Fix grouped_gemm_splitk kernels on MI300. (#694)

* replace amd_buffer_atomic_add with hip_atomic_add

* fix grouped_gemm_splitk kernels on mi300

* fix syntax

* revert experimental atomic_add changes

---------
Co-authored-by: Jing Zhang <jizhan@amd.com>

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

@@ -34,7 +34,8 @@ __global__ void
         kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
                                        const index_t group_count)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
     __shared__ uint8_t p_shared[shared_size];