Hi Daniel,

Please can you take a look at and evaluate the attached patch for
grib_api provided by MPI Hamburg.

My understanding is that the patch replaces the grib_api encoding engine
with a C version of that used in GRIBEX developed at MPI Hamburg. This
allows for separation of integer and floating point operations and a
more efficient use of the P6 (and P7) pipelines. They claim to see a
factor o4 4x speed up for encoding on P6 and ~2x on P7.

Also, please can you send a "contributors' licence" to Uwe Schulzweida
<uwe.schulzweida@zmaw.de> and Luis Kornblueh <luis.kornblueh@zmaw.de>.

Thanks

Paul

-------- Original Message --------
Subject: gribapi patch
Date: Wed, 09 May 2012 14:27:25 +0200
From: Uwe Schulzweida <uwe.schulzweida@zmaw.de>
To: Paul.Dando@ecmwf.int
CC: Luis Kornblueh <luis.kornblueh@zmaw.de>

Hi Paul,

Please find attached a performance patch for GRIB_API version 1.9.16.
With this patch the GRIB2 encoding is a factor of 4 faster on our IBM.
The changes can be activated with the preprocessor directive: USE_CGRIBEX

Best regards,
Uwe