Resolution of issue #153: Add compiler warning on comparing int and size_t (#212)

* Turning compare warnings on

* Cleaning part I

* Cleaning part II

* Explicit static_cast to ck::type_convert

* Resolving large tensor size issue.

* format

* revert change to tensor descriptor; promote lementSpaceSize to 64bit

* use integer value for GEMM test

* Review remarks

* Review remarks + issues with (un)signed arithmetic

* Format fix

* Format

* Clang-format.

* fix 2gb limit issue
Co-authored-by: Chao Liu <chao.liu2@amd.com>
Co-authored-by: Adam Osewski <aosewski@amd.com>

library/src/host_tensor/host_tensor.cpp

@@ -25,7 +25,7 @@ std::size_t HostTensorDescriptor::GetElementSize() const
 std::size_t HostTensorDescriptor::GetElementSpace() const
 {
     std::size_t space = 1;
-    for(int i = 0; i < mLens.size(); ++i)
+    for(std::size_t i = 0; i < mLens.size(); ++i)
     {
         space += (mLens[i] - 1) * mStrides[i];
     }
@@ -68,7 +68,7 @@ void ostream_HostTensorDescriptor(const HostTensorDescriptor& desc, std::ostream
 // FIXME: remove
 void bf16_to_f32_(const Tensor<ck::bhalf_t>& src, Tensor<float>& dst)
 {
-    for(int i = 0; i < src.mData.size(); ++i)
+    for(std::size_t i = 0; i < src.mData.size(); ++i)
         dst.mData[i] = ck::type_convert<float>(src.mData[i]);
 }
 #endif

library/src/utility/conv_fwd_util.cpp

@@ -71,11 +71,12 @@ ConvParams::ConvParams(ck::index_t n_dim,
       input_left_pads(left_pads),
       input_right_pads(right_pads)
 {
-    if(filter_spatial_lengths.size() != num_dim_spatial ||
-       input_spatial_lengths.size() != num_dim_spatial ||
-       conv_filter_strides.size() != num_dim_spatial ||
-       conv_filter_dilations.size() != num_dim_spatial ||
-       input_left_pads.size() != num_dim_spatial || input_right_pads.size() != num_dim_spatial)
+    if(ck::type_convert<ck::index_t>(filter_spatial_lengths.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(input_spatial_lengths.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(conv_filter_strides.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(conv_filter_dilations.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(input_left_pads.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(input_right_pads.size()) != num_dim_spatial)
     {
         throw(
             std::runtime_error("ConvParams::GetOutputSpatialLengths: "
@@ -85,11 +86,12 @@ ConvParams::ConvParams(ck::index_t n_dim,
 
 std::vector<ck::index_t> ConvParams::GetOutputSpatialLengths() const
 {
-    if(filter_spatial_lengths.size() != num_dim_spatial ||
-       input_spatial_lengths.size() != num_dim_spatial ||
-       conv_filter_strides.size() != num_dim_spatial ||
-       conv_filter_dilations.size() != num_dim_spatial ||
-       input_left_pads.size() != num_dim_spatial || input_right_pads.size() != num_dim_spatial)
+    if(ck::type_convert<ck::index_t>(filter_spatial_lengths.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(input_spatial_lengths.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(conv_filter_strides.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(conv_filter_dilations.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(input_left_pads.size()) != num_dim_spatial ||
+       ck::type_convert<ck::index_t>(input_right_pads.size()) != num_dim_spatial)
     {
         throw(
             std::runtime_error("ConvParams::GetOutputSpatialLengths: "

profiler/include/profile_convnd_bwd_data_impl.hpp

@@ -222,7 +222,7 @@ static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
 {
     float max_diff = 1e-6;
 
-    for(int i = 0; i < ref.mData.size(); ++i)
+    for(std::size_t i = 0; i < ref.mData.size(); ++i)
     {
         float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
         if(max_diff < diff)
@@ -236,16 +236,16 @@ template <typename DataType>
 void show_data_nhwc_layout(Tensor<DataType>& nhwc)
 {
     std::cout << "[";
-    for(int n = 0; n < nhwc.mDesc.GetLengths()[0]; n++)
+    for(int n = 0; n < ck::type_convert<int>(nhwc.mDesc.GetLengths()[0]); n++)
     {
         std::cout << "[";
-        for(int hi = 0; hi < nhwc.mDesc.GetLengths()[2]; hi++)
+        for(int hi = 0; hi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[2]); hi++)
         {
             std::cout << "[";
-            for(int wi = 0; wi < nhwc.mDesc.GetLengths()[3]; wi++)
+            for(int wi = 0; wi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[3]); wi++)
             {
                 std::cout << "[";
-                for(int c = 0; c < nhwc.mDesc.GetLengths()[1]; c++)
+                for(int c = 0; c < ck::type_convert<int>(nhwc.mDesc.GetLengths()[1]); c++)
                 {
                     std::cout << static_cast<float>(nhwc(n, c, hi, wi)) << "  ";
                 }

profiler/include/profile_grouped_gemm_impl.hpp

@@ -50,12 +50,12 @@ void profile_grouped_gemm_impl(int do_verification,
                                int init_method,
                                bool do_log,
                                int nrepeat,
-                               std::vector<int> Ms,
-                               std::vector<int> Ns,
-                               std::vector<int> Ks,
-                               std::vector<int> StrideAs,
-                               std::vector<int> StrideBs,
-                               std::vector<int> StrideCs)
+                               const std::vector<int>& Ms,
+                               const std::vector<int>& Ns,
+                               const std::vector<int>& Ks,
+                               const std::vector<int>& StrideAs,
+                               const std::vector<int>& StrideBs,
+                               const std::vector<int>& StrideCs)
 {
     auto f_host_tensor_descriptor =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
@@ -71,7 +71,7 @@ void profile_grouped_gemm_impl(int do_verification,
             }
         };
 
