MIOpen
MIOpen copied to clipboard
AMD's Machine Intelligence Library
Create gTests to cover cases without saved Mean and saved InvVar, equivalent to e.g. the command below ./bin/MIOpenDriver bnormbfp16 -n 1 -c 2 -H 24 -W 16 -m 1 --forw...
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3 contains a bug and is being fixed in CK. We are disabling it until it is fixed in CK. This is a workaround, and an issue was created for...
Enable NHWC support for BN, and include performance fixes.
This adds MIOpenDriver conv* argument `--output_type`/`-M`, that supports values `same` and `fp32` and thus allows launching convolution with float output data type despite input and weights types being different. `ConvArgsForMIOpenDriver`...
Additional MIOPEN_FIND_MODE = 6 (TrustVerify) This mode extends DynamicHybrid Running with TrustVerify will first attempt to load tuning results from system resources If no solution is returned tuning will be...
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3 kernel has a bug and is being fixed in CK. We are temporarily disabling it. This is a workaround and should be undone when CK is updated.
## Scope While testing the variant 0 for bnorm backward spatial single, we observed that the error tolerance in MIOpenDriver is sub-optimal. For instance, the following results were observed ```shell...
- Add Where operation with contiguous backward kernel. - Add driver and gtest for kernel. - MIOpen performs better if: - Input, other and condition tensors have the same shape...
I recently updated my AI workflow to ROCm 6.3.2 on Arch Linux, and found that some PyTorch operations were crashing with "MIOpen Error: tensor_shape_variable needs to be an array". With...
Addition of CI targets for navi3, navi4, and gfx94x. Remove old targets that are no longer supported
- jenkins file only builds these targets for my branch at this time. Will move this to develop when we go to merge