Add Windows 11 on ARM to CI

.github/workflows/public.continuous.yml

@@ -113,4 +113,42 @@ jobs:
           cmake -D BUILD_TBB_FROM_SOURCE=ON -D EMBREE_EXTRA_OPTIONS="-DBUILD_TESTING=ON -DEMBREE_TUTORIALS=ON -DEMBREE_ISPC_SUPPORT=ON -DEMBREE_TESTING_INTENSITY=3" ../superbuild
           cmake --build .
           cd embree/build
-          ctest
+          ctest
+
+  windows-11-arm:
+    runs-on: windows-11-arm
+    steps:
+      - name: Checkout Repository
+        uses: actions/checkout@v4
+
+      - name: Build and Run
+        shell: pwsh
+        run: |
+          $opts  = "-DBUILD_TESTING=ON"
+          $opts += " -DEMBREE_TUTORIALS=ON"
+          $opts += " -DEMBREE_ISPC_SUPPORT=OFF"
+          $opts += " -DEMBREE_TASKING_SYSTEM=INTERNAL"
+          $opts += " -DEMBREE_TESTING_INTENSITY=2"
+
+          cmake -B build -G "Visual Studio 17 2022" -A ARM64 $opts.Split(" ") .
+          cmake --build build --config Release
+          ctest --test-dir build -C Release --output-on-failure
+
+  windows-11:
+    runs-on: windows-latest
+    steps:
+      - name: Checkout Repository
+        uses: actions/checkout@v4
+
+      - name: Build and Run
+        shell: pwsh
+        run: |
+          $opts  = "-DBUILD_TESTING=ON"
+          $opts += " -DEMBREE_TUTORIALS=ON"
+          $opts += " -DEMBREE_ISPC_SUPPORT=OFF"
+          $opts += " -DEMBREE_TASKING_SYSTEM=INTERNAL"
+          $opts += " -DEMBREE_TESTING_INTENSITY=2"
+
+          cmake -B build -G "Visual Studio 17 2022" $opts.Split(" ") .
+          cmake --build build --config Release
+          ctest --test-dir build -C Release --output-on-failure

common/simd/arm/simd_wrapper_types.h

@@ -0,0 +1,53 @@
+// Copyright 2009-2024 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include <immintrin.h>
+#include "../../sys/platform.h"
+
+namespace embree
+{
+#if defined(_MSC_VER) && defined(_ARM64_)
+  // On ARM64 MSVC, __m128 and __m128i are both aliased to __n128 in arm_neon.h,
+  // causing C++ overload ambiguity. We wrap them in unique types to disambiguate.
+
+  struct __m128_wrapper {
+    __m128 data;
+    __forceinline __m128_wrapper() {}
+    __forceinline __m128_wrapper(__m128 v) : data(v) {}
+    __forceinline operator __m128() const { return data; }
+    __forceinline operator __m128&() { return data; }
+  };
+
+  struct __m128i_wrapper {
+    __m128i data;
+    __forceinline __m128i_wrapper() {}
+    __forceinline __m128i_wrapper(__m128i v) : data(v) {}
+    __forceinline operator __m128i() const { return data; }
+    __forceinline operator __m128i&() { return data; }
+  };
+
+  struct __m128d_wrapper {
+    __m128d data;
+    __forceinline __m128d_wrapper() {}
+    __forceinline __m128d_wrapper(__m128d v) : data(v) {}
+    __forceinline operator __m128d() const { return data; }
+    __forceinline operator __m128d&() { return data; }
+  };
+
+#else
+  // On other platforms, use identity "wrappers" that compile away to nothing
+  template<typename T> struct identity_wrapper {
+    T data;
+    __forceinline identity_wrapper() {}
+    __forceinline identity_wrapper(T v) : data(v) {}
+    __forceinline operator T() const { return data; }
+    __forceinline operator T&() { return data; }
+  };
+
+  using __m128_wrapper = identity_wrapper<__m128>;
+  using __m128i_wrapper = identity_wrapper<__m128i>;
+  using __m128d_wrapper = identity_wrapper<__m128d>;
+#endif
+}

