c = 0; c < N_CHARS; c++) {
|
|
if (m->remap[c] == chars[i]) {
|
|
t[c] = unaligned_load_u16(reinterpret_cast<const u8*>(&states[i])) & STATE_MASK;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
t[TOP] = aux->top & STATE_MASK;
|
|
}
|
|
|
|
void describeEdge(FILE *f, const u16 *t, u16 i) {
|
|
for (u16 s = 0; s < N_CHARS; s++) {
|
|
if (!t[s]) {
|
|
continue;
|
|
}
|
|
|
|
u16 ss;
|
|
for (ss = 0; ss < s; ss++) {
|
|
if (t[s] == t[ss]) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (ss != s) {
|
|
continue;
|
|
}
|
|
|
|
CharReach reach;
|
|
for (ss = s; ss < 256; ss++) {
|
|
if (t[s] == t[ss]) {
|
|
reach.set(ss);
|
|
}
|
|
}
|
|
|
|
fprintf(f, "%u -> %u [ label = \"", i, t[s]);
|
|
|
|
describeClass(f, reach, 5, CC_OUT_DOT);
|
|
|
|
fprintf(f, "\" ];\n");
|
|
}
|
|
}
|
|
|
|
void dumpAccelText(FILE *f, const union AccelAux *accel) {
|
|
switch(accel->accel_type) {
|
|
case ACCEL_NONE:
|
|
break;
|
|
case ACCEL_VERM:
|
|
fprintf(f, ":V");
|
|
break;
|
|
case ACCEL_DVERM:
|
|
fprintf(f, ":VV");
|
|
break;
|
|
case ACCEL_VERM_NOCASE:
|
|
fprintf(f, ":VN");
|
|
break;
|
|
case ACCEL_DVERM_NOCASE:
|
|
fprintf(f, ":VVN");
|
|
break;
|
|
case ACCEL_SHUFTI:
|
|
fprintf(f, ":S");
|
|
break;
|
|
case ACCEL_DSHUFTI:
|
|
fprintf(f, ":SS");
|
|
break;
|
|
case ACCEL_TRUFFLE:
|
|
fprintf(f, ":M");
|
|
break;
|
|
case ACCEL_TRUFFLE_WIDE:
|
|
fprintf(f, ":MM");
|
|
break;
|
|
default:
|
|
fprintf(f, ":??");
|
|
break;
|
|
}
|
|
}
|
|
|
|
void dumpAccelDot(FILE *f, u16 i, const union AccelAux *accel) {
|
|
switch(accel->accel_type) {
|
|
case ACCEL_NONE:
|
|
break;
|
|
case ACCEL_VERM:
|
|
case ACCEL_VERM_NOCASE:
|
|
case ACCEL_DVERM:
|
|
case ACCEL_DVERM_NOCASE:
|
|
fprintf(f, "%u [ color = forestgreen style=diagonals];\n", i);
|
|
break;
|
|
case ACCEL_SHUFTI:
|
|
case ACCEL_DSHUFTI:
|
|
case ACCEL_TRUFFLE:
|
|
case ACCEL_TRUFFLE_WIDE:
|
|
fprintf(f, "%u [ color = darkgreen style=diagonals ];\n", i);
|
|
break;
|
|
default:
|
|
fprintf(f, "%u [ color = yellow style=diagonals ];\n", i);
|
|
break;
|
|
}
|
|
}
|
|
|
|
static
|
|
void describeNode(const NFA *n, const mcclellan *m, u16 i, FILE *f) {
|
|
const mstate_aux *aux = getAux(n, i);
|
|
|
|
bool isSherman = m->sherman_limit && i >= m->sherman_limit;
|
|
|
|
fprintf(f, "%u [ width = 1, fixedsize = true, fontsize = 12, "
|
|
"label = \"%u%s\" ]; \n", i, i, isSherman ? "w":"");
|
|
|
|
if (aux->accel_offset) {
|
|
dumpAccelDot(f, i, reinterpret_cast<const union AccelAux *>
|
|
(reinterpret_cast<const char *>(m) + aux->accel_offset));
|
|
}
|
|
|
|
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 != i) {
|
|
fprintf(f, "%u -> %u [color = darkgoldenrod weight=0.1 ]\n", i,
|
|
aux->top);
|
|
}
|
|
|
|
if (i == m->start_anchored) {
|
|
fprintf(f, "STARTA -> %u [color = blue ]\n", i);
|
|
}
|
|
|
|
if (i == m->start_floating) {
|
|
fprintf(f, "STARTF -> %u [color = red ]\n", i);
|
|
}
|
|
|
|
if (isSherman) {
|
|
const char *winfo_base = reinterpret_cast<const char *>(n) + m->sherman_offset;
|
|
const char *state_base
|
|
= winfo_base + SHERMAN_FIXED_SIZE * (i - m->sherman_limit);
|
|
assert(state_base < reinterpret_cast<const char *>(m) + m->length - sizeof(NFA));
|
|
UNUSED u8 type = *(reinterpret_cast<const u8 *>(state_base + SHERMAN_TYPE_OFFSET));
|
|
assert(type == SHERMAN_STATE);
|
|
fprintf(f, "%u [ fillcolor = lightblue style=filled ];\n", i);
|
|
u16 daddy = *(reinterpret_cast<const u16 *>(state_base + SHERMAN_DADDY_OFFSET));
|
|
if (daddy) {
|
|
fprintf(f, "%u -> %u [ color=royalblue style=dashed weight=0.1]\n",
|
|
i, daddy);
|
|
}
|
|
}
|
|
}
|
|
|
|
void dumpDotPreambleDfa(FILE *f) {
|
|
dumpDotPreamble(f);
|
|
|
|
// DFA specific additions.
