Merge branch 'develop' of https://github.com/ROCmSoftwarePlatform/AMDMIGraphX into nhwc_workaround

.githooks/pre-commit

@@ -4,62 +4,7 @@
 # are installed, and if so, uses the installed version to format
 # the staged changes.
 
-base=clang-format-10
-format=""
-yapf_base=yapf
-yapf_format=""
-
-use_yapf=true
-
-# Redirect output to stderr.
-exec 1>&2
-
-# check if clang-format is installed
-type "$base" >/dev/null 2>&1 && format="$base"
-
-# no versions of clang-format are installed
-if [ -z "$format" ]
-then
-    echo "$base is not installed. Pre-commit hook will not be executed."
-    exit 0
-fi
-
-# check if yapf is installed
-type "$yapf_base" >/dev/null 2>&1 && yapf_format="$yapf_base"
-
-# no versions of yapf are installed
-if [ -z "$yapf_format" ]
-then
-    echo "$yapf_base is not installed. Pre-commit hook for python files will not be executed"
-    use_yapf=false
-fi
-
 # Do everything from top - level
 cd $(git rev-parse --show-toplevel)
 
-if git rev-parse --verify HEAD >/dev/null 2>&1
-then
-    against=HEAD
-else
-    # Initial commit: diff against an empty tree object
-    against=16bbb57
-fi
-
-# do the formatting
-for file in $(git diff-index --cached --name-only $against | grep -E '\.h$|\.hpp$|\.cpp$|\.cl$|\.c$|\.h\.in$|\.hpp\.in$|\.cpp\.in$|\.py$')
-do
-    if [ -e "$file" ]
-    then
-        if [ $(echo $file | grep -E '\.py$') ]
-        then
-            if $use_yapf
-            then
-                echo "$yapf_format $file"
-                "$yapf_format" -i "$file"
-            fi
-        else
-            echo "$format $file"
-            "$format" -i -style=file "$file"
-        fi
-    fi
-done
+python3 tools/format.py -q -i HEAD

.github/workflows/ci.yaml

@@ -144,6 +144,8 @@ jobs:
     runs-on: ROCM-Ubuntu
     steps:
     - uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
 
     # In this step, this action saves a list of existing images,
     # the cache is created without them in the post run.
@@ -164,19 +166,8 @@ jobs:
       shell: bash -c "docker run -i -v=$GITHUB_WORKSPACE:/data -w /data migraphx bash < {0}"
       run: |
         set -e
-        find . -iname '*.h' \
-              -o -iname '*.hpp' \
-              -o -iname '*.cpp' \
-              -o -iname '*.h.in' \
-              -o -iname '*.hpp.in' \
-              -o -iname '*.cpp.in' \
-              -o -iname '*.cl' \
-              -o -iname '*.c' \
-          | grep -v 'build/' \
-          | xargs -n 1 -P 1 -I{} -t sh -c 'clang-format-10 -style=file {} | diff - {}'
-          find . -iname '*.py' \
-          | grep -v 'build/'  \
-          | xargs -n 1 -P 1 -I{} -t sh -c 'yapf {} | diff - {}'
+        git config --global --add safe.directory /data
+        python3 tools/format.py origin/${{ github.event_name == 'pull_request' && github.base_ref || 'develop' }}
 
   pyflakes:
     runs-on: ubuntu-20.04
@@ -323,6 +314,7 @@ jobs:
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       run: |
+        set +x
         gh extension install actions/gh-actions-cache --pin v1.0.1
         gh actions-cache delete ${{ steps.ccache_restore.outputs.cache-matched-key }} --confirm
 
@@ -439,6 +431,7 @@ jobs:
       env:
         GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       run: |
+        set +x
         gh extension install actions/gh-actions-cache
         gh actions-cache delete ${{ steps.ccache_restore_fpga.outputs.cache-matched-key }} --confirm
       continue-on-error: true

.github/workflows/performance.yaml

@@ -12,7 +12,7 @@ on:
       rocm_release:
         description: ROCm Version
         required: true
-        default: '5.5'
+        default: '5.6'
       performance_reports_repo:
         description: Repository where performance reports are stored
         required: true
@@ -42,13 +42,15 @@ on:
         required: true
         default: '-r'
 
