Merge pull request #123 from r-devulap/macos-np

Fix more build issues on macOS

Author: Raghuveer Devulapalli

src/avx512-64bit-common.h

@@ -912,6 +912,19 @@ struct zmm_vector<uint64_t> {
                 left_addr, right_addr, k, reg);
     }
 };
+
+/*
+ * workaround on 64-bit macOS which defines size_t as unsigned long and defines
+ * uint64_t as unsigned long long, both of which are 8 bytes
+ */
+#if defined(__APPLE__) && defined(__x86_64__)
+static_assert(sizeof(size_t) == sizeof(uint64_t),
+              "Size of size_t and uint64_t are not the same");
+template <>
+struct zmm_vector<size_t> : public zmm_vector<uint64_t> {
+};
+#endif
+
 template <>
 struct zmm_vector<double> {
     using type_t = double;

src/xss-common-argsort.h

@@ -549,19 +549,10 @@ avx512_argsort(T *arr, arrsize_t *arg, arrsize_t arrsize, bool hasnan = false)
                                               ymm_vector<T>,
                                               zmm_vector<T>>::type;
 
-/* Workaround for NumPy failed build on macOS x86_64: implicit instantiation of
- * undefined template 'zmm_vector<unsigned long>'*/
-#ifdef __APPLE__
-    using argtype =
-            typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
-                                      ymm_vector<uint32_t>,
-                                      zmm_vector<uint64_t>>::type;
-#else
     using argtype =
             typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
                                       ymm_vector<arrsize_t>,
                                       zmm_vector<arrsize_t>>::type;
-#endif
 
     if (arrsize > 1) {
         if constexpr (std::is_floating_point_v<T>) {
@@ -595,18 +586,10 @@ avx2_argsort(T *arr, arrsize_t *arg, arrsize_t arrsize, bool hasnan = false)
                                               avx2_half_vector<T>,
                                               avx2_vector<T>>::type;
 
-#ifdef __APPLE__
-    using argtype =
-            typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
-                                      avx2_half_vector<uint32_t>,
-                                      avx2_vector<uint64_t>>::type;
-#else
     using argtype =
             typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
                                       avx2_half_vector<arrsize_t>,
                                       avx2_vector<arrsize_t>>::type;
-#endif
-
     if (arrsize > 1) {
         if constexpr (std::is_floating_point_v<T>) {
             if ((hasnan) && (array_has_nan<vectype>(arr, arrsize))) {
@@ -643,19 +626,10 @@ X86_SIMD_SORT_INLINE void avx512_argselect(T *arr,
                                               ymm_vector<T>,
                                               zmm_vector<T>>::type;
 
-/* Workaround for NumPy failed build on macOS x86_64: implicit instantiation of
- * undefined template 'zmm_vector<unsigned long>'*/
-#ifdef __APPLE__
-    using argtype =
-            typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
-                                      ymm_vector<uint32_t>,
-                                      zmm_vector<uint64_t>>::type;
-#else
     using argtype =
             typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
                                       ymm_vector<arrsize_t>,
                                       zmm_vector<arrsize_t>>::type;
-#endif
 
     if (arrsize > 1) {
         if constexpr (std::is_floating_point_v<T>) {
@@ -692,17 +666,10 @@ X86_SIMD_SORT_INLINE void avx2_argselect(T *arr,
                                               avx2_half_vector<T>,
                                               avx2_vector<T>>::type;
 
-#ifdef __APPLE__
-    using argtype =
-            typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
-                                      avx2_half_vector<uint32_t>,
-                                      avx2_vector<uint64_t>>::type;
-#else
     using argtype =
             typename std::conditional<sizeof(arrsize_t) == sizeof(int32_t),
                                       avx2_half_vector<arrsize_t>,
                                       avx2_vector<arrsize_t>>::type;
-#endif
 
     if (arrsize > 1) {
         if constexpr (std::is_floating_point_v<T>) {

src/xss-network-keyvaluesort.hpp

@@ -442,7 +442,7 @@ bitonic_fullmerge_n_vec(typename keyType::reg_t *keys,
 
 template <typename keyType, typename indexType, int numVecs>
 X86_SIMD_SORT_INLINE void argsort_n_vec(typename keyType::type_t *keys,
-                                        typename indexType::type_t *indices,
+                                        arrsize_t *indices,
                                         int N)
 {
     using kreg_t = typename keyType::reg_t;
@@ -587,7 +587,7 @@ X86_SIMD_SORT_INLINE void kvsort_n_vec(typename keyType::type_t *keys,
 
 template <typename keyType, typename indexType, int maxN>
 X86_SIMD_SORT_INLINE void argsort_n(typename keyType::type_t *keys,
-                                    typename indexType::type_t *indices,
+                                    arrsize_t *indices,
                                     int N)
 {
     static_assert(keyType::numlanes == indexType::numlanes,