CUDA Kernel Error: invalid device pointer

[d0PING]

moreover is that f****** intel compiler???
better for code optimization but sux for reversing
more over
00434F94 |. 66:0F6F0D D0354700 movdqa xmm1 , dqword ptr [004735D0h]
00434F9C |. 66:0F7FBC24 C0010000 movdqa dqword ptr [esp+000001C0h] , xmm7
00434FA5 |. 66:0F6F3D 40354700 movdqa xmm7 , dqword ptr [00473540h]
00434FAD |. 66:0F7F4C24 30 movdqa dqword ptr [esp+30h] , xmm1
00434FB3 |. 66:0F6F0D C0354700 movdqa xmm1 , dqword ptr [004735C0h]
00434FBB |. 66:0F7F4C24 40 movdqa dqword ptr [esp+40h] , xmm1
00434FC1 |. 66:0F6F0D B0354700 movdqa xmm1 , dqword ptr [004735B0h]
00434FC9 |. 66:0F7F4C24 50 movdqa dqword ptr [esp+50h] , xmm1
00434FCF |. 66:0F6F0D A0354700 movdqa xmm1 , dqword ptr [004735A0h]
00434FD7 |. 66:0F7F4C24 60 movdqa dqword ptr [esp+60h] , xmm1
00434FDD |. 66:0F6F0D 90354700 movdqa xmm1 , dqword ptr [00473590h]
00434FE5 |. 66:0F7F4C24 70 movdqa dqword ptr [esp+70h] , xmm1

global variable---aka [00473590h]---> located not less than memory page for local aka ---dqword ptr [esp+70h]
needs for optimization anyway what's not so much hot-spot

[BarsMonster]

man, that's the biggest sse2 code in one function i'd ever seen ( with pc gamess as the exeption)

Actually, it's so big, that Intel C++ v11 was hanging forever trying to compile more than 60% of it
Now they've fixed this issue

[d0PING]

>>>>Actually, it's so big, that Intel C++ v11 was hanging forever trying to compile more than 60% of it

v11 too fat
moreover as you could see in my previously post it puts local variables into stack, if that's too much hanging and death -guaranted
but that's not too much:
00435506 |. 66:0F6F9424 20020000 |movdqa xmm2 , dqword ptr [esp+00000220h]

so maybe some buggy, fat and crazy compiler .thanks God they had made it free for non-comercial.
had you used intrinsics or assembly insertion ?

[d0PING]

>>>rotate?
yep.that's true aka (x<<a)||(x>>(0x20-a))

[BarsMonster]

>>>>Actually, it's so big, that Intel C++ v11 was hanging forever trying to compile more than 60% of it

v11 too fat
moreover as you could see in my previously post it puts local variables into stack, if that's too much hanging and death -guaranted
but that's not too much:
00435506 |. 66:0F6F9424 20020000 |movdqa xmm2 , dqword ptr [esp+00000220h]

so maybe some buggy, fat and crazy compiler .thanks God they had made it free for non-comercial.
had you used intrinsics or assembly insertion ?

intrinsics. I assure you, decision with stack is 100% correct

[d0PING]

>>>>intrinsics. I assure you, decision with stack is 100% correct
I assure you, all compilers put local variables into stack and that's correct,and it has to be so...
but where was bug got some errata from these guys,...
p.s.these new compilers so intelectual what they even checks for stack overflows , but there's another ways...

[BarsMonster]

>>>>intrinsics. I assure you, decision with stack is 100% correct
I assure you, all compilers put local variables into stack and that's correct,and it has to be so...
but where was bug got some errata from these guys,...
p.s.these new compilers so intelectual what they even checks for stack overflows , but there's another ways...

Hmm, I lost you Where is the problem then?

[d0PING]

>>>Hmm, I lost you Where is the problem then?
you hadn't (i guess ).
where's the problem?
what exactly i can tell you is that the problem not in code implementation if it's compiled ,but i hadn't said linking;
compiler hangs as you'd told and yours code that's all information and you wanna me to be more specific
maybe it could be some really scary compiler bug, so post it to intel guys
or processor bug (that's more interesting) then read some errata on this stuff(if exists )(for example if cpu calls simd comand of non-aligned state(intel hates that )it(cpu)raises exception which not sends to os) and do you know what does it mean for us(users) yes shit).
but that's only (if i'm not mistaken) on x86 not x86-64 and on c2d qx6xxx and some xeons but that's for old revisions aka (versions for software)
can you just compile simd-code with intel compiler, after that just link this .obj file with msvc or anything (the best linker i'd ever seen--ulink)...
or write all simd on fasm
so compiler -is not a problem...

