Merge branch 'develop' into resize_op

.github/workflows/ci.yaml

@@ -465,7 +465,7 @@ jobs:
     - name: Upload code coverage
       if: "matrix.configuration == 'codecov'"
       env:
-        CODECOV_TOKEN: "8545af1c-f90b-4345-92a5-0d075503ca56"
+        CODECOV_TOKEN: "f5d5a10b-3177-4c76-b25f-9b1c2f165e8b"
       run: |
         sudo apt-get install -y lcov
         cd build

Jenkinsfile

@@ -22,6 +22,8 @@ def rocmtestnode(Map conf) {
         def cmd = """
             ulimit -c unlimited
             echo "leak:dnnl::impl::malloc" > suppressions.txt
+            echo "leak:libtbb.so" >> suppressions.txt
+            cat suppressions.txt
             export LSAN_OPTIONS="suppressions=\$(pwd)/suppressions.txt"
             export MIGRAPHX_GPU_DEBUG=${gpu_debug}
             export CXX=${compiler}
@@ -134,12 +136,14 @@ rocmtest clang_debug: rocmnode('mi100+') { cmake_build ->
     }
 }, mlir_debug: rocmnode('mi100+') { cmake_build ->
     stage('MLIR Debug') {
-        withEnv(['MIGRAPHX_ENABLE_EXTRA_MLIR=1']) {
+        withEnv(['MIGRAPHX_ENABLE_EXTRA_MLIR=1', 'MIGRAPHX_MLIR_USE_SPECIFIC_OPS=fused,attention,convolution,dot']) {
             def sanitizers = "undefined"
             // Note: the -fno-sanitize= is copied from upstream LLVM_UBSAN_FLAGS.
             def debug_flags_cxx = "-g -O2 -fsanitize=${sanitizers} -fno-sanitize=vptr,function -fno-sanitize-recover=${sanitizers}"
             def debug_flags = "-g -O2 -fsanitize=${sanitizers} -fno-sanitize=vptr -fno-sanitize-recover=${sanitizers}"
             def gpu_targets = getgputargets()
+            // Since the purpose of this run verify all things MLIR supports,
+            // enabling all possible types of offloads
             cmake_build(flags: "-DCMAKE_BUILD_TYPE=debug -DMIGRAPHX_ENABLE_PYTHON=Off -DMIGRAPHX_ENABLE_MLIR=On -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags_cxx}' -DCMAKE_C_FLAGS_DEBUG='${debug_flags}' -DGPU_TARGETS='${gpu_targets}'")
         }
     }

docs/.doxygen/Doxyfile

@@ -28,7 +28,14 @@ MACRO_EXPANSION = YES
 
 OUTPUT_DIRECTORY = docBin
 
-PREDEFINED = DOXYGEN
+PREDEFINED = \
+    DOXYGEN \
+    MIGRAPHX_EXPORT= \
+    MIGRAPHX_API_EXPORT= \
+    MIGRAPHX_GPU_EXPORT= \
+    MIGRAPHX_CPU_EXPORT= \
+    MIGRAPHX_ONNX_EXPORT= \
+    MIGRAPHX_TF_EXPORT= \
 
 PROJECT_NAME = MIGraphX
 

docs/.sphinx/requirements.txt

@@ -21,7 +21,7 @@ charset-normalizer==3.1.0
     # via requests
 click==8.1.3
     # via sphinx-external-toc
-cryptography==41.0.4
+cryptography==41.0.6
     # via pyjwt
 deprecated==1.2.13
     # via pygithub
@@ -89,7 +89,7 @@ requests==2.28.2
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core==0.28.0
+rocm-docs-core==0.30.0
     # via -r requirements.in
 smmap==5.0.0
     # via gitdb

docs/dev/data.rst

@@ -5,26 +5,36 @@ shape
 -----
 
 .. doxygenstruct:: migraphx::internal::shape
+   :members:
+   :undoc-members:
 
 literal
 -------
 
 .. doxygenstruct:: migraphx::internal::literal
+   :members:
+   :undoc-members:
 
 argument
 --------
 
 .. doxygenstruct:: migraphx::internal::argument
+   :members:
+   :undoc-members:
 
 raw_data
 --------
 
 .. doxygenstruct:: migraphx::internal::raw_data
+   :members:
+   :undoc-members:
 
-.. doxygenfunction:: migraphx::internal::visit_all
+.. doxygenfunction:: template<class T, class ...Ts> auto migraphx::internal::visit_all(T &&x, Ts&&... xs)
 
 
 tensor_view
 -----------
 
 .. doxygenstruct:: migraphx::internal::tensor_view
+   :members:
+   :undoc-members:

docs/dev/dev_intro.rst

@@ -18,8 +18,8 @@ Directions for building MIGraphX from source can be found in the main README fil
 Adding Two Literals
 --------------------
 
-A program is a collection of modules, which are collections of instructions to be executed when calling `eval <migraphx::program::eval>`.
-Each instruction has an associated `operation <migraphx::operation>` which represents the computation to be performed by the instruction.
+A program is a collection of modules, which are collections of instructions to be executed when calling :cpp:any:`eval <migraphx::internal::program::eval>`.
+Each instruction has an associated :cpp:any:`operation <migraphx::internal::operation>` which represents the computation to be performed by the instruction.
 
