Question

8. Draw a diagram of the CUDA device memory model for programmers for the allocation, movement, and usage of the various memory types of a device. Explain how CUDA API cudaMemcpy0 functions as data transfer between memories with all its arguments You have to create 6 blocks and every block should contain 8 threads. With the help of a block diagram show how you can create this using 2D Grid of 3D Block. Also mention blockId of each block and threadld of each thread. How will you calculate the threadId of each thread? 3 9. 3

Answer 1

cudaMemcpy(void *dst, void *src, size_t nbytes, enum cudaMemcpyKind direction)

GPU Block (o,o) Shared Memory RegisteysRegistens Thr eadkThrread Share4 Menn ork- Registens iste s (o,o) LocalLocal Memory Memory Memory MemDv Lobal Memory Constant Memory lextuye Memory DA Memory Modet

dst is pointer to destination

src is pointer to data that is going to be copied(source)

direction specifies location of source and destination, ex- host to device, device to host, deice to device