GPU L14: Memory скачать в хорошем качестве

GPU L14: Memory

00:04:34.318,00:04:37.318 M Venkat Rahul cs18b026: 1..3... and so on 00:04:50.759,00:04:53.759 Rigved Sah cs20m053: Odd numbers in jumbled order if n greaterthan 32 00:05:10.121,00:05:13.121 Nishant Prabhu me17b084: Odd numbers until n or n-1 00:05:41.969,00:05:44.969 Aditya Agrawal CS20S026: Sir, why did you use extern keyword here? 00:06:55.284,00:06:58.284 Sheera Shamsu CS20D001: Sir I didn't understand [[[1,n, n*sizeof(int)]]] 00:07:02.943,00:07:05.943 Sheera Shamsu CS20D001: What is third parameter 00:07:03.980,00:07:06.980 Aditya Agrawal CS20S026: I see. Thank you sir 00:07:20.244,00:07:23.244 Rigved Sah cs20m053: If we use s[10] instead of s[] will it allocate 10 bytes or n bytes? 00:07:54.061,00:07:57.061 Shouvick Mondal: if n=5, and extern _shared_ int s[10]; what would happen? 00:07:54.216,00:07:57.216 Vignesh S me17b078: what if we need 2 dynamic shared memory? how do you do the kernel launch? 00:08:02.404,00:08:05.404 Sheera Shamsu CS20D001: So if it exceeds maximum shared memory size 00:10:55.799,00:10:58.799 Anumala Venu Madhava Reddy cs18b051: why didn't we pass n as an argument and declare _shared_ int s[n] ? 00:12:23.729,00:12:26.729 Parth S. Shah ee17b059: Sir if that's the case for multiple arrays then they must be of the same type? 00:16:10.869,00:16:13.869 Rupesh Nasre.: k[[[1, N, 3*n*sizeof(int)]]](...); 00:16:45.878,00:16:48.878 Rupesh Nasre.: int *s1 = s; int *s2 = s + n; 00:17:06.994,00:17:09.994 Buddhavarapu Venkata Surya Sudheendra cs18b006: So that means we can only have one extern shared variable field 00:17:18.728,00:17:21.728 Parth S. Shah ee17b059: that answers my question also then. Thank you Sir. 00:18:01.587,00:18:04.587 Rupesh Nasre.: char *s2 = (char *)(s+n); 00:20:34.445,00:20:37.445 Nishant Prabhu me17b084: What's the use case of dynamic shared memory against static one? For example what benefit do we miss here if the memory was static? 00:21:40.534,00:21:43.534 Nishant Prabhu me17b084: Oh ok. Thank you sir. 00:22:07.146,00:22:10.146 Sumit Negi cs20m067: Do we use any method to free the dynamic memory that is created here like we do in c , c++ by using free method. 00:23:27.778,00:23:30.778 Shouvick Mondal: cudaFree() 00:28:00.919,00:28:03.919 Buddhavarapu Venkata Surya Sudheendra cs18b006: Can we access the dynamic shared mem called into one kernel from another kernel 00:29:26.454,00:29:29.454 P Sai Venkat Kushal ee17b141: Is "banks" an abstraction or physically present im gpu? 00:29:53.638,00:29:56.638 P Sai Venkat Kushal ee17b141: Yes 00:30:00.123,00:30:03.123 Buddhavarapu Venkata Surya Sudheendra cs18b006: If it is malloc then yes 00:30:33.533,00:30:36.533 Buddhavarapu Venkata Surya Sudheendra cs18b006: Ok 00:31:32.555,00:31:35.555 Jeeva Keshav S ee20s057: sir, can we say that bank conflicts lead to thread divergence? 00:34:48.187,00:34:51.187 Jeeva Keshav S ee20s057: sir, different warps accessing adjacent elements of the same bank can happen simultaneously? 00:35:51.120,00:35:54.120 Parth S. Shah ee17b059: How is that exception handled in hardware? If we do not want sequentialisation then all the threads access the word in parallel right? Does that mean each bank has 32 parallel buses? 00:36:25.370,00:36:28.370 Shouvick Mondal: It is like scanning the bank? 00:38:33.897,00:38:36.897 Sumit Negi cs20m067: what is the bank size as it seems to hamper parallelism. 00:42:55.986,00:42:58.986 P Sai Venkat Kushal ee17b141: 2nd and 3rd 00:43:00.234,00:43:03.234 Nishant Prabhu me17b084: 2, 3 00:43:03.414,00:43:06.414 Betapudi Sai Chaitanya cs18b053: none 00:43:03.977,00:43:06.977 Sumit Negi cs20m067: 2nd 3rd 00:43:08.625,00:43:11.625 Shubham Mohan Randive cs20m064: 2nd and 3rd 00:43:12.667,00:43:15.667 Rigved Sah cs20m053: None 00:43:13.291,00:43:16.291 Kshitij Bipin Deogade cs17b104: none 00:43:14.508,00:43:17.508 Nistala Krishna Vamsi ee20s025: 2,3 00:43:14.744,00:43:17.744 Arihant Samar cs18b052: none 00:43:15.074,00:43:18.074 Parth S. Shah ee17b059: None 00:43:15.433,00:43:18.433 M Venkat Rahul cs18b026: none of them? 00:43:17.620,00:43:20.620 Piyush Avinash Chincholikar cs20m045: No 00:43:22.282,00:43:25.282 Patlolla Bharath Simha Reddy cs18b034: none 00:45:23.102,00:45:26.102 Jeeva Keshav S ee20s057: there is no conflict here because all the threads are from same warp right sir? 00:56:32.182,00:56:35.182 Kshitij Bipin Deogade cs17b104: How to assign numbers to meta if it has size greaterthan 1 00:56:32.482,00:56:35.482 P Sai Venkat Kushal ee17b141: Can we keep meta as a variable size array 00:57:37.682,00:57:40.682 Kshitij Bipin Deogade cs17b104: Ok