mirror of
https://github.com/VectorCamp/vectorscan.git
synced 2025-06-28 16:41:01 +03:00
49eb18ee4f
This optimization is based on the thread https://twitter.com/Danlark1/status/1539344279268691970 and uses shift right and narrow by 4 instruction https://developer.arm.com/documentation/ddi0596/2020-12/SIMD-FP-Instructions/SHRN--SHRN2--Shift-Right-Narrow--immediate-- To achieve that, I needed to redesign a little movemask into comparemask and have an additional step towards mask iteration. Our benchmarks showed 10-15% improvement on average for long matches.
311 lines
12 KiB
C++
311 lines
12 KiB
C++
/*
|
|
* Copyright (c) 2017, Intel Corporation
|
|
* Copyright (c) 2020-2021, VectorCamp PC
|
|
*
|
|
* 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.
|
|
*/
|
|
|
|
/* SIMD engine agnostic noodle scan parts */
|
|
|
|
#include "util/supervector/supervector.hpp"
|
|
#include "util/supervector/casemask.hpp"
|
|
|
|
static really_really_inline
|
|
hwlm_error_t single_zscan(const struct noodTable *n,const u8 *d, const u8 *buf,
|
|
Z_TYPE z, size_t len, const struct cb_info *cbi) {
|
|
while (unlikely(z)) {
|
|
Z_TYPE pos = JOIN(findAndClearLSB_, Z_BITS)(&z) >> Z_POSSHIFT;
|
|
size_t matchPos = d - buf + pos;
|
|
DEBUG_PRINTF("match pos %zu\n", matchPos);
|
|
hwlmcb_rv_t rv = final(n, buf, len, n->msk_len != 1, cbi, matchPos);
|
|
RETURN_IF_TERMINATED(rv);
|
|
}
|
|
return HWLM_SUCCESS;
|
|
}
|
|
|
|
static really_really_inline
|
|
hwlm_error_t double_zscan(const struct noodTable *n,const u8 *d, const u8 *buf,
|
|
Z_TYPE z, size_t len, const struct cb_info *cbi) {
|
|
while (unlikely(z)) {
|
|
Z_TYPE pos = JOIN(findAndClearLSB_, Z_BITS)(&z) >> Z_POSSHIFT;
|
|
size_t matchPos = d - buf + pos - 1;
|
|
DEBUG_PRINTF("match pos %zu\n", matchPos);
|
|
hwlmcb_rv_t rv = final(n, buf, len, true, cbi, matchPos);
|
|
RETURN_IF_TERMINATED(rv);
|
|
}
|
|
return HWLM_SUCCESS;
|
|
}
|
|
|
|
|
|
template<uint16_t S>
|
|
static really_inline
|
|
hwlm_error_t scanSingleShort(const struct noodTable *n, const u8 *buf,
|
|
SuperVector<S> caseMask, SuperVector<S> mask1,
|
|
const struct cb_info *cbi, size_t len, size_t start,
|
|
size_t end) {
|
|
const u8 *d = buf + start;
|
|
DEBUG_PRINTF("start %zu end %zu\n", start, end);
|
|
const size_t l = end - start;
|
|
DEBUG_PRINTF("l = %ld\n", l);
|
|
//assert(l <= 64);
|
|
if (!l) {
|
|
return HWLM_SUCCESS;
|
|
}
|
|
|
|
SuperVector<S> v = SuperVector<S>::Zeroes();
|
|
memcpy(&v.u, d, l);
|
|
|
|
typename SuperVector<S>::comparemask_type mask =
|
|
SINGLE_LOAD_MASK(l * SuperVector<S>::mask_width());
|
|
v = v & caseMask;
|
|
typename SuperVector<S>::comparemask_type z = mask & mask1.eqmask(v);
|
|
z = SuperVector<S>::iteration_mask(z);
|
|
|
|
return single_zscan(n, d, buf, z, len, cbi);
|
|
}
|
|
|
|
// The short scan routine. It is used both to scan data up to an
|
|
// alignment boundary if needed and to finish off data that the aligned scan
|
|
// function can't handle (due to small/unaligned chunk at end)
|
|
template<uint16_t S>
|
|
static really_inline
|
|
