[VL] Results are mismatch with the vanilla Spark when cast(sum(decmial(20,4)), float)

[kecookier]

Backend

VL (Velox)

Bug description

The type of column ind_order_amount in SrcTable is decimal(20,4), and the type of column week_ind_order_amount
in TableTarget is float.

The following SQL snippet shows that the calculation of columweek_ind_order_amount for Gluten is inconsistent with the Vanilla version, as detailed in the screenshot.

gluten branch https://github.com/apache/incubator-gluten/tree/branch-1.1

INSERT OVERWRITE TABLE `TableTarget` PARTITION (dt='2024-02-29')
SELECT 
a.deal_id,
SUM(a.quantity) AS week_quantity,
SUM(a.ind_quantity) AS week_ind_quantity,
SUM(a.ind_order_amount) AS week_ind_order_amount,
SUM(a.user_cnt) AS week_user_cnt,
FROM
(
  SELECT deal_id,
  quantity,
  ind_quantity,
  ind_order_amount,
  user_cnt,
  FROM SrcTable
  WHERE partition_date BETWEEN '2024-01-31' AND '2024-02-29'
    AND deal_id IS NOT NULL
) a
GROUP BY deal_id;

Spark version

None

Spark configurations

No response

System information

No response

Relevant logs

No response

[kecookier]

cc @zhouyuan @rui-mo Could you please help me take a look at this issue?

[kecookier]

SrcTable test data

deal_id | ind_order_amount
------------+--------------------------
  952732041 |                2599.0000
  952732041 |                2599.0000
  952732041 |                   0.0000
  952732041 |               10396.0000
  952732041 |                5198.0000
  952732041 |                2599.0000
  952732041 |                   0.0000
  952732041 |                   0.0000
  952732041 |                2599.0000
  952732041 |                2599.0000

Expect 28589.0
Actual 28588.998

[rui-mo]

@kecookier Do you find this issue is caused by cast(decimal as float)? If so, could you provide a decimal value before cast so I can reproduce the result difference? Thanks.

[kecookier]

@kecookier Do you find this issue is caused by cast(decimal as float)? If so, could you provide a decimal value before cast so I can reproduce the result difference? Thanks.

Hi @rui-mo , You can use the following SQL to reproduce the issue.

SELECT cast(sum(cast(col as decimal(20, 4))) as float) FROM VALUES (2599.0000), (2599.0000), (2599.0000), (2599.0000), (2599.0000), (10396.0000), (5198.0000) AS tab(col);

[kecookier]

I've identified the error code, and I will submit a velox PR as soon as possible.

[kecookier]

I've identified the error code, and I will submit a velox PR as soon as possible.

cc @rui-mo