[gmes-user] The first posting

[전경원 <kw...@gi...>]

Hi. I am the developer of GMES, Kyungwon Chun.

>From now on, I'll report the development progress and plan of GMES, through
this mailing list to guide any one who is interested.

These days, I've tried to accelerate the time-stepping speed. The direction
of work is adjusting virtual method level to reduce overhead of vtable look
up and increasing memory locality for the cache hit ratio. The PwMaterial
abstract base class lost its virtual method, 'update', and the 'update_all'
method became the new virtual method. This change reduce the number of
vtable lookup to just one for PwMaterial, for one time-step.

Also, there is a change in attributes of 'PwMaterial' class. It just saves
sequence of index, and its concrete class save the sequence of
'PwMaterialParam'. These sequences use std::vector as a container which
provides a contiguous storage. To make the contiguous storage of
'PwMaterialParam', GMES stores 'PwMaterialParam' instances, not the pointer
of it.

These changes increased the stepping speed almost 5 times. Now, the source
in the trunk only reflects the virtual function changes. The memory
locality fixes will be updated soon, also.

See you.