-concurrency: "perftest-${{ github.head_ref ||  github.base_ref || 'schedule' }}"
+concurrency: 
+  group: "perftest-${{ github.head_ref ||  github.base_ref || 'schedule' }}"
+  cancel-in-progress: true
 
 jobs:
   release:
     uses: ROCmSoftwarePlatform/migraphx-benchmark/.github/workflows/perf-test.yml@main
     with:
-      rocm_release: ${{ github.event.inputs.rocm_release || '5.5' }}
+      rocm_release: ${{ github.event.inputs.rocm_release || '5.6' }}
       result_number: ${{ github.event.inputs.result_number || '10' }}
       flags: ${{ github.event.inputs.flags || '-r' }}
       performance_reports_repo: ${{ github.event.inputs.performance_reports_repo || 'ROCmSoftwarePlatform/migraphx-reports' }}

.gitignore

@@ -70,5 +70,13 @@ docs/_images
 docs/_static
 docs/_templates
 docs/.doxygen/docBin
+docs/.doxygen/DoxygenWarningLog.txt
 docs/_doxygen
+docs/html
 /_readthedocs
+
+# JetBrains config directories (ignoring symlinks)
+.idea/
+cmake-build*/
+build*/
+

Dockerfile

@@ -19,7 +19,6 @@ RUN sh -c "echo 'Package: *\nPin: release o=repo.radeon.com\nPin-priority: 600'
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
     apt-utils \
     build-essential \
-    clang-format-10 \
     cmake \
     curl \
     doxygen \
@@ -97,7 +96,8 @@ RUN /download_models.sh && rm /download_models.sh
 # Install latest ccache version
 RUN cget -p $PREFIX install facebook/zstd@v1.4.5 -X subdir -DCMAKE_DIR=build/cmake
 RUN cget -p $PREFIX install ccache@v4.1 -DENABLE_TESTING=OFF
-RUN cget -p /opt/cmake install kitware/cmake@v3.24.3
+RUN cget -p /opt/cmake install kitware/cmake@v3.26.4
+
 
 COPY ./test/onnx/.onnxrt-commit /
 

Jenkinsfile

@@ -15,13 +15,11 @@ def rocmtestnode(Map conf) {
         def compiler = bconf.get("compiler", "/opt/rocm/llvm/bin/clang++")
         def flags = bconf.get("flags", "")
         def gpu_debug = bconf.get("gpu_debug", "0")
-        def hiprtc_workarounds = bconf.get("hiprtc_workarounds", "0")
         def cmd = """
             ulimit -c unlimited
             echo "leak:dnnl::impl::malloc" > suppressions.txt
             export LSAN_OPTIONS="suppressions=\$(pwd)/suppressions.txt"
             export MIGRAPHX_GPU_DEBUG=${gpu_debug}
-            export MIGRAPHX_ENABLE_HIPRTC_WORKAROUNDS=${hiprtc_workarounds}
             export CXX=${compiler}
             export CXXFLAGS='-Werror'
             env
@@ -90,9 +88,9 @@ def rocmnodename(name) {
     } else if(name == "navi21") {
         node_name = "${rocmtest_name} && navi21";
     } else if(name == "mi100+") {
-        node_name = "${rocmtest_name} && (gfx908 || gfx90a)";
+        node_name = "${rocmtest_name} && (gfx908 || gfx90a) && !vm";
     } else if(name == "cdna") {
-        node_name = "${rocmtest_name} && (gfx908 || gfx90a || vega)";
+        node_name = "${rocmtest_name} && (gfx908 || gfx90a || vega) && !vm";
     } else if(name == "nogpu") {
         node_name = "${rocmtest_name} && nogpu";
     }
@@ -109,7 +107,7 @@ rocmtest clang_debug: rocmnode('cdna') { cmake_build ->
     stage('hipRTC Debug') {
         def sanitizers = "undefined"
         def debug_flags = "-g -O2 -fsanitize=${sanitizers} -fno-sanitize-recover=${sanitizers}"
-        cmake_build(flags: "-DCMAKE_BUILD_TYPE=debug -DMIGRAPHX_ENABLE_PYTHON=Off -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags}' -DCMAKE_C_FLAGS_DEBUG='${debug_flags}' -DMIGRAPHX_USE_HIPRTC=On", gpu_debug: true, hiprtc_workarounds: true)
+        cmake_build(flags: "-DCMAKE_BUILD_TYPE=debug -DMIGRAPHX_ENABLE_PYTHON=Off -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags}' -DCMAKE_C_FLAGS_DEBUG='${debug_flags}' -DMIGRAPHX_USE_HIPRTC=On", gpu_debug: true)
     }
 }, clang_release: rocmnode('cdna') { cmake_build ->
     stage('Hip Clang Release') {

