Merge remote-tracking branch 'origin/develop' into transformer-opts

.github/workflows/ci.yaml

@@ -15,7 +15,8 @@ jobs:
 
     steps:
     - name: Free space
-      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia*
+      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia* /usr/local/lib/android
+
     - uses: actions/checkout@v2
 
     # In this step, this action saves a list of existing images,
@@ -63,7 +64,7 @@ jobs:
 
     steps:
     - name: Free space
-      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia*
+      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia* /usr/local/lib/android
     - uses: actions/checkout@v2
 
     # In this step, this action saves a list of existing images,
@@ -108,7 +109,7 @@ jobs:
 
     steps:
     - name: Free space
-      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia*
+      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia* /usr/local/lib/android
     - uses: actions/checkout@v2
 
     # In this step, this action saves a list of existing images,
@@ -143,7 +144,7 @@ jobs:
 
     steps:
     - name: Free space
-      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia*
+      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia* /usr/local/lib/android
     - uses: actions/checkout@v2
     - name: Set up Python
       uses: actions/setup-python@v2
@@ -182,7 +183,7 @@ jobs:
 
     steps:
     - name: Free space
-      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia*
+      run: sudo rm -rf /usr/local/android /usr/share/dotnet /usr/local/share/boost /opt/ghc /usr/local/share/chrom* /usr/share/swift /usr/local/julia* /usr/local/lib/android
     - uses: actions/checkout@v2
     - name: Set up Python
       uses: actions/setup-python@v2

CMakeLists.txt

@@ -93,11 +93,14 @@ rocm_enable_clang_tidy(
         modernize-*
         performance-*
         readability-*
-        -bugprone-signed-char-misuse
+        -bugprone-easily-swappable-parameters
+        -bugprone-implicit-widening-of-multiplication-result
         -bugprone-macro-parentheses
+        -bugprone-signed-char-misuse
         # Disable the aliased reserved identifiers
         -cert-dcl37-c
         -cert-dcl51-cpp
+        -cert-err33-c
         -cert-str34-c
         # Disable all alpha checks by default
         -clang-analyzer-alpha*
@@ -127,6 +130,7 @@ rocm_enable_clang_tidy(
         -cppcoreguidelines-pro-type-union-access
         -cppcoreguidelines-pro-type-vararg
         -cppcoreguidelines-special-member-functions
+        -cppcoreguidelines-virtual-class-destructor
         -google-readability-*
         -google-runtime-int
         -google-runtime-references
@@ -144,8 +148,10 @@ rocm_enable_clang_tidy(
         -readability-convert-member-functions-to-static
         -readability-else-after-return
         -readability-function-cognitive-complexity
+        -readability-identifier-length
         -readability-named-parameter
         -readability-redundant-string-init
+        -readability-suspicious-call-argument
         -readability-uppercase-literal-suffix
         -*-avoid-c-arrays
         -*-explicit-constructor
@@ -216,6 +222,7 @@ rocm_enable_cppcheck(
         CPPCHECK=1
         __device__=
         __host__=
+        __global__=
 )
 
 enable_testing()

Dockerfile

-FROM ubuntu:18.04
+FROM ubuntu:20.04
 
 ARG PREFIX=/usr/local
 
@@ -6,7 +6,7 @@ ARG PREFIX=/usr/local
 RUN dpkg --add-architecture i386
 
 # Add rocm repository
-RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/4.5/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
+RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.0.2/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
@@ -16,16 +16,12 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     cmake \
     curl \
     doxygen \
-    g++-5 \
     g++-7 \
     gdb \
     git \
     lcov \
     locales \
     pkg-config \
-    python \
-    python-dev \
-    python-pip \
     python3 \
     python3-dev \
     python3-pip \

cmake/Embed.cmake

@@ -94,5 +94,6 @@ function(add_embed_library EMBED_NAME)
     generate_embed_source(${EMBED_NAME} SRC ${SRC_FILE} HEADER ${HEADER_FILE} OBJECTS ${OUTPUT_FILES} SYMBOLS ${SYMBOLS})
     add_library(${EMBED_NAME} STATIC ${OUTPUT_FILES} "${SRC_FILE}")
     target_include_directories(${EMBED_NAME} PUBLIC "${EMBED_DIR}/include")
+    target_compile_options(${EMBED_NAME} PRIVATE -Wno-reserved-identifier)
     set_target_properties(${EMBED_NAME} PROPERTIES POSITION_INDEPENDENT_CODE On)
 endfunction()

cmake/EnableCompilerWarnings.cmake

@@ -96,6 +96,7 @@ else()
                 -Wno-gnu-zero-variadic-macro-arguments
                 -Wno-missing-prototypes
                 -Wno-nested-anon-types
+                -Wno-option-ignored
                 -Wno-padded
                 -Wno-shorten-64-to-32
                 -Wno-sign-conversion

doc/src/reference/py.rst

@@ -146,6 +146,13 @@ module
     :param list[module] mod_args: optional list of module arguments to the operator.
     :rtype instruction
 
