diff --git a/include/cutlass/arch/barrier.h b/include/cutlass/arch/barrier.h index 402ef50cbb..b08ee546d3 100644 --- a/include/cutlass/arch/barrier.h +++ b/include/cutlass/arch/barrier.h @@ -360,7 +360,7 @@ struct ClusterBarrier { uint32_t smem_addr = cute::cast_smem_ptr_to_uint(smem_ptr); asm volatile( "{\n\t" - "mbarrier.ival.shared::cta.b64 [%0]; \n\t" + "mbarrier.inval.shared::cta.b64 [%0]; \n\t" "}" : : "r"(smem_addr));