GitLab

Following up on issue #1166 and PR #1220. Using the same approach as in #1220 for parallelizing the eval calls, we can significantly reduce the time spent on eliminate_contiguous pass.

Addressing issue #1166 - propagate_constant pass currently uses a recursive approach to find all instructions in a module that can be evaluated to a literal and performs the replacement in the same call.

New approach:

Perform single pass though instructions in the module to determine which instructions can be evaluated
Evaluate selected instructions in parallel
Replace the selected instructions with the corresponding literal

* Upgrade to cppcheck 2.8

* Improve applicable batched gemms for bert

Remove std references in runtime compilation since these are not available when using hiprtc and the headers may not be available on the system

* Fuse gemm add with pointwise fusions

As described in #1196, the ONNX mean parser does not work correctly for integral types. This update fixes the issue by handling integral types separately, where summation is performed before division. Additional test cases have also been added for handling integral types.

For clarity on kernel names found when profiling. The new names are set to the order of the ops being compiled. For example: add + relu = add_relu_kernel.

Updated variable names according to #1193

Fuse layernorm and added triadd_layernorm fusion.  This is a prep performance booster

Expose add_literal method in C/C++ api

Improves performance for add_gelu.  In bert it is 4x faster and for mul_add it is 50% faster than what we current have.

Move to CI containers to rocm 5.0.2
upgrade to 20.04
free up some more file space in github action environments

Fixes the #error when using cppcheck. This no longer suppresses cppcheck errors when including those errors. This fixes the cppcheck errors that was there already.

Helps avoid dangling references. This also deprecates the constructors that didnt take a lifetime annotation since its ambiguous the lifetime.

Add ref and gpu implementations for ONNX op GatherND

Resolves #1032

With reductions such as {2048, 2, 1456} on axes 1, this is 23x faster than using our new block_reduce, and its even over 100x faster than our original reduce_sum:

# lane
gpu::code_object[code_object=13736,symbol_name=kernel,global=2981888,local=1024,]: 0.0672928ms
# block
gpu::code_object[code_object=13800,symbol_name=kernel,global=39321600,local=64,]: 1.46072ms
# original
gpu::reduce_sum[axes={1}]: 6.73456ms
There is some basic logic to pick between lane and block reduce automatically.

* expose get_queue method

Implements the ReverseSequence ONNX operator as a parser.

This parser can only handle a constant sequence_lens input. This is the same as what is handled for TensorRT as far as I can tell.
We could handle a variable sequence_lens input; that would require ref and GPU implementations of the operator.
The ONNX backend tests are disabled because this does not handle variable sequence_lens.

Refactored the reference implementation of pooling to something like what was done for roialign. Moved the reference implementation of pooling from targets/ref/lowering.cpp to pooling.hpp.
Removed cpu_pooling, instead using reference pooling in pooling.hpp
Added reference implementation of Lp Norm pooling and the global version
Added tests for the Lp Norm Pooling

There is significant improvement on larger tensors with half almost 50% faster:

lens: [1024, 384, 768]
gpu::code_object[code_object=13832,symbol_name=kernel,global=39321600,local=256,]: 1.16685ms
gpu::reduce_sum[axes={2}]: 1.73126ms
Also for non-trivial layouts this can sometimes be over 2x faster:

lens: [64, 1024, 768, 4]
gpu::code_object[code_object=13832,symbol_name=kernel,global=39321600,local=256,]: 1.1706ms
gpu::reduce_sum[axes={1}]: 2.63375ms
Of course if the stride becomes larger this speed improvement diminishes due to poor memory access patterns. A lane_reduce instead of a block_reduce is needed for such type of kernels. I plan to address that in a future PR.

Finally, this also includes a MIGRAPHX_GPU_DUMP_ASM env variable which will print out the assembly when the kernel compiles.

Issue 1127 Updates the math.hpp header file to perform overloads of various standard functions (ops) for the hip half2 type. The half2 type is two 16-bit floats packed into a 32-bit number and therefore the overloads act on vectors of sizes that are multiples of 2. They are invoked in runtime compilation any time one of the ops is called on a tensor declared with the data type shape::half_type.

Defined new template, made instances of the template for those math operations that the hip library contains, added verify tests for the sqrt operator for three cases:

tensor size not divisible by 2
tensor size divisible by 2 but not by 4
tensor size divisible by 4

also added the PYTHON_DISABLE_VERSIONS cmake variable to disable python versions.

out-of-bounds access when generate uses nonpacked tensors and add some additional asserts for gpu memory.

ref implementation of the gemm op is sequential, this PR is to parallelize the gemm computation in the ref implementation.

Change the "scatter" struct and op to a base/child set of three: scatter_none, scatter_add, scatter_mul to mirror Onnx' ScatterElements op. and its three reduction options. (Onnx Scatter op is deprecated and is equivalent to scatter_none.)

Provides both a reference op. and update to Onnx parsing. Tests updated and new test case added.

When create a tensor_view with vector date type, the last dimension of the shape should be divided by the vec_size.