 We start with a snippet of the simple ``add_two_literals()`` function::
 
@@ -41,14 +41,14 @@ We start with a snippet of the simple ``add_two_literals()`` function::
     auto result = p.eval({}).back();
     std::cout << "add_two_literals: 1 + 2 = " << result << "\n";
 
-We start by creating a simple ``migraphx::program`` object and then getting a pointer to the main module of it.
+We start by creating a simple :cpp:any:`migraphx::program <migraphx::internal::program>` object and then getting a pointer to the main module of it.
 The program is a collection of ``modules`` that start executing from the main module, so instructions are added to the modules rather than directly onto the program object.
-We then use the `add_literal <migraphx::program::add_literal>` function to add an instruction that stores the literal number ``1`` while returning an `instruction_ref <migraphx::instruction_ref>`.
-The returned `instruction_ref <migraphx::instruction_ref>` can be used in another instruction as an input.
-We use the same `add_literal <migraphx::program::add_literal>` function to add a ``2`` to the program.
+We then use the :cpp:any:`add_literal <migraphx::internal::program::add_literal>` function to add an instruction that stores the literal number ``1`` while returning an :cpp:any:`instruction_ref <migraphx::internal::instruction_ref>`.
+The returned :cpp:any:`instruction_ref <migraphx::internal::instruction_ref>` can be used in another instruction as an input.
+We use the same :cpp:any:`add_literal <migraphx::internal::program::add_literal>` function to add a ``2`` to the program.
 After creating the literals, we then create the instruction to add the numbers together.
-This is done by using the `add_instruction <migraphx::program::add_instruction>` function with the ``"add"`` `operation <migraphx::program::operation>` created by `make_op <migraphx::program::make_op>` along with the previous `add_literal` `instruction_ref <migraphx::instruction_ref>` for the input arguments of the instruction.
-Finally, we can run this `program <migraphx::program>` by compiling it for the reference target (CPU) and then running it with `eval <migraphx::program::eval>`
+This is done by using the :cpp:any:`add_instruction <migraphx::internal::program::add_instruction>` function with the ``"add"`` :cpp:any:`operation <migraphx::internal::program::operation>` created by :cpp:any:`make_op <migraphx::internal::program::make_op>` along with the previous `add_literal` :cpp:any:`instruction_ref <migraphx::internal::instruction_ref>` for the input arguments of the instruction.
+Finally, we can run this :cpp:any:`program <migraphx::internal::program>` by compiling it for the reference target (CPU) and then running it with :cpp:any:`eval <migraphx::internal::program::eval>`
 The result is then retreived and printed to the console.
 
 We can compile the program for the GPU as well, but the file will have to be moved to the ``test/gpu/`` directory and the correct target must be included::
@@ -76,8 +76,8 @@ We can modify the program to take an input parameter ``x``, as seen in the ``add
     p.compile(migraphx::ref::target{});
 
 This adds a parameter of type ``int32``, and compiles it for the CPU.
-To run the program, we need to pass the parameter as a ``parameter_map`` when we call `eval <migraphx::program::eval>`.
-We create the ``parameter_map`` by setting the ``x`` key to an `argument <migraphx::argument>` object with an ``int`` data type::
+To run the program, we need to pass the parameter as a ``parameter_map`` when we call :cpp:any:`eval <migraphx::internal::program::eval>`.
+We create the ``parameter_map`` by setting the ``x`` key to an :cpp:any:`argument <migraphx::internal::argument>` object with an ``int`` data type::
 
     // create a parameter_map object for passing a value to the "x" parameter
     std::vector<int> data = {4};
@@ -92,7 +92,7 @@ We create the ``parameter_map`` by setting the ``x`` key to an `argument <migrap
 Handling Tensor Data
 ---------------------
 
-In the previous examples we have only been dealing with scalars, but the `shape <migraphx::shape>` class can describe multi-dimensional tensors.
+In the previous examples we have only been dealing with scalars, but the :cpp:any:`shape <migraphx::internal::shape>` class can describe multi-dimensional tensors.
 For example, we can compute a simple convolution::
 
     migraphx::program p;
@@ -109,7 +109,7 @@ For example, we can compute a simple convolution::
 
 Here we create two parameters for both the ``input`` and ``weights``.
 In the previous examples, we created simple literals, however, most programs will take data from allocated buffers (usually on the GPU).
-In this case, we can create `argument <migraphx::argument>` objects directly from the pointers to the buffers::
+In this case, we can create :cpp:any:`argument <migraphx::internal::argument>` objects directly from the pointers to the buffers::
 
     // Compile the program
     p.compile(migraphx::ref::target{});
@@ -133,8 +133,8 @@ In this case, we can create `argument <migraphx::argument>` objects directly fro
 
     EXPECT(migraphx::verify::verify_rms_range(results_vector, sol));
 
-An `argument <migraphx::argument>` can handle memory buffers from either the GPU or the CPU.
-By default when running the `program <migraphx::program>`, buffers are allocated on the corresponding target.
+An :cpp:any:`argument <migraphx::internal::argument>` can handle memory buffers from either the GPU or the CPU.
+By default when running the :cpp:any:`program <migraphx::internal::program>`, buffers are allocated on the corresponding target.
 When compiling for the CPU, the buffers by default will be allocated on the CPU.
 When compiling for the GPU, the buffers by default will be allocated on the GPU.
 With the option ``offload_copy=true`` set while compiling for the GPU, the buffers will be located on the CPU.
@@ -143,7 +143,7 @@ With the option ``offload_copy=true`` set while compiling for the GPU, the buffe
 Importing From ONNX
 --------------------
 
