MIOpen
MIOpen copied to clipboard
ROCmSoftwarePlatform
AMD's Machine Intelligence Library
PyTorch wheels are self-contained wrt ROCm (and hence MIOpen) libraries, bundling them as part of the wheel. When running workloads with the PyTorch wheel, we observe slow warmup times and...
I found `miopen::deref(&dropoutDesc) = new miopen::DropoutDescriptor();` in `rnn.cpp`, but didn't known where to delete it. Can you help me determine if this is a bug. https://github.com/ROCmSoftwarePlatform/MIOpen/blob/4949d29f70b6ac26b834713493d9e33759ba5f41/src/rnn.cpp#L282-L371
Right now, we have several problems with the current API design. I would post the as comments bellow to ease later referencing as suggested by @atamazov.
- [x] As a first step (workaround), #981 must be reverted. - Done in #1360 - [ ] Full-blown solution TBD. More info: https://ontrack-internal.amd.com/browse/SWDEV-304151#comment-8282503 /cc @sunway513 for awareness. I do...
hi, I was using a pooling kernel, I notice there are some not used arguments in forward and backward functions, as in below: ```c++ miopenStatus_t PoolingDescriptor::Forward(Handle& handle, const void* alpha,...
@zjing14 On April 7 2021 https://ontrack-internal.amd.com/browse/SWDEV-271887 was set to implemented state. It needs to find out ROCm release the fix is included in, and disable the W/A starting from that...
``` ./bin/MIOpenDriver poolfp16 -M 0 -n 32 -c 192 -H 27 -W 27 -y 3 -x 3 -p 0 -q 0 -v 2 -u 2 -m max -F 1 -t...
convfp16 -n 128 -c 512 -H 28 -W 28 -k 128 -y 1 -x 1 -p 0 -q 0 -u 1 -v 1 -l 1 -j 1 -m conv -g...
Current MIOpenGeMM takes convolution descriptions (configs) as input to solve GeMM problems (Hip-backend). We need to add GeMM solvers for GeMM problems. **MIOpen:** Pure GeMM solvers APIs that call pure...