set occupancy to 1 to solve scratchdata and agpr issues

include/ck/tensor_operation/gpu/device/impl/device_batched_multihead_attention_backward_xdl_cshuffle_v2.hpp

@@ -15,7 +15,7 @@
 #include "ck/tensor_operation/gpu/device/masking_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_v3.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_v2.hpp"
 #include "ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
@@ -47,7 +47,8 @@ template <typename GridwiseGemm,
           bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+    // __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, 1)
 #endif
         kernel_batched_gemm_softmax_gemm_xdl_cshuffle_v2(
             const DataType* __restrict__ p_a_grid,