RE: [PATCH] CPM2: Fix totally bogus alignment handing of dpalloc

Hi,

Some results: alignments are now respected but it seems that the method
I took from linux 2.4 is more efficient in terms of memory consumption.
I will provide a complete patch today with the same mechanism for
rh_alloc_fixed.

Fabien



Allocation method with offset (patch from myself)
rh_alloc - Required size =3D 0040, align =3D 0008
rh_alloc - Block size =3D 3F80, start =3D 00000080 - Off =3D 0000
rh_alloc - Alloc size =3D 0040, start =3D 00000080

rh_alloc - Required size =3D 0008, align =3D 0010
rh_alloc - Block size =3D 3F40, start =3D 000000c0 - Off =3D 0000
rh_alloc - Alloc size =3D 0010, start =3D 000000c0

rh_alloc - Required size =3D 1800, align =3D 0020
rh_alloc - Block size =3D 3F30, start =3D 000000d0 - Off =3D 0010
rh_alloc - Alloc size =3D 1810, start =3D 000000e0

rh_alloc - Required size =3D 0108, align =3D 0040
rh_alloc - Block size =3D 2720, start =3D 000018e0 - Off =3D 0020
rh_alloc - Alloc size =3D 0160, start =3D 00001900

rh_alloc - Required size =3D 0040, align =3D 0010
rh_alloc - Block size =3D 25C0, start =3D 00001a40 - Off =3D 0000
rh_alloc - Alloc size =3D 0040, start =3D 00001a40

rh_alloc - Required size =3D 0020, align =3D 0020
rh_alloc - Block size =3D 2580, start =3D 00001a80 - Off =3D 0000
rh_alloc - Alloc size =3D 0020, start =3D 00001a80

rh_alloc - Required size =3D 0010, align =3D 0010
rh_alloc - Block size =3D 2560, start =3D 00001aa0 - Off =3D 0000
rh_alloc - Alloc size =3D 0010, start =3D 00001aa0

rh_alloc - Required size =3D 000C, align =3D 0040
rh_alloc - Block size =3D 2550, start =3D 00001ab0 - Off =3D 0010
rh_alloc - Alloc size =3D 0050, start =3D 00001ac0


Allocation method without offset (Patch from Pantelis)
rh_alloc - Required size =3D 0040, align =3D 0008
rh_alloc - Block size =3D 3F80, start =3D 00000080
rh_alloc - Alloc size =3D 0047, start =3D 00000080

rh_alloc - Required size =3D 0008, align =3D 0010
rh_alloc - Block size =3D 3F39, start =3D 000000c7
rh_alloc - Alloc size =3D 0017, start =3D 000000d0

rh_alloc - Required size =3D 1800, align =3D 0020
rh_alloc - Block size =3D 3F22, start =3D 000000de
rh_alloc - Alloc size =3D 181F, start =3D 000000e0

rh_alloc - Required size =3D 0108, align =3D 0040
rh_alloc - Block size =3D 2703, start =3D 000018fd
rh_alloc - Alloc size =3D 0147, start =3D 00001900

rh_alloc - Required size =3D 0040, align =3D 0010
rh_alloc - Block size =3D 25BC, start =3D 00001a44
rh_alloc - Alloc size =3D 004F, start =3D 00001a50

rh_alloc - Required size =3D 0020, align =3D 0020
rh_alloc - Block size =3D 256D, start =3D 00001a93
rh_alloc - Alloc size =3D 003F, start =3D 00001aa0

rh_alloc - Required size =3D 0010, align =3D 0010
rh_alloc - Block size =3D 252E, start =3D 00001ad2
rh_alloc - Alloc size =3D 001F, start =3D 00001ae0

rh_alloc - Required size =3D 000C, align =3D 0040
rh_alloc - Block size =3D 250F, start =3D 00001af1
rh_alloc - Alloc size =3D 004B, start =3D 00001b00

-----Original Message-----
From: linuxppc-embedded-bounces@ozlabs.org
[mailto:linuxppc-embedded-bounces@ozlabs.org] On Behalf Of Clement
Fabien
Sent: jeudi 24 novembre 2005 12:26
To: Pantelis Antoniou; Kumar Gala; Marcelo Tosatti; linuxppc-embedded
Subject: RE: [PATCH] CPM2: Fix totally bogus alignment handing of
dpalloc

Here is a patch that builds.


-----Original Message-----
From: Pantelis Antoniou [mailto:panto@intracom.gr]=20
Sent: jeudi 24 novembre 2005 12:04
To: Clement Fabien; Kumar Gala; Marcelo Tosatti; linuxppc-embedded
Subject: [PATCH] CPM2: Fix totally bogus alignment handing of dpalloc

Hi all.

Alignments are currently handled bogusly for cpm2.

Please test.

Regards

Pantelis

Clement Fabien wrote:

Hi,

Some results: alignments are now respected but it seems that the method
I took from linux 2.4 is more efficient in terms of memory consumption.
I will provide a complete patch today with the same mechanism for
rh_alloc_fixed.

Fabien

Please don't bother.

The other method messes up the deallocations, you just have to live
with the wasted space.

With rh_alloc_fixed you're suppossed to know your alignment, the align
argument is bogus.

Pantelis