Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 11 May 2012 17:30:47 +0200
From: Samuele Giovanni Tonon <>
Subject: Re: OpenCL SWAP64 on 7970

On 05/11/2012 11:26 AM, myrice wrote:
> Hi, ALL
> On Sun, May 6, 2012 at 11:43 PM, jfoug <> wrote:
>> If n already is a ulong, then the proper casting would be:
>>  ((n & ((ulong)0xff00UL)) << 24)
>> Another option may be:
>> ((n<<24)&0xff00000000UL)
> It is very weird that non of these worked. After tested for a long
> time, I even not sure about the problem.
> My situation is: while I commented out all the thing inside my opencl
> kernel only left SWAP64(ctx.H[0]), the SWAPT64 is right. However, when
> I left some more code in sha512, the problem of SWAP64(ctx.H[0])
> occurs as I stated. Please notice that, ctx.H[0] is assigned to an
> int64 constant in the beginning and untouched in the following code.

i tested on 6970 and i got a self_test failed :

../run/john -test -fo=xsha512-opencl
OpenCL platform 0: AMD Accelerated Parallel Processing, 2 device(s).
Using device 0: Cayman
Compilation log: LOOP UNROLL: pragma unroll (line 174)
    Unrolled as requested!
LOOP UNROLL: pragma unroll (line 179)
    Unrolled as requested!
LOOP UNROLL: pragma unroll (line 194)
    Unrolled as requested!

Local work size = 256
Global work size = 2097152
Benchmarking: Mac OS X 10.7+ salted SHA-512 [OpenCL]... FAILED

maybe i'm totally wrong but have you tried wzyx notation or - better -
variable.s76543210 ?

(you can find examples here :

page 162 and on )


Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.