재사용 가능한 CUDA_CHECK 매크로

Writing an if-statement after every CUDA call clutters your code fast. A reusable CUDA_CHECK macro fixes that with one clean line. 🧹

You want to wrap any call, grab its return code, and bail out loudly if it failed. One macro can do this everywhere.

Inside, the macro stores the call result in a local cudaError_t so it can be inspected once without calling twice.

It then tests the code against cudaSuccess. If they match, nothing happens and your program flows on as normal.

On failure it prints a readable message using __FILE__ and __LINE__, so you jump straight to the offending line.

The body is wrapped in a do-while(0). This makes the macro a single statement that works safely even inside an unbraced if.

Put it together and you get a compact, reusable safety net you can paste into any CUDA project header.

Use it on any call that returns a code: cudaMalloc, cudaMemcpy, cudaFree. One wrapper guards them all uniformly.

Kernels return nothing, so wrap cudaGetLastError and a sync after the launch instead of wrapping the launch line itself.

Drop the macro in a shared header so every file uses the same check. Consistency here saves hours of confusing debugging.

The big win is failing fast: the instant a call breaks, you get a precise file, line, and message instead of silent garbage.

Test your grasp of the CUDA_CHECK macro.

A CUDA_CHECK macro captures the code, compares to success, and prints file and line on failure. Define it once, use it everywhere. 🎉