hip-clang.docker

@@ -12,7 +12,6 @@ RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.6
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
     apt-utils \
     build-essential \
-    clang-format-10 \
     cmake \
     curl \
     doxygen \

requirements.txt

@@ -24,7 +24,7 @@
 google/protobuf@v3.11.0 -DCMAKE_POSITION_INDEPENDENT_CODE=On -X subdir -Dprotobuf_BUILD_TESTS=Off
 nlohmann/json@v3.8.0
 live-clones/blaze@v3.8 -X header -DHEADER_DIR=blaze -H sha256:d0ff011f47538285178908ea5f2cab46bb6a8f55b1edb6e03224a82dbc1a3212
-ROCmSoftwarePlatform/half@rocm-5.4.2
+ROCmSoftwarePlatform/half@rocm-5.6.0
 pybind/pybind11@d159a563383d10c821ba7b2a71905d1207db6de4 --build
 msgpack/msgpack-c@cpp-3.3.0 -DMSGPACK_BUILD_TESTS=Off
 sqlite3@3.17 -DCMAKE_POSITION_INDEPENDENT_CODE=On

src/CMakeLists.txt

@@ -134,6 +134,7 @@ register_migraphx_ops(
     cosh
     cos
     dequantizelinear
+    dimensions_of
     div
     dot
     elu

src/common.cpp

@@ -150,8 +150,8 @@ insert_common_args(module& m, instruction_ref ins, std::vector<instruction_ref>
         auto c_type       = compute_common_types(input_shapes);
         auto c_dyn_dims   = compute_common_dyn_dims(input_shapes);
 
-        // following should work for a static or dynamic shape
-        if(inputs[0]->get_shape().dyn_dims() != c_dyn_dims)
+        auto s0 = inputs[0]->get_shape();
+        if(not s0.dynamic() or s0.dyn_dims() != c_dyn_dims)
         {
             inputs[0] = m.insert_instruction(
                 ins, make_op("multibroadcast", {{"out_dyn_dims", to_value(c_dyn_dims)}}), inputs);
@@ -159,7 +159,8 @@ insert_common_args(module& m, instruction_ref ins, std::vector<instruction_ref>
         std::transform(inputs.begin() + 1, inputs.end(), inputs.begin() + 1, [&](auto input) {
             // uses previous input to avoid recalculating the common shape from the
             // full set of input shapes at runtime
-            if(input->get_shape().dyn_dims() != c_dyn_dims)
+            auto s = input->get_shape();
+            if(not s.dynamic() or s.dyn_dims() != c_dyn_dims)
             {
                 return m.insert_instruction(
                     ins,

src/generate.cpp

@@ -26,7 +26,7 @@
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 
-argument fill_argument(shape s, unsigned long value)
+argument fill_argument(shape s, double value)
 {
     argument result;
     if(s.type() == shape::tuple_type)

src/include/migraphx/allocation_model.hpp

@@ -96,7 +96,7 @@ struct allocation_model
     {
         using std::swap;
         auto* derived = this->any_cast<PrivateDetailTypeErasedT>();
-        if(derived and private_detail_te_handle_mem_var.unique())
+        if(derived and private_detail_te_handle_mem_var.use_count() == 1)
         {
             *derived = std::forward<PrivateDetailTypeErasedT>(value);
         }
@@ -267,7 +267,7 @@ struct allocation_model
     private_detail_te_handle_base_type& private_detail_te_get_handle()
     {
         assert(private_detail_te_handle_mem_var != nullptr);
-        if(not private_detail_te_handle_mem_var.unique())
+        if(private_detail_te_handle_mem_var.use_count() > 1)
             private_detail_te_handle_mem_var = private_detail_te_handle_mem_var->clone();
         return *private_detail_te_handle_mem_var;
     }

src/include/migraphx/concat_opt.hpp

@@ -88,7 +88,7 @@ struct concat_optimization
     {
         using std::swap;
         auto* derived = this->any_cast<PrivateDetailTypeErasedT>();
-        if(derived and private_detail_te_handle_mem_var.unique())
+        if(derived and private_detail_te_handle_mem_var.use_count() == 1)
         {
             *derived = std::forward<PrivateDetailTypeErasedT>(value);
         }
@@ -233,7 +233,7 @@ struct concat_optimization
     private_detail_te_handle_base_type& private_detail_te_get_handle()
     {
         assert(private_detail_te_handle_mem_var != nullptr);
-        if(not private_detail_te_handle_mem_var.unique())
+        if(private_detail_te_handle_mem_var.use_count() > 1)
             private_detail_te_handle_mem_var = private_detail_te_handle_mem_var->clone();
         return *private_detail_te_handle_mem_var;
     }