-A `program <migraphx::program>` can be built directly from an onnx file using the MIGraphX ONNX parser.
+A :cpp:any:`program <migraphx::internal::program>` can be built directly from an onnx file using the MIGraphX ONNX parser.
 This makes it easier to use neural networks directly from other frameworks.
 In this case, there is an ``parse_onnx`` function::
 

docs/dev/env_vars.rst

@@ -4,13 +4,13 @@ Environment Variables
 For parsing
 ---------------
 
-**MIGRAPHX_TRACE_ONNX_PARSER**
+.. envvar:: MIGRAPHX_TRACE_ONNX_PARSER
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print debugging traces for the onnx parser.
 Prints: initializers (if used), ONNX node operators, added MIGraphX instructions
 
-**MIGRAPHX_DISABLE_FP16_INSTANCENORM_CONVERT**
+.. envvar:: MIGRAPHX_DISABLE_FP16_INSTANCENORM_CONVERT
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Disables the conversion from fp16 to fp32 for the InstanceNormalization ONNX operator that MIGX does as a workaround for accuracy issues with reduce_mean/variance.
@@ -20,16 +20,16 @@ See ``parse_instancenorm.cpp`` for more details.
 Matchers
 ------------
 
-**MIGRAPHX_TRACE_MATCHES**
+.. envvar:: MIGRAPHX_TRACE_MATCHES
 
 Set to "1" to print the matcher that matches an instruction and the matched instruction.
 Set to "2" and use the ``MIGRAPHX_TRACE_MATHCES_FOR`` flag to filter out results.
 
-**MIGRAPHX_TRACE_MATCHES_FOR**
+.. envvar:: MIGRAPHX_TRACE_MATCHES_FOR
 
 Set to the name of any matcher and only traces for that matcher will be printed out.
 
-**MIGRAPHX_VALIDATE_MATCHES**
+.. envvar:: MIGRAPHX_VALIDATE_MATCHES
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Validate the module after finding the matches (runs ``module.validate()``).
@@ -37,7 +37,7 @@ Validate the module after finding the matches (runs ``module.validate()``).
 Program Execution 
 ---------------------
 
-**MIGRAPHX_TRACE_EVAL**
+.. envvar:: MIGRAPHX_TRACE_EVAL
 
 Set to "1", "2", or "3" to use.
 "1" prints the instruction run and the time taken.
@@ -48,7 +48,7 @@ Set to "1", "2", or "3" to use.
 Program Verification
 ------------------------
 
-**MIGRAPHX_VERIFY_ENABLE_ALLCLOSE**
+.. envvar:: MIGRAPHX_VERIFY_ENABLE_ALLCLOSE
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Uses ``allclose`` with the given ``atol`` and ``rtol`` for verifying ranges with ``driver verify`` or the tests that use ``migraphx/verify.hpp``.
@@ -57,76 +57,76 @@ Uses ``allclose`` with the given ``atol`` and ``rtol`` for verifying ranges with
 Pass debugging or Pass controls
 -----------------------------------
 
-**MIGRAPHX_TRACE_ELIMINATE_CONTIGUOUS**
+.. envvar:: MIGRAPHX_TRACE_ELIMINATE_CONTIGUOUS
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Debug print the instructions that have input ``contiguous`` instructions removed.
 
-**MIGRAPHX_DISABLE_POINTWISE_FUSION**
+.. envvar:: MIGRAPHX_DISABLE_POINTWISE_FUSION
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Disables the ``fuse_pointwise`` compile pass.
 
-**MIGRAPHX_DEBUG_MEMORY_COLORING**
+.. envvar:: MIGRAPHX_DEBUG_MEMORY_COLORING
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print debug statements for the ``memory_coloring`` pass.
 
-**MIGRAPHX_TRACE_SCHEDULE**
+.. envvar:: MIGRAPHX_TRACE_SCHEDULE
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print debug statements for the ``schedule`` pass.
 
-**MIGRAPHX_TRACE_PROPAGATE_CONSTANT**
+.. envvar:: MIGRAPHX_TRACE_PROPAGATE_CONSTANT
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Traces instructions replaced with a constant.
 
-**MIGRAPHX_INT8_QUANTIZATION_PARAMS**
+.. envvar:: MIGRAPHX_INT8_QUANTIZATION_PARAMS
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print the quantization parameters in only the main module.
 
-**MIGRAPHX_DISABLE_DNNL_POST_OPS_WORKAROUND**
+.. envvar:: MIGRAPHX_DISABLE_DNNL_POST_OPS_WORKAROUND
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Disable the DNNL post ops workaround.
 
-**MIGRAPHX_DISABLE_MIOPEN_FUSION**
+.. envvar:: MIGRAPHX_DISABLE_MIOPEN_FUSION
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Disable MIOpen fusions.
 
-**MIGRAPHX_DISABLE_SCHEDULE_PASS**
+.. envvar:: MIGRAPHX_DISABLE_SCHEDULE_PASS
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Disable the ``schedule`` pass.
 
-**MIGRAPHX_DISABLE_REDUCE_FUSION**
+.. envvar:: MIGRAPHX_DISABLE_REDUCE_FUSION
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Disable the ``fuse_reduce`` pass.
 
-**MIGRAPHX_ENABLE_NHWC**
+.. envvar:: MIGRAPHX_ENABLE_NHWC
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Enable the ``layout_nhwc`` pass.
 
-**MIGRAPHX_ENABLE_CK**
+.. envvar:: MIGRAPHX_ENABLE_CK
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Enable using the Composable Kernels library.
 Should be used in conjunction with ``MIGRAPHX_DISABLE_MLIR=1``.
 
