[jira] [Assigned] (IMPALA-6380) Replace FunctionContext::IsArgConstant(int i) with constant during codegen

["Michael Ho (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/IMPALA-6380?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Michael Ho reassigned IMPALA-6380:
----------------------------------

    Assignee:     (was: Zoram Thanga)

> Replace FunctionContext::IsArgConstant(int i) with constant during codegen
> --------------------------------------------------------------------------
>
>                 Key: IMPALA-6380
>                 URL: https://issues.apache.org/jira/browse/IMPALA-6380
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Backend
>    Affects Versions: Impala 2.12.0
>            Reporter: Michael Ho
>            Priority: Minor
>              Labels: codegen, newbie
>
> There are few places in the string related built-in functions which check whether the input argument is constant per row:
> {noformat}
> TimestampVal TimestampFunctions::ToTimestamp(FunctionContext* context,
>     const StringVal& date, const StringVal& fmt) {
>   if (fmt.is_null || fmt.len == 0) {
>     TimestampFunctions::ReportBadFormat(context, fmt, false);
>     return TimestampVal::null();
>   }
>   if (date.is_null || date.len == 0) return TimestampVal::null();
>   void* state = context->GetFunctionState(FunctionContext::THREAD_LOCAL);
>   DateTimeFormatContext* dt_ctx = reinterpret_cast<DateTimeFormatContext*>(state);
>   if (!context->IsArgConstant(1)) {
>      dt_ctx->Reset(reinterpret_cast<const char*>(fmt.ptr), fmt.len);
>      if (!TimestampParser::ParseFormatTokens(dt_ctx)) {
>        ReportBadFormat(context, fmt, false);
>        return TimestampVal::null();
>      }
>   }
>   const TimestampValue& tv = TimestampValue::Parse(
>       reinterpret_cast<const char*>(date.ptr), date.len, *dt_ctx);
>   TimestampVal tv_val;
>   tv.ToTimestampVal(&tv_val);
>   return tv_val;
> }
> {noformat}
> {noformat}
> StringVal StringFunctions::BTrimString(FunctionContext* ctx,
>     const StringVal& str, const StringVal& chars_to_trim) {
>   if (str.is_null) return StringVal::null();
>   bitset<256>* unique_chars = reinterpret_cast<bitset<256>*>(
>       ctx->GetFunctionState(FunctionContext::THREAD_LOCAL));
>   // When 'chars_to_trim' is unique for each element (e.g. when 'chars_to_trim'
>   // is each element of a table column), we need to prepare a bitset of unique
>   // characters here instead of using the bitset from function context.
>   if (!ctx->IsArgConstant(1)) {
>     unique_chars->reset();
>     DCHECK(chars_to_trim.len != 0 || chars_to_trim.is_null);
>     for (int32_t i = 0; i < chars_to_trim.len; ++i) {
>       unique_chars->set(static_cast<int>(chars_to_trim.ptr[i]), true);
>     }
>   }
> ...
> ...
> {noformat}
> it seems generally useful to extend {{FunctionContextImpl::GetConstFnAttr()}} to also return whether the i-th argument to a UDF is constant or not and then replace these calls with constants during codegen.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org