Use default arch alignment as default alignment for xsimd::aligned_allocator

That way we match the behavior of xsimd::batch.

Author: serge-sans-paille

benchmark/xsimd_benchmark.hpp

@@ -24,7 +24,7 @@ namespace xsimd
     using duration_type = std::chrono::duration<double, std::milli>;
 
     template <class T>
-    using bench_vector = std::vector<T, xsimd::aligned_allocator<T, default_arch::alignment()>>;
+    using bench_vector = std::vector<T, xsimd::aligned_allocator<T>>;
 
     template <class T>
     void init_benchmark(bench_vector<T>& lhs, bench_vector<T>& rhs, bench_vector<T>& res, size_t size)

examples/mandelbrot.cpp

@@ -290,7 +290,7 @@ int main()
     const float y1 = 1;
     const int maxIters = 256;
 
-    std::vector<int, xsimd::aligned_allocator<int, xsimd::default_arch::alignment()>> buf(width * height);
+    std::vector<int, xsimd::aligned_allocator<int>> buf(width * height);
 
     auto bencher = pico_bench::Benchmarker<milliseconds> { 64, seconds { 10 } };
 

include/xsimd/memory/xsimd_aligned_allocator.hpp

@@ -39,7 +39,7 @@ namespace xsimd
      * @tparam T type of objects to allocate.
      * @tparam Align alignment in bytes.
      */
-    template <class T, size_t Align>
+    template <class T, size_t Align = std::conditional<std::is_same<unsupported, default_arch>::value, generic, default_arch>::type::alignment()>
     class aligned_allocator
     {
     public: