fixed host tensor size

profiler/include/profiler/profile_gemm_universal_impl.hpp

@@ -105,7 +105,7 @@ bool profile_gemm_universal_impl(int do_verification,
     const auto c_element_op = CElementOp{};
 
     DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n_permute.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n_permute.mDesc.GetElementSpaceSize() / (is_same_v<BDataType, ck::pk_i4_t> ? 2 : 1));
     DeviceMem c_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
 
     a_device_buf.ToDevice(a_m_k.mData.data());