gcc benchmark build fix; gcc perf exceedingly poor due to aliasing

LoopModels · Nov 27, 2023 · 6b56376 · 6b56376
1 parent f332351
commit 6b56376
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/benchmark/dual_benchmark.cpp b/benchmark/dual_benchmark.cpp
@@ -39,7 +39,7 @@ template <typename T, ptrdiff_t N, bool SIMDArry = false> struct ManualDual {
 
 template <std::floating_point T, ptrdiff_t N> struct ManualDual<T, N, false> {
   T value;
-  poly::simd::Vec<N, T> partials;
+  poly::simd::Vec<ptrdiff_t(std::bit_ceil(size_t(N))), T> partials;
 };
 template <std::floating_point T, ptrdiff_t N> struct ManualDual<T, N, true> {
   T value;

diff --git a/include/Math/Dual.hpp b/include/Math/Dual.hpp
@@ -110,7 +110,7 @@ template <class T, ptrdiff_t N> struct Dual<T, N, false> {
   constexpr Dual() = default;
   constexpr Dual(T v) : val(v) {}
   constexpr Dual(T v, ptrdiff_t n) : val(v) { partials[n] = T{1}; }
-  constexpr Dual(T v, SVector<T, N> g) : val(v) { partials << g; }
+  constexpr Dual(T v, SVector<T, N> g) : val(v), partials(g) {}
   constexpr Dual(std::integral auto v) : val(v) {}
   constexpr Dual(std::floating_point auto v) : val(v) {}
   constexpr auto value() -> T & { return val; }
@@ -123,7 +123,7 @@ template <class T, ptrdiff_t N> struct Dual<T, N, false> {
   [[nodiscard]] constexpr auto gradient(ptrdiff_t i) const -> const T & {
     return partials[i];
   }
-  [[gnu::always_inline]] constexpr auto operator-() const & -> Dual {
+  [[gnu::always_inline]] constexpr auto operator-() const -> Dual {
     return {-val, -partials};
   }
   [[gnu::always_inline]] constexpr auto