+.. py:method:: add_literal(data)
+
+    Adds constant or literal data of provided shape into the module from python buffer which includes numpy array.    
+
+    :param py::buffer data: Python buffer or numpy array 
+    :rtype instruction 
+
 .. py:method:: add_parameter(name, shape)
     
     Adds a parameter to the module with provided name and shape.

examples/nlp/python_bert_squad/BERT-Squad.ipynb

@@ -62,7 +62,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!wget -nc https://github.com/onnx/models/raw/master/text/machine_comprehension/bert-squad/model/bertsquad-10.onnx"
+    "!wget -nc https://github.com/onnx/models/blob/main/text/machine_comprehension/bert-squad/model/bertsquad-10.onnx"
    ]
   },
   {

examples/nlp/python_bert_squad/README.md

@@ -23,7 +23,7 @@ unzip uncased_L-12_H-768_A-12.zip
 ```
 5) Get BERT ONNX model (bertsquad-10.onnx):
 ```
-wget https://github.com/onnx/models/raw/master/text/machine_comprehension/bert-squad/model/bertsquad-10.onnx
+wget https://github.com/onnx/models/blob/main/text/machine_comprehension/bert-squad/model/bertsquad-10.onnx
 ```
 6) Run the inference, it will compile and run the model on three questions and small data provided in `inputs.json`:
 ```

hip-clang.docker

-FROM ubuntu:18.04
+FROM ubuntu:20.04
 
 ARG PREFIX=/usr/local
 
@@ -6,7 +6,7 @@ ARG PREFIX=/usr/local
 RUN dpkg --add-architecture i386
 
 # Add rocm repository
-RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/4.5/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
+RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.0.2/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
@@ -20,9 +20,6 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     git \
     lcov \
     pkg-config \
-    python \
-    python-dev \
-    python-pip \
     python3 \
     python3-dev \
     python3-pip \

src/api/api.cpp

@@ -1072,6 +1072,22 @@ migraphx_module_add_instruction_with_mod_args(migraphx_instruction_t* out,
     return api_error_result;
 }
 
+extern "C" migraphx_status migraphx_module_add_literal(migraphx_instruction_t* out,
+                                                       migraphx_module_t module,
+                                                       const_migraphx_shape_t shape,
+                                                       const char* buffer)
+{
+    auto api_error_result = migraphx::try_([&] {
+        if(module == nullptr)
+            MIGRAPHX_THROW(migraphx_status_bad_param, "Bad parameter module: Null pointer");
+        if(shape == nullptr)
+            MIGRAPHX_THROW(migraphx_status_bad_param, "Bad parameter shape: Null pointer");
+        *out = allocate<migraphx_instruction_t>(
+            (module->object).add_literal((shape->object), (buffer)));
+    });
+    return api_error_result;
+}
+
 extern "C" migraphx_status migraphx_module_add_parameter(migraphx_instruction_t* out,
                                                          migraphx_module_t module,
                                                          const char* name,

src/api/include/migraphx/migraphx.h

@@ -258,6 +258,11 @@ migraphx_status migraphx_module_add_instruction_with_mod_args(migraphx_instructi
                                                               migraphx_instructions_t args,
                                                               migraphx_modules_t module_refs);
 
+migraphx_status migraphx_module_add_literal(migraphx_instruction_t* out,
+                                            migraphx_module_t module,
+                                            const_migraphx_shape_t shape,
+                                            const char* buffer);
+
 migraphx_status migraphx_module_add_parameter(migraphx_instruction_t* out,
                                               migraphx_module_t module,
                                               const char* name,

src/api/include/migraphx/migraphx.hpp

@@ -314,6 +314,7 @@ struct interface_base : Base
                 T** y = reinterpret_cast<T**>(out);
                 T* x  = reinterpret_cast<T*>(input);
                 assert(x != nullptr and y != nullptr and *y == nullptr);
+                // cppcheck-suppress useSmartPointer
                 *y = new T(*x); // NOLINT
             });
         };
@@ -339,6 +340,7 @@ struct interface_base : Base
     template <class T, class Setter, class F>
     void set_auto_fp(Setter setter, F f)
     {
+        // cppcheck-suppress constParameter
         return set_fp<T>(setter, [=](T& obj, auto out, auto... xs) {
             auto_invoke(f, out, obj, auto_convert_param(rank<2>{}, xs)...);
         });
@@ -760,6 +762,15 @@ struct module
         return instruction(op_ins, own{});
     }
 
