Daniel Lowell

@amcamd Can you test the two configs above. @singvision is seeing a segfault. It is pointing to rocBLAS, but it could be the way MIOpen is configuring the parameters.

Same here. From: [TF_issue#8049](https://github.com/tensorflow/tensorflow/issues/8049#issuecomment-290316926) I used: ``` output, _cell_state = rnn.static_rnn(cell, x, dtype=tf.float32) ``` Instead of ``` output, _cell_state = rnn.rnn(cell, x, dtype=tf.float32) ``` And I got pass that section.

Where do you have HCC installed in that docker? Actually, can you post a link to your Dockerfile?

@pfultz2 We'll need to come up with a fix, or a workaround for this. We will get flooded by a tidal wave of complaints from existing Docker infrastructure.

Hmmm, looks like it.

@whchung Do you expect more PRs are needed to fill out this feature? I will need to bump the revision # when we get close to branching for ROCm 2.8.

@pfultz2 @whchung Good to go? Still needed?

All issues I've discovered are local to `MIO_BN_VARIANT=1` kernel. From the error output: ``` root@hpe-sjc2-05:~/dMIOpen/src/kernels# /opt/rocm/bin/clang-ocl -DMIOPEN_USE_FP16=1 -DMIOPEN_USE_FP32=0 -DMIOPEN_USE_FPMIX=0 -DMIO_BN_USESAVED=1 -DMIO_BN_N=256 -DMIO_BN_C=64 -DMIO_BN_HW=3136 -DMIO_BN_NHW=802816 -DMIO_BN_CHW=200704 -DMIO_BN_NCHW=51380224 -DMIO_BN_LDS_SIZE=1024 -DMIO_BN_LDSGCN_SIZE=16 -DMIO_BN_VARIANT=1 -DMIO_WAVESIZE=64...

To reproduce: `docker pull rocm/miopen-private:alex_datatype_fix` Then launch the docker container: `docker run -it --network host -v /home/miopenpdb:/data --device=/dev/kfd --device=/dev/dri --group-add video ocm/miopen-private:alex_datatype_fix` Go to: `cd /root/dFin/_hip` Create a file `fin_input.json`...

To reproduce minimally: ``` /opt/rocm/bin/clang-ocl -DMIOPEN_USE_FP16=1 -DMIOPEN_USE_FP32=0 -DMIOPEN_USE_FPMIX=0 -DMIO_BN_USESAVED=1 -DMIO_BN_N=256 -DMIO_BN_C=64 -DMIO_BN_HW=3136 -DMIO_BN_NHW=802816 -DMIO_BN_CHW=200704 -DMIO_BN_NCHW=51380224 -DMIO_BN_LDS_SIZE=1024 -DMIO_BN_LDSGCN_SIZE=16 -DMIO_BN_VARIANT=1 -DMIO_WAVESIZE=64 -DMIO_BN_GRP0=1024 -DMIO_BN_GRP1=1 -DMIO_BN_GRP2=1 -DMIO_LAYOUT_NHWC=0 -DMIO_BN_GFX103X=0 -mcpu=gfx908 -Wno-everything -mcode-object-version=4 MIOpenBatchNormBwdSpatial.cl -o /tmp/miopen-MIOpenBatchNormBwdSpatial.cl-31a6-867b-afc8-20da/MIOpenBatchNormBwdSpatial.cl.o...