POC zero left padding for PODArray

[amosbird]

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

Category (leave one):

• Performance Improvement

Short description (up to few sentences):

Zero left padding PODArray so that -1 element is always valid and zeroed. It's used for branchless Offset access.

simple tests show 1%-5% improvements

Before padding

Running: left pad test
[
{
    "hostname": "201.nobida.cn",
    "main_metric": "min_time",
    "num_cores": 20,
    "num_threads": 40,
    "ram": 134768111616,
    "runs": [
        {
            "min_time": 0.086000,
            "query": "SELECT max(length(MobilePhoneModel)) FROM hashfile"
        },
        {
            "min_time": 0.102000,
            "query": "SELECT max(length(Params)) FROM hashfile"
        },
        {
            "min_time": 0.441000,
            "query": "SELECT max(length(Title)) FROM hashfile"
        },
        {
            "min_time": 0.143000,
            "query": "SELECT max(length(URLDomain)) FROM hashfile"
        },
        {
            "min_time": 0.120000,
            "query": "SELECT max(length(PageCharset)) FROM hashfile"
        },
        {
            "min_time": 0.429000,
            "query": "SELECT max(length(Referer)) FROM hashfile"
        },
        {
            "min_time": 0.504000,
            "query": "SELECT max(length(URL)) FROM hashfile"
        },
        {
            "min_time": 0.075000,
            "query": "SELECT max(length(UTMSource)) FROM hashfile"
        }
    ],
    "server_version": "18.16.0",
    "test_name": "left pad test",
    "time": "2018-12-24 22:45:10"
}
]

After padding

Running: left pad test
[
{
    "hostname": "201.nobida.cn",
    "main_metric": "min_time",
    "num_cores": 20,
    "num_threads": 40,
    "ram": 134768111616,
    "runs": [
        {
            "min_time": 0.083000,
            "query": "SELECT max(length(MobilePhoneModel)) FROM hashfile"
        },
        {
            "min_time": 0.098000,
            "query": "SELECT max(length(Params)) FROM hashfile"
        },
        {
            "min_time": 0.399000,
            "query": "SELECT max(length(Title)) FROM hashfile"
        },
        {
            "min_time": 0.139000,
            "query": "SELECT max(length(URLDomain)) FROM hashfile"
        },
        {
            "min_time": 0.116000,
            "query": "SELECT max(length(PageCharset)) FROM hashfile"
        },
        {
            "min_time": 0.424000,
            "query": "SELECT max(length(Referer)) FROM hashfile"
        },
        {
            "min_time": 0.492000,
            "query": "SELECT max(length(URL)) FROM hashfile"
        },
        {
            "min_time": 0.072000,
            "query": "SELECT max(length(UTMSource)) FROM hashfile"
        }
    ],
    "server_version": "18.16.0",
    "test_name": "left pad test",
    "time": "2018-12-24 22:47:23"
}
]

[alesapin]

@amosbird What is 'hashfile' you referenced in your test?

[amosbird]

@alesapin It's a data file shared by @alexey-milovidov

It has the following schema and I use it for hash testing

DESCRIBE TABLE hashfile

┌─name─────────────┬─type───┬─default_type─┬─default_expression─┬─comment_expression─┬─codec_expression─┐
│ MobilePhoneModel │ String │              │                    │                    │                  │
│ Params           │ String │              │                    │                    │                  │
│ Title            │ String │              │                    │                    │                  │
│ URLDomain        │ String │              │                    │                    │                  │
│ PageCharset      │ String │              │                    │                    │                  │
│ Referer          │ String │              │                    │                    │                  │
│ URL              │ String │              │                    │                    │                  │
│ UTMSource        │ String │              │                    │                    │                  │
└──────────────────┴────────┴──────────────┴────────────────────┴────────────────────┴──────────────────┘

[alexey-milovidov]

Selected columns from test.hits.

[alexey-milovidov]

@alesapin If you substitute test.hits for hashfile it will work.