GitLab

Add ref and gpu implementations for ONNX op GatherND

Resolves #1032

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

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.

Add exclusive and reverse modes to gpu implementation of prefix_scan_sum, which completes support for ONNX op CumSum

Changed the pooling values for two structures from strings to specialized enum classes. Many test and operator parsing changes to support this. Introduces one new source file, op_enums.cpp.

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

Implements the IsNaN operator, ref, gpu, and onnx parser.

Make doc/CMakeLists.txt standalone
Switch to use rocm-cmake modules for document generation
Add CONFIGURE_DEPENDS to file(GLOB) so it will update without an explicit cmake run
Add STRINGS property for build type to make it easier to switch build types with ccmake
Various fixes and improvements

There is now a MIGRAPHX_DISABLE_POINTWISE_FUSION to disable it

Changed MessagePack file extensions to mxr.

allow nonstd shape for the arg ops, non-standard shapes include broadcast, slice and transpose

Make clip a pointwise op

Fix pointwise compile error with half sqrt 

Resolves a problem in parsing the ssd-10 model.

The problem is, after inserting contiguous in the auto_contiguous pass, standard output shape of some operators becomes non-standard. Then, if the next operator requires standard input shape, an exception is throw.

For example, if we pass the following model:
Input (standard shape) -> transpose (transposed) -> softmax (transposed) -> transpose (standard) -> gather.
It works fine, and no contiguous is required.

In the auto_contiguous pass, a contiguous is inserted after the first transpose. Then we need to replace the first transpose with the contiguous and recompute all shapes. When it comes to the gather operator, its input is a transposed shape, and an exception is thrown.

The solution is in the recompute_shape() function. If it is called by the auto_contiguous pass and shape of an instruction is changed, and the shape is non_standard, we do not recompute shape of its output. The reason is: since its output shape is non_standard, a contiguous op will be added after the instruction, which will recompute shape for later operators.

This PR is to turn on a few gemm unit test with int8 input datatype. Before rocm4.4, int8 input data type requires matrix size to be no less than 4 in rocblas implementation. Because of this limitation, we turned off a few gemm unit tests with int8 input data type.

This limitation is removed in rocm4.4, so after we upgrade to rocm4.5, we can turn on these unit tests. Also we change to unit test conv_bn_add to adding instructions to module instead of program.
Co-authored-by: kahmed10 <15948690+kahmed10@users.noreply.github.com>

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>

Checks wavefront size, then changes implementation and number of threads for DPP reduce

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.

* add topk operator doe ref, cpu and gpu
* Hash modules for quicker lookup of modules
* add onnx unit test
* add unit tests for the topk operator
Co-authored-by: Paul <pfultz2@yahoo.com>
Co-authored-by: mvermeulen <5479696+mvermeulen@users.noreply.github.com>

* rename broadcast and multibroadcast output_lens attribute to out_lens attribute, and change tests and source code to reflect the same

* change the reshape attribute from dims to out_lens

* change transpose attribute's name from dims to perm to reflect better meaning

* use permutation instead of perm for transpose

clang formaating

* use dims instead of out_lens for reshape

clang formatting

Co-authored-by: Chris Austen <causten@users.noreply.github.com>

* check for divisor encodable or not, fallback if needed

* verify test for retinaface case

* Add operators, refactor parsers, add rewrite passes, add tests

* Formatting

* Fix cppcheck

* Review comments

* Formatting

* Combine rewrite passes

* Formatting

* Add ref implementations

* Formatting

* Review comments

* Formatting

* Tidy warnings

* Apply review comments

* Formatting

* Fix CI error

* Formatting

* Increase code coverage

* Formatting

* Move broadcasting of scales and zero points to onnx parser

* Formatting

* Allow for x and zero_point to have different types in quantizelinear; fix zero_point default type

* Formatting

* Increase code coverage

* Formatting

* Switch certain variables to int64_t

* Formatting

* Fix overflow in implicit constant conversion

* Formatting

* Increase code coverage

* Formatting

* Remove operators.hpp from includes in tf_test.cpp

* Formatting

* Add conversion for int32 input to quantizelinear and add test case; remove operators.hpp from onnx_test.cpp includes

* Formatting

* Switch dequantizelinear math from int32 to float

* Formatting

* Remove changes to operators.hpp

* Simplify apply_quantizelinear

* Formatting

* Add verify test for int32 data

* Add rewrite_quantization back to CMakeLists

* Add build for ubuntu 20.04

* Fix ambiguous overload resolution with stream

* Fix warning

* Capture by value

* Format

* Add initial scan operator

* Formatting

* Fix with a working test

* Fix bugs

* Formatting

* Formatting

* Simplify

* Formatting

* Use non-power of 2 for test

* Make pointer
Co-authored-by: mvermeulen <5479696+mvermeulen@users.noreply.github.com>

* Add preallocate method

* Add preallocate_param pass

* Preallocate buffers on the cpu

* Formatting

* Preallocate on the gpu

* Add missing cpp file

* Formatting

* Add lifetime function

* Formatting

* Always allocate

* Fix tidy warning

* Add const

* Add missing lifetime annotations
Co-authored-by: mvermeulen <5479696+mvermeulen@users.noreply.github.com>

* Improve handling of exceptions in test driver

* Formatting

* Auto print exception

* Formatting

* Fork each test case

* Formatting

* Exclude gcc 5 debug build

* Fix tidy issues

* Add color

* Formatting

* Create driver class

* Formatting

* Customize test_case names

* Formatting

* Report status from forked processes

* Formatting

* Update the verify driver

* Formatting

* Print out failed tests

* Formatting

* Fix tidy issues

* Formatting

* Expect passing

* Improve failure reporting on non-linux systems

* Fix ifdef

* Flush code code cov

* Formatting

* Fix tidy

* Check if weak symbols is linked

* Formatting

* Add continue flag

* Formatting

* Set exe name

* Use stringstream and use quotes
Co-authored-by: mvermeulen <5479696+mvermeulen@users.noreply.github.com>

* slice+step+reverse: parsing + onnxfile + test

* cleanup

* updates for step operator, abs axis

* test updates

* updates to tests

* slice+reverse verify test

* verify test for slice+step+reverse

* clang format

* reverse with lens

* simplify normalize_compute_shape

* step op: normalization for axes

* clang format

* change of order: fixing wrong functionality in some cases

* test for slice,reverse,step

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

* Add definitions for all pointwise operators

* Formatting

* Add cpp generator class

* Formatting

* Move compilation to core

* Formatting

* Add clock to tmp name

* Add dynamic loader

* Formatting

* Add tests for code gen

* Formatting

* Add test for literals

* Formatting

* Use with_char

* Add missing header

* Fix mismerge

* Ignore tidy warning

* Fxx gcc 5 errors

* Apply fixits

* Skip signed bitwise of status

* Remove unused parameters

* Explicitly add c++14 flag

* Fix tidy warning

* unify the compute function signature

* clang format

* make another change

* unify the compute function

* clang format

* remove unnecessary code

* more refinement about the operator compute funciton

* clang format

* add an overload function

* clang format

* add support for axes inputs for sequeeze/unsqueeze/reduce_sum

* clang format

* fix build problems

* backup code changes

* clang format

* Add tuple type to shape class

* Formatting

* fix a bug in parsing quantizelinear operator

* clang format

* fix a cppcheck error

* disable different versions of unit tests for different onnx version

* clang format

* upgrade onnx to 1.8

* update onnx to 1.8.1

* disable two more real models

* clang format

* Make data member private

* Formatting

* Add sub arguments

* Formatting

* Trun clang format off

* Disable clang-format

* fix review comments

* fix the function of assign axes in parsing the squeeze operator

* add unit tests and fix a bug

* clang format

* fix review comments

* clang format

* fix a build error

* backup code changes

* clang format

* add more unit tests and add parsing opset version

* clang format

* Improve visiting tuples

* Formatting

* fix cppcheck error

* adding installing the onnx package

* resolve no protobuf compiler

* add an inline subgraph pass

* clang format

* Add more argument tests

* Formatting

* Handle tuple in load

* Formatting

* code backup

* clang format

* Remove .o files

* Add tuple type to api

* Formatting

* fix build errors

* clang format

* code backup

* code backup

* add unit tests for the inline subgraph

* clang format

* refine the inline subgraph and parse if operator

* clang format

* fix cppcheck issue

* clang format

* add unit test for inline subgraph pass

* clang format

* fix format issue

* remove the context from the if operator

* clang format

* simplify the compute functions

* Fix tidy warnings

* fix cppcheck error

* clang format

* fix cppcheck error

* Fix tidy warnings

* fix a cppcheck error

* clang format

* Add a test for share method

* Formatting

* Add a test cpp_type

* add unit tests for more code coverage

* clang format

* add unit tests to have more code coverage

* clang format

* try a comment in jenkins build

* include the install onnnx line

* code backup

* reorder the dependenciesd installed

* refine dockerfile

* fix review comments

* clang format

* remove unnecessary overload function

* fix cppcheck error

* change back the argument test

* Suppress tidy warning

* add the operator get_tuple_elem

* clang format

* add get_tuple_elem to operator include file

* chang if to support multiple operation outputs

* clang format

* optimize inline subgraph

* clang format

* code backup

* clang format

* fix bug

* refine unit tests for tuple output of the if operator

* clang format

* refine a instruction replacement code

* add a unit test and sort all the unit tests alphabetically

* fix cppcheck error

* add more unit tests for multiple op outputs

* clang format

* fix cppcheck error

* Update pass manager to get modules after every pass

* more unit test to cover more scenarios

* clang format

* fixed a bug in a unit test

* add more tests

* clang format

* add more unit tests to have more code coverage

* fix a bug in a unit test

* Add program overload for module

* Formatting

* Hash modules for quicker lookup of modules

* Bump file version

* Add methods to remove modules

* Formatting

* add the tuple type to the support list

* Eliminate unused modules

* Formatting

* Fix test errors

* Foramtting

* Fix tidy issues

* fix problem related to inline subgraph

* clang format

* fix review comments

* fix review comments

* fix review comments

* fix review comments

* clang format

* fix a unit test

* one more code change

* remove an optimization related to the if operator

* clang format

* fix review comments
Co-authored-by: Paul <pfultz2@yahoo.com>
Co-authored-by: mvermeulen <5479696+mvermeulen@users.noreply.github.com>