src/include/migraphx/context.hpp

@@ -118,7 +118,7 @@ struct context
     {
         using std::swap;
         auto* derived = this->any_cast<PrivateDetailTypeErasedT>();
-        if(derived and private_detail_te_handle_mem_var.unique())
+        if(derived and private_detail_te_handle_mem_var.use_count() == 1)
         {
             *derived = std::forward<PrivateDetailTypeErasedT>(value);
         }
@@ -373,7 +373,7 @@ struct context
     private_detail_te_handle_base_type& private_detail_te_get_handle()
     {
         assert(private_detail_te_handle_mem_var != nullptr);
-        if(not private_detail_te_handle_mem_var.unique())
+        if(private_detail_te_handle_mem_var.use_count() > 1)
             private_detail_te_handle_mem_var = private_detail_te_handle_mem_var->clone();
         return *private_detail_te_handle_mem_var;
     }

src/include/migraphx/generate.hpp

@@ -117,14 +117,14 @@ auto generate_tensor_data(const migraphx::shape& s, unsigned long seed = 0)
 }
 
 template <class T>
-auto fill_tensor_data(const migraphx::shape& s, unsigned long value = 0)
+auto fill_tensor_data(const migraphx::shape& s, double value = 0)
 {
     auto result = make_shared_array<T>(s.element_space());
     std::generate(result.get(), result.get() + s.element_space(), [=] { return value; });
     return result;
 }
 
-MIGRAPHX_EXPORT argument fill_argument(shape s, unsigned long value = 0);
+MIGRAPHX_EXPORT argument fill_argument(shape s, double value = 0);
 
 MIGRAPHX_EXPORT argument generate_argument(shape s, unsigned long seed = 0);
 

src/include/migraphx/marker.hpp

@@ -80,7 +80,7 @@ struct marker
     {
         using std::swap;
         auto* derived = this->any_cast<PrivateDetailTypeErasedT>();
-        if(derived and private_detail_te_handle_mem_var.unique())
+        if(derived and private_detail_te_handle_mem_var.use_count() == 1)
         {
             *derived = std::forward<PrivateDetailTypeErasedT>(value);
         }
@@ -233,7 +233,7 @@ struct marker
     private_detail_te_handle_base_type& private_detail_te_get_handle()
     {
         assert(private_detail_te_handle_mem_var != nullptr);
-        if(not private_detail_te_handle_mem_var.unique())
+        if(private_detail_te_handle_mem_var.use_count() > 1)
             private_detail_te_handle_mem_var = private_detail_te_handle_mem_var->clone();
         return *private_detail_te_handle_mem_var;
     }

src/include/migraphx/op/convert.hpp

@@ -66,7 +66,19 @@ struct convert : unary<convert>
         auto type = target_type;
         return [type](auto x) {
             auto y = x;
-            shape::visit(type, [&](auto as) { y = as(x); });
+            shape::visit(type, [&](auto as) {
+                // clamping value between target_type's max and min doesn't work for NaNs,
+                if(std::isnan(x))
+                {
+                    y = as.nan();
+                }
+                else
+                {
+                    // clamp overflowing/underflowing values to min()/max() instead of +/-infinity
+                    // during downcasting
+                    y = std::min(std::max(as(x), as.min()), as.max());
+                }
+            });
             return y;
         };
     }

src/include/migraphx/op/dimensions_of.hpp

+/*
+ * 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.
+ */
+#ifndef MIGRAPHX_GUARD_OPERATORS_DIMENSIONS_OF_HPP
+#define MIGRAPHX_GUARD_OPERATORS_DIMENSIONS_OF_HPP
+
+#include <migraphx/check_shapes.hpp>
+#include <migraphx/argument.hpp>
+#include <migraphx/dyn_output.hpp>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+namespace op {
+
+/**
+ * Returns the dimensions of the input argument from starting axis to ending axis.
+ * Atleast `end` must be set to use this operator (set `end` to ndim for default ONNX behavior of
+ * `Shape` operator) This should only be used for dynamic shapes as this can be simplified to a
+ * literal for static shapes.
+ */
+struct dimensions_of
+{
+    std::size_t start = 0;
+    std::size_t end   = 0;
+
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return pack(f(self.start, "start"), f(self.end, "end"));
+    }
+
+    std::string name() const { return "dimensions_of"; }
+
+    shape compute_shape(const std::vector<shape>& inputs) const
+    {
+        check_shapes{inputs, *this, true}.has(1);
+        if(start >= end)
+        {
+            MIGRAPHX_THROW("DIMENSIONS_OF: start >= end. start = " + std::to_string(start) +
+                           ", end = " + std::to_string(end));
+        }
+        return shape{shape::int64_type, {end - start}};
+    }
+
+    argument compute(const shape& output_shape, std::vector<argument> args) const
+    {
+        argument result{output_shape};
+        auto input_lens = args[0].get_shape().lens();
+        result.visit([&](auto output) {
+            std::copy(input_lens.cbegin() + start, input_lens.cbegin() + end, output.begin());
+        });
+        return result;
+    }
+};
+
+} // namespace op
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/include/migraphx/op/multibroadcast.hpp

