-
@tridao
Hello!
I am currently working with the selective scan forward kernel, specifically the step h_t = A*h_{t-1} + Bx, where h_0 is currently set to 0. I would like to modify this behavior to …
-
Is there a way to get the output stream?
I'd like to buffer some video for manual playing, currently I use `--sout #std{mux=ts,access=file,dst=-}` to output the data to the console and capture it m…
-
This is preliminary and will be updated or discarded as comments come in.
## Current Issues
1) To maximize performance, we need to be able to utilize shared memory buffers more generically as te…
-
Hi, I've just created a small project ([link to the project](https://github.com/Yanksi/cute_mma)) by modifying the `sgemm_sm80` example. What I was doing was trying to make use of the tensor cores for…
-
CHIRRTL memories were supposed to have an `undefined`, `old`, or `new` read-under-write modifiers after the memory declaration. E.g.:
```
smem mem1 : UInt [8] old
smem mem2 : UInt [8] new
smem m…
-
In many CUDA related project, we can see the line of code as the following:
```
extern "C" __device__ uint32_t __nvvm_get_smem_pointer(void *ptr);
```
It is used to convert the shared memory addre…
-
RT，tile_to_shape这个函数的作用是什么
-
the `smem_start` in `fb_fix_screeninfo` is `unsigned long` type, you should use `%lu` instead `%d` in `printf` for `unsigned long`. Don't you notice the value of `smem_start` in your output is a negat…
-
The following scripts need `root` privs to run `smem`:
- `tests/metrics/density/docker_memory_usage.sh.in`
- `tests/metrics/smem_monitor.sh`
These scripts should contain a check like the follow…
-
I have another question about MMult_cuda_12.cu
Honestly, I don't know how to overlap the share2register and computing process. Is it the asm(PTX) that make them run parallelly? The instructions are s…