v4.2 tag release. (#2638)

examples/88_hopper_fmha/kernel/fmha_kernel_tma.hpp

@@ -137,6 +137,9 @@ struct FmhaKernelTma {
   }
 
   CUTLASS_DEVICE void operator()(const Params &params, char* smem) {
+#if ! defined(CUTLASS_ARCH_MMA_SM90A_ENABLED)
+    printf("ERROR : Arch conditional MMA instruction used without targeting appropriate compute capability. Aborting.\n");
+#else
     TileScheduler tile_scheduler{params.tile_scheduler};
 
     // Shared memory.
@@ -216,6 +219,7 @@ struct FmhaKernelTma {
       result, typename CollectiveMainloop::TiledMmaPV{},
       params.problem_size, params.epilogue,
       epi_load_pipeline, storage.epilogue);
+#endif
   }
 };
 

examples/88_hopper_fmha/kernel/fmha_kernel_tma_warpspecialized.hpp

@@ -160,6 +160,9 @@ struct FmhaKernelTmaWarpSpecialized {
   }
 
   CUTLASS_DEVICE void operator()(const Params &params, char* smem) {
+#if ! defined(CUTLASS_ARCH_MMA_SM90A_ENABLED)
+    printf("ERROR : Arch conditional MMA instruction used without targeting appropriate compute capability. Aborting.\n");
+#else
 
     enum class WarpGroupRole {
       Producer = 0,
@@ -412,6 +415,7 @@ struct FmhaKernelTmaWarpSpecialized {
         if constexpr (kIsEpilogueLocked) ; math_wg_order_barrier.arrive();
       }
     }
+#endif
   }
 };