GPU L11: Memory

00:03:07.728,00:03:10.728 Jeeva Keshav S ee20s057: sir, here, we take an assumption that we launch 32 threads right? 00:05:29.903,00:05:32.903 Jeeva Keshav S ee20s057: okay sir 00:05:52.973,00:05:55.973 Amalan S EE20D408: histogram 00:12:29.890,00:12:32.890 Rupesh Nasre.: struct AoS { int x, y, z; }; 00:12:45.601,00:12:48.601 Rupesh Nasre.: struct AoS aos[N]; 00:13:16.536,00:13:19.536 M Venkat Rahul cs18b026: int SOAx[32];int SOAy[32];int SOAz[32]; 00:16:47.803,00:16:50.803 Buddhavarapu Venkata Surya Sudheendra cs18b006: We only have one node in SoA case right, so where do we have the same attribute of the next node. 00:17:24.955,00:17:27.955 Kshitij Bipin Deogade cs17b104: Why is SoA performance better? 00:21:41.822,00:21:44.822 SHETH DEV YASHPAL CS17B106: cudamemcopy(&nodes.alla[i], &nodes[i].a) for all i from 1 to N 00:22:29.102,00:22:32.102 SHETH DEV YASHPAL CS17B106: same as what he is saying, my syntax may be wrong 00:44:40.103,00:44:43.103 M Venkat Rahul cs18b026: sir then what does __global__ do different from __device__ 00:47:02.873,00:47:05.873 M Venkat Rahul cs18b026: i dont know sir 00:47:03.106,00:47:06.106 NIKAM ASHUTOSH SHASHIKANT ee16b143: Sir, Why did we use nngpu and not cudaMalloc(&nn, ...)? 00:47:30.283,00:47:33.283 M Venkat Rahul cs18b026: oh okay sir 00:47:34.570,00:47:37.570 M Venkat Rahul cs18b026: thank you 00:50:20.615,00:50:23.615 Akash Haridas ae17b020: can we just iterate through the linked list and memcpy each element? what is the advantage of the recursive method? 00:50:45.377,00:50:48.377 Akash Haridas ae17b020: ok thank you 00:51:54.673,00:51:57.673 Nistala Krishna Vamsi ee20s025: Will cudamalloc copy a struct which containing pointer and array properly