hwlm_error_t scanSingleUnaligned(const struct noodTable *n, const u8 *buf,
|
|
SuperVector<S> caseMask, SuperVector<S> mask1,
|
|
const struct cb_info *cbi, size_t len, size_t offset,
|
|
size_t start,
|
|
size_t end) {
|
|
const u8 *d = buf + offset;
|
|
DEBUG_PRINTF("start %zu end %zu offset %zu\n", start, end, offset);
|
|
const size_t l = end - start;
|
|
DEBUG_PRINTF("l = %ld\n", l);
|
|
assert(l <= 64);
|
|
if (!l) {
|
|
return HWLM_SUCCESS;
|
|
}
|
|
size_t buf_off = start - offset;
|
|
typename SuperVector<S>::comparemask_type mask =
|
|
SINGLE_LOAD_MASK(l * SuperVector<S>::mask_width())
|
|
<< (buf_off * SuperVector<S>::mask_width());
|
|
SuperVector<S> v = SuperVector<S>::loadu(d) & caseMask;
|
|
typename SuperVector<S>::comparemask_type z = mask & mask1.eqmask(v);
|
|
z = SuperVector<S>::iteration_mask(z);
|
|
|
|
return single_zscan(n, d, buf, z, len, cbi);
|
|
}
|
|
|
|
template<uint16_t S>
|
|
static really_inline
|
|
hwlm_error_t scanDoubleShort(const struct noodTable *n, const u8 *buf,
|
|
SuperVector<S> caseMask, SuperVector<S> mask1, SuperVector<S> mask2,
|
|
const struct cb_info *cbi, size_t len, size_t start, size_t end) {
|
|
const u8 *d = buf + start;
|
|
DEBUG_PRINTF("start %zu end %zu\n", start, end);
|
|
const size_t l = end - start;
|
|
assert(l <= S);
|
|
if (!l) {
|
|
return HWLM_SUCCESS;
|
|
}
|
|
SuperVector<S> v = SuperVector<S>::Zeroes();
|
|
memcpy(&v.u, d, l);
|
|
v = v & caseMask;
|
|
|
|
typename SuperVector<S>::comparemask_type mask =
|
|
DOUBLE_LOAD_MASK(l * SuperVector<S>::mask_width());
|
|
typename SuperVector<S>::comparemask_type z1 = mask1.eqmask(v);
|
|
typename SuperVector<S>::comparemask_type z2 = mask2.eqmask(v);
|
|
typename SuperVector<S>::comparemask_type z =
|
|
mask & (z1 << (SuperVector<S>::mask_width())) & z2;
|
|
z = SuperVector<S>::iteration_mask(z);
|
|
|
|
return double_zscan(n, d, buf, z, len, cbi);
|
|
}
|
|
|
|
template<uint16_t S>
|
|
static really_inline
|
|
hwlm_error_t scanDoubleUnaligned(const struct noodTable *n, const u8 *buf,
|
|
SuperVector<S> caseMask, SuperVector<S> mask1, SuperVector<S> mask2,
|
|
const struct cb_info *cbi, size_t len, size_t offset, size_t start, size_t end) {
|
|
const u8 *d = buf + offset;
|
|
DEBUG_PRINTF("start %zu end %zu offset %zu\n", start, end, offset);
|
|
const size_t l = end - start;
|
|
assert(l <= S);
|
|
if (!l) {
|
|
return HWLM_SUCCESS;
|
|
}
|
|
SuperVector<S> v = SuperVector<S>::loadu(d) & caseMask;
|
|
size_t buf_off = start - offset;
|
|
typename SuperVector<S>::comparemask_type mask =
|
|
DOUBLE_LOAD_MASK(l * SuperVector<S>::mask_width())
|
|
<< (buf_off * SuperVector<S>::mask_width());
|
|
typename SuperVector<S>::comparemask_type z1 = mask1.eqmask(v);
|
|
typename SuperVector<S>::comparemask_type z2 = mask2.eqmask(v);
|
|
typename SuperVector<S>::comparemask_type z =
|
|
mask & (z1 << SuperVector<S>::mask_width()) & z2;
|
|
z = SuperVector<S>::iteration_mask(z);
|
|
|
|
return double_zscan(n, d, buf, z, len, cbi);
|
|
}
|
|
|
|
template <uint16_t S>
|
|
static really_inline
|
|
