MIOpen
MIOpen copied to clipboard
ROCmSoftwarePlatform
AMD's Machine Intelligence Library
I get the above warning however, a file called gfx90878.db is available. I have also copied this file with the name gfx90878.kdb to no avail. Please advise.
I suggest to mark exported functions with C linkage as `noexcept` for better c++ support. Most of them wrapped in `miopen::try_` internally, so it is possible to add `noexcept` to...
Found on Navi 21 with ROCm 5.2.3 release: ```bash $ MIOPEN_LOG_LEVEL=5 ./bin/MIOpenDriver conv -x 5 -y 5 -W 28 -H 28 -c 192 -n 16 -k 32 -p 2 -q...
The environment is a computing cluster: Slurm 20.11.3 MI50 GPUs PyTorch 1.12.0 ROCM 5.2.0 Code (test.py): **************************************** import torch net = torch.nn.Conv2d(2, 28, 3).cuda() inp = torch.randn(20, 2, 50, 50).cuda()...
```sh export MIOPEN_FIND_MODE=1 export MIOPEN_DEBUG_FIND_ONLY_SOLVER=GemmFwdRest ./bin/MIOpenDriver convfp16 -W 1760 -H 1760 -c 128 -n 1 -k 16 -x 3 -y 3 -p 0 -q 0 -u 1 -v 1 -F...
The following in gfx906_64.HIP.fdb.txt is 3d convolutions，but using `NCHW` layout.  Thus I parse the config to `MIOpenDriver` command, then i got an error as follows: ```sh what(): /xxx/code/miopen-3x3-1x1-verfiy/driver/tensor_driver.hpp:124: unmatched...
This adds batchnorm+activation solver to find 2.0 fusion. Most of the PR lines are two tests: na_(training/inferences)_find2. They are copies of find 1.0 tests with different underlying API.
#1612 has to be reverted due to the following issue: ``` myst v0.15.1: MdParserConfig(renderer='sphinx', commonmark_only=False, enable_extensions=['dollarmath'], dmath_allow_labels=True, dmath_allow_space=True, dmath_allow_digits=True, dmath_double_inline=False, update_mathjax=True, mathjax_classes='tex2jax_process|mathjax_process|math|output_area', disable_syntax=[], url_schemes=['http', 'https', 'mailto', 'ftp'], heading_anchors=None, heading_slug_func=None, html_meta=[],...
[Keywords]: test; gfx1030; [Description]: Random Memory access faults on gfx1030. 5 different PRs failed at a random stage, but always with gfx1030. http://micimaster.amd.com/blue/organizations/jenkins/MLLibs%2FMIOpen/detail/test-int8-mlir-nonxdlops/2/pipeline log info ```log Full Tests I /...
Discussion opened at https://github.com/ROCmSoftwarePlatform/MIOpen/pull/1539#discussion_r881996230. Maybe we have a problem. /cc @junliume @jerryyin