-    int group_count = Ms.size();
+    std::size_t group_count = Ms.size();
 
     if(!(group_count == Ns.size() && group_count == Ks.size() && group_count == StrideAs.size() &&
          group_count == StrideBs.size() && group_count == StrideCs.size()))
@@ -83,7 +83,7 @@ void profile_grouped_gemm_impl(int do_verification,
     std::vector<Tensor<BDataType>> b_k_n;
     std::vector<Tensor<CDataType>> c_m_n_device_results;
 
-    for(int i = 0; i < Ms.size(); i++)
+    for(std::size_t i = 0; i < group_count; i++)
     {
         a_m_k.push_back(
             Tensor<ADataType>(f_host_tensor_descriptor(Ms[i], Ks[i], StrideAs[i], ALayout{})));
@@ -144,7 +144,7 @@ void profile_grouped_gemm_impl(int do_verification,
 
     gemm_shapes.reserve(group_count);
 
-    for(int i = 0; i < group_count; i++)
+    for(std::size_t i = 0; i < group_count; i++)
     {
         a_device_buf.emplace_back(
             std::make_unique<DeviceMem>(sizeof(ADataType) * a_m_k[i].mDesc.GetElementSpace()));
@@ -234,7 +234,7 @@ void profile_grouped_gemm_impl(int do_verification,
             float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat);
 
             std::size_t flop = 0, num_btype = 0;
-            for(int i = 0; i < gemm_shapes.size(); i++)
+            for(std::size_t i = 0; i < gemm_shapes.size(); i++)
             {
                 flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
 
@@ -258,7 +258,7 @@ void profile_grouped_gemm_impl(int do_verification,
 
             if(do_verification)
             {
-                for(int i = 0; i < gemm_shapes.size(); i++)
+                for(std::size_t i = 0; i < gemm_shapes.size(); i++)
                 {
 
                     c_device_buf[i]->FromDevice(c_m_n_device_results[i].mData.data());

profiler/src/profile_reduce.cpp

@@ -186,7 +186,7 @@ class AppArgs
 
     int processArgs(int argc, char* argv[])
     {
-        unsigned int ch;
+        int ch;
 
         optind++; // to skip the "reduce" module name
 

test/gemm_split_k/gemm_split_k.cpp

@@ -45,7 +45,7 @@ static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
 {
     float max_diff = 1e-6;
 
-    for(int i = 0; i < ref.mData.size(); ++i)
+    for(std::size_t i = 0; i < ref.mData.size(); ++i)
     {
         float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
         if(max_diff < diff)

test/grouped_gemm/grouped_gemm_fp16.cpp

@@ -104,7 +104,7 @@ bool TestGroupedGemm(DeviceGroupedGemmPtr_& groupedGemmPtr)
     b_tensors_device.reserve(group_count);
     c_tensors_device.reserve(group_count);
 
-    for(int i = 0; i < gemm_shapes.size(); i++)
+    for(std::size_t i = 0; i < gemm_shapes.size(); i++)
     {
         a_tensors.emplace_back(Tensor<ADataType>(f_host_tensor_descriptor(
             gemm_shapes[i].M, gemm_shapes[i].K, gemm_shapes[i].StrideA, ALayout{})));
@@ -119,7 +119,7 @@ bool TestGroupedGemm(DeviceGroupedGemmPtr_& groupedGemmPtr)
         b_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
     }
 
-    for(int i = 0; i < gemm_shapes.size(); i++)
+    for(std::size_t i = 0; i < gemm_shapes.size(); i++)
     {
         a_tensors_device.emplace_back(
             std::make_unique<DeviceMem>(sizeof(ADataType) * a_tensors[i].mDesc.GetElementSize()));
@@ -147,7 +147,7 @@ bool TestGroupedGemm(DeviceGroupedGemmPtr_& groupedGemmPtr)
 
     invoker_ptr->Run(argument_ptr.get());
 
-    for(int i = 0; i < gemm_shapes.size(); i++)
+    for(std::size_t i = 0; i < gemm_shapes.size(); i++)
     {
         c_tensors_device[i]->FromDevice(c_device_tensors[i].mData.data());
 

test/reduce/reduce_no_index.cpp

@@ -460,7 +460,7 @@ class SimpleAppArgs
 
     int processArgs(int argc, char* argv[])
     {
-        unsigned int ch;
+        int ch;
 
         while(1)
         {

test/reduce/reduce_util.hpp

@@ -9,7 +9,7 @@ namespace reduce_util {
 template <typename T>
 void to_f32_vector(const Tensor<T>& src, Tensor<float>& dst)
 {
-    for(int i = 0; i < src.mData.size(); ++i)
+    for(std::size_t i = 0; i < src.mData.size(); ++i)
         dst.mData[i] = type_convert<float>(src.mData[i]);
 }
 

test/reduce/reduce_with_index.cpp

@@ -463,7 +463,7 @@ class SimpleAppArgs
 
     int processArgs(int argc, char* argv[])
     {
-        unsigned int ch;
+        int ch;
 
         while(1)
         {