iree

iree copied to clipboard

This gives us an interface for creating a logical device from a set of physical hip devices. In a future PR I plan on removing the normal hip_device ut for...

AWoloszyn

Add conversions for 1x1 conv_2d to matmul

3

Convert 1x1 conv_2d to `linalg.matmul` ops when the HW dimensions are dynamic and convert `linalg.conv_2d_nhwc_hwcf` when the N dimension is not 1. No change to `linalg.conv_2d_nchw_fchw` currently (see linked issue...

IanWood1

Rename `iree-hip-` compiler flags to `iree-rocm-` when they apply to codegen.

3

The various ROCM codegen flags like `iree-hip-target` and `iree-hip-enable-ukernels` are not HIP-specific and need to be renamed back to ROCM. The AMDGPU target uses the same ROCM codegen and we...

benvanik

HoistIntoGlobals fails to hoist constantOp

2

### What happened? running the pass --iree-util-hoist-into-globals fails to hoist constants into globals. Changing one NON-const op, namely tensor. expand_shape to tensor.reshape make the constants get hoisted. ### Steps to...

ziereis

pashu123

[compiler] strip execution context affinities in const eval

2

During compile-time constant evaluation in pass iree-consteval-jit-globals it does not make sense to assign device/queue affinities. We will be compiling and executing it on the compilation host. The JITed IR...

sogartar

When importing onnx model, do I convert onnx to torch dialect?

1


pyl3000

[compiler][flow] Move cast, reshape and bitcast after transfer op

3

We got incoming IR of the form ```mlir %cast = tensor.cast %0 : tensor to tensor

sogartar

[gpu] 'func.func' op uses 401920 bytes of shared memory; exceeded the limit of 65536 bytes

7

### What happened? For the given IR ``` module { func.func @main_graph(%arg0: !torch.vtensor, %arg1: !torch.vtensor, %arg2: !torch.vtensor, %arg3: !torch.vtensor, %arg4: !torch.vtensor, %arg5: !torch.vtensor, %arg6: !torch.vtensor, %arg7: !torch.vtensor, %arg8: !torch.vtensor, %arg9:...

pdhirajkumarprasad

This PR teaches attention decomposition to set attributes for attention matmuls by passing attribute dictionaries to iree_linalg_ext.online_attention operation. This allows us to further control codegen of matmuls (generally the root...

Groverkss