-**MIGRAPHX_DISABLE_MLIR** 
+.. envvar:: MIGRAPHX_DISABLE_MLIR*
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Disable using the rocMLIR library.
 
-**MIGRAPHX_ENABLE_EXTRA_MLIR**
+.. envvar:: MIGRAPHX_ENABLE_EXTRA_MLIR
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Enables additional opportunities to use MLIR that may improve performance.
 
-**MIGRAPHX_COPY_LITERALS**
+.. envvar:: MIGRAPHX_COPY_LITERALS
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Use ``hip_copy_to_gpu`` with a new ``literal`` instruction rather than use ``hip_copy_literal{}``.
@@ -134,22 +134,22 @@ Use ``hip_copy_to_gpu`` with a new ``literal`` instruction rather than use ``hip
 Compilation traces
 ----------------------
 
-**MIGRAPHX_TRACE_FINALIZE**
+.. envvar:: MIGRAPHX_TRACE_FINALIZE
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Debug print instructions during the ``module.finalize()`` step.
 
-**MIGRAPHX_TRACE_COMPILE**
+.. envvar:: MIGRAPHX_TRACE_COMPILE
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print trace information for the graph compilation process.
 
-**MIGRAPHX_TRACE_PASSES**
+.. envvar:: MIGRAPHX_TRACE_PASSES
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print the compile pass and the program after the pass.
 
-**MIGRAPHX_TIME_PASSES**
+.. envvar:: MIGRAPHX_TIME_PASSES
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Time the compile passes.
@@ -158,77 +158,77 @@ Time the compile passes.
 GPU Kernels JIT compilation debugging (applicable for both hiprtc and hipclang)
 -----------------------------------------
 
-**MIGRAPHX_TRACE_CMD_EXECUTE**
+.. envvar:: MIGRAPHX_TRACE_CMD_EXECUTE
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print commands executed by the MIGraphX ``process``.
 
-**MIGRAPHX_TRACE_HIPRTC**
+.. envvar:: MIGRAPHX_TRACE_HIPRTC
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print HIPRTC options and C++ file executed.
 
-**MIGRAPHX_DEBUG_SAVE_TEMP_DIR**
+.. envvar:: MIGRAPHX_DEBUG_SAVE_TEMP_DIR
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Make it so the created temporary directories are not deleted.
 
-**MIGRAPHX_GPU_DEBUG**
+.. envvar:: MIGRAPHX_GPU_DEBUG
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Internally, this adds the option ``-DMIGRAPHX_DEBUG`` when compiling GPU kernels. It enables assertions and capture of source locations for the errors. 
 
-**MIGRAPHX_GPU_DEBUG_SYM**
+.. envvar:: MIGRAPHX_GPU_DEBUG_SYM
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Adds the option ``-g`` when compiling HIPRTC.
 
-**MIGRAPHX_GPU_DUMP_SRC**
+.. envvar:: MIGRAPHX_GPU_DUMP_SRC
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Dump the HIPRTC source files compiled.
 
-**MIGRAPHX_GPU_DUMP_ASM**
+.. envvar:: MIGRAPHX_GPU_DUMP_ASM
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Dump the hip-clang assembly.
 
-**MIGRAPHX_GPU_OPTIMIZE**
+.. envvar:: MIGRAPHX_GPU_OPTIMIZE
 
 Set the optimization mode for GPU compile (``-O`` option).
 Defaults to ``-O3``.
 
-**MIGRAPHX_GPU_COMPILE_PARALLEL**
+.. envvar:: MIGRAPHX_GPU_COMPILE_PARALLEL
 
 Set to the number of threads to use.
 Compile GPU code in parallel with the given number of threads.
 
-**MIGRAPHX_TRACE_NARY**
+.. envvar:: MIGRAPHX_TRACE_NARY
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print the ``nary`` device functions used.
 
-**MIGRAPHX_ENABLE_HIPRTC_WORKAROUNDS**
+.. envvar:: MIGRAPHX_ENABLE_HIPRTC_WORKAROUNDS
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Enable HIPRTC workarounds for bugs in HIPRTC.
 
-**MIGRAPHX_USE_FAST_SOFTMAX**
+.. envvar:: MIGRAPHX_USE_FAST_SOFTMAX
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Use the fast softmax optimization.
 
-**MIGRAPHX_ENABLE_NULL_STREAM**
+.. envvar:: MIGRAPHX_ENABLE_NULL_STREAM
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Allow using null stream for miopen and hipStream.
 
-**MIGRAPHX_NSTREAMS**
+.. envvar:: MIGRAPHX_NSTREAMS
 
 Set to the number of streams to use.
 Defaults to 1.
 
-**MIGRAPHX_TRACE_BENCHMARKING**
+.. envvar:: MIGRAPHX_TRACE_BENCHMARKING
 
 Set to "1" to print benchmarching trace.
 Set to "2" to print benchmarching trace with more detail.
@@ -236,45 +236,49 @@ Set to "2" to print benchmarching trace with more detail.
 MLIR vars
 -------------
 
-**MIGRAPHX_TRACE_MLIR**
+.. envvar:: MIGRAPHX_TRACE_MLIR
 
 Set to "1" to trace MLIR and print any failures.
 Set to "2" to additionally print all MLIR operations.
 
-**MIGRAPHX_MLIR_USE_SPECIFIC_OPS**
+.. envvar:: MIGRAPHX_MLIR_USE_SPECIFIC_OPS
 
 Set to the name of the operations you want to always use MLIR regardless of GPU architecture.
 Accepts a list of operators separated by commas (ex: "fused", "convolution", "dot").
 
