Cuda uncomprensible error

Hi all, I’m just working with the convfft2d example that comes with the SDK examples. I’ve implemented a little alghoritm for my thesis that uses the cufft libraries. But if i load images with size major of 256 x 256 pixel and use with it a kernel 256x256 I obtain an unexplicable error :unsure: .

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fc34..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fc34..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fc34..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fbc8..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fbc8..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fbd0..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fc44..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fc44..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fa6c..

First-chance exception at 0x7c812a7b in GMIrda.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012f778..

The code that give this error is a part of the convfft2d provided by example that in standalone mode runs without problems with data and kernel size of 1024 x 1024.

The error is generated by

CUDA_SAFE_CALL( cudaMallocArray(&a_Data,   &float2tex,   DATA_W,   DATA_H));

Looks like full memory but if I call

cuMemGetInfo(&freem,&total);

i obtain that plus of 74% of my dram is free :blink: !!! I dont understand! Removing the CUDA_SAFE_CALL( … ) ambient the malloc runs but the CUFFT library exit with error :/ .

I’m disperate, I don’t understand what is the problem . If anyone could help me, I will thank him/her a lot!!! :tearywave:

Bye, pioniere

Hi,

did you solve your problem and how? I have exactly the same error message, even the memory address reported is close to the one you have, I’m not running the fft example though. this has been bugging me for a few days now and I still can’t figure it out… there are some comments on the problem in the forums but I haven’t been able to make anything out of them, the examples run fine on my machine so there must be some issue with my code, all I’m doing is cudamalloc and everything else seems to be the same as the examples.

Hello,

Did you find out what happened? I am having the same issue now.

Thanks,

JL

I got a bunch of those “First-chance exception” errors. It caused my code to stop working. After a hunt at what I changed, I realized in my case, I had accidentally added in -arch sm_13 as a compiler option, but my card only supports compute capability 1.1. So when I changed the compiler option to -arch sm_11, everything worked great.

I don’t know if this works for other peoples situations, but it worked for me.

HI,
I got also these exceptions, when my program called the first time a cuda instruction (in my case “cudaEventCreate(&start)”). Nevertheless my program seemed to work fine, but I was little bit scared.

So after some investigations (ok 1day :wacko:), I found out, that in my case cublas caused these errors. After removing the cublas.lib from the linker and the cublas.h from the #include, the errors disappear.

I hope someone will help this.

HI,
I got also these exceptions, when my program called the first time a cuda instruction (in my case “cudaEventCreate(&start)”). Nevertheless my program seemed to work fine, but I was little bit scared.

So after some investigations (ok 1day :wacko:), I found out, that in my case cublas caused these errors. After removing the cublas.lib from the linker and the cublas.h from the #include, the errors disappear.

I hope someone will help this.