Shayan Pardis

In the current implementation, in order to add noise on multiple qubits, we manually run a separate Noise kernel for each qubit (so `length(noiseop.indices)` are being run. In order to...

Each Boolean value takes 1 byte (but we are using 1 bit of information here). By using UInt32 and bitmask, we will improve the memory by 8x. Also, this becomes...

memory batching

1

Right now, we are loading the whole Stabilizer to the GPU memory. Depending on the user's GPU, it might not always fit. A 2000 gate Pauli_Frame with 2^19 trajectories takes...

document that CUDA.@sync is necessary for timing. I removed all unnecessary synchronizations for the sake of improving performance. as a result, when using apply! on gpu, the result is computed...