Fix __dp4a call

codeplaysoftware · Nov 29, 2024 · 6fa4521 · 6fa4521
1 parent 03e7fad
commit 6fa4521
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/include/cutlass/numeric_conversion.h b/include/cutlass/numeric_conversion.h
@@ -3664,7 +3664,11 @@ struct NumericArrayConverter<float, int8_t, N, Round> {
 
     CUTLASS_PRAGMA_UNROLL
     for (int ii = 0; ii < PackedResultType::kElements; ++ii) {
+#if defined(__CUDA_ARCH__)
       t[ii] = __dp4a(x, mask[ii], 0);
+#else
+      t[ii] = x * mask[ii];
+#endif
       r[ii] = static_cast<float>(t[ii]);
     }
   #endif