-
## Description
Previously, I remember I can use `--exportLayerInfo` to dump the comprehensive layerwise info of the engine, including the precision of the layer, and the IO tensor datatype and layo…
-
`simulation_period` is a parameter which users frequently type on the command-line, but it's relatively lengthy to type. Should we use a shorthand phrase instead, such as `T` or `duration`?
Please …
-
I adapted from the CUTLASS conv2d example, and tried to change it to run with float16 and on sm86:
Environment:
CUDA Version: 12.0
Device: A10
```
using ElementAccumulator = float; …
-
In some of our kernels (namely, the `Model::interact` kernels in physics) only an arbitrary subset of the threads will be active in that particular kernel. There are two main ways to operate such kern…
-
I tried to preform batch 1D FIR filter by adapting conv2d_fprop_tensorop example, given a 2D continuous row major data layout for (M,N),which has M entry of N length 1D signal, A 1D filter of length S…
-
Hello,
I am trying to use the sim time in my plugin but I'm not sure how to access it. Based on the wiki I should use "settings.simtclock" I think, however, the only attribute that I am able to ac…
-
I cloned the FBK-fairseq repo (https://github.com/hlt-mt/FBK-fairseq.git), installed it following the instructions [here](https://github.com/hlt-mt/FBK-fairseq.git), and tried to run the EDATT model o…
-
In #5227, I am adding a verifier for an invariant of TritonGPU IR. The invariant is not new, it's just that we didn't check it before.
The invariant is, if a tensor has a blocked layout with `warp…
-
I'm getting `LLVM ERROR: operation destroyed but still has uses` when running `gpu-opt` pipeline. The erroring pass is `ConvertFuncToGPUPass`.
error msg:
```
../test.mlir:374:13: error: 'scf.for…
-
Hi Yanyan. Thanks for your excellent work!
I tried to install the latest Spconv (cu113) with pip and run my task in an image with PyTorch 1.12.1 (PyTorch official image). And I encounter the follow…