It would be absurd to install postgres for this functionality but if it's available it offers at least two attractive alternatives. I'll just put my test file here:

#!/bin/sh

echo "OPTION 1: DATA FULLY IMPORTED in temp table"

time (
  < INPUTFILE.csv psql -c "
  drop table if exists t;
  create temporary table t(i1 int, i2 int, n3 numeric);
  copy t from stdin with (format csv, delimiter ',');
  copy (
         select i1,i2,sum(n3) from t group by i1,i2 order by i1,i2
  ) to stdout with(format csv, delimiter ',');
"
)

echo "-- OPTION 2: DATA read via Foreign Data Wrapper (file_fdw)"

echo " drop foreign table if exists inputfile cascade;
       drop server                  pgcsv     cascade; " | psql -qX

time (
echo "
  create server        pgcsv foreign data wrapper file_fdw;
  create foreign table inputfile (
    i1 int,
    i2 int,
    n3 numeric
) server pgcsv
options ( filename '/tmp/INPUTFILE.csv', format 'csv' );
copy (
       select i1,i2,sum(n3) from inputfile group by i1,i2 order by i1,
+i2 )
  to stdout with(format csv, delimiter ',')
" | psql 
)
[download]

I thought it was interesting to see the different timings:

-- OPTION 1: DATA FULLY IMPORTED in temp table
real    0m0.048s
user    0m0.002s
sys     0m0.006s

-- OPTION 2: DATA read via file_fdw (foreign data wrapper)
real    0m0.038s
user    0m0.002s
sys     0m0.006s

-- Your program
real    0m0.051s
user    0m0.048s
sys     0m0.003s
[download]

Foreign Tables, via Foreign Data Wrapper file_fdw, are handy in the gray area between file and database. There is the next step towards full database via a materialized view (of a foreign table) which offers more real db-features (indexing, for one).