simdjson-simdjson/stage1_find_marks.cpp

#include <x86intrin.h>
#include <assert.h>

#include "common_defs.h"
#include "simdjson_internal.h"

using namespace std;

// a straightforward comparison of a mask against input. 5 uops; would be cheaper in AVX512.
really_inline u64 cmp_mask_against_input(m256 input_lo, m256 input_hi, m256 mask) {
    m256 cmp_res_0 = _mm256_cmpeq_epi8(input_lo, mask);
    u64 res_0 = (u32)_mm256_movemask_epi8(cmp_res_0);
    m256 cmp_res_1 = _mm256_cmpeq_epi8(input_hi, mask);
    u64 res_1 = _mm256_movemask_epi8(cmp_res_1);
    return res_0 | (res_1 << 32);
}

/*never_inline*/ bool find_structural_bits(const u8 * buf, size_t len, ParsedJson & pj) {
    // Useful constant masks
    const u64 even_bits = 0x5555555555555555ULL;
    const u64 odd_bits = ~even_bits;

    // for now, just work in 64-byte chunks
    // we have padded the input out to 64 byte multiple with the remainder being zeros

    // persistent state across loop
    u64 prev_iter_ends_odd_backslash = 0ULL; // either 0 or 1, but a 64-bit value
    u64 prev_iter_inside_quote = 0ULL; // either all zeros or all ones
    u64 prev_iter_ends_pseudo_pred = 0ULL;

    for (size_t idx = 0; idx < len; idx+=64) {
        __builtin_prefetch(buf + idx + 128);
#ifdef DEBUG
        cout << "Idx is " << idx << "\n";
        for (u32 j = 0; j < 64; j++) {
            char c = *(buf+idx+j);
            if (isprint(c)) {
                cout << c;
            } else {
                cout << '_';
            }
        }
        cout << "|  ... input\n";
#endif
        m256 input_lo = _mm256_load_si256((const m256 *)(buf + idx + 0));
        m256 input_hi = _mm256_load_si256((const m256 *)(buf + idx + 32));

        ////////////////////////////////////////////////////////////////////////////////////////////
        //     Step 1: detect odd sequences of backslashes
        ////////////////////////////////////////////////////////////////////////////////////////////

        u64 bs_bits = cmp_mask_against_input(input_lo, input_hi, _mm256_set1_epi8('\\'));
        dumpbits(bs_bits, "backslash bits");
        u64 start_edges = bs_bits & ~(bs_bits << 1);
        dumpbits(start_edges, "start_edges");

        // flip lowest if we have an odd-length run at the end of the prior iteration
        u64 even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
        u64 even_starts = start_edges & even_start_mask;
        u64 odd_starts = start_edges & ~even_start_mask;

        dumpbits(even_starts, "even_starts");
        dumpbits(odd_starts, "odd_starts");

        u64 even_carries = bs_bits + even_starts;

        u64 odd_carries;
        // must record the carry-out of our odd-carries out of bit 63; this indicates whether the
        // sense of any edge going to the next iteration should be flipped
        bool iter_ends_odd_backslash = __builtin_uaddll_overflow(bs_bits, odd_starts, &odd_carries);

        odd_carries |= prev_iter_ends_odd_backslash; // push in bit zero as a potential end
                                                     // if we had an odd-numbered run at the end of
                                                     // the previous iteration
        prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1ULL : 0x0ULL;

        dumpbits(even_carries, "even_carries");
        dumpbits(odd_carries, "odd_carries");

        u64 even_carry_ends = even_carries & ~bs_bits;
        u64 odd_carry_ends = odd_carries & ~bs_bits;
        dumpbits(even_carry_ends, "even_carry_ends");
        dumpbits(odd_carry_ends, "odd_carry_ends");

        u64 even_start_odd_end = even_carry_ends & odd_bits;
        u64 odd_start_even_end = odd_carry_ends & even_bits;
        dumpbits(even_start_odd_end, "esoe");
        dumpbits(odd_start_even_end, "osee");

        u64 odd_ends = even_start_odd_end | odd_start_even_end;
        dumpbits(odd_ends, "odd_ends");

        ////////////////////////////////////////////////////////////////////////////////////////////
        //     Step 2: detect insides of quote pairs
        ////////////////////////////////////////////////////////////////////////////////////////////

        u64 quote_bits = cmp_mask_against_input(input_lo, input_hi, _mm256_set1_epi8('"'));
        quote_bits = quote_bits & ~odd_ends;
        dumpbits(quote_bits, "quote_bits");
        u64 quote_mask = _mm_cvtsi128_si64(_mm_clmulepi64_si128(_mm_set_epi64x(0ULL, quote_bits),
                                                                _mm_set1_epi8(0xFF), 0));
        quote_mask ^= prev_iter_inside_quote;
        prev_iter_inside_quote = (u64)((s64)quote_mask>>63);
        dumpbits(quote_mask, "quote_mask");

