Fix avx512 16-bit symbol output

lin-toto · Apr 17, 2023 · 7344a71 · 7344a71
1 parent 0654ee1
commit 7344a71
Showing 1 changed file with 7 additions and 2 deletions.
diff --git a/include/recoil/simd/rans_decoder_avx512_32x16n.h b/include/recoil/simd/rans_decoder_avx512_32x16n.h
@@ -55,8 +55,13 @@ namespace Recoil {
         }
 
         inline void writeResult(const u32x16 symbolsSimd, ValueType *ptr) override {
-            auto sym = _mm512_cvtepi32_epi8(symbolsSimd);
-            _mm_storeu_si128(reinterpret_cast<__m128i*>(ptr), sym);
+            if constexpr (sizeof(ValueType) == 1) {
+                auto sym = _mm512_cvtepi32_epi8(symbolsSimd);
+                _mm_storeu_si128(reinterpret_cast<__m128i *>(ptr), sym);
+            } else {
+                auto sym = _mm512_cvtepi32_epi16(symbolsSimd);
+                _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), sym);
+            }
         }
     };