hwlm_error_t scanSingleMain(const struct noodTable *n, const u8 *buf,
|
|
size_t len, size_t offset,
|
|
SuperVector<S> caseMask, SuperVector<S> mask1,
|
|
const struct cb_info *cbi) {
|
|
size_t start = offset + n->msk_len - 1;
|
|
size_t end = len;
|
|
|
|
const u8 *d = buf + start;
|
|
const u8 *e = buf + end;
|
|
DEBUG_PRINTF("start %p end %p \n", d, e);
|
|
assert(d < e);
|
|
if (e - d < S) {
|
|
return scanSingleShort(n, buf, caseMask, mask1, cbi, len, start, end);
|
|
}
|
|
if (d + S <= e) {
|
|
// peel off first part to cacheline boundary
|
|
const u8 *d1 = ROUNDUP_PTR(d, S);
|
|
DEBUG_PRINTF("until aligned %p \n", d1);
|
|
if (scanSingleUnaligned(n, buf, caseMask, mask1, cbi, len, start, start, d1 - buf) == HWLM_TERMINATED) {
|
|
return HWLM_TERMINATED;
|
|
}
|
|
d = d1;
|
|
|
|
size_t loops = (end - (d - buf)) / S;
|
|
DEBUG_PRINTF("loops %ld \n", loops);
|
|
|
|
for (size_t i = 0; i < loops; i++, d+= S) {
|
|
DEBUG_PRINTF("d %p \n", d);
|
|
const u8 *base = ROUNDUP_PTR(d, 64);
|
|
// On large packet buffers, this prefetch appears to get us about 2%.
|
|
__builtin_prefetch(base + 256);
|
|
|
|
SuperVector<S> v = SuperVector<S>::load(d) & caseMask;
|
|
typename SuperVector<S>::comparemask_type z = mask1.eqmask(v);
|
|
z = SuperVector<S>::iteration_mask(z);
|
|
|
|
hwlm_error_t rv = single_zscan(n, d, buf, z, len, cbi);
|
|
RETURN_IF_TERMINATED(rv);
|
|
}
|
|
}
|
|
|
|
DEBUG_PRINTF("d %p e %p \n", d, e);
|
|
// finish off tail
|
|
size_t s2End = ROUNDDOWN_PTR(e, S) - buf;
|
|
if (s2End == end) {
|
|
return HWLM_SUCCESS;
|
|
}
|
|
|
|
return scanSingleUnaligned(n, buf, caseMask, mask1, cbi, len, end - S, s2End, len);
|
|
}
|
|
|
|
template <uint16_t S>
|
|
static really_inline
|
|
hwlm_error_t scanDoubleMain(const struct noodTable *n, const u8 *buf,
|
|
size_t len, size_t offset,
|
|
SuperVector<S> caseMask, SuperVector<S> mask1, SuperVector<S> mask2,
|
|
const struct cb_info *cbi) {
|
|
// we stop scanning for the key-fragment when the rest of the key can't
|
|
// possibly fit in the remaining buffer
|
|
size_t end = len - n->key_offset + 2;
|
|
|
|
size_t start = offset + n->msk_len - n->key_offset;
|
|
|
|
typename SuperVector<S>::comparemask_type lastz1{0};
|
|
|
|
const u8 *d = buf + start;
|
|
const u8 *e = buf + end;
|
|
DEBUG_PRINTF("start %p end %p \n", d, e);
|
|
assert(d < e);
|
|
if (e - d < S) {
|
|
return scanDoubleShort(n, buf, caseMask, mask1, mask2, cbi, len, d - buf, end);
|
|
}
|
|
if (d + S <= e) {
|
|
// peel off first part to cacheline boundary
|
|
const u8 *d1 = ROUNDUP_PTR(d, S) + 1;
|
|
DEBUG_PRINTF("until aligned %p \n", d1);
|
|
if (scanDoubleUnaligned(n, buf, caseMask, mask1, mask2, cbi, len, start, start, d1 - buf) == HWLM_TERMINATED) {
|
|
return HWLM_TERMINATED;
|
|
}
|
|
d = d1 - 1;
|
|
|
|
size_t loops = (end - (d - buf)) / S;
|
|
DEBUG_PRINTF("loops %ld \n", loops);
|
|
|
|
for (size_t i = 0; i < loops; i++, d+= S) {
|
|
DEBUG_PRINTF("d %p \n", d);
|
|
const u8 *base = ROUNDUP_PTR(d, 64);
|
|
// On large packet buffers, this prefetch appears to get us about 2%.
