AN65974 stream transfer problems

I've been trying to use AN65974, Designing with the Cypress Ez-USB FX3 DVK and Xilinx FPGA ZC702 Evaluation Board.  I've modified the verilog code for FPGA and firmware for FX3 , but I have trouble with stream IN & OUT transfer. The problems are as follows.

1. CYU3P_PIB_ERR_THR3_RD_UNDERRUN error message when I tried stream out transfer on USB control center.

- I was attached my firmware code included my GPIF II code.

- I used GPIF 32bit bus and MANUAL DMA channel for stream IN/OUT transfers.

- I was transfer out 16384 byte from USB to FPGA through FX3 on USB Control Center. (#define CY_FX_SLFIFO_DMA_BUF_COUNT_U_2_P == 1)

- I was found CYU3P_PIB_ERR_THR3_RD_UNDERRUN on UART terminal.

- I think that my data was transfer to FPGA because I identified SLRD# = Low and SLOE# = LOW at Oscilloscope.

- So, I can't understand CYU3P_PIB_ERR_THR3_RD_UNDERRUN Error message.
- Why does that error message occur ? Is there any transfer problem except the error message ?

2. output previous data from DMA buffer when I tried stream in transfer on USB control center.

- Now my FPGA operate that register of FPGA is written by the data through transfer out and the data read from the register is send to USB through transfer in.

- I was transfer in 16384 byte from FPGA to USB through FX3 on USB Control Center. (#define CY_FX_SLFIFO_DMA_BUF_COUNT_P_2_U == 1)

- But I can read the previous data at DMA buffer, not the current data,  through transfer IN.

- If I would like to read the current data, I must read one more through transfer in.

- For example

     1. I was TRANSFER OUT "A" data(16384 byte) to FPGA.

     2. I was TRANSFER IN the data from FPGA. But The data read from FPGA was default data of FPGA.

     3. I was 1 more TRANSFER IN the data from FPGA. The data I read was "A" data.

     4. The next I was TRANSFER OUT "B" data(16384 byte) to FPGA.

     5. I was TRANSFER IN the data from FPGA. The data I read was "A" data.

     6. I was 1 more TRANSFER IN the data from FPGA. The data I read was "B" data.

- I think the cause of problem is due to DMA buffer or GPIF state. ( I can't identify that GPIF state machine enter WRITE STATE through TRANSFER IN. )

- How can I read the current data at once ?