microsoft · orlmon01 · Dec 19, 2025 · Dec 19, 2025 · Jan 7, 2026 · Jan 12, 2026
diff --git a/cmake/CMakeLists.txt b/cmake/CMakeLists.txt
@@ -638,6 +638,7 @@ else()
   check_cxx_compiler_flag(-Wcatch-value HAS_CATCH_VALUE)
   check_cxx_compiler_flag(-Wclass-memaccess HAS_CLASS_MEMACCESS)
   check_cxx_compiler_flag(-Wcharacter-conversion HAS_CHARACTER_CONVERSION)
+  check_cxx_compiler_flag(-Wno-error=character-conversion HAS_NO_ERROR_CHARACTER_CONVERSION)
   check_cxx_compiler_flag(-Wdangling-reference HAS_DANGLING_REFERENCE)
   check_cxx_compiler_flag(-Wdeprecated-anon-enum-enum-conversion HAS_DEPRECATED_ANON_ENUM_ENUM_CONVERSION)
   check_cxx_compiler_flag(-Wdeprecated-builtins HAS_DEPRECATED_BUILTINS)

diff --git a/cmake/onnxruntime_unittests.cmake b/cmake/onnxruntime_unittests.cmake
@@ -50,6 +50,13 @@ function(filter_test_srcs test_srcs_var)
 endfunction()
 
 set(disabled_warnings)
+
+function(onnxruntime_disable_gtest_character_conversion_as_error target_name)
+  if (HAS_NO_ERROR_CHARACTER_CONVERSION)
+    target_compile_options(${target_name} PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:-Wno-error=character-conversion>")
+  endif()
+endfunction()
+
 function(AddTest)
   cmake_parse_arguments(_UT "DYN" "TARGET" "LIBS;SOURCES;DEPENDS;TEST_ARGS" ${ARGN})
   list(REMOVE_DUPLICATES _UT_SOURCES)
@@ -170,9 +177,7 @@ function(AddTest)
     if (${HAS_NOERROR})
       target_compile_options(${_UT_TARGET} PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:-Wno-error=uninitialized>")
     endif()
-    if (${HAS_CHARACTER_CONVERSION})
-      target_compile_options(${_UT_TARGET} PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:-Wno-error=character-conversion>")
-    endif()
+    onnxruntime_disable_gtest_character_conversion_as_error(${_UT_TARGET})
   endif()
 
   set(TEST_ARGS ${_UT_TEST_ARGS})
@@ -847,9 +852,7 @@ if(MSVC)
                 "$<$<NOT:$<COMPILE_LANGUAGE:CUDA>>:/wd6326>")
 else()
   target_include_directories(onnxruntime_test_utils PRIVATE ${CMAKE_CURRENT_BINARY_DIR} ${ONNXRUNTIME_ROOT})
-  if (HAS_CHARACTER_CONVERSION)
-    target_compile_options(onnxruntime_test_utils PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:-Wno-error=character-conversion>")
-  endif()
+  onnxruntime_disable_gtest_character_conversion_as_error(onnxruntime_test_utils)
 endif()
 if (onnxruntime_USE_NCCL)
   target_include_directories(onnxruntime_test_utils PRIVATE ${NCCL_INCLUDE_DIRS})

diff --git a/docs/ContribOperators.md b/docs/ContribOperators.md
@@ -3569,7 +3569,6 @@ This version of the operator has been available since version 1 of the 'com.micr
 ### <a name="com.microsoft.NhwcFusedConv"></a><a name="com.microsoft.nhwcfusedconv">**com.microsoft.NhwcFusedConv**</a>
 
   NhwcFusedConv is a Conv operator with optional activation and add operators fused in.
-  Only has fp16 implementation as of 2023/04/15.
 
 #### Version
 
@@ -3600,26 +3599,26 @@ This version of the operator has been available since version 1 of the 'com.micr
 
 <dl>
 <dt><tt>X</tt> : T</dt>
-<dd></dd>
+<dd>Input activation tensor in channels-last layout. For 2D convolution this is [N, H, W, C], where N is batch size, H/W are spatial dimensions, and C is the number of input channels.</dd>
 <dt><tt>W</tt> : T</dt>
-<dd></dd>
+<dd>Convolution weight tensor in the standard ONNX Conv filter layout [M, C/group, kH, kW], where M is the number of output channels.</dd>
 <dt><tt>B</tt> (optional) : T</dt>
-<dd></dd>
+<dd>Optional 1D bias tensor of shape [M].</dd>
 <dt><tt>Z</tt> (optional) : T</dt>
-<dd>Tensor to be added to the output, must be the same shape and format as the output tensor.</dd>
+<dd>Optional residual/add tensor in the same channels-last layout and shape as the output tensor Y. For 2D convolution this is [N, out_H, out_W, M].</dd>
 </dl>
 
 #### Outputs
 
 <dl>
 <dt><tt>Y</tt> : T</dt>
-<dd></dd>
+<dd>Output tensor in channels-last layout. For 2D convolution this is [N, out_H, out_W, M], where M is the number of output channels.</dd>
 </dl>
 
 #### Type Constraints
 
 <dl>
-<dt><tt>T</tt> : tensor(float16)</dt>
+<dt><tt>T</tt> : tensor(float16), tensor(float)</dt>
 <dd>Constrain input and output types to float tensors</dd>
 </dl>
 

diff --git a/onnxruntime/contrib_ops/cpu/cpu_contrib_kernels.cc b/onnxruntime/contrib_ops/cpu/cpu_contrib_kernels.cc
@@ -20,6 +20,9 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1,
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, EmbedLayerNormalization);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, ExpandDims);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, FusedConv);
+#ifdef USE_KLEIDIAI
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, NhwcFusedConv);
+#endif
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, FusedGemm);
 #if !defined(DISABLE_GENERATION_OPS)
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, GreedySearch);
@@ -313,6 +316,9 @@ Status RegisterCpuContribKernels(KernelRegistry& kernel_registry) {
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, EmbedLayerNormalization)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, ExpandDims)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, FusedConv)>,
+#ifdef USE_KLEIDIAI
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, NhwcFusedConv)>,
+#endif
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, FusedGemm)>,
 #if !defined(DISABLE_GENERATION_OPS)
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, GreedySearch)>,

diff --git a/onnxruntime/contrib_ops/cpu/fused_conv.cc b/onnxruntime/contrib_ops/cpu/fused_conv.cc
@@ -26,5 +26,18 @@ ONNX_CPU_OPERATOR_TYPED_MS_KERNEL(
         .TypeConstraint("T", DataTypeImpl::GetTensorType<float>()),
     FusedConvFloat);
 
+#ifdef USE_KLEIDIAI
+ONNX_CPU_OPERATOR_TYPED_MS_KERNEL(
+    NhwcFusedConv,
+    1,
+    float,
+    KernelDefBuilder()
+        // Allow the optional "sum" input (index 3) to be reused as the output buffer (index 0),
+        // consistent with the FusedConv kernel registration.
+        .MayInplace(3, 0)
+        .TypeConstraint("T", DataTypeImpl::GetTensorType<float>()),
+    FusedConvFloat);
+#endif
+
 }  // namespace contrib
 }  // namespace onnxruntime