dying for the disassembler in toolkit 4.0!

Can some registered developer please put a link to the disassembler binary(exe) here for me? Or send it to my email hyq.neuron@gmail.com? Thanks a lot!

I tried to register but failed because I don’t have an institution email address. I tried other disassemblers. decuda is too old for sm20, and the nvdis never gives me anything near to what I’m be expecting… Somebody please help!

nv50dis/nvc0dis are working fine here. How is the result different from what you expect?

em… I’ve been using VS all my life so I don’t have much knowledge about linking… but nvdis works on .cubin right? Yet it doesn’t matter how I change my kernel, it always gives the same output, which doesn’t even nearly match the code in my kernel.

for example, a simple testing kernel, also the only one in the file main.cu

global void test1(float *f)



float f2 = 0;

for(int i=0; i<100; i++)


*f = f2;


gets this output (compiled with optimization disabled, and the kernel’s been tested to produces 200 for *f)

PS C:\Document\Program Projects\Visual Studio 2008\Projects\CUDA\smalltests> nvdis .\main.sm_20.cubin

ELF File…

00000000: 2800440400005de4 mov b32 $r1 c1[0x100]

00000008: 2800400080001de4 mov b32 $r0 c0[0x20]

00000010: 190b200000009de2 mov b32 $r2 0x42c80000

00000018: 9000000000009c85 st b32 wb g[$r0+0] $r2

00000020: 8000000000001de7 exit

Is there something wrong with what I’m doing? Or can you please upload the disassembler binaries you are using?

Indeed this seems wrong.

I’m on Linux/Mac OS and my installations are a bit older - I am using nv50dis/nvc0dis together with [font=“Courier New”]elfToCubin.py --nouveau[/font].


We prefer institutional email addresses, but if you don’t have one, we allow registered developers with non-institutional email addresses. However, we ask that you provide some valid information about yourself and what you work on. Please resubmit and fill it the rest of the information completely.

I’m a high school student with only a bunch of personal projects… anyway, I’ve resubmitted the application, with detailed information about the projects that I’ve been working on. Please please let me through

BTW, I though NVidia employees rarely reply to posts in this CUDA forum. You surprised me.

don’t I have 2300 posts or something? :)

(oh 2864. okay, I was off)

Ha ha yes you are the exception! But I was just thinking where are the so many CUDA engineers from Nvidia… are they all busy with work or do they hate their work so much that when they’re off from work they try their best to run away from any CUDA stuff?

A lot of hardware implementation related questions here could have been easily answered by them. Or is it a corporate policy that engineers are not allowed to reveal the details of the hardware? hmm… I just love how Intel does it :D