Fast path for stringifying float point numbers

[amosbird]

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

Performance test float_parsing is used.

Without this patch

[
{
    "hostname": "dell123",
    "main_metric": "avg_rows_per_second",
    "num_cores": 20,
    "num_threads": 40,
    "parameters": {
        "expr": ["toString(1 / rand())", "toString(rand() / 0xFFFFFFFF)", "toString(0xFFFFFFFF / rand())", "toString(number)", "toString(number % 10)", "toString(rand())", "toString(rand64())", "concat(toString(rand(1)), '.', toString(rand(2)))", "concat(toString(rand(1)), 'e', toString(rand(2) % 100))", "concat(toString(rand64(1)), toString(rand64(2)), toString(rand64(3)))"]
    },
    "ram": 134774140928,
    "runs": [
        {
            "avg_rows_per_second": 5492364.093416,
            "parameters": {
                "expr": "toString(1 / rand())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(1 / rand())))"
        },
        {
            "avg_rows_per_second": 5613593.130320,
            "parameters": {
                "expr": "toString(rand() / 0xFFFFFFFF)"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(rand() / 0xFFFFFFFF)))"
        },
        {
            "avg_rows_per_second": 5378318.817115,
            "parameters": {
                "expr": "toString(0xFFFFFFFF / rand())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(0xFFFFFFFF / rand())))"
        },
        {
            "avg_rows_per_second": 31540174.865385,
            "parameters": {
                "expr": "toString(number)"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(number)))"
        },
        {
            "avg_rows_per_second": 58654804.116776,
            "parameters": {
                "expr": "toString(number % 10)"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(number % 10)))"
        },
        {
            "avg_rows_per_second": 24805574.955150,
            "parameters": {
                "expr": "toString(rand())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(rand())))"
        },
        {
            "avg_rows_per_second": 14271656.427017,
            "parameters": {
                "expr": "toString(rand64())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(rand64())))"
        },
        {
            "avg_rows_per_second": 9723083.398888,
            "parameters": {
                "expr": "concat(toString(rand(1)), '.', toString(rand(2)))"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(concat(toString(rand(1)), '.', toString(rand(2)))))"
        },
        {
            "avg_rows_per_second": 12161248.746597,
            "parameters": {
                "expr": "concat(toString(rand(1)), 'e', toString(rand(2) % 100))"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(concat(toString(rand(1)), 'e', toString(rand(2) % 100))))"
        },
        {
            "avg_rows_per_second": 4682649.135879,
            "parameters": {
                "expr": "concat(toString(rand64(1)), toString(rand64(2)), toString(rand64(3)))"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(concat(toString(rand64(1)), toString(rand64(2)), toString(rand64(3)))))"
        }
    ],
    "server_version": "1.1.54394",
    "test_name": "float_parsing",
    "time": "2018-07-17 19:50:48"
}
]

With this patch:

[
{
    "hostname": "dell123",
    "main_metric": "avg_rows_per_second",
    "num_cores": 20,
    "num_threads": 40,
    "parameters": {
        "expr": ["toString(1 / rand())", "toString(rand() / 0xFFFFFFFF)", "toString(0xFFFFFFFF / rand())", "toString(number)", "toString(number % 10)", "toString(rand())", "toString(rand64())", "concat(toString(rand(1)), '.', toString(rand(2)))", "concat(toString(rand(1)), 'e', toString(rand(2) % 100))", "concat(toString(rand64(1)), toString(rand64(2)), toString(rand64(3)))"]
    },
    "ram": 134774140928,
    "runs": [
        {
            "avg_rows_per_second": 5787884.707826,
            "parameters": {
                "expr": "toString(1 / rand())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(1 / rand())))"
        },
        {
            "avg_rows_per_second": 6120520.732772,
            "parameters": {
                "expr": "toString(rand() / 0xFFFFFFFF)"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(rand() / 0xFFFFFFFF)))"
        },
        {
            "avg_rows_per_second": 5718844.456722,
            "parameters": {
                "expr": "toString(0xFFFFFFFF / rand())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(0xFFFFFFFF / rand())))"
        },
        {
            "avg_rows_per_second": 32744994.161772,
            "parameters": {
                "expr": "toString(number)"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(number)))"
        },
        {
            "avg_rows_per_second": 56835546.048325,
            "parameters": {
                "expr": "toString(number % 10)"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(number % 10)))"
        },
        {
            "avg_rows_per_second": 26416653.881761,
            "parameters": {
                "expr": "toString(rand())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(rand())))"
        },
        {
            "avg_rows_per_second": 14899093.542744,
            "parameters": {
                "expr": "toString(rand64())"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(rand64())))"
        },
        {
            "avg_rows_per_second": 9866366.888363,
            "parameters": {
                "expr": "concat(toString(rand(1)), '.', toString(rand(2)))"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(concat(toString(rand(1)), '.', toString(rand(2)))))"
        },
        {
            "avg_rows_per_second": 12581788.724010,
            "parameters": {
                "expr": "concat(toString(rand(1)), 'e', toString(rand(2) % 100))"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(concat(toString(rand(1)), 'e', toString(rand(2) % 100))))"
        },
        {
            "avg_rows_per_second": 5093015.983612,
            "parameters": {
                "expr": "concat(toString(rand64(1)), toString(rand64(2)), toString(rand64(3)))"
            },
            "query": "SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(concat(toString(rand64(1)), toString(rand64(2)), toString(rand64(3)))))"
        }
    ],
    "server_version": "1.1.54394",
    "test_name": "float_parsing",
    "time": "2018-07-17 20:17:14"
}
]

The only degradation SQL is SELECT count() FROM system.numbers WHERE NOT ignore(toFloat64(toString(number % 10)))

[alexey-milovidov]

Thank you! It looks clearly better.