CUTLASS 3.5.0 (#1411)

examples/53_hopper_gemm_permute/53_hopper_gemm_permute.cu

@@ -393,7 +393,8 @@ private:
     ElementB, StrideB, 128 / cutlass::sizeof_bits<ElementB>::value,
     ElementAccumulator,
     TileShape, ClusterShape,
-    cutlass::gemm::collective::StageCountAutoCarveout<sizeof(typename CollectiveEpilogue::SharedStorage)>,
+    cutlass::gemm::collective::StageCountAutoCarveout<
+      static_cast<int>(sizeof(typename CollectiveEpilogue::SharedStorage))>,
     cutlass::gemm::collective::KernelScheduleAuto
   >::CollectiveOp;
 
@@ -403,7 +404,8 @@ private:
     ElementB, StrideBPermute, 128 / cutlass::sizeof_bits<ElementB>::value,
     ElementAccumulator,
     TileShapePermute, ClusterShape,
-    cutlass::gemm::collective::StageCountAutoCarveout<sizeof(typename CollectiveEpiloguePermute::SharedStorage)>,
+    cutlass::gemm::collective::StageCountAutoCarveout<
+      static_cast<int>(sizeof(typename CollectiveEpiloguePermute::SharedStorage))>,
     cutlass::gemm::collective::KernelScheduleAuto
   >::CollectiveOp;
 

examples/53_hopper_gemm_permute/permute_kernel.cuh

@@ -37,7 +37,7 @@
 #include "cutlass/layout/matrix.h"
 #include "cutlass/tensor_view.h"
 #include "cutlass/fast_math.h"
-#include "cute/numeric/uint128.hpp"
+#include "cute/numeric/numeric_types.hpp"
 
 namespace example
 {