geofflangdale

uint64_t neonmovemask_bulk(uint8x16_t p0, uint8x16_t p1, uint8x16_t p2, uint8x16_t p3) {
  const uint8x16_t bitmask1 = { 0x01, 0x10, 0x01, 0x10, 0x01, 0x10, 0x01, 0x10,
                                0x01, 0x10, 0x01, 0x10, 0x01, 0x10, 0x01, 0x10};
  const uint8x16_t bitmask2 = { 0x02, 0x20, 0x02, 0x20, 0x02, 0x20, 0x02, 0x20,
                                0x02, 0x20, 0x02, 0x20, 0x02, 0x20, 0x02, 0x20};
  const uint8x16_t bitmask3 = { 0x04, 0x40, 0x04, 0x40, 0x04, 0x40, 0x04, 0x40,
                                0x04, 0x40, 0x04, 0x40, 0x04, 0x40, 0x04, 0x40};
  const uint8x16_t bitmask4 = { 0x08, 0x80, 0x08, 0x80, 0x08, 0x80, 0x08, 0x80,
                                0x08, 0x80, 0x08, 0x80, 0x08, 0x80, 0x08, 0x80};

geofflangdale

uint64_t neonmovemask_bulk(uint8x16_t p0, uint8x16_t p1, uint8x16_t p2, uint8x16_t p3) {
  const uint8x16_t bitmask = { 0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80,
                               0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80};
  uint8x16_t t0 = vandq_u8(p0, bitmask);
  uint8x16_t t1 = vandq_u8(p1, bitmask);
  uint8x16_t t2 = vandq_u8(p2, bitmask);
  uint8x16_t t3 = vandq_u8(p3, bitmask);
  uint8x16_t sum0 = vpaddq_u8(t0, t1);
  uint8x16_t sum1 = vpaddq_u8(t2, t3);
  sum0 = vpaddq_u8(sum0, sum1);

geofflangdale

        ////////////////////////////////////////////////////////////////////////////////////////////
        //     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");
        // pdep pattern is alternating 0 and 1 bits, starting with 0 or 1 depending on whether
        // we're in a quote-pair from the previous iteration
        u64 pdep_pattern = even_bits ^ prev_iter_inside_quote;

geofflangdale

really_inline uint64_t find_quote_mask_and_bits(
    __m256i input_lo, __m256i input_hi, uint64_t odd_ends,
    uint64_t &prev_iter_inside_quote, uint64_t &quote_bits) {
  quote_bits =
      cmp_mask_against_input(input_lo, input_hi, _mm256_set1_epi8('"'));
  quote_bits = quote_bits & ~odd_ends;
  uint64_t 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;
  // right shift of a signed value expected to be well-defined and standard

geofflangdale

template <typename T>
void match_multiple_smh(T & smh, std::vector<u8 *> & buffers, std::vector<size_t> & lengths,
                       std::vector<u32> & results) {
        u32 i = 0;
#ifndef NO_UNROLL
        for (; i+7 < buffers.size(); i+=8) {
            results[i+0] = smh.match(buffers[i+0], lengths[i+0]); LFENCE
            results[i+1] = smh.match(buffers[i+1], lengths[i+1]); LFENCE
            results[i+2] = smh.match(buffers[i+2], lengths[i+2]); LFENCE
            results[i+3] = smh.match(buffers[i+3], lengths[i+3]); LFENCE

geofflangdale

template <typename T>
void match_multiple_smh_latency_test(T & smh, std::vector<u8 *> & buffers, std::vector<size_t> & lengths,
                       std::vector<u32> & results) {
        u32 i = 0;
        u32 tmp = 0;
#ifndef NO_UNROLL
        // NOTE: experimental code only. Note that the addition of 'tmp' - being the id of a possible
        // match - could take us RIGHT outside our buffer if we actually matched something. We aren't
        // in this particular run, but so it goes. Saner would be to build up an all-zero id vector
        for (; i+7 < buffers.size(); i+=8) {

geofflangdale

109 111 117 115:101   0   0   0|  0   0   0   0:  0   0   0   0|109 111 117 115:101   0   0   0|  0   0   0   0:  0   0   0   0| input
  0   1   2   3:  4 128   0   1|  2   3   4 128:  0   1   2 128|  0   1   2 128:128 128 128 128|128 128 128 128:128 128 128 128| shuf_mask
109 111 117 115:101   0 109 111|117 115 101   0:109 111 117   0|109 111 117   0:  0   0   0   0|  0   0   0   0:  0   0   0   0| shuf result
109 111 111 115:101 255 109 111|117 115 101 255: 99  97 116 255|100 111 103 255:255 255 255 255|255 255 255 255:255 255 255 255| cmp_mask
255 255   0 255:255   0 255 255|255 255 255   0:  0   0   0   0|  0 255   0   0:  0   0   0   0|  0   0   0   0:  0   0   0   0| cmp result

11_11_11111______1______________________________________________ input to gpr-smh
_____1_____1___1___1____________________________________________ hi
1_____1_____1___1_______________________________________________ low
__111______11___11______________________________________________ after_add

geofflangdale

struct SIMD_SMH_PART {
    m256 shuf_mask;
    m256 cmp_mask;
    m256 and_mask; // not yet used
    m256 sub_mask; // not yet used
    u32 doit(m256 d) {
        return _mm256_movemask_epi8(
                   _mm256_cmpeq_epi8(_mm256_shuffle_epi8(d, shuf_mask),
                                     cmp_mask));
    }   

geofflangdale

$ sudo nice --20 taskset -c 1 ./sheng

Sheng
0/1  1/1  2/1  3/1  4/1  5/1  6/1  7/1  8/1  9/1  
10/1  11/1  12/1  13/1  14/1  15/1  16/1  17/1  18/1  19/1  
20/1  21/1  22/1  23/1  24/1  25/2  26/3  27/4  28/5  29/5  
30/5  31/5  32/5  33/5  34/5  35/5  36/5  37/5  38/5  39/5  
40/5  41/5  42/5  43/5  44/5  45/5  46/5  47/5  48/5  49/5  
50/5  51/5  52/5  53/5  54/5  55/5  56/5  57/5  58/5  59/5  
60/6  61/7  62/8  63/9  64/10  65/1  66/1  67/1  68/1  

geofflangdale

struct Sheng {
    typedef m128 State;
    m128 transitions[256];
    State start_state;

    Sheng(std::vector<std::tuple<u32, u32, u8>> & trans_vec, u8 start_state_, u8 default_state) {
        // fill all transitions with default state
        for (u32 i = 0; i < 256; ++i) {
            transitions[i] = _mm_set1_epi8(default_state);
        }