common/simd/arm/sse2neon.h

@@ -81,7 +81,11 @@
 #define _sse2neon_likely(x) __builtin_expect(!!(x), 1)
 #define _sse2neon_unlikely(x) __builtin_expect(!!(x), 0)
 #else /* non-GNU / non-clang compilers */
+#if defined(_MSC_VER)
+#pragma message("Macro name collisions may happen with unsupported compiler.")
+#else
 #warning "Macro name collisions may happen with unsupported compiler."
+#endif
 #ifndef FORCE_INLINE
 #define FORCE_INLINE static inline
 #endif

common/simd/vboolf4_sse2.h

@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include "arm/simd_wrapper_types.h"
+
 #define vboolf vboolf_impl
 #define vboold vboold_impl
 #define vint vint_impl
@@ -23,8 +25,8 @@ namespace embree
     typedef vint4   Int;
     typedef vfloat4 Float;
 
-    enum  { size = 4 };            // number of SIMD elements
-    union { __m128 v; int i[4]; }; // data
+    enum  { size = 4 };                      // number of SIMD elements
+    union { __m128_wrapper v; int i[4]; }; // data
 
     ////////////////////////////////////////////////////////////////////////////////
     /// Constructors, Assignment & Cast Operators
@@ -35,10 +37,10 @@ namespace embree
     __forceinline vboolf4& operator =(const vboolf4& other) { v = other.v; return *this; }
 
     __forceinline vboolf(__m128 input) : v(input) {}
-    __forceinline operator const __m128&() const { return v; }
+    __forceinline operator const __m128&() const { return v.data; }
     #if !defined(__EMSCRIPTEN__)
-    __forceinline operator const __m128i() const { return _mm_castps_si128(v); }
-    __forceinline operator const __m128d() const { return _mm_castps_pd(v); }
+    __forceinline operator const __m128i() const { return _mm_castps_si128(v.data); }
+    __forceinline operator const __m128d() const { return _mm_castps_pd(v.data); }
     #endif
 
     __forceinline vboolf(bool a)
@@ -52,7 +54,7 @@ namespace embree
 
     /* return int32 mask */
     __forceinline __m128i mask32() const { 
-      return _mm_castps_si128(v);
+      return _mm_castps_si128(v.data);
     }
 
     ////////////////////////////////////////////////////////////////////////////////
@@ -66,7 +68,7 @@ namespace embree
     /// Array Access
     ////////////////////////////////////////////////////////////////////////////////
 
-    __forceinline bool operator [](size_t index) const { assert(index < 4); return (_mm_movemask_ps(v) >> index) & 1; }
+    __forceinline bool operator [](size_t index) const { assert(index < 4); return (_mm_movemask_ps(v.data) >> index) & 1; }
     __forceinline int& operator [](size_t index)       { assert(index < 4); return i[index]; }
   };
 

common/simd/vfloat4_sse2.h

@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include "arm/simd_wrapper_types.h"
+
 #define vboolf vboolf_impl
 #define vboold vboold_impl
 #define vint vint_impl
@@ -23,8 +25,8 @@ namespace embree
     typedef vint4   Int;
     typedef vfloat4 Float;
 
-    enum  { size = 4 };                        // number of SIMD elements
-    union { __m128 v; float f[4]; int i[4]; }; // data
+    enum  { size = 4 };                              // number of SIMD elements
+    union { __m128_wrapper v; float f[4]; int i[4]; }; // data
 
     ////////////////////////////////////////////////////////////////////////////////
     /// Constructors, Assignment & Cast Operators
@@ -37,8 +39,8 @@ namespace embree
     __forceinline vfloat4& operator =(const vfloat4& other) { v = other.v; return *this; }
 
