초기화에 cudaMemset 사용하기

A buffer from cudaMalloc holds leftover bytes, not zeros. Before you trust it, you often want to initialize the whole thing. 🧼

cudaMemset fills a device buffer with one repeated byte value. It is the GPU cousin of the C library memset.

You give cudaMemset a device pointer, the byte value to write, and how many bytes to fill. That is the whole call.

Passing 0 as the value clears the buffer. Every byte becomes zero, which is the most common initialization you will reach for.

Watch out: cudaMemset repeats a single byte. Asking for the value 1 fills 0x01 bytes, not the integer 1 in every slot.

Because all-zero bytes equal the number 0 for ints and floats, zeroing always works as you expect. Other values rarely do.

cudaMemset runs entirely on the GPU. You skip building a host array and copying it just to clear a device buffer. 🚀

Histograms and sums need a clean start. Use cudaMemset to zero those counters before the kernel begins adding to them.

Like other CUDA calls, cudaMemset returns a status. Check it to catch a bad pointer or a size that is too large.

To fill a buffer with a true number like 3.14, cudaMemset will not help. Launch a small kernel or copy a prepared host array.

For clearing buffers, cudaMemset is the quickest tool you have: one line, no host data, and it runs on the device. 👍

Let us check what cudaMemset actually writes into memory.

You learned cudaMemset fills a device buffer byte by byte on the GPU, ideal for zeroing arrays and accumulators in one call. 🎉