MIOpen
MIOpen copied to clipboard
ROCmSoftwarePlatform
AMD's Machine Intelligence Library
This PR implement [torchvision.ops.sigmoid_focal_loss](https://pytorch.org/vision/main/generated/torchvision.ops.sigmoid_focal_loss.html) operation. There is no constraint here, MIOpen is faster than ROCm in all cases. - [x] Added SigmoidFocalLoss operation with forward and backward kernels. - [x]...
- Added RoPE(rotary position embedding) operation and kernel with solver - Added driver test and gtest for RoPE - Compared to ROCm pytorch(no fusion), there is a performance improvement rope...
The value in find options was completely ignored.
- Allows to use `boost:filesystem` instead of `std::filesystem` via CMake flag. - This is to workaround runtime issues (SIGSEGV) related to misconfigured/buggy std::filesystem on some host configurations (e.g. CentOS 8...
This PR focuses on converting the Batch Norm Forward Inference Spatial and Per Activation kernels from OpenCL to HIP. This conversion is a part of the broader initiative to translate...
Remove default datatype and fail when datatype is not set. Fix tests. _Originally posted by @atamazov in https://github.com/ROCm/MIOpen/pull/3123#discussion_r1697486280_
If convolution perf and find dbs have not been started loading yet, handle constructor starts a separate thread that does that. When a db is requested in some library call...
Copied from LLVM cmake variables LLVM_PARALLEL{COMPILE,LINK}_JOBS, adds the option for the ninja generator of splitting the jobs pool between the compile and link jobs. This allows the user to fine...
_The internals of our library went some major changes recently with solver registry, invokers etc. It would be helpful to have a design document to understand what the new design...
Can we do tuning using API? Lets say if I want to tune fused solver or regular solver. I can do: MIOPEN_FIND_ENFORCE=3 MIOPEN_FIND_MODE=1 ./bin/MIOpenDriver CBAInfer -F 4 -n 128 -c...