     __forceinline vfloat(__m128 a) : v(a) {}
-    __forceinline operator const __m128&() const { return v; }
-    __forceinline operator       __m128&()       { return v; }
+    __forceinline operator const __m128&() const { return v.data; }
+    __forceinline operator       __m128&()       { return v.data; }
 
     __forceinline vfloat(float a) : v(_mm_set1_ps(a)) {}
     __forceinline vfloat(float a, float b, float c, float d) : v(_mm_set_ps(d, c, b, a)) {}

common/simd/vint4_sse2.h

@@ -3,6 +3,7 @@
 
 #pragma once
 
+#include "arm/simd_wrapper_types.h"
 #include "../math/emath.h"
 
 #define vboolf vboolf_impl
@@ -25,8 +26,8 @@ namespace embree
     typedef vint4   Int;
     typedef vfloat4 Float;
 
-    enum  { size = 4 };             // number of SIMD elements
-    union { __m128i v; int i[4]; }; // data
+    enum  { size = 4 };               // number of SIMD elements
+    union { __m128i_wrapper v; int i[4]; }; // data
 
     ////////////////////////////////////////////////////////////////////////////////
     /// Constructors, Assignment & Cast Operators
@@ -37,8 +38,8 @@ namespace embree
     __forceinline vint4& operator =(const vint4& a) { v = a.v; return *this; }
 
     __forceinline vint(__m128i a) : v(a) {}
-    __forceinline operator const __m128i&() const { return v; }
-    __forceinline operator       __m128i&()       { return v; }
+    __forceinline operator const __m128i&() const { return v.data; }
+    __forceinline operator       __m128i&()       { return v.data; }
 
     __forceinline vint(int a) : v(_mm_set1_epi32(a)) {}
     __forceinline vint(int a, int b, int c, int d) : v(_mm_set_epi32(d, c, b, a)) {}

common/simd/vuint4_sse2.h

@@ -3,6 +3,7 @@
 
 #pragma once
 
+#include "arm/simd_wrapper_types.h"
 #include "../math/emath.h"
 
 #define vboolf vboolf_impl
@@ -26,7 +27,7 @@ namespace embree
     typedef vfloat4 Float;
 
     enum  { size = 4 }; // number of SIMD elements
-    union { __m128i v; unsigned int i[4]; }; // data
+    union { __m128i_wrapper v; unsigned int i[4]; }; // data
 
     ////////////////////////////////////////////////////////////////////////////////
     /// Constructors, Assignment & Cast Operators
@@ -37,8 +38,8 @@ namespace embree
     __forceinline vuint4& operator =(const vuint4& a) { v = a.v; return *this; }
 
     __forceinline vuint(const __m128i a) : v(a) {}
-    __forceinline operator const __m128i&() const { return v; }
-    __forceinline operator       __m128i&()       { return v; }
+    __forceinline operator const __m128i&() const { return v.data; }
+    __forceinline operator       __m128i&()       { return v.data; }
 
 
     __forceinline vuint(unsigned int a) : v(_mm_set1_epi32(a)) {}

common/sys/intrinsics.h

@@ -9,7 +9,7 @@
 #include <intrin.h>
 #endif
 
-#if defined(__ARM_NEON)
+#if defined(__ARM_NEON) || defined(__aarch64__)
 #include "../simd/arm/emulation.h"
 #else
 #include <immintrin.h>

common/sys/platform.h

@@ -57,6 +57,11 @@
 #define __X86_ASM__
 #endif
 
+/* normalize ARM64 platform macro */
+#if defined(_M_ARM64) || defined(_M_ARM64EC)
+#define __aarch64__
+#endif
+
 /* detect 64 bit platform */
 #if defined(__X86_64__) || defined(__aarch64__)
 #define __64BIT__

superbuild/CMakeLists.txt

@@ -3,7 +3,7 @@
 
 ## Global settings ##
 
-cmake_minimum_required(VERSION 3.1)
+cmake_minimum_required(VERSION 3.5)
 
 set(CMAKE_MODULE_PATH
   ${CMAKE_MODULE_PATH}