Merge pull request #952 from anutosh491/Introducing_wasm_based_IS

Initial Implementation for the new WASM based instruction set
xtensor-stack · Oct 25, 2023 · f27bbb8 · f27bbb8
2 parents cc14c23 + 1402e0e
commit f27bbb8
Show file tree

Hide file tree

Showing 8 changed files with 1,585 additions and 3 deletions.
diff --git a/include/xsimd/arch/generic/xsimd_generic_arithmetic.hpp b/include/xsimd/arch/generic/xsimd_generic_arithmetic.hpp
@@ -127,6 +127,20 @@ namespace xsimd
             return { res_r, res_i };
         }
 
+        // hadd
+        template <class A, class T, class /*=typename std::enable_if<std::is_integral<T>::value, void>::type*/>
+        inline T hadd(batch<T, A> const& self, requires_arch<generic>) noexcept
+        {
+            alignas(A::alignment()) T buffer[batch<T, A>::size];
+            self.store_aligned(buffer);
+            T res = 0;
+            for (T val : buffer)
+            {
+                res += val;
+            }
+            return res;
+        }
+
         // incr
         template <class A, class T>
         inline batch<T, A> incr(batch<T, A> const& self, requires_arch<generic>) noexcept
@@ -172,6 +186,23 @@ namespace xsimd
         {
             return add(self, other); // no saturated arithmetic on floating point numbers
         }
+        template <class A, class T, class /*=typename std::enable_if<std::is_integral<T>::value, void>::type*/>
+        inline batch<T, A> sadd(batch<T, A> const& self, batch<T, A> const& other, requires_arch<generic>) noexcept
+        {
+            if (std::is_signed<T>::value)
+            {
+                auto mask = (other >> (8 * sizeof(T) - 1));
+                auto self_pos_branch = min(std::numeric_limits<T>::max() - other, self);
+                auto self_neg_branch = max(std::numeric_limits<T>::min() - other, self);
+                return other + select(batch_bool<T, A>(mask.data), self_neg_branch, self_pos_branch);
+            }
+            else
+            {
+                const auto diffmax = std::numeric_limits<T>::max() - self;
+                const auto mindiff = min(diffmax, other);
+                return self + mindiff;
+            }
+        }
         template <class A>
         inline batch<double, A> sadd(batch<double, A> const& self, batch<double, A> const& other, requires_arch<generic>) noexcept
         {
@@ -184,6 +215,19 @@ namespace xsimd
         {
             return sub(self, other); // no saturated arithmetic on floating point numbers
         }
+        template <class A, class T, class /*=typename std::enable_if<std::is_integral<T>::value, void>::type*/>
+        inline batch<T, A> ssub(batch<T, A> const& self, batch<T, A> const& other, requires_arch<generic>) noexcept
+        {
+            if (std::is_signed<T>::value)
+            {
+                return sadd(self, -other);
+            }
+            else
+            {
+                const auto diff = min(self, other);
+                return self - diff;
+            }
+        }
         template <class A>
         inline batch<double, A> ssub(batch<double, A> const& self, batch<double, A> const& other, requires_arch<generic>) noexcept
         {

diff --git a/include/xsimd/arch/xsimd_generic_fwd.hpp b/include/xsimd/arch/xsimd_generic_fwd.hpp
@@ -31,6 +31,12 @@ namespace xsimd
         inline batch_bool<T, A> gt(batch<T, A> const& self, batch<T, A> const& other, requires_arch<generic>) noexcept;
         template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
         inline batch<T, A> mul(batch<T, A> const& self, batch<T, A> const& other, requires_arch<generic>) noexcept;
+        template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
+        inline batch<T, A> sadd(batch<T, A> const& self, batch<T, A> const& other, requires_arch<generic>) noexcept;
+        template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
+        inline batch<T, A> ssub(batch<T, A> const& self, batch<T, A> const& other, requires_arch<generic>) noexcept;
+        template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
+        inline T hadd(batch<T, A> const& self, requires_arch<generic>) noexcept;
 
     }
 }

diff --git a/include/xsimd/arch/xsimd_isa.hpp b/include/xsimd/arch/xsimd_isa.hpp
@@ -80,6 +80,10 @@
 #include "./xsimd_sve.hpp"
 #endif
 
+#if XSIMD_WITH_WASM
+#include "./xsimd_wasm.hpp"
+#endif
+
 // Must come last to have access to all conversion specializations.
 #include "./xsimd_generic.hpp"