Merge branch 'dyn_broadcast' of github.com:ROCmSoftwarePlatform/AMDMIGraphX into dyn_unsqueeze

.github/workflows/ci.yaml

@@ -11,7 +11,7 @@ jobs:
         with:
           access_token: ${{ github.token }}
   tidy:
-    runs-on: ubuntu-18.04
+    runs-on: ubuntu-20.04
 
     steps:
     - name: Free space
@@ -61,7 +61,7 @@ jobs:
         make -j2 -k onnx-proto tf-proto tidy
 
   cppcheck:
-    runs-on: ubuntu-18.04
+    runs-on: ubuntu-20.04
 
     steps:
     - name: Free space
@@ -106,7 +106,7 @@ jobs:
         make -j2 cppcheck
 
   format:
-    runs-on: ubuntu-18.04
+    runs-on: ubuntu-20.04
 
     steps:
     - name: Free space
@@ -142,7 +142,7 @@ jobs:
           | xargs -n 1 -P 1 -I{} -t sh -c 'yapf {} | diff - {}'
 
   pyflakes:
-    runs-on: ubuntu-18.04
+    runs-on: ubuntu-20.04
 
     steps:
     - name: Free space
@@ -163,7 +163,7 @@ jobs:
         mypy tools/api.py
 
   licensing:
-    runs-on: ubuntu-18.04
+    runs-on: ubuntu-20.04
 
     steps:
     - name: Free space
@@ -190,7 +190,6 @@ jobs:
     strategy:
       matrix:
         os:
-          - ubuntu-18.04
           - ubuntu-20.04
         configuration:
           - debug
@@ -204,7 +203,7 @@ jobs:
     - name: Set up Python
       uses: actions/setup-python@v2
       with:
-        python-version: 3.6
+        python-version: 3.7
     - name: Cache dependencies
       # Ignore the failure of a step and avoid terminating the job.
       continue-on-error: true
@@ -287,7 +286,6 @@ jobs:
     strategy:
       matrix:
         os:
-          - ubuntu-18.04
           - ubuntu-20.04
         configuration:
           - debug
@@ -301,7 +299,7 @@ jobs:
     - name: Set up Python
       uses: actions/setup-python@v2
       with:
-        python-version: 3.6
+        python-version: 3.7
     - name: Cache dependencies
       # Ignore the failure of a step and avoid terminating the job.
       continue-on-error: true

CMakeLists.txt

@@ -114,6 +114,7 @@ rocm_enable_clang_tidy(
         hicpp-signed-bitwise
         llvm-namespace-comment
         misc-*
+	-misc-confusable-identifiers
         modernize-*
         performance-*
         readability-*

Dockerfile

@@ -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/5.0.2/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
+RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.3/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
@@ -71,7 +71,7 @@ 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
+RUN cget -p $PREFIX install ccache@v4.1 -DENABLE_TESTING=OFF
 
 # Install newer cmake for onnx runtime
 RUN cget -p /opt/cmake install kitware/cmake@v3.13.4

dev-requirements.txt

@@ -23,7 +23,7 @@
 #####################################################################################
 ROCmSoftwarePlatform/rocm-recipes
 facebook/zstd@v1.4.5 -X subdir -DCMAKE_DIR=build/cmake
-ccache@v4.1
+ccache@v4.1 -DENABLE_TESTING=OFF
 pcre,pfultz2/pcre@8.45 -H sha256:d6f7182602a775a7d500a0cedca6449af0400c6493951513046d17615ed0bf11
 danmar/cppcheck@2.9 -DHAVE_RULES=1
 RadeonOpenCompute/rocm-cmake@1ebf7e7bc61bb5e949c171562b421264065230a7 --build

hip-clang.docker

@@ -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/5.0.2/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
+RUN sh -c 'echo deb [arch=amd64 trusted=yes] http://repo.radeon.com/rocm/apt/5.3/ ubuntu main > /etc/apt/sources.list.d/rocm.list'
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \

src/fuse_pointwise.cpp

@@ -39,7 +39,7 @@ static literal get_scalar(instruction_ref ins)
     if(ins->name() == "contiguous")
         return get_scalar(ins->inputs().front());
     const auto& s = ins->get_shape();
-    if(not(s.elements() == 1 or s.scalar()))
+    if(s.elements() != 1 && not(s.scalar()))
         return {};
     if(not ins->can_eval())
         return {};

src/include/migraphx/op/broadcast.hpp

@@ -86,6 +86,7 @@ struct broadcast
             shape output{t, broadcast_lens, std::move(bcast_strides)};
             if(output.elements() < s0.elements())
             {
+                // don't think this can occur?
                 MIGRAPHX_THROW("BROADCAST: output size must be greater than or equal to s0 size");
             }
             return output;
@@ -125,9 +126,6 @@ struct broadcast
             std::vector<size_t> bcast_strides(s1.ndim(), 0);
             std::copy(s0.strides().begin(), s0.strides().end(), bcast_strides.begin() + axis);
             shape output{t, s1.lens(), std::move(bcast_strides)};
-            if(output.elements() < s0.elements())
-                MIGRAPHX_THROW(
-                    "BROADCAST_2in: output size must be greater than or equal to s0 size");
             return output;
         }
     }

src/include/migraphx/op/convolution.hpp

@@ -58,8 +58,8 @@ struct convolution
 
     void check_attribute_size() const
     {
-        if(not((padding.size() == stride.size() or (padding.size() / 2) == stride.size()) and
-               stride.size() == dilation.size()))
+        if((padding.size() != stride.size() and (padding.size() / 2) != stride.size()) or
+           stride.size() != dilation.size())
         {
             MIGRAPHX_THROW("CONVOLUTION: inconsistent attribute sizes");
         }
@@ -74,7 +74,8 @@ struct convolution
         // num of dims of input and attribute should match
         const auto input_size   = inputs[0].max_lens().size();
         const auto padding_size = padding.size();
-        if(not(input_size == padding_size / 2 + 2 or input_size == padding_size + 2))
+
+        if(input_size != padding_size / 2 + 2 && input_size != padding_size + 2)
         {
             MIGRAPHX_THROW("CONVOLUTION: input and attribute size mismatch!");
         }

src/include/migraphx/op/deconvolution.hpp

@@ -61,8 +61,8 @@ struct deconvolution
 
     void check_attribute_size() const
     {
-        if(not((padding.size() == stride.size() or (padding.size() / 2) == stride.size()) and
-               stride.size() == dilation.size()))
+        if((padding.size() != stride.size() and (padding.size() / 2) != stride.size()) or
+           stride.size() != dilation.size())
         {
             MIGRAPHX_THROW("deconvolution: inconsistent attribute sizes");
         }

src/include/migraphx/op/pooling.hpp

@@ -64,8 +64,8 @@ struct pooling
 
     void check_attribute_size() const
     {
-        if(not((padding.size() == stride.size() or (padding.size() / 2) == stride.size()) and
-               stride.size() == lengths.size()))
+        if((padding.size() != stride.size() and (padding.size() / 2) != stride.size()) or
+           stride.size() != lengths.size())
         {
             MIGRAPHX_THROW("POOLING: inconsistent attribute sizes");
         }
@@ -83,7 +83,7 @@ struct pooling
         size_t kdims      = input_lens.size() - 2;
         auto input_size   = inputs[0].lens().size();
         auto padding_size = padding.size();
-        if(not(input_size == padding_size / 2 + 2 or input_size == padding_size + 2))
+        if(input_size != padding_size / 2 + 2 and input_size != padding_size + 2)
         {
             MIGRAPHX_THROW("POOLING: input and attribute size mismatch!");
         }

src/include/migraphx/op/quant_convolution.hpp

@@ -63,8 +63,8 @@ struct quant_convolution
 
     void check_attribute_size() const
     {
-        if(not((padding.size() == stride.size() or (padding.size() / 2) == stride.size()) and
-               stride.size() == dilation.size()))
+        if((padding.size() != stride.size() and (padding.size() / 2) != stride.size()) or
+           stride.size() != dilation.size())
         {
             MIGRAPHX_THROW("QUANT_CONVOLUTION: inconsistent attribute sizes");
         }