-**MIGRAPHX_MLIR_TUNING_DB**
+.. envvar:: MIGRAPHX_MLIR_TUNING_DB
 
 Set to the path of the MLIR tuning database to load.
 
-**MIGRAPHX_MLIR_TUNING_CFG**
+.. envvar:: MIGRAPHX_MLIR_TUNING_CFG
 
 Set to the path of the tuning configuration.
 Appends to tuning cfg file that could be used with rocMLIR tuning scripts.
 
-**MIGRAPHX_MLIR_TUNE_EXHAUSTIVE**
+.. envvar:: MIGRAPHX_MLIR_TUNE_EXHAUSTIVE
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Do exhaustive tuning for MLIR.
 
+.. envvar:: MIGRAPHX_MLIR_TUNE_LIMIT
+
+Set to an integer greater than 1.
+Limits the number of solutions that MLIR will use for tuning.
 
 CK vars
 -----------
 
-**MIGRAPHX_LOG_CK_GEMM**
+.. envvar:: MIGRAPHX_LOG_CK_GEMM
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Print Composable Kernels GEMM traces.
 
-**MIGRAPHX_CK_DEBUG**
+.. envvar:: MIGRAPHX_CK_DEBUG
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Always add the ``-DMIGRAPHX_CK_CHECK=1`` for compiling Composable Kernels operators.
 
-**MIGRAPHX_TUNE_CK**
+.. envvar:: MIGRAPHX_TUNE_CK
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Use tuning for Composable Kernels.
@@ -282,19 +286,19 @@ Use tuning for Composable Kernels.
 Testing 
 ------------
 
-**MIGRAPHX_TRACE_TEST_COMPILE**
+.. envvar:: MIGRAPHX_TRACE_TEST_COMPILE
 
 Set to the target that you want to trace the compilation of (ex. "gpu", "cpu").
 Prints the compile trace for the given target for the verify tests.
 This flag shouldn't be used in conjunction with ``MIGRAPHX_TRACE_COMPILE``.
 For the verify tests only use ``MIGRAPHX_TRACE_TEST_COMPILE``.
 
-**MIGRAPHX_TRACE_TEST**
+.. envvar:: MIGRAPHX_TRACE_TEST
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Prints the reference and target programs even if the verify passed successfully.
 
-**MIGRAPHX_DUMP_TEST**
+.. envvar:: MIGRAPHX_DUMP_TEST
 
 Set to "1", "enable", "enabled", "yes", or "true" to use.
 Dumps verify tests to ``.mxr`` files.

docs/dev/operators.rst

@@ -5,6 +5,8 @@ operation
 ---------
 
 .. doxygenstruct:: migraphx::internal::operation
+   :members:
+   :undoc-members:
 
 .. doxygenfunction:: migraphx::internal::is_context_free
 
@@ -14,3 +16,5 @@ operators
 ---------
 
 .. doxygennamespace:: migraphx::internal::op
+   :members:
+   :undoc-members:

docs/dev/pass.rst

@@ -5,63 +5,82 @@ pass
 ----
 
 .. doxygenstruct:: migraphx::internal::pass
+   :members:
+   :undoc-members:
 
 dead_code_elimination
 ---------------------
 
 .. doxygenstruct:: migraphx::internal::dead_code_elimination
+   :members:
+   :undoc-members:
 
 eliminate_common_subexpression
 ------------------------------
 
 .. doxygenstruct:: migraphx::internal::eliminate_common_subexpression
+   :members:
+   :undoc-members:
 
 eliminate_concat
 ----------------
 
 .. doxygenstruct:: migraphx::internal::eliminate_concat
+   :members:
+   :undoc-members:
 
 eliminate_contiguous
 --------------------
 
 .. doxygenstruct:: migraphx::internal::eliminate_contiguous
+   :members:
+   :undoc-members:
 
 eliminate_identity
 ------------------
 
 .. doxygenstruct:: migraphx::internal::eliminate_identity
+   :members:
+   :undoc-members:
 
 eliminate_pad
 -------------
 
 .. doxygenstruct:: migraphx::internal::eliminate_pad
+   :members:
+   :undoc-members:
 
 propagate_constant
 ------------------
 
 .. doxygenstruct:: migraphx::internal::propagate_constant
-
-rewrite_batchnorm
------------------
-
-.. doxygenstruct:: migraphx::internal::rewrite_batchnorm
+   :members:
+   :undoc-members:
 
 rewrite_rnn
 -----------
 
 .. doxygenstruct:: migraphx::internal::rewrite_rnn
+   :members:
+   :undoc-members:
 
 schedule
 --------
 
 .. doxygenstruct:: migraphx::internal::schedule
+   :members:
+   :undoc-members:
 
 simplify_algebra
 ----------------
 
 .. doxygenstruct:: migraphx::internal::simplify_algebra
+   :members:
+   :undoc-members:
 
 simplify_reshapes
 -----------------
 
 .. doxygenstruct:: migraphx::internal::simplify_reshapes
+   :members:
+   :undoc-members:

docs/dev/program.rst

@@ -5,6 +5,8 @@ instruction
 -----------
 
 .. doxygenstruct:: migraphx::internal::instruction
+   :members:
+   :undoc-members:
 
 instruction_ref
 ---------------
@@ -17,6 +19,8 @@ program
 -------
 
 .. doxygenstruct:: migraphx::internal::program
+   :members:
+   :undoc-members:
 
 parse_onnx
 ----------

docs/dev/targets.rst

@@ -5,14 +5,20 @@ target
 ------
 
 .. doxygenstruct:: migraphx::internal::target
+   :members:
+   :undoc-members:
 