[BarsMonster]

Well, I was compiling fine with Intel C++ v10
Also, that bug is fixed now in v11

[d0PING]

i had just tried sse2 version on PIII, wanted to see something, but nothing just crash, as it has to be, but where's nice messagebox???
put some cpuid check

[d0PING]

is there some technical report on upgrade(aka what's new... )

[d0PING]

>>>put some cpuid check
everything is ok.with that
00452A71 |. 9D popfd
00452A72 |. 33C0 xor eax , eax
00452A74 |. 0FA2 cpuid
00452A76 |. 8945 F4 mov dword ptr [ebp-0Ch] , eax
00452A79 |. 895D E8 mov dword ptr [ebp-18h] , ebx
00452A7C |. 8955 EC mov dword ptr [ebp-14h] , edx
00452A7F |. 894D F0 mov dword ptr [ebp-10h] , ecx
00452A82 |. B8 01000000 mov eax , 00000001h
00452A87 |. 0FA2 cpuid
00452A89 |. 8955 FC mov dword ptr [ebp-04h] , edx
00452A8C |. 8945 F8 mov dword ptr [ebp-08h] , eax
00452A8F |> 5B pop ebx
00452A90 |. F745 FC 000000>test dword ptr [ebp-04h] , 04000000h
00452A97 |. 74 0E je 00452AA7h ; bars.00452AA7
00452A99 |. E8 5EFFFFFF call 004529FCh ; bars.004529FC
00452A9E |. 85C0 test eax , eax
00452AA0 |. 74 05 je 00452AA7h ; bars.00452AA7
00452AA2 |. 33C0 xor eax , eax
00452AA4 |. 40 inc eax
00452AA5 |. EB 02 jmp 00452AA9h ; bars.00452AA9


but on my PIII it doesn't executes;), so that's maybe mine...

[d0PING]

have you used vtune for perfomance counts;
interesting is simd per clock;
and L1 hit;L1miss...
was some simd part virtual?(function for c++)
sinhf() and a lot of different linked shit: your compiler decided to link it (your code) with some .lib was there (compiler keys) any -fastmat or something with math
had you used only sse2 or others sseX (looks funny)
please give compilers keys

[BarsMonster]

compiler options is a total mess -Ox is most important
0.9 gonna be 100% different, and I am scared with 3Mb binary :-S
Yes, sin and virtual methods used where needed.

[d0PING]

sin in md5 weird
no,obviously you was used something pow() maybe to find all possible passwords or rounding functions, but not all in this .lib
>>>>0.9 gonna be 100% different, and I am scared with 3Mb binary :-S
...now compiler decided to link your code with not only one lib yours code part in this binary i guess 20-30% -not much
>>>-Ox man make some experiments...
>>>vitual that's for you easy to declare it with virtual but for compiler and reversers... -total shit

[BarsMonster]

sin in md5 weird

Probably that's a surprise for you, but MD5 is based on trigonometric functions
Seriously, MD5 round add constants are calculated through sin function.

[d0PING]

what realy????my God!!!what's next??? i'm near suicide
now seriously... or no a tale aka skazka
once upon a time... before 1995, but not after, there was some man who decided to make some hash (or give some work for bruteforce programmers)
... and made md5 there to assure some peoples he decided to put in to this algorithm some magic numbers obtained from some function (firstly he didn't know which to use), but there was one obstacle this function must be in some range, he decided [0,1] which function could return values in this range
after 0xFFFFFFFF bottles of russian vodka ,he got some overflow and felt very bad he decided to get some funny and went to public...
...what he was doing there is not so interesting, but he maid a sin with (i guess) some girl
after that he liked that public...and always make sin() to get some funny.The End.
P.S.:in all 4 round of md5 it uses const addition from table obtained from K=2^32 * abs (sin (i))-aka "casesensetive" and adds it with array from our incoming message aka f(password)=(padding,etc) and funny functions ( digest message).
...with trigonometry functions we won't had any nice brootforcer and used ours gpu for quake and cs as it has to be...
...so, firstly it's constant no need to call each time this funcion, even to fill array,so
for example these one for FF():
0xD76AA478L
0xE8C7B756L
0x242070DBL
...
so no need to call sin()... and tan() and acos()... and even hypotf()...(shit that it that )it's from your static linked code
next costants you'll get after sending me one million dollars...
so after you'd used pow() for example this .lib was linked with your code
...to long p.s. but for example md5 compiled with extremelly stupid compiler; check this (the same FF):

00411811 8B45 E0 MOV EAX,DWORD PTR SS:[EBP-20]
00411814 3345 D4 XOR EAX,DWORD PTR SS:[EBP-2C]
00411817 2345 EC AND EAX,DWORD PTR SS:[EBP-14]
0041181A 3345 D4 XOR EAX,DWORD PTR SS:[EBP-2C]
0041181D 8B4D 0C MOV ECX,DWORD PTR SS:[EBP+C]
00411820 0301 ADD EAX,DWORD PTR DS:[ECX]
00411822 8B55 F8 MOV EDX,DWORD PTR SS:[EBP-8]
00411825 8D8402 78A46AD7 LEA EAX,DWORD PTR DS:[EDX+EAX+D76AA478]<-in total this's fast command and costant in edx our part of pass, in eax-part of digest
0041182C 8945 F8 MOV DWORD PTR SS:[EBP-8],EAX
0041182F 8B4D F8 MOV ECX,DWORD PTR SS:[EBP-8]
00411832 C1E1 07 SHL ECX,7
00411835 8B55 F8 MOV EDX,DWORD PTR SS:[EBP-8]
00411838 C1EA 19 SHR EDX,19
0041183B 0BCA OR ECX,EDX
0041183D 894D F8 MOV DWORD PTR SS:[EBP-8],ECX
00411840 8B45 F8 MOV EAX,DWORD PTR SS:[EBP-8]
00411843 0345 EC ADD EAX,DWORD PTR SS:[EBP-14]
00411846 8945 F8 MOV DWORD PTR SS:[EBP-8],EAX
this is as it has not be!!!!Carefully!!!