src/load_save.cpp

@@ -25,7 +25,6 @@
 #include <migraphx/file_buffer.hpp>
 #include <migraphx/json.hpp>
 #include <migraphx/msgpack.hpp>
-#include <migraphx/file_buffer.hpp>
 #include <fstream>
 
 namespace migraphx {

src/module.cpp

@@ -34,7 +34,6 @@
 #include <migraphx/pass_manager.hpp>
 #include <migraphx/make_op.hpp>
 #include <migraphx/register_target.hpp>
-#include <migraphx/make_op.hpp>
 #include <migraphx/json.hpp>
 #include <iostream>
 #include <sstream>

src/onnx/conv.cpp

@@ -30,7 +30,7 @@ namespace onnx {
 
 void recalc_conv_attributes(value& v, size_t kdims)
 {
-    if(not(v["padding"].size() == kdims or v["padding"].size() == kdims * 2))
+    if(v["padding"].size() != kdims and v["padding"].size() != kdims * 2)
     {
         v["padding"].resize(kdims);
         std::fill_n(v["padding"].begin(), kdims, 0);

src/rewrite_rnn.cpp

@@ -46,9 +46,6 @@
 #include <migraphx/iterator_for.hpp>
 #include <migraphx/dfor.hpp>
 #include <migraphx/ranges.hpp>
-#include <migraphx/op/common.hpp>
-#include <migraphx/op/rnn_var_sl_last_output.hpp>
-#include <migraphx/op/rnn_variable_seq_lens.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {

src/simplify_algebra.cpp

@@ -827,7 +827,7 @@ MIGRAPHX_PRED_MATCHER(horiz_conv_dot, instruction_ref ins)
     };
     auto dots  = std::count_if(ins->outputs().begin(), ins->outputs().end(), pred("dot"));
     auto convs = std::count_if(ins->outputs().begin(), ins->outputs().end(), pred("convolution"));
-    return not(dots < 2 and convs < 2);
+    return (dots >= 2 or convs >= 2);
 }
 
 struct find_conv_dot_horiz_fusion

src/targets/cpu/target.cpp

@@ -41,7 +41,6 @@
 #include <migraphx/rewrite_quantization.hpp>
 #include <migraphx/rewrite_rnn.hpp>
 #include <migraphx/schedule.hpp>
-#include <migraphx/memory_coloring.hpp>
 #include <migraphx/simplify_algebra.hpp>
 #include <migraphx/simplify_qdq.hpp>
 #include <migraphx/simplify_reshapes.hpp>

src/targets/gpu/hip.cpp

@@ -183,8 +183,8 @@ argument register_on_gpu(const argument& arg)
 {
     auto arg_shared = arg.share();
     auto p          = register_on_gpu(arg_shared.data(), arg_shared.get_shape().bytes());
-    return {arg_shared.get_shape(),
-            [p, a = std::move(arg_shared)]() mutable { return get_device_ptr(p.get()); }};
+    auto s          = arg_shared.get_shape();
+    return {s, [p, a = std::move(arg_shared)]() mutable { return get_device_ptr(p.get()); }};
 }
 
 argument to_gpu(const argument& arg, bool host)

test/gpu/literal.cpp

@@ -48,4 +48,4 @@ void gpu_literal_test()
     }
 }
 
-int main() { gpu_literal_test(); }
+int main() { gpu_literal_test(); } // NOLINT (bugprone-exception-escape)

test/gpu/quantization.cpp

@@ -30,7 +30,6 @@
 #include <migraphx/ref/target.hpp>
 #include <migraphx/gpu/target.hpp>
 #include <migraphx/verify.hpp>
-#include <migraphx/quantization.hpp>
 #include <migraphx/dead_code_elimination.hpp>
 #include <migraphx/propagate_constant.hpp>
 #include <migraphx/pass_manager.hpp>