Investigate performance of CASE WHEN expression

[andygrove]

What is the problem the feature request solves?

Given this SQL query:

select
    sum(case when ws_wholesale_cost > 10 then 1 else 0 end) as a,
    sum(case when ws_wholesale_cost > 20 then 1 else 0 end) as b,
    sum(case when ws_wholesale_cost > 30 then 1 else 0 end) as c
from web_sales;

The query time with Comet disabled is much better than when Comet is enabled.

Disabled:

[1.5233631134033203, 0.35130977630615234, 0.3343064785003662, 0.35425901412963867, 0.31323719024658203]

Enabled:

[2.219223737716675, 0.886965274810791, 0.8125569820404053, 0.8285551071166992, 0.7927875518798828]

28 of the 99 TPC-DS queries use CASE WHEN, so optimizing this could have a nice impact on overall performance of that benchmark.

Describe the potential solution

No response

Additional context

No response

[andygrove]

I have already started experimenting with some optimizations for this

[andygrove]

We can specialize for certain cases, such as apache/datafusion#11484

[andygrove]

Fixed in apache/datafusion#11484 and #672