|
|
__builtin_prefetch(base + 256);
|
|
|
|
SuperVector<S> v = SuperVector<S>::load(d) & caseMask;
|
|
typename SuperVector<S>::comparemask_type z1 = mask1.eqmask(v);
|
|
typename SuperVector<S>::comparemask_type z2 = mask2.eqmask(v);
|
|
typename SuperVector<S>::comparemask_type z =
|
|
(z1 << SuperVector<S>::mask_width() | lastz1) & z2;
|
|
lastz1 = z1 >> (Z_SHIFT * SuperVector<S>::mask_width());
|
|
z = SuperVector<S>::iteration_mask(z);
|
|
|
|
hwlm_error_t rv = double_zscan(n, d, buf, z, len, cbi);
|
|
RETURN_IF_TERMINATED(rv);
|
|
}
|
|
if (loops == 0) {
|
|
d = d1;
|
|
}
|
|
}
|
|
// finish off tail
|
|
size_t s2End = ROUNDDOWN_PTR(e, S) - buf;
|
|
if (s2End == end) {
|
|
return HWLM_SUCCESS;
|
|
}
|
|
return scanDoubleUnaligned(n, buf, caseMask, mask1, mask2, cbi, len, end - S, d - buf, end);
|
|
}
|
|
|
|
// Single-character specialisation, used when keyLen = 1
|
|
static really_inline
|
|
hwlm_error_t scanSingle(const struct noodTable *n, const u8 *buf, size_t len,
|
|
size_t start, bool noCase, const struct cb_info *cbi) {
|
|
if (!ourisalpha(n->key0)) {
|
|
noCase = 0; // force noCase off if we don't have an alphabetic char
|
|
}
|
|
|
|
const SuperVector<VECTORSIZE> caseMask{noCase ? getCaseMask<VECTORSIZE>() : SuperVector<VECTORSIZE>::Ones()};
|
|
const SuperVector<VECTORSIZE> mask1{getMask<VECTORSIZE>(n->key0, noCase)};
|
|
|
|
return scanSingleMain(n, buf, len, start, caseMask, mask1, cbi);
|
|
}
|
|
|
|
|
|
static really_inline
|
|
hwlm_error_t scanDouble(const struct noodTable *n, const u8 *buf, size_t len,
|
|
size_t start, bool noCase, const struct cb_info *cbi) {
|
|
|
|
const SuperVector<VECTORSIZE> caseMask{noCase ? getCaseMask<VECTORSIZE>() : SuperVector<VECTORSIZE>::Ones()};
|
|
const SuperVector<VECTORSIZE> mask1{getMask<VECTORSIZE>(n->key0, noCase)};
|
|
const SuperVector<VECTORSIZE> mask2{getMask<VECTORSIZE>(n->key1, noCase)};
|
|
|
|
return scanDoubleMain(n, buf, len, start, caseMask, mask1, mask2, cbi);
|
|
}
|