|
|
fprintf(f, "STARTF [style=invis];\n");
|
|
fprintf(f, "STARTA [style=invis];\n");
|
|
fprintf(f, "0 [style=invis];\n");
|
|
}
|
|
|
|
static
|
|
void nfaExecMcClellan16_dumpDot(const NFA *nfa, FILE *f) {
|
|
assert(nfa->type == MCCLELLAN_NFA_16);
|
|
const mcclellan *m = reinterpret_cast<const mcclellan *>(getImplNfa(nfa));
|
|
|
|
dumpDotPreambleDfa(f);
|
|
|
|
u16 sherman_ceil = m->has_wide == 1 ? m->wide_limit : m->state_count;
|
|
for (u16 i = 1; i < sherman_ceil; i++) {
|
|
describeNode(nfa, m, i, f);
|
|
|
|
u16 t[ALPHABET_SIZE];
|
|
|
|
mcclellanGetTransitions(nfa, i, t);
|
|
|
|
describeEdge(f, t, i);
|
|
}
|
|
|
|
fprintf(f, "}\n");
|
|
}
|
|
|
|
static
|
|
void nfaExecMcClellan8_dumpDot(const NFA *nfa, FILE *f) {
|
|
assert(nfa->type == MCCLELLAN_NFA_8);
|
|
const mcclellan *m = reinterpret_cast<const mcclellan *>(getImplNfa(nfa));
|
|
|
|
dumpDotPreambleDfa(f);
|
|
|
|
for (u16 i = 1; i < m->state_count; i++) {
|
|
describeNode(nfa, m, i, f);
|
|
|
|
u16 t[ALPHABET_SIZE];
|
|
|
|
mcclellanGetTransitions(nfa, i, t);
|
|
|
|
describeEdge(f, t, i);
|
|
}
|
|
|
|
fprintf(f, "}\n");
|
|
}
|
|
|
|
static
|
|
void dumpAccelMasks(FILE *f, const mcclellan *m, const mstate_aux *aux) {
|
|
fprintf(f, "\n");
|
|
fprintf(f, "Acceleration\n");
|
|
fprintf(f, "------------\n");
|
|
|
|
u16 sherman_ceil = m->has_wide == 1 ? m->wide_limit : m->state_count;
|
|
for (u16 i = 0; i < sherman_ceil; i++) {
|
|
if (!aux[i].accel_offset) {
|
|
continue;
|
|
}
|
|
|
|
const AccelAux *accel = reinterpret_cast<const AccelAux *>(reinterpret_cast<const char *>(m)
|
|
+ aux[i].accel_offset);
|
|
fprintf(f, "%05hu ", i);
|
|
dumpAccelInfo(f, *accel);
|
|
}
|
|
}
|
|
|
|
void describeAlphabet(FILE *f, const mcclellan *m) {
|
|
map<u8, CharReach> rev;
|
|
|
|
for (u16 i = 0; i < N_CHARS; i++) {
|
|
rev[m->remap[i]].clear();
|
|
}
|
|
|
|
for (u16 i = 0; i < N_CHARS; i++) {
|
|
rev[m->remap[i]].set(i);
|
|
}
|
|
|
|
map<u8, CharReach>::const_iterator it;
|
|
fprintf(f, "\nAlphabet\n");
|
|
for (it = rev.begin(); it != rev.end(); ++it) {
|
|
fprintf(f, "%3hhu: ", it->first);
|
|
describeClass(f, it->second, 10240, CC_OUT_TEXT);
|
|
fprintf(f, "\n");
|
|
}
|
|
fprintf(f, "\n");
|
|
}
|
|
|
|
static
|
|
void dumpCommonHeader(FILE *f, const mcclellan *m) {
|
|
fprintf(f, "report: %u, states: %u, length: %u\n", m->arb_report,
|
|
m->state_count, m->length);
|
|
fprintf(f, "astart: %hu, fstart: %hu\n", m->start_anchored,
|
|
m->start_floating);
|
|
fprintf(f, "single accept: %d, has_accel: %d\n",
|
|
!!(int)m->flags & MCCLELLAN_FLAG_SINGLE, m->has_accel);
|
|
}
|
|
|
|
static
|
|
void dumpTransitions(FILE *f, const NFA *nfa, const mcclellan *m,
|
|
const mstate_aux *aux) {
|
|
u16 sherman_ceil = m->has_wide == 1 ? m->wide_limit : m->state_count;
|
|