[BarsMonster]

That's a really nice story

[d0PING]

had you used any sin() in your code to generate this table

[BarsMonster]

had you used any sin() in your code to generate this table

I guess so :

for(int i=0;i<64;i++)
{
int tmp = (int)(__int64)floorl(fabsl(sinl(i + 1))*(powl(2.0,32)));
set32(md5_const_sse2,tmp);
}

[d0PING]

yep, ,but no need!!!
anyway, that's only for runtime on all perfomance that doesn't make any infuence on total perfomance;
it's more better cause firstly i thought why in place adding constant it(pointer) got some shit(no gived constants) in that addresses;
carefully you'll show all yours code what'll you sell?
what's more interesting on yours code implementation?
had you done reversing of last round?

[d0PING]

i'd just read subject :CUDA Kernel Error: invalid device pointer ;
and question on this theme: is it ok, what cudaGetDeviceCount() from cudart.dll and even cuDeviceGetCount() from nvcuda.dll returns 1 on s3 savage without any nvidia stuff

[BarsMonster]

i'd just read subject :CUDA Kernel Error: invalid device pointer ;
and question on this theme: is it ok, what cudaGetDeviceCount() from cudart.dll and even cuDeviceGetCount() from nvcuda.dll returns 1 on s3 savage without any nvidia stuff

Well, maybe yes if that's emulation build.

[d0PING]

>>>Well, maybe yes if that's emulation build.
man,i have got(friend with 8800gt gived)cudart.dll and drivers package
Driver was established, cudart.dll in archive with prog. stupig code with messagebox on 'found ' or 'nofound'
00401000 >/$ 68 65104000 PUSH 401065 ; /FileName = "cudart.dll"
00401005 |. FF15 22114000 CALL DWORD PTR DS:[401122] ; \LoadLibraryA
0040100B |. 83C4 04 ADD ESP,4
0040100E |. 68 70104000 PUSH 401070 ; /ProcNameOrOrdinal = "cudaGetDeviceCount"
00401013 |. 50 PUSH EAX ; |hModule
00401014 |. FF15 26114000 CALL DWORD PTR DS:[401126] ; \GetProcAddress
0040101A |. 83C4 08 ADD ESP,8
0040101D |. 68 83104000 PUSH 401083
00401022 |. FFD0 CALL EAX
00401024 |. 833D 83104000 >CMP DWORD PTR DS:[401083],0
0040102B |. 74 1C JE SHORT 00401049 ; cuda.00401049
0040102D |. 6A 00 PUSH 0 ; /Style = MB_OK|MB_APPLMODAL
0040102F |. 68 87104000 PUSH 401087 ; |Title = "information!"
00401034 |. 68 94104000 PUSH 401094 ; |Text = "Device found!!!"
00401039 |. 6A 00 PUSH 0 ; |hOwner = NULL
0040103B |. FF15 66114000 CALL DWORD PTR DS:[401166] ; \MessageBoxA
00401041 |. 6A 00 PUSH 0 ; /ExitCode = 0
00401043 |. FF15 1E114000 CALL DWORD PTR DS:[40111E] ; \ExitProcess
00401049 |> 6A 00 PUSH 0 ; /Style = MB_OK|MB_APPLMODAL
0040104B |. 68 87104000 PUSH 401087 ; |Title = "information!"
00401050 |. 68 A4104000 PUSH 4010A4 ; |Text = "Device not found!!!"
00401055 |. 6A 00 PUSH 0 ; |hOwner = NULL
00401057 |. FF15 66114000 CALL DWORD PTR DS:[401166] ; \MessageBoxA
0040105D |. 6A 00 PUSH 0 ; /ExitCode = 0
0040105F \. FF15 1E114000 CALL DWORD PTR DS:[40111E] ; \ExitProcess
emulation, this is great so i don't need any alpha testing and hope beta too
if you got some installed, please give some link to emulation stuff (.dlls&driver set(.sys, .ini to him))
p.s.:watch attachment

[BarsMonster]

Attachment works, says device found on my uber 8400GS
Emulation does not need DLLs at all, AFAIK, it just compiles program so that everything runs on CPU.
Fortunately, never used that.