GitLab

* Add driver flag "--exhaustive-tune" to enable tuning, add support for the same in C/C++ and python API

* Changed the allocates to occur in the submodules
  * Incomplete, as the use_local_alloc variable in module does not work
  properly
* added a hip::sync_stream before the return
* not sure why the hip::sync_stream gets rid of the dangling reference
error (code-wise it's because hip::sync_stream's output alias is -1)

JIT implementation of the gather operator
Added a few more unit tests to this one as well since I saw some odd behavior during bring up.

Can be enabled via environment variable MIGRAPHX_ENABLE_NHWC

updated GPU pad to now use JIT version.
added range functions for JIT kernels.

Refactor dynamic compute
- add a compute_output_shape object that implicitly converts to a new dyn_output or shape object
- dyn_output object can handle computing the static output shape of an operator given the input arguments shapes
  change an operator's compute function to argument compute(const dyn_output& dyn_out, std::vector<argument> args) to 
  use dyn_output object

Dynamic ref unary functions
-  Included these changes to have an example of the refactored dynamic compute being used
-  Changes to unary base class to handle dynamic shapes
-  Changed elu and leaky_relu to use unary base class and pointwise JIT

* use find2.0 for the convolution
Co-authored-by: Vasilii Filippov <DrizztDoUrden@users.noreply.github.com>
Co-authored-by: Chris Austen <causten@users.noreply.github.com>

Rewrites the TF batch norm like operators to other MIGX operators
Removes the code related to batch_norm_inference

Adds two methods for the custom_ops virtual class.

bool runs_on_offload_target(), if the custom op runs directly on the gpu then it should be set to true. in this case, custom op expects its parameters to reside in GPU memory and writes output to the GPU memory. If it is set to false then, custom op expects it's parameter to reside on the host and puts back the result into the host memory.

output_alias, if output of the custom op is aliasing the input buffer. i.e. interpreting the same input buffer with differnet shape and strides.

Update as_vector() in C++ API to handle non-standard shapes. It required exposing element_index to space_index conversion method for the shape class.

test_gpu_pack_int8_args fails on gfx908 machine, because it doesn't set compute_fp32 flag correctly. This PR fixes the test such that it checks for the device-name, and rocblas-versions and sets this flag accordingly.

* Remove device functions
* Update tests

* Implement concat using jit compilation

* Remove unused headers

Using not and or improves readability. The cppcheck rule will help ensure we are doing it consistently.

*In the verification tests, check that saving and reloading the program is the same program. This also fixes serialization to always load instructions in the same order. There is also fixes for deconv and quant_conv which didn't save the solution id, and was broken for serialization.

* Add softmax kernel

Updated each source file in the repo with the existing license.

* remove code for allocation of C param in dot lowering

* formatting
Co-authored-by: Paul Fultz II <pfultz2@yahoo.com>

* add allocate op header

* formatting

* add replace_allocate pass

* formatting

* move output param to remove_allocate pass

* formatting

* fix bugs in replace_allocate pass

* formatting

* fix verify if tests

* formatting

* move if op logic

* formatting

* cleanup lowering

* cleanup lowering

* formatting

* fix tidy

* formatting

* fix tidy

* add cpu allocate check

* formatting

* change cpu allocate in pass

* formatting

* add some tests for replace_allocate pass

* formatting

* pass by ref

* fix run_pass

* formatting

* update variable name for module

* update dce to use contains() and fix tidy

* formatting

* update cppcheck

* add if test

* formatting

* add if test

* rename var to mod_output_names

* formatting

* remove conditional

* update allocate op and tests

* formatting

* update replace_allocate tests

* update create_output_names() and conditional in replace_allocate

* formatting

* remove extra variable in replace_allocate

* update tools script for allocation_model
Co-authored-by: Umang Yadav <29876643+umangyadav@users.noreply.github.com>
Co-authored-by: Chris Austen <causten@users.noreply.github.com>
Co-authored-by: Paul Fultz II <pfultz2@yahoo.com>

prioritizing int8 over int8x4 when it is applicable
Amend return to continue in apply loop
Adding error handling in case int8x4 compilation failed
Co-authored-by: Paul Fultz II <pfultz2@yahoo.com>

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.

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.

This adds the infrastructure so we can compile everything in parallel, whereas before only pointwise kernels were compiled in parallel. This will also directly integrate with lowering and the gpu-driver. The kernels for pointwise and roialign are using this infrastructure. Scatternd is not since it does require standard shape.

This also makes it easier to add new runtime compiled kernels in the future.

better performance doing it this way

Add onnx parser and ref and gpu implementations of ONNX op ScatterND

Update the base version of clang-format from 5.0 to 10.0

This enables the pointwise fusions using the MIGRAPHX_ENABLE_POINTWISE_FUSION env variable. Its disabled by default since MIOpen fusions need to be refactored.

This also adds a compile_ops pass to compile the pointwise modules. All tests except test_gpu_fast_math passes with MIGRAPHX_ENABLE_POINTWISE_FUSION=1 set.

This PR is the ref implementation of the nonmaxsuppression operator. It always returns the max possible output shape, which is the problem tracked in issue #948.

GPU implementation of the roialign operator, using the jit approach to reduce the lib size.

Implementation of the roialign operator. For now, we have only the ref implementation. When we run a model on the GPU, we fall back the execution to use the ref implementation.

This PR is for the nonzero operator with static output shape.  
Co-authored-by: Paul Fultz II <pfultz2@yahoo.com>
Co-authored-by: mvermeulen <5479696+mvermeulen@users.noreply.github.com>

Previously dot operator was defined as C = alpha * A . B + beta * C where * is scalar multiplication and . is dot product or matrix multiplication depending on dimension of the inputs.

Aim is to have the definition of dot operator as C = A . B without having alpha or beta.

In order to achieve the same effect as alpha and beta (1) it multiplies the one of the inputs to the dot operator with alpha value. (2) if beta is present then, multiplies the C with beta and then adds into the output from step 1.

Add multinomial op to onnx parser with ref and GPU implementations.

The onnx parser inserts a literal of shape {batch_size, sample_size} with random values in the range [0, 1) and inserts existing ops to compute the cumulative density function. The multinomial operator multiplies the random values by the sum of the CDF and returns the index of the first element of the CDF that is greater than the result, representing samples randomly drawn from [0, class_size) that follow the log-probability distribution.

Resolves #821
Co-authored-by: Shucai Xiao <shucai@gmail.com>

This reverts commit 9e43cb8b.

This PR aims to remove alpha and beta attributes from dot operator completely.

Previously dot operator was defined as C = alpha * A . B + beta * C where * is scalar multiplication and . is dot product or matrix multiplication depending on dimension of the inputs.

Aim is to have the definition of dot operator as C = A . B without having alpha or beta.

In order to achieve the same effect as alpha and beta (1) it multiplies the one of the inputs to the dot operator with alpha value. (2) if beta is present then, multiplies the C with beta and then adds into the output from step 1.

Add Loop operator for opset version 13.
Notes: 1) Default max iteration number is 10 if no max iteration number is provided
2) To change the max iter number, a user can set the max_loop_iterations in the onnx_option struct when parsing a model.
3) The returned shape of the scan output is from the max_loop_iterations even the actual loop num is less than that. This issue also applies to other operators like NonZero and NonMaxSuppression. A issue #948 is created to track this and to be resolved later.
Co-authored-by: Paul <pfultz2@yahoo.com>
Co-authored-by: mvermeulen <5479696+mvermeulen@users.noreply.github.com>

Implement the Where operator for the CPU and GPU.  This is for better performance.