spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From tejasapatil <...@git.apache.org>
Subject [GitHub] spark pull request #17184: [SPARK-19843] [SQL] UTF8String => (int / long) co...
Date Wed, 08 Mar 2017 05:00:01 GMT
Github user tejasapatil commented on a diff in the pull request:

    https://github.com/apache/spark/pull/17184#discussion_r104843817
  
    --- Diff: common/unsafe/src/main/java/org/apache/spark/unsafe/types/UTF8String.java ---
    @@ -897,41 +898,52 @@ public long toLong() {
             break;
           }
     
    -      int digit = getDigit(b);
    +      int digit;
    +      if (b >= '0' && b <= '9') {
    +        digit = b - '0';
    +      } else {
    +        return false;
    +      }
    +
           // We are going to process the new digit and accumulate the result. However, before
doing
           // this, if the result is already smaller than the stopValue(Long.MIN_VALUE / radix),
then
    -      // result * 10 will definitely be smaller than minValue, and we can stop and throw
exception.
    +      // result * 10 will definitely be smaller than minValue, and we can stop.
           if (result < stopValue) {
    -        throw new NumberFormatException(toString());
    +        return false;
           }
     
           result = result * radix - digit;
           // Since the previous result is less than or equal to stopValue(Long.MIN_VALUE
/ radix), we
    -      // can just use `result > 0` to check overflow. If result overflows, we should
stop and throw
    -      // exception.
    +      // can just use `result > 0` to check overflow. If result overflows, we should
stop.
           if (result > 0) {
    -        throw new NumberFormatException(toString());
    +        return false;
           }
         }
     
         // This is the case when we've encountered a decimal separator. The fractional
         // part will not change the number, but we will verify that the fractional part
         // is well formed.
         while (offset < numBytes) {
    -      if (getDigit(getByte(offset)) == -1) {
    -        throw new NumberFormatException(toString());
    +      byte currentByte = getByte(offset);
    +      if (currentByte < '0' || currentByte > '9') {
    +        return false;
           }
           offset++;
         }
     
         if (!negative) {
           result = -result;
           if (result < 0) {
    -        throw new NumberFormatException(toString());
    +        return false;
           }
         }
     
    -    return result;
    +    toLongResult.value = result;
    +    return true;
    +  }
    +
    +  public static class IntWrapper {
    --- End diff --
    
    Using IntWrapper for integers gives better perf. If we use LongWrapper for integers, there
would be a conversion needed from long -> int. Its not that big of a difference but given
that ints are used heavily in workloads, I dont want to leave that behind.
    
    Here is microbenchmark result for current approach VS using LongWrapper everywhere
    
    ```
    conversion to int:                       Best/Avg Time(ms)    Rate(M/s)   Per Row(ns)
  Relative
    ------------------------------------------------------------------------------------------------
    IntWrapper                                  20397 / 20564         26.3          38.0 
     1.0X
    LongWrapper                                 20855 / 21530         25.7          38.8 
     1.0X
    ```


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org


Mime
View raw message