Windows Build Fixes (openvinotoolkit#645)

* [NVIDIA] Fix CudaUnitTests target for Windows * [NVIDIA] Fix CUDA::math::pow function * [NVIDIA] Fix Windows build error in convert_benchmark.cpp --------- Co-authored-by: Nadezhda Ageeva <[email protected]>
redradist · May 4, 2023 · b2147a3 · b2147a3
1 parent e5b0fe6
commit b2147a3
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 26 deletions.
diff --git a/modules/nvidia_plugin/src/cuda/math.cuh b/modules/nvidia_plugin/src/cuda/math.cuh
@@ -83,7 +83,7 @@ inline __device__ T exp(T x) {
 
 template <typename T>
 inline __device__ T pow(T x, T y) {
-    return ::pow(x, y);
+    return static_cast<T>(powf(static_cast<float>(x), static_cast<float>(y)));
 }
 
 template <typename T>
@@ -103,11 +103,6 @@ inline __device__ __half round(__half x) {
     return ::round(static_cast<float>(x));
 }
 
-template <>
-inline __device__ __half pow<__half>(__half x, __half y) {
-    return powf(static_cast<float>(x), static_cast<float>(y));
-}
-
 #if defined(CUDA_HAS_HALF_MATH)
 inline __device__ __half floor(__half x) { return ::hfloor(x); }
 
@@ -171,11 +166,6 @@ inline __device__ __nv_bfloat16 round(__nv_bfloat16 x) {
     return ::round(static_cast<float>(x));
 }
 
-template <>
-inline __device__ __nv_bfloat16 pow<__nv_bfloat16>(__nv_bfloat16 x, __nv_bfloat16 y) {
-    return powf(static_cast<float>(x), static_cast<float>(y));
-}
-
 #if defined(CUDA_HAS_BF16_MATH)
 inline __device__ __nv_bfloat16 floor(__nv_bfloat16 x) { return ::hfloor(x); }
 

diff --git a/modules/nvidia_plugin/tests/unit/CMakeLists.txt b/modules/nvidia_plugin/tests/unit/CMakeLists.txt
@@ -36,7 +36,7 @@ addIeTargetTest(
             gtest_main
             gmock
             ngraphFunctions
-            commonTestUtils_s
+            commonTestUtils
         ADD_CPPLINT
         LABELS
             CUDA

diff --git a/modules/nvidia_plugin/tests/unit/convert_benchmark.cpp b/modules/nvidia_plugin/tests/unit/convert_benchmark.cpp
@@ -49,20 +49,20 @@ TEST_F(ConvertTest, DISABLED_benchmark) {
         emptyTensor, emptyMapping, emptyTensor, emptyMapping, threadContext, token, profiler};
 
     using Type_t = ov::element::Type_t;
-    static constexpr auto supported_types = {Type_t::boolean,
-                                             Type_t::bf16,
-                                             Type_t::f16,
-                                             Type_t::f32,
-                                             Type_t::f64,
-                                             Type_t::i8,
-                                             Type_t::i16,
-                                             Type_t::i32,
-                                             Type_t::i64,
-                                             /*Type_t::u1, convert doesn't support it*/
-                                             Type_t::u8,
-                                             Type_t::u16,
-                                             Type_t::u32,
-                                             Type_t::u64};
+    constexpr Type_t supported_types[] = {Type_t::boolean,
+                                          Type_t::bf16,
+                                          Type_t::f16,
+                                          Type_t::f32,
+                                          Type_t::f64,
+                                          Type_t::i8,
+                                          Type_t::i16,
+                                          Type_t::i32,
+                                          Type_t::i64,
+                                          /*Type_t::u1, convert doesn't support it*/
+                                          Type_t::u8,
+                                          Type_t::u16,
+                                          Type_t::u32,
+                                          Type_t::u64};
     for (auto inputIdx : supported_types) {
         for (auto outputIdx : supported_types) {
             const auto inputType = Type_t(static_cast<std::underlying_type<Type_t>::type>(inputIdx));