+    template <typename T>
+    instruction add_literal(const migraphx::shape& s, T* buffer)
+    {
+        migraphx_instruction_t literal_ins;
+        const auto* buffer_ptr = reinterpret_cast<const char*>(buffer);
+        call(&migraphx_module_add_literal, &literal_ins, mm.get(), s.get_handle_ptr(), buffer_ptr);
+        return instruction(literal_ins, own{});
+    }
+
     instruction add_parameter(const std::string& name, shape s)
     {
         migraphx_instruction_t param_ins;

src/api/migraphx.py

@@ -212,6 +212,9 @@ def module(h):
                         module_refs='std::vector<migraphx::module*>'),
              fname='add_instruction',
              returns='migraphx::instruction_ref')
+    h.method('add_literal',
+             api.params(shape='const migraphx::shape&', buffer='const char*'),
+             returns='migraphx::instruction_ref')
     h.method('add_parameter',
              api.params(name='const char*', shape='const migraphx::shape&'),
              returns='migraphx::instruction_ref')

src/driver/marker_roctx.cpp

@@ -17,7 +17,7 @@ class marker_roctx
     std::function<int(const char*)> sym_roctx_range_push;
     std::function<int()> sym_roctx_range_pop;
 
-    uint64_t range_id;
+    uint64_t range_id = 0;
 
     public:
     marker_roctx()

src/include/migraphx/filesystem.hpp

@@ -3,7 +3,10 @@
 
 #include <migraphx/config.hpp>
 
-#if defined(__has_include) && !defined(CPPCHECK)
+#if defined(CPPCHECK)
+#define MIGRAPHX_HAS_FILESYSTEM 1
+#define MIGRAPHX_HAS_FILESYSTEM_TS 1
+#elif defined(__has_include)
 #if __has_include(<filesystem>) && __cplusplus >= 201703L
 #define MIGRAPHX_HAS_FILESYSTEM 1
 #else

src/include/migraphx/optional.hpp

@@ -3,7 +3,10 @@
 
 #include <migraphx/config.hpp>
 
-#if defined(__has_include) && !defined(CPPCHECK)
+#if defined(CPPCHECK)
+#define MIGRAPHX_HAS_OPTIONAL 1
+#define MIGRAPHX_HAS_OPTIONAL_TS 1
+#elif defined(__has_include)
 #if __has_include(<optional>) && __cplusplus >= 201703L
 #define MIGRAPHX_HAS_OPTIONAL 1
 #else

src/onnx/parse_mean.cpp

@@ -24,13 +24,16 @@ struct parse_mean : op_parser<parse_mean>
         auto divisor = info.add_literal(
             migraphx::literal{migraphx::shape{args[0]->get_shape().type()}, {num_data}});
 
-        return std::accumulate(args.begin(), args.end(), args[0], [&](auto& mean, auto& data_i) {
-            // Pre-divide each tensor element-wise by n to reduce risk of overflow during summation
-            data_i = info.add_broadcastable_binary_op("div", data_i, divisor);
-
-            if(data_i != args[0])
-                return info.add_broadcastable_binary_op("add", mean, data_i);
-            return data_i;
+        // TODO: Only divide when using floating-point
+        return std::accumulate(args.begin() + 1,
+                               args.end(),
+                               info.add_broadcastable_binary_op("div", args[0], divisor),
+                               [&](auto mean, auto data_i) {
+                                   // Pre-divide each tensor element-wise by n to reduce risk of
+                                   // overflow during summation
+                                   auto div =
+                                       info.add_broadcastable_binary_op("div", data_i, divisor);
+                                   return info.add_broadcastable_binary_op("add", mean, div);
                                });
     }
 };

src/py/migraphx_py.cpp

@@ -273,6 +273,14 @@ MIGRAPHX_PYBIND11_MODULE(migraphx, m)
             py::arg("op"),
             py::arg("args"),
             py::arg("mod_args") = std::vector<migraphx::module*>{})
+        .def(
+            "add_literal",
+            [](migraphx::module& mm, py::buffer data) {
+                py::buffer_info info = data.request();
+                auto literal_shape   = to_shape(info);
+                return mm.add_literal(literal_shape, reinterpret_cast<char*>(info.ptr));
+            },
+            py::arg("data"))
         .def(
             "add_parameter",
             [](migraphx::module& mm, const std::string& name, const migraphx::shape shape) {

src/simplify_algebra.cpp

@@ -995,7 +995,7 @@ struct find_split_transpose
         auto axis = any_cast<op::slice>(slc->get_operator()).axes.front();
         auto it   = std::find(perm.begin(), perm.end(), axis);
         assert(it != perm.end());
-        auto axis_new = static_cast<int64_t>(std::distance(perm.begin(), it));
+        int64_t axis_new = std::distance(perm.begin(), it);
 
         for(auto in : split_outputs)
         {

src/targets/cpu/include/migraphx/cpu/parallel.hpp

@@ -7,7 +7,16 @@
 #ifdef MIGRAPHX_DISABLE_OMP
 #include <migraphx/par_for.hpp>
 #else
+
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wreserved-identifier"
+#endif
 #include <omp.h>
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
+
 #endif
 
 namespace migraphx {