Avoid using _mm512_storeu_epi16()

mstembera · vondele · commit 0383670cd556 · 2025-08-16T11:46:39.000+02:00
gcc 9 & 10 do not provide this function, instead use the generic _mm512_storeu_si512 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95483 https://tests.stockfishchess.org/actions?max_actions=1&action=&user=&text=&before=1755266300.455175&run_id= closes #6231 No functional change
diff --git a/src/nnue/layers/affine_transform_sparse_input.h b/src/nnue/layers/affine_transform_sparse_input.h
@@ -104,7 +104,7 @@ void find_nnz(const std::int32_t* RESTRICT input,
 
         // Avoid _mm512_mask_compressstoreu_epi16() as it's 256 uOps on Zen4
         __m512i nnz = _mm512_maskz_compress_epi16(nnzMask, base);
-        _mm512_storeu_epi16(out + count, nnz);
+        _mm512_storeu_si512(out + count, nnz);
 
         count += popcount(nnzMask);
         base = _mm512_add_epi16(base, increment);
