[Common] Fix "0" literal for compilation (#2934)

cyanguwa · web-flow · commit 82ace626294c · 2026-04-27T18:55:30.000-07:00
diff --git a/transformer_engine/common/fused_attn/flash_attn.cu b/transformer_engine/common/fused_attn/flash_attn.cu
@@ -311,7 +311,7 @@ __device__ __forceinline__ void permute_vec_loop(const T *__restrict__ in, T *__
       const size_t s_local = w / pad_elems;
       const size_t s_i = s_begin + s_local;
       const size_t d_off = D + (w % pad_elems);
-      out[out_base + s_i * D_out + d_off] = static_cast<T>(0);
+      out[out_base + s_i * D_out + d_off] = static_cast<T>(0.f);
     }
   }
 }

Original file line number	Diff line number	Diff line change
`@@ -311,7 +311,7 @@ __device__ __forceinline__ void permute_vec_loop(const T __restrict__ in, T __`
`311`	`311`	`const size_t s_local = w / pad_elems;`
`312`	`312`	`const size_t s_i = s_begin + s_local;`
`313`	`313`	`const size_t d_off = D + (w % pad_elems);`
`314`		`- out[out_base + s_i * D_out + d_off] = static_cast<T>(0);`
	`314`	`+ out[out_base + s_i * D_out + d_off] = static_cast<T>(0.f);`
`315`	`315`	`}`
`316`	`316`	`}`
`317`	`317`	`}`