        // How do we build up a user traversable data structure
        // first, do a 'shufti' to detect structural JSON characters
        // they are { 0x7b } 0x7d : 0x3a [ 0x5b ] 0x5d , 0x2c
        // these go into the first 3 buckets of the comparison (1/2/4)

        // we are also interested in the four whitespace characters
        // space 0x20, linefeed 0x0a, horizontal tab 0x09 and carriage return 0x0d
        // these go into the next 2 buckets of the comparison (8/16)
        const m256 low_nibble_mask = _mm256_setr_epi8(
        //  0                           9  a   b  c  d
            16, 0, 0, 0, 0, 0, 0, 0, 0, 8, 12, 1, 2, 9, 0, 0,
            16, 0, 0, 0, 0, 0, 0, 0, 0, 8, 12, 1, 2, 9, 0, 0
        );
        const m256 high_nibble_mask = _mm256_setr_epi8(
        //  0     2   3     5     7
            8, 0, 18, 4, 0, 1, 0, 1, 0, 0, 0, 3, 2, 1, 0, 0,
            8, 0, 18, 4, 0, 1, 0, 1, 0, 0, 0, 3, 2, 1, 0, 0
        );

        m256 structural_shufti_mask = _mm256_set1_epi8(0x7);
        m256 whitespace_shufti_mask = _mm256_set1_epi8(0x18);

        m256 v_lo = _mm256_and_si256(
                        _mm256_shuffle_epi8(low_nibble_mask, input_lo),
                        _mm256_shuffle_epi8(high_nibble_mask,
                           _mm256_and_si256(_mm256_srli_epi32(input_lo, 4), _mm256_set1_epi8(0x7f))));

        m256 v_hi = _mm256_and_si256(
                        _mm256_shuffle_epi8(low_nibble_mask, input_hi),
                        _mm256_shuffle_epi8(high_nibble_mask,
                           _mm256_and_si256(_mm256_srli_epi32(input_hi, 4), _mm256_set1_epi8(0x7f))));
        m256 tmp_lo = _mm256_cmpeq_epi8(_mm256_and_si256(v_lo, structural_shufti_mask),
                                        _mm256_set1_epi8(0));
        m256 tmp_hi = _mm256_cmpeq_epi8(_mm256_and_si256(v_hi, structural_shufti_mask),
                                        _mm256_set1_epi8(0));

        u64 structural_res_0 = (u32)_mm256_movemask_epi8(tmp_lo);
        u64 structural_res_1 = _mm256_movemask_epi8(tmp_hi);
        u64 structurals =  ~(structural_res_0 | (structural_res_1 << 32));

        // this additional mask and transfer is non-trivially expensive, unfortunately
        m256 tmp_ws_lo = _mm256_cmpeq_epi8(_mm256_and_si256(v_lo, whitespace_shufti_mask),
                                        _mm256_set1_epi8(0));
        m256 tmp_ws_hi = _mm256_cmpeq_epi8(_mm256_and_si256(v_hi, whitespace_shufti_mask),
                                        _mm256_set1_epi8(0));

        u64 ws_res_0 = (u32)_mm256_movemask_epi8(tmp_ws_lo);
        u64 ws_res_1 = _mm256_movemask_epi8(tmp_ws_hi);
        u64 whitespace =  ~(ws_res_0 | (ws_res_1 << 32));

        dumpbits(structurals, "structurals");
        dumpbits(whitespace, "whitespace");

        // mask off anything inside quotes
        structurals &= ~quote_mask;

        // add the real quote bits back into our bitmask as well, so we can
        // quickly traverse the strings we've spent all this trouble gathering
        structurals |= quote_bits;

        // Now, establish "pseudo-structural characters". These are non-whitespace characters
        // that are (a) outside quotes and (b) have a predecessor that's either whitespace or a structural
        // character. This means that subsequent passes will get a chance to encounter the first character
        // of every string of non-whitespace and, if we're parsing an atom like true/false/null or a number
        // we can stop at the first whitespace or structural character following it.

        // a qualified predecessor is something that can happen 1 position before an
        // psuedo-structural character
        u64 pseudo_pred = structurals | whitespace;
        dumpbits(pseudo_pred, "pseudo_pred");
        u64 shifted_pseudo_pred = (pseudo_pred << 1) | prev_iter_ends_pseudo_pred;
        dumpbits(shifted_pseudo_pred, "shifted_pseudo_pred");
        prev_iter_ends_pseudo_pred = pseudo_pred >> 63;
        u64 pseudo_structurals = shifted_pseudo_pred & (~whitespace) & (~quote_mask);
        dumpbits(pseudo_structurals, "pseudo_structurals");
        dumpbits(structurals, "final structurals without pseudos");
        structurals |= pseudo_structurals;
        dumpbits(structurals, "final structurals and pseudo structurals");

        // now, we've used our close quotes all we need to. So let's switch them off
        // they will be off in the quote mask and on in quote bits.
        structurals &= ~(quote_bits & ~quote_mask);
        dumpbits(structurals, "final structurals and pseudo structurals after close quote removal");
        *(u64 *)(pj.structurals + idx/8) = structurals;
    }
    return true;
}