for (u16 i = 0; i < sherman_ceil; i++) {
|
|
fprintf(f, "%05hu", i);
|
|
if (aux[i].accel_offset) {
|
|
dumpAccelText(f, reinterpret_cast<const union AccelAux *>(reinterpret_cast<const char *>(m) +
|
|
aux[i].accel_offset));
|
|
}
|
|
|
|
u16 trans[ALPHABET_SIZE];
|
|
mcclellanGetTransitions(nfa, i, trans);
|
|
|
|
int rstart = 0;
|
|
u16 prev = 0xffff;
|
|
for (int j = 0; j < N_CHARS; j++) {
|
|
u16 curr = trans[j];
|
|
if (curr == prev) {
|
|
continue;
|
|
}
|
|
|
|
if (prev != 0xffff) {
|
|
if (j == rstart + 1) {
|
|
fprintf(f, " %02x->%hu", rstart, prev);
|
|
} else {
|
|
fprintf(f, " [%02x - %02x]->%hu", rstart, j - 1, prev);
|
|
}
|
|
}
|
|
|
|
prev = curr;
|
|
rstart = j;
|
|
}
|
|
if (N_CHARS == rstart + 1) {
|
|
fprintf(f, " %02x->%hu", rstart, prev);
|
|
} else {
|
|
fprintf(f, " [%02x - %02x]->%hu", rstart, N_CHARS - 1, prev);
|
|
}
|
|
fprintf(f, "\n");
|
|
}
|
|
}
|
|
|
|
static
|
|
void nfaExecMcClellan16_dumpText(const NFA *nfa, FILE *f) {
|
|
assert(nfa->type == MCCLELLAN_NFA_16);
|
|
const mcclellan *m = reinterpret_cast<const mcclellan *>(getImplNfa(nfa));
|
|
const mstate_aux *aux =
|
|
reinterpret_cast<const mstate_aux *>(reinterpret_cast<const char *>(nfa) + m->aux_offset);
|
|
|
|
fprintf(f, "mcclellan 16\n");
|
|
dumpCommonHeader(f, m);
|
|
fprintf(f, "sherman_limit: %d, sherman_end: %d\n", (int)m->sherman_limit,
|
|
(int)m->sherman_end);
|
|
fprintf(f, "\n");
|
|
|
|
describeAlphabet(f, m);
|
|
dumpTransitions(f, nfa, m, aux);
|
|
dumpAccelMasks(f, m, aux);
|
|
|
|
fprintf(f, "\n");
|
|
dumpTextReverse(nfa, f);
|
|
}
|
|
|
|
static
|
|
void nfaExecMcClellan8_dumpText(const NFA *nfa, FILE *f) {
|
|
assert(nfa->type == MCCLELLAN_NFA_8);
|
|
const mcclellan *m = reinterpret_cast<const mcclellan *>(getImplNfa(nfa));
|
|
const mstate_aux *aux =
|
|
reinterpret_cast<const mstate_aux *>(reinterpret_cast<const char *>(nfa) + m->aux_offset);
|
|
|
|
fprintf(f, "mcclellan 8\n");
|
|
dumpCommonHeader(f, m);
|
|
fprintf(f, "accel_limit: %hu, accept_limit %hu\n", m->accel_limit_8,
|
|
m->accept_limit_8);
|
|
fprintf(f, "\n");
|
|
|
|
describeAlphabet(f, m);
|
|
dumpTransitions(f, nfa, m, aux);
|
|
dumpAccelMasks(f, m, aux);
|
|
|
|
fprintf(f, "\n");
|
|
dumpTextReverse(nfa, f);
|
|
}
|
|
|
|
void nfaExecMcClellan16_dump(const NFA *nfa, const string &base) {
|
|
assert(nfa->type == MCCLELLAN_NFA_16);
|
|
nfaExecMcClellan16_dumpText(nfa, StdioFile(base + ".txt", "w"));
|
|
nfaExecMcClellan16_dumpDot(nfa, StdioFile(base + ".dot", "w"));
|
|
}
|
|
|
|
void nfaExecMcClellan8_dump(const NFA *nfa, const string &base) {
|
|
assert(nfa->type == MCCLELLAN_NFA_8);
|
|
nfaExecMcClellan8_dumpText(nfa, StdioFile(base + ".txt", "w"));
|
|
nfaExecMcClellan8_dumpDot(nfa, StdioFile(base + ".dot", "w"));
|
|
}
|
|
|
|
} // namespace ue2
|