find query is too slow

[meftunca]

I am sharing 2 different scenarios and performance results below. I expected it to be faster than Eve. But as far as I understand, simd can't run it properly.

My Code (version 1)

static std::vector<char> whiteSpaceTable = {' ',  '\t', '\n', '\r', '\f',
                                            '\b', '\v', '\a', '\0'};

inline bool isScalarWhiteSpace(const char& c) {
    switch (c) {
        case ' ':
        case '\t':
        case '\n':
        case '\r':
        case '\f':
        case '\b':
        case '\v':
        case '\a':
        case '\0':
            return true;
        default:
            return false;
    }

    // return std::find(whiteSpaceTable.begin(), whiteSpaceTable.end(), c) !=
    //        whiteSpaceTable.end();
}

EVE_FORCEINLINE bool isWhiteSpace(const char& c) {
    return eve::algo::find_if(whiteSpaceTable, [c](auto x) {
               return x == c;
           }) != whiteSpaceTable.end();
}

Result: (version 1)

Run on (12 X 2600 MHz CPU s)
CPU Caches:
  L1 Data 32 KiB
  L1 Instruction 32 KiB
  L2 Unified 256 KiB (x6)
  L3 Unified 12288 KiB
Load Average: 3.12, 2.70, 2.66
----------------------------------------------------------------------------------------------
Benchmark                                    Time             CPU   Iterations UserCounters...
----------------------------------------------------------------------------------------------
Scalar isWs Benchmark                   335334 ns       335209 ns         2049 FileSize=631.514k bytes_per_second=1.75455G/s items_per_second=1.88394G/s
Eve isWs Benchmark                     2026291 ns      2021680 ns          353 FileSize=631.514k bytes_per_second=297.9M/s items_per_second=312.371M/s

My Code (version 2)

static std::vector<char> whiteSpaceTable = {' ',  '\t', '\n', '\r', '\f',
                                            '\b', '\v', '\a', '\0'};

inline bool isScalarWhiteSpace(const char& c) {
       return std::find(whiteSpaceTable.begin(), whiteSpaceTable.end(), c) !=
            whiteSpaceTable.end();
}

EVE_FORCEINLINE bool isWhiteSpace(const char& c) {
    return eve::algo::find_if(whiteSpaceTable, [c](auto x) {
               return x == c;
           }) != whiteSpaceTable.end();
}

Result: (version 2)

Run on (12 X 2600 MHz CPU s)
CPU Caches:
  L1 Data 32 KiB
  L1 Instruction 32 KiB
  L2 Unified 256 KiB (x6)
  L3 Unified 12288 KiB
Load Average: 4.08, 3.22, 2.88
----------------------------------------------------------------------------------------------
Benchmark                                    Time             CPU   Iterations UserCounters...
----------------------------------------------------------------------------------------------
Scalar isWs Benchmark                  1951788 ns      1950786 ns          346 FileSize=631.514k bytes_per_second=308.726M/s items_per_second=323.723M/s
Eve isWs Benchmark                     1895297 ns      1894380 ns          358 FileSize=631.514k bytes_per_second=317.919M/s items_per_second=333.362M/s

My Code (version 3)

static std::vector<char> whiteSpaceTable = {' ',  '\t', '\n', '\r', '\f',
                                            '\b', '\v', '\a', '\0'};

inline bool isScalarWhiteSpace(const char& c) {
     return std::any_of(whiteSpaceTable.begin(), whiteSpaceTable.end(),
                       [c](auto x) { return x == c; });
}

EVE_FORCEINLINE bool isWhiteSpace(const char& c) {
    return eve::algo::any_of(whiteSpaceTable, [c](auto x) {
               return x == c;
           }) ;
}

Result: (version 3)

Run on (12 X 2600 MHz CPU s)
CPU Caches:
  L1 Data 32 KiB
  L1 Instruction 32 KiB
  L2 Unified 256 KiB (x6)
  L3 Unified 12288 KiB
Load Average: 4.08, 3.22, 2.88
----------------------------------------------------------------------------------------------
Benchmark                                    Time             CPU   Iterations UserCounters...
----------------------------------------------------------------------------------------------
Scalar isWs Benchmark                  2626064 ns      2625623 ns          265 FileSize=631.514k bytes_per_second=229.377M/s items_per_second=240.52M/s
Eve isWs Benchmark                     1345726 ns      1345100 ns          510 FileSize=631.514k bytes_per_second=447.743M/s items_per_second=469.492M/s

[DenisYaroshevskiy]

It's not really a find though there are things you can do. What's your context?

[meftunca]

Hi Denis, I am writing a JSON parser. I'm trying to speed up block mappings on certain tables. In parallel with these processes, I am trying to learn eve on the one hand.

What I want to do is space characters, numeric characters and null | false | Quickly matching constant expressions such as true.

[meftunca]

I getting some error

error: no matching constructor for initialization of 'eve::wide<std::int8_t, eve::fixed<16>>' (aka 'wide<signed char, fixed<16>>')
    eve::wide<std::int8_t, eve::fixed<16>> eveWhiteSpaceTable = {

[DenisYaroshevskiy]

It's probably the =
https://godbolt.org/z/8zP1scvqn

[meftunca]

@DenisYaroshevskiy Dude the result is awesome :D

Run on (12 X 2600 MHz CPU s)
CPU Caches:
  L1 Data 32 KiB
  L1 Instruction 32 KiB
  L2 Unified 256 KiB (x6)
  L3 Unified 12288 KiB
Load Average: 3.20, 2.93, 2.74
------------------------------------------------------------------------------------------------
Benchmark                                      Time             CPU   Iterations UserCounters...
------------------------------------------------------------------------------------------------
Scalar isWs Benchmark                    2721073 ns      2720988 ns          255 FileSize=631.514k bytes_per_second=221.338M/s items_per_second=232.09M/s
Eve isWs Benchmark                        431218 ns       430912 ns         1662 FileSize=631.514k bytes_per_second=1.36488G/s items_per_second=1.46553G/s

[meftunca]

I've just realized. Eve can never be as fast as switch-case. Can I improve the performance of eve::any?

[DenisYaroshevskiy]

Eve::any is 2 instructions.
Do you mean eve::algo::any_of?

What is it you are trying to solve