Multiplying two float variables by CUDA

Question

I have a really interesting problem, but I solve it in 3 hours, and I just can’t understand what is happening and why it does not work. I tried google, but without any results.

I am programming a CUDA program. I have this very simple piece of code:

__global__ void calcErrorOutputLayer_kernel(*arguments...*) { int idx = blockIdx.x * blockDim.x + threadIdx.x; float gradient; float derivation; derivation = pow((2/(pow(euler, neuron_device[startIndex + idx].outputValue) + pow(euler, -neuron_device[startIndex + idx].outputValue))), 2); gradient = (backVector_device[idx] - neuron_device[startIndex + idx].outputValue); gradient = gradient * derivation; //this line doesn't work gradient = gradient * 2.0; //this line works 

ok, so the gradient is calculated correctly and also output. but when a line arrives where these two variables should be unrelated, nothing happens (the gradient value does not change), and on the next line the CUDA debugger tells me that: “derivation” does not matter at the target location "

gradient * 2.0 works correctly and changes the gradient value 2 times.

Can anybody help me?

Answer 1

 a = pow(euler, neuron_device[startIndex + idx].outputValue); b = pow(euler, -neuron_device[startIndex + idx].outputValue); derivation = pow((2/(a + b),2); 

Pow gives an error when: If the base is negative and the exponent is not an integral value , or if base is zero , and the exponent is negative , a domain error occurred by setting the global errno variable to EDOM.

My guest is that you get problems with accuracy, and both "a" and "b" are equal == 0. Probably you get output = 0 or "inf".

Can you swap float for paired?