@@ -69,7 +69,7 @@ struct multibroadcast
 
         auto make_bcast_strides = [&](std::vector<std::size_t> bcast_lens, std::size_t offset) {
             std::vector<size_t> bcast_strides(bcast_lens.size(), 0);
-            for(std::ptrdiff_t i = s0.lens().size() - 1; i >= 0; i--)
+            for(std::ptrdiff_t i = s0.ndim() - 1; i >= 0; i--)
             {
                 if(bcast_lens[i + offset] == s0.lens()[i])
                 {
@@ -84,13 +84,13 @@ struct multibroadcast
             if(s0.dynamic())
                 MIGRAPHX_THROW(
                     "MULTIBROADCAST: Single dynamic input shape not supported.  Use two inputs.");
-            if(s0.lens().size() > output_lens.size())
+            if(s0.ndim() > output_lens.size())
             {
                 MIGRAPHX_THROW("MULTIBROADCAST: input dimensions should <= output size");
             }
 
-            auto offset = output_lens.size() - s0.lens().size();
-            for(std::ptrdiff_t i = s0.lens().size() - 1; i >= 0; i--)
+            auto offset = output_lens.size() - s0.ndim();
+            for(std::ptrdiff_t i = s0.ndim() - 1; i >= 0; i--)
             {
                 if(output_lens[i + offset] != s0.lens()[i] and s0.lens()[i] != 1)
                 {
@@ -119,7 +119,7 @@ struct multibroadcast
             {
                 // output_lens will not be set for 2+ input version
                 auto bcast_lens    = compute_common_lens(inputs);
-                auto offset        = bcast_lens.size() - s0.lens().size();
+                auto offset        = bcast_lens.size() - s0.ndim();
                 auto bcast_strides = make_bcast_strides(bcast_lens, offset);
                 return {t, std::move(bcast_lens), std::move(bcast_strides)};
             }

src/include/migraphx/op/prefix_scan_op.hpp

 /*
  * The MIT License (MIT)
  *
- * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
+ * 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
@@ -21,6 +21,7 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  * THE SOFTWARE.
  */
+
 #ifndef MIGRAPHX_GUARD_OPERATORS_SCAN_OP_HPP
 #define MIGRAPHX_GUARD_OPERATORS_SCAN_OP_HPP
 
@@ -37,6 +38,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
+/**
+ * Parent struct for prefix scan operations.  A prefix scan is equivalent to the C++
+ * std::exclusive_scan or std::inclusive_scan.  Given a list of numbers, a prefix scan
+ * sum op returns an equal size list of running totals of the values.  Other operations
+ * besides addition can be supported by their own child ops.
+ */
 template <class Derived>
 struct prefix_scan_op : op_name<Derived>
 {
@@ -60,9 +67,13 @@ struct prefix_scan_op : op_name<Derived>
 
     shape normalize_compute_shape(std::vector<shape> inputs) const
     {
-        check_shapes{inputs, *this}.has(1);
+        check_shapes{inputs, *this, true}.has(1);
         auto s = inputs.front();
-        if(s.broadcasted())
+        if(s.dynamic())
+        {
+            return s;
+        }
+        else if(s.broadcasted())
         {
             return {s.type(), s.lens()};
         }
@@ -72,8 +83,9 @@ struct prefix_scan_op : op_name<Derived>
         }
     }
 
-    argument compute(const shape& output_shape, std::vector<argument> args) const
+    argument compute(const dyn_output& dyn_out, std::vector<argument> args) const
     {
+        shape output_shape(dyn_out.computed_shape);
         argument result{output_shape};
         auto s = args[0].get_shape();
         if(s == output_shape)