GPU L47: Dynamic Parallelism Memory and Synchronization

00:05:34.357,00:05:37.357
Betapudi Sai Chaitanya cs18b053: what is the use of 2nd syncthreads

00:06:55.626,00:06:58.626
NIKAM ASHUTOSH SHASHIKANT ee16b143: It will be visible to all threads even without syncthreads. Because end of kernel is also a global barrier.

00:31:03.839,00:31:06.839
Mohit Singla cs17b113: can you repeat point once again

00:33:14.263,00:33:17.263
Kshitij Bipin Deogade cs17b104: This point only holds for default streams right?

00:33:20.240,00:33:23.240
Mohit Singla cs17b113: yes sir

[anantjain]

If a block (or tile) of global memory (i.e matrix) is copied into local memory (in registers) of thread. And, After that, a device function is called containing that block (in local memory of thread) as parameter.

Can we call a kernel function from that device function with that block as parameter? Since that block is a pointer to local memory of the thread of the kernel.

In such cases, What can we do to operate (child) kernel function on that block from that device function called by parent (kernel ) function?

[tejpavangarapati]

the cds written below child_launch kernel, wont it keep on waiting till all kernels are complete, but the parent_launch kernel will complete only if cds is over, isnt there a deadlock situation here?