Adds more supported types to arithmetic reductions

Permits `float` accumulation type with 64 bit integer and unsigned integer inouts to prevent unnecessary copies on devices that don't support double precision
IntelPython · Nov 8, 2023 · 375bbde · 375bbde
1 parent ddccf5c
commit 375bbde
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/dpctl/tensor/libtensor/include/kernels/reductions.hpp b/dpctl/tensor/libtensor/include/kernels/reductions.hpp
@@ -2806,10 +2806,12 @@ struct TypePairSupportDataForSumReductionTemps
 
         // input int64_t
         td_ns::TypePairDefinedEntry<argTy, std::int64_t, outTy, std::int64_t>,
+        td_ns::TypePairDefinedEntry<argTy, std::int64_t, outTy, float>,
         td_ns::TypePairDefinedEntry<argTy, std::int64_t, outTy, double>,
 
         // input uint64_t
         td_ns::TypePairDefinedEntry<argTy, std::uint64_t, outTy, std::uint64_t>,
+        td_ns::TypePairDefinedEntry<argTy, std::uint64_t, outTy, float>,
         td_ns::TypePairDefinedEntry<argTy, std::uint64_t, outTy, double>,
 
         // input half
@@ -3077,10 +3079,12 @@ struct TypePairSupportDataForProductReductionTemps
 
         // input int64_t
         td_ns::TypePairDefinedEntry<argTy, std::int64_t, outTy, std::int64_t>,
+        td_ns::TypePairDefinedEntry<argTy, std::int64_t, outTy, float>,
         td_ns::TypePairDefinedEntry<argTy, std::int64_t, outTy, double>,
 
         // input uint32_t
         td_ns::TypePairDefinedEntry<argTy, std::uint64_t, outTy, std::uint64_t>,
+        td_ns::TypePairDefinedEntry<argTy, std::uint64_t, outTy, float>,
         td_ns::TypePairDefinedEntry<argTy, std::uint64_t, outTy, double>,
 
         // input half