 gpu::target
 -----------
 
 .. doxygenstruct:: migraphx::internal::gpu::target
+   :members:
+   :undoc-members:
 
 cpu::target
 -----------
 
 .. doxygenstruct:: migraphx::internal::cpu::target
+   :members:
+   :undoc-members:
 

docs/reference/cpp.rst

@@ -8,45 +8,65 @@ shape
 .. doxygenenum:: migraphx_shape_datatype_t
 
 .. doxygenstruct:: migraphx::shape
+   :members:
+   :undoc-members:
 
 argument
 --------
 
 .. doxygenstruct:: migraphx::argument
+   :members:
+   :undoc-members:
 
 target
 ------
 
 .. doxygenstruct:: migraphx::target
+   :members:
+   :undoc-members:
 
 program
 -------
 
 .. doxygenstruct:: migraphx::program_parameter_shapes
+   :members:
+   :undoc-members:
 
 .. doxygenstruct:: migraphx::program_parameters
+   :members:
+   :undoc-members:
 
 .. doxygenstruct:: migraphx_compile_options
+   :members:
+   :undoc-members:
 
 .. doxygenstruct:: migraphx::program
+   :members:
+   :undoc-members:
 
 quantize
 --------
 
 .. doxygenstruct:: migraphx::quantize_op_names
+   :members:
+   :undoc-members:
 
 .. doxygenfunction:: migraphx::quantize_fp16(const program&)
 
 .. doxygenfunction:: migraphx::quantize_fp16(const program&, const quantize_op_names&)
 
 .. doxygenstruct:: migraphx::quantize_int8_options
+   :members:
+   :undoc-members:
 
-.. doxygenfunction:: migraphx::quantize_int8
+.. doxygenfunction::migraphx::quantize_int8
 
 parse_onnx
 ----------
 
 .. doxygenstruct:: migraphx::onnx_options
+   :members:
+   :undoc-members:
 
 .. doxygenfunction:: migraphx::parse_onnx(const char *)
 
@@ -63,16 +83,18 @@ parse_onnx
 load
 ----
 
-.. doxygenstruct:: migraphx_file_options
+.. doxygenstruct:: migraphx::file_options
+   :members:
+   :undoc-members:
 
 .. doxygenfunction:: migraphx::load(const char *)
 
-.. doxygenfunction:: migraphx::load(const char *, migraphx_file_options)
+.. doxygenfunction:: migraphx::load(const char *, const file_options&)
 
 save
 ----
 
 .. doxygenfunction:: migraphx::save(const program&, const char *)
 
-.. doxygenfunction:: migraphx::save(const program&, const char *, migraphx_file_options)
+.. doxygenfunction:: migraphx::save(const program&, const char *, const file_options&)
 

examples/README.md

@@ -6,4 +6,5 @@ This directory contains examples of common use cases for MIGraphX.
 ## Examples:
 - [MIGraphX usage and utilities](./migraphx)
 - [Vision inference examples](./vision)
-- [Natural language inference examples](./nlp)
\ No newline at end of file
+- [Natural language inference examples](./nlp)
+- [Diffusion inference examples](./diffusion)

examples/diffusion/README.md

+# Diffusion Inference Examples
+
+- [Python Stable Diffusion 2.1](./python_stable_diffusion_21)

examples/diffusion/python_stable_diffusion_21/README.md

+# Stable Diffusion 2.1
+
+This version was tested with [rocm 5.7](https://github.com/ROCmSoftwarePlatform/AMDMIGraphX/tree/rocm-5.7.0) revision.
+
+## Jupyter notebook
+
+There is a dedicated step-by-step notebook. See [sd21.ipynb](./sd21.ipynb)
+
+## Console application
+
+To run the console application, follow these steps below.
+
+Setup python environment
+
+```bash
+# this will require the python venv to installed (e.g. apt install python3.8-venv)
+python3 -m venv sd_venv
+. sd_venv/bin/activate
+```
+
+Install dependencies
+
+```bash
+pip install -r requirements.txt
+```
+
+Use MIGraphX Python Module
+
+```bash
+export PYTHONPATH=/opt/rocm/lib:$PYTHONPATH
+```
+
+Get models with optimum
+
+```bash
+optimum-cli export onnx --model stabilityai/stable-diffusion-2-1 models/sd21-onnx
+```
+*Note: `models/sd21-onnx` will be used in the scripts.*
+
+Run the text-to-image script with the following example prompt and seed:
+
+```bash
+python txt2img.py --prompt "a photograph of an astronaut riding a horse" --seed 13 --output astro_horse.jpg
+```
+*Note: The first run will compile the models and cache them to make subsequent runs faster.*
+
+The result should look like this:
+
+![example_output.jpg](./example_output.jpg)
+
+## Gradio application
+
+Note: requires `Console application` to work
+
+Install gradio dependencies
+
+```bash
+pip install -r gradio_requirements.txt
+```
+
+Usage
+
+```bash
+python gradio_app.py
+```
+
+This will load the models (which can take several minutes), and when the setup is ready, starts a server on `http://127.0.0.1:7860`.

examples/diffusion/python_stable_diffusion_21/gradio_app.py

+#####################################################################################
+# The MIT License (MIT)
+#
+# Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+# THE SOFTWARE.
+#####################################################################################
+
+from txt2img import StableDiffusionMGX
+import gradio as gr
+
+
+def main():
+    # Note: This will load the models, which can take several minutes
+    sd = StableDiffusionMGX()
+
+    def gr_wrapper(prompt, negative_prompt, steps, seed, scale):
+        result = sd.run(str(prompt), str(negative_prompt), int(steps),
+                        int(seed), float(scale))
+        return StableDiffusionMGX.convert_to_rgb_image(result)
+
+    demo = gr.Interface(
+        gr_wrapper,
+        [
+            gr.Textbox(value="a photograph of an astronaut riding a horse",
+                       label="Prompt"),
+            gr.Textbox(value="", label="Negative prompt (Optional)"),
+            gr.Slider(1, 100, step=1, value=20, label="Number of steps"),
+            gr.Textbox(value=13, label="Random seed"),
+            gr.Slider(1, 20, step=0.1, value=7.0, label="Guidance scale"),
+        ],
+        "image",
+    )
+    demo.launch()
+
+
+if __name__ == "__main__":
+    main()

examples/diffusion/python_stable_diffusion_21/gradio_reqirements.txt

+#####################################################################################
+# The MIT License (MIT)
+#
+# Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+# THE SOFTWARE.
+#####################################################################################
+-f requirements.txt
+gradio
\ No newline at end of file

examples/diffusion/python_stable_diffusion_21/requirements.txt

+#####################################################################################
+# The MIT License (MIT)
+#
+# Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in
+# all copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+# THE SOFTWARE.
+#####################################################################################
+accelerate
+diffusers
+optimum[onnxruntime]
+transformers
\ No newline at end of file

examples/diffusion/python_stable_diffusion_21/sd21.ipynb

+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#  The MIT License (MIT)\n",
+    "#\n",
+    "#  Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.\n",
+    "#\n",
+    "#  Permission is hereby granted, free of charge, to any person obtaining a copy\n",
+    "#  of this software and associated documentation files (the 'Software'), to deal\n",
+    "#  in the Software without restriction, including without limitation the rights\n",
+    "#  to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n",
+    "#  copies of the Software, and to permit persons to whom the Software is\n",
+    "#  furnished to do so, subject to the following conditions:\n",
+    "#\n",
+    "#  The above copyright notice and this permission notice shall be included in\n",
+    "#  all copies or substantial portions of the Software.\n",
+    "#\n",
+    "#  THE SOFTWARE IS PROVIDED 'AS IS', WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n",
+    "#  IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n",
+    "#  FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE\n",
+    "#  AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n",
+    "#  LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n",
+    "#  OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n",
+    "#  THE SOFTWARE."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Stable Diffusion 2.1\n",
+    "\n",
+    "The following example will show how to run `Stable Diffusion 2.1` with `MIGraphX`."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Install the required dependencies."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Install dependencies\n",
+    "!pip install optimum[onnxruntime] transformers diffusers accelerate"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We will use optimum to generate the onnx files."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# export models\n",
+    "!optimum-cli export onnx --model stabilityai/stable-diffusion-2-1 models/sd21-onnx"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Now it is time to load these models with python.\n",
+    "\n",
+    "First, we make sure that MIGraphX module is found in the python path."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import sys\n",
+    "mgx_lib_path = \"/opt/rocm/lib/\" # or \"/code/AMDMIGraphX/build/lib/\"\n",
+    "if mgx_lib_path not in sys.path:\n",
+    "    sys.path.append(mgx_lib_path)\n",
+    "import migraphx as mgx"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Next, a helper method to load and cache the models.\n",
+    "\n",
+    "This will use the `models/sd21-onnx` path. If you changed it, make sure to update here as well."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "# helper for model loading\n",
+    "def load_mgx_model(name, shapes):\n",
+    "    file = f\"models/sd21-onnx/{name}/model\"\n",
+    "    print(f\"Loading {name} model from {file}\")\n",
+    "    if os.path.isfile(f\"{file}.mxr\"):\n",
+    "        print(f\"Found mxr, loading it...\")\n",
+    "        model = mgx.load(f\"{file}.mxr\", format=\"msgpack\")\n",
+    "    elif os.path.isfile(f\"{file}.onnx\"):\n",
+    "        print(f\"Parsing from onnx file...\")\n",
+    "        model = mgx.parse_onnx(f\"{file}.onnx\", map_input_dims=shapes)\n",
+    "        model.compile(mgx.get_target(\"gpu\"))\n",
+    "        print(f\"Saving {name} model to mxr file...\")\n",
+    "        mgx.save(model, f\"{file}.mxr\", format=\"msgpack\")\n",
+    "    else:\n",
+    "        print(f\"No {name} model found. Please verify the path is correct and re-try, or re-download model.\")\n",
+    "        os.exit(1)\n",
+    "    return model"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "With that, we can load the models. This could take several minutes."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "text_encoder = load_mgx_model(\"text_encoder\", {\"input_ids\": [1, 77]})"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "unet = load_mgx_model(\n",
+    "        \"unet\", {\n",
+    "            \"sample\": [1, 4, 64, 64],\n",
+    "            \"encoder_hidden_states\": [1, 77, 1024],\n",
+    "            \"timestep\": [1],\n",
+    "        })"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "vae = load_mgx_model(\"vae_decoder\", {\"latent_sample\": [1, 4, 64, 64]})"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Import the remaining packages."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from diffusers import EulerDiscreteScheduler\n",
+    "from transformers import CLIPTokenizer\n",
+    "import torch\n",
+    "import numpy as np\n",
+    "from tqdm.auto import tqdm\n",
+    "from PIL import Image"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Time to load the scheduler and tokenizer from the original source."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "model_id = \"stabilityai/stable-diffusion-2-1\"\n",
+    "scheduler = EulerDiscreteScheduler.from_pretrained(model_id,\n",
+    "                                                   subfolder=\"scheduler\")\n",
+    "tokenizer = CLIPTokenizer.from_pretrained(model_id, subfolder=\"tokenizer\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Next, we will define all the steps one by one, to make the last step short and simple.\n",
+    "\n",
+    "The first step will be to tokenize the user prompt. It will make a `(1, 77)` shaped `input_ids`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def tokenize(input):\n",
+    "    return tokenizer([input],\n",
+    "                     padding=\"max_length\",\n",
+    "                     max_length=tokenizer.model_max_length,\n",
+    "                     truncation=True,\n",
+    "                     return_tensors=\"np\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Optional\n",
+    "test_tk = tokenize(\"test tokenizer to see the tokens\")\n",
+    "test_tk.input_ids.shape"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We run the tokenized prompt through the `Text Encoder` model. It expects the `(1, 77)` data as `int32`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Optional\n",
+    "text_encoder.get_parameter_shapes()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def get_embeddings(input):\n",
+    "    return np.array(\n",
+    "        text_encoder.run({\"input_ids\": input.input_ids.astype(np.int32)\n",
+    "                          })[0]).astype(np.float32)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Optional\n",
+    "test_emb = get_embeddings(tokenize(\"test tokenizer to see the tokens\"))\n",
+    "test_emb.shape"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "The other input of the model is latent representation (pure noise). It will be transformed into a 512x512 image later.\n",
+    "The last input will be the timestep."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def generate_latents(seed):\n",
+    "    return torch.randn(\n",
+    "        (1, 4, 64, 64),\n",
+    "        generator=torch.manual_seed(seed),\n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Optional\n",
+    "test_latents = generate_latents(42)\n",
+    "latents.shape"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Now we add two helpers to access and convert from torch to numpy with the proper datatype."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def get_scaled_sample(latents, t):\n",
+    "    return scheduler.scale_model_input(latents, t).numpy().astype(np.float32)\n",
+    "\n",
+    "\n",
+    "def get_timestep(t):\n",
+    "    return np.atleast_1d(t.numpy().astype(np.int64))  # convert 0D -> 1D"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "The UNet model will be run in a loop. It will predict the noise residual."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Optional\n",
+    "unet.get_parameter_shapes()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def denoise(sample, embeddings, timestep):\n",
+    "    return np.array(\n",
+    "        unet.run({\n",
+    "            \"sample\": sample,\n",
+    "            \"encoder_hidden_states\": embeddings,\n",
+    "            \"timestep\": timestep\n",
+    "        })[0])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Helpers to do the classifier-free guidance and computing the previous noisy sample."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def perform_guidance(noise_pred_uncond, noise_pred_text, scale):\n",
+    "    return noise_pred_uncond + scale * (noise_pred_text - noise_pred_uncond)\n",
+    "\n",
+    "def compute_previous(noise_pred, t, latents):\n",
+    "    # compute the previous noisy sample x_t -> x_t-1\n",
+    "    return scheduler.step(noise_pred, t, latents).prev_sample\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Scale and decode the image latents with VAE."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def scale_denoised(latents):\n",
+    "    return 1 / 0.18215 * latents\n",
+    "\n",
+    "\n",
+    "def decode(latents):\n",
+    "    return np.array(\n",
+    "        vae.run({\"latent_sample\": latents.numpy().astype(np.float32)})[0])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "And lastly, we need to convert it to an image to display or save."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def convert_to_rgb_image(image):\n",
+    "    image = np.clip(image / 2 + 0.5, 0, 1)\n",
+    "    image = np.transpose(image, (0, 2, 3, 1))\n",
+    "    images = (image * 255).round().astype(\"uint8\")\n",
+    "    return Image.fromarray(images[0])\n",
+    "\n",
+    "def save_image(pil_image, filename=\"output.png\"):\n",
+    "    pil_image.save(filename, format=\"png\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Feel free to play around with these params."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "prompt = \"a photograph of an astronaut riding a horse\"\n",
+    "negative_prompt = \"\"\n",
+    "steps = 20\n",
+    "seed = 13\n",
+    "scale = 7.0"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "And now, to put everything together and run the whole pipeline:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "scheduler.set_timesteps(steps)\n",
+    "\n",
+    "text_input, uncond_input = tokenize(prompt), tokenize(negative_prompt)\n",
+    "text_embeddings, uncond_embeddings = get_embeddings(\n",
+    "    text_input), get_embeddings(uncond_input)\n",
+    "latents = generate_latents(seed) * scheduler.init_noise_sigma\n",
+    "\n",
+    "for t in tqdm(scheduler.timesteps):\n",
+    "    sample = get_scaled_sample(latents, t)\n",
+    "    timestep = get_timestep(t)\n",
+    "\n",
+    "    noise_pred_uncond = denoise(sample, uncond_embeddings, timestep)\n",
+    "    noise_pred_text = denoise(sample, text_embeddings, timestep)\n",
+    "\n",
+    "    noise_pred = perform_guidance(noise_pred_uncond, noise_pred_text, scale)\n",
+    "    latents = compute_previous(torch.from_numpy(noise_pred), t, latents)\n",
+    "\n",
+    "latents = scale_denoised(latents)\n",
+    "result = decode(latents)\n",
+    "image = convert_to_rgb_image(result)\n",
+    "\n",
+    "# show the image\n",
+    "image"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "If you like the generated image, save it with the following:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "save_image(image, \"output.png\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "sd_venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.12"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}