-
## Question
I would like to understand more on how TRT under the hood converts Q/DQ models to use actual INT8 kernels. Specifically, how are the scale factors being propagated. Specifically I wonde…
-
The `reference_accesses` methods for an LFRic kernel adds the arguments to the access list, but (esp. in the case of fields) does not provide any index information. As a result of this, the dependency…
hiker updated
7 months ago
-
I haven't see the complete documentation of this project. But the binding for cuTensor is not found in the repo. I wonder if there is any plan for supporting general tensor contractions?
-
## CVE-2020-12652 - Medium Severity Vulnerability
Vulnerable Library - linuxlinux-4.6
The Linux Kernel
Library home page: https://mirrors.edge.kernel.org/pub/linux/kernel/v4.x/?wsslib=linux
Found i…
-
### Bug description
When running the provided code as a standalone executable, a CUDA illegal memory access is reported. Using compute-sanitizer, I could pinpoint this to an illegal shared memory a…
-
## CVE-2020-12652 - Medium Severity Vulnerability
Vulnerable Library - linuxlinux-4.6
The Linux Kernel
Library home page: https://mirrors.edge.kernel.org/pub/linux/kernel/v4.x/?wsslib=linux
Found i…
-
# Overview
With progress on MLIR based kernel generators like [tkf](https://github.com/nod-ai/SHARK-Turbine/tree/tkf?tab=readme-ov-file) (this might be an out of date link), we have an opportunity …
-
Due to interest expressed by ECP application users in trying the RAJA SYCL back-end, we deploy features in development in RAJA releases as they become ready. For the initial release (v0.14.0 -- we are…
-
nvFuser generated code for a fusion block present in DiT has worse than expected performance. The subgraph is performing a `LayerNorm + + Mul + Add + Add` computation as shown in the code below. nvFus…
-
### 🚀 The feature, motivation and pitch
Here is some brief context:
Right now inductor decides loop ordering before fusion. That can lose some fusion opportunities